Get Goalside

Extra time for in-play time

Sun, 28 Jun 2026 00:00:00 GMT

Let's run things back.

In 2023, Get Goalside put its analysis where its attitude was and ran a couple of small-scale studies on 'ball-in-play' time. In 'Stop the clock', we confirmed our priors (which is always a little disconcerting): the 'worryingly low' levels of in-play time that top leagues were experiencing at the time (around 54 minutes) were, in fact, within normal ranges.

Since then, in-play time has risen, and we'll talk about that later on. But another, more important thing has happened: Statsbomb's released more data. So it's time for some updated analysis.

The data we'll be looking at here is a big set of women's football data: the 2023 NWSL season, 2023/24 seasons of the English, Italian, German, and Spanish top-flights, and, for an international flavour, the 2023 World Cup and 2025 Euros.

A headline? The WSL is even more in-play-pilled than FIFA.

But first, let's take a look at the most basic-but-fascinating chart of the day. The entire women's dataset* (two German teams aside) sits outside the interquartile range of the men's 2015/16 leagues* for total match-clock time.

*I'll generally refer to this as 'this dataset' or 'the 2023/24 leagues' from here on, and refer to 'men's 2015/16 leagues' when talking about the Big Five European leagues from Statsbomb's dataset.

There is an in-play difference but the big story is that these matches take much more time for the marginal increase in in-play time. This is easier to see below, where we see that the interquartile range of in-play time percentage is actually slightly lower than the men's 2015/16 league dataset.

As you'll have picked up from the team labels on show, there's a big competition difference here within the more recent women's football dataset. Two competitions are particularly stark: the 2023 World Cup, and the 2023/24 English WSL.

The World Cup's uniqueness is no surprise, as it was an outlier in the previous work, but the WSL is eye-catching. The median ball-in-play time of the competitions is as follows:

2023/24 English WSL: 61 minutes, 35 seconds
2023 World Cup (normal time matches only): 57 minutes, 52 seconds
2023 American NWSL: 57 minutes, 28 seconds
2023/24 Spanish Liga F: 56 minutes, 37 seconds
2023/24 Italian Serie A: 56 minutes, 31 seconds
2023/24 German Bundesliga: 56 minutes, 4 seconds
2025 European Championship: 54 minutes, 30 seconds

Plotting the data match-by-match, you can see just how different the WSL is to the other competitions in the dataset

This is partly because WSL matches that season had a far higher percentage of in-play time than other competitions. Interestingly, the three competitions in this dataset which had VAR (2023 World Cup, 2023 NWSL, 2025 Euros) had the lowest percentage of game-time in play, although the match-level distributions generally don't look that different.

Unsurprisingly, this extra in-play time does not necessarily translate to extra excitement. WSL teams (from 2023/24) may average far more in-play time than the other teams in the dataset, but spend a very similar amount of time in the final third. (We'll look at league competitions only from here, for the larger sample sizes and smaller quality gap between teams).

It's worth noting here that, so far, the team-level data has been figures for 'matches involving this team'. If you think (or scroll) back to earlier scatterplots you'll see that USWNT games in the 2023 World Cup averaged a very low percentage of in-play time. That doesn't mean the US caused the low percentage. In two of their three group games (their knock-out penalty loss to Sweden is excluded from the analysis as it went to extra-time), they played Vietnam and Portugal, two low-in-play teams.

There are some teams whose matches see an interesting difference when they're at home compared to away. Given the link between in-play time and restarts, it seems plausible that these teams have a distinct enough change of strategy when home/away that it shows up on the in-play stopwatch.

In many cases, you'll notice that the value when they're at home is higher than when away. Sevilla are the most dramatic, presumably a case of them seeking to play a possession-based style at home and being content to junk things up on the road. But for some the difference is in the reverse direction: teams more likely to sit back in a more passive defensive block and allow home teams to possess the ball uncontested?

If we split the final third time into the team's own time in the final third, we see a different picture for the WSL sides. Poor Bristol. (They were duly relegated that season).

Some other notable things from this chart: Barcelona's complete Liga F dominance; Arsenal and Manchester City both having higher final third shares than actual league champions, Chelsea (in Emma Hayes' final season there); NWSL's parity showing – pretty much all teams sitting in the 40-60% range.

The NWSL is a little weird in this dataset in terms of final third time though. Or, some of its teams were. There were six teams in the dataset who averaged 2+ minutes more time in the final third at home than on the road: three from the NWSL.

It would be interesting to follow-up this data with points, expected goals, and amount of restarts, but then we'd be here all day.

But, back to the question referenced right at the start of this piece: why is in-play time higher than the men's 2015/16 dataset? Particularly if the share of in-play time is lower?

Tendencies around added time looks like part of it (I mean, it'd have to be, given that the total match-clock time has increased). Now, the event data doesn't include the amount of time added by the refereeing team: only when match periods ended.

Although VAR doesn't fully explain this (as several 2023/24 women's leagues didn't have it), the increase to substitution numbers could be part of it. But only part. Substitutions rarely happen in the first half, and yet the trend of added time played shifts from overwhelmingly under two minutes in the men's 2015/16 leagues to a wider 1-4 minute spread in the women's 2023-2024 leagues. You'd need to examine this in more depth, but it seems plausible that subs, VAR, and a change of refereeing accountancy has changed these figures.

My assumption is that this is more to do with the datespan than gender of the competitions. There's no theory I can think of for why women's football would be stoppage time-d differently to men's football, and the 2023-2024 seasons followed the public fretting about time-wasting.

And, while I could also look up the situation on the Continent, I know off the dome that the PGMOL is responsible for officiating both the Premier League and the WSL in England. (Different officials, true, but same organisation).

In a circular way, this draws back to the very first chart: this dataset of women's football (in 2023-2024) has higher ball-in-play time than the men's dataset (in 2015/16), but coming through more time added on, not through a higher share of the game spent in-play. Except the WSL which seems to do both. Never let it be said that the English don't care about value for money.

Notes

A note on averages

In the charts with interquartile ranges, these are IQRs of the teams' averages, rather than match-level. This was mostly for convenience at the time which I didn't think it worthwhile to redo on a match level. I'd also argue it's interesting to have averages taken from both a match level and a team level because many fans will primarily experience their team's matches. Although the IQR of team-level data is an IQR of a set of median values, this probably captures the 'majority range of fan experience' better than an IQR of match-level values.

That's bait

Sun, 14 Jun 2026 00:00:00 GMT

I've had an unusual disinclination to write blogs lately.

It's not for a lack of material. My LinkedIn feed throws shiny posts at me, trying to bait me into writing, but no. Enough stuff has built up now to make for a good grab-bag, though, so here we go with the links and segues...

In case you don't read to the end, have a good (northern hemisphere) summer. Get Goalside will probably be back in inboxes at some point. Feel free to get in touch in the meantime.

Statsbomb released data from 2023-2024 for five top women's leagues: the WSL (England), Liga F (Spain), Frauen Bundesliga (Germany), Serie A (Italy), and the NWSL (USA). We like this, this is a good thing. Related, the lines to submit proposals for this year's Hudl Performance Insights conference are now open.

US Soccer changed the paradigm by giving Mauricio Pochettino a laptop to use during water breaks in pre-World Cup friendlies, putting him an air-conditioned office away from becoming a Daily Mail bête noire. He used it to show players video clips, but this is a step towards the dream of teams being able to see their own beautiful data in real-time.

Speaking of seeing beautiful data, the BBC and FIFA are both determined to make 'fetch' happen, offering 3D versions of World Cup matches. This follows shortly after La Liga turned some Real Madrid highlights into EA FC. In the latter two cases, video game engines are at-hand to render pose data (the BBC are working with immersiv).

I still don't really understand what these are for, other than trying to work out how to make the tech work as a product. I get it, it feels like there's something to be had there, but, personally, I've got a mental block around near-real rendering of actually-real football. I'd rather the Australian Open Mii's or the NFL's Disney collab. Perhaps this is like agentic AI, where the gains will be made in the professional space before being brought over to the consumer world in better packaging.

If you've got ideas for what near-real 3D rendering of football can be used for, let me know. Also let me know if you've got ideas/opinions about what I should do with my data-and-game-model AI assistant project that I spoke about a couple of weeks ago.

Back in the real world, The Cutback published a report on next steps for women's football - 'From Visibility to Viability: A blueprint for the next phase of women's football' - based on interviews with execs and experts across the game. It's not strictly 'football analytics', but growth of women's sport and women in sport more generally are all good things worthy of Get Goalside's digital ink.

In a similar 'slightly tangential but Get Goalside wants to mention' vein, Thom Lawrence has moved on from Statsbomb/Hudl. One of the 'analytics Twitter' cadre, his work and attitude (curious, built on strong technical foundations, funny) has been very influential on me. I'm sure a lot of others share the same view, and we wish him well with his deserved break and with whatever comes next.

Another figure who's moved on to something new is Arsenal's former Head of Football Platforms, Danny Karbassiyoon. He's started a company, Sentrum, whose website tagline is 'The decision layer for' and then a rolling series of job titles. Alas, that activates Get Goalside's 2023 trap card 'So... everyone's a decision-maker now'.

However, the real reason this is bait is that the Sentrum content marketing is what Get Goalside would write if Get Goalside was a credible industry figure:

Two of the clubs I spoke to had actually tried [to build data infrastructure internally]. In both cases, smart people with real resources took a run at centralising their data infrastructure internally. Both eventually stopped.

The reasons were consistent. Building the initial version is possible. Maintaining it alongside actually running a football club is the part that breaks. Data providers update their APIs. New sources get added. Staff turn over and institutional knowledge walks out the door.

Speaking of staff turnover, King's League has made ripple-sized headlines this week for cutting jobs as part of a restructure, the extent and nature of which is disputed by a statement from King's League employees. Like 3D-rendering of real football, King's League has been a long-time Get Goalside case of 'this is an interesting development but there are dots I can't quite connect'.

Now, onto a competition which does seemingly know how to sustain itself (although has also pushed through expansion plans), this summer's World Cup will feature FIFA Power Rankings. This possession value model-sounding set of rankings will also be "the first major global collaboration with Aramco, FIFA's exclusive Energy Partner". Some may be surprised at this partnership, but not Get Goalside! Retro is firmly in, and this is merely a retro callback to the Opta Castrol Index of the late 2000s/early 2010s. You think solar can power these numbers? Pah!

The rankings will mean that "[p]erformance will no longer be judged by opinion alone," according to Arsène Wenger in his role as FIFA’s Chief of Global Football Development, who will presumably be expecting social media to be a quiet place during the tournament. Interesting too is the fact that "[s]cores will be made available no later than four hours after the final whistle of each game", giving an insight into FIFA's data collection and processing pipeline.

Related to FIFA's data collection processing, a paper on generating event data automatically from broadcast tracking data was released by a FIFA team last month. I've not yet read it in full, but we like this, this is a good thing.

Another paper that I've not yet read in full but am eagerly looking forward to is– [puts finger to ear] sorry, I'm being informed that Hudl's Defensive Responsibility metric does not come with a free white paper. Still, there's a video with a nice animation. According to it, the new model:

[P]rovides a better way to measure each player's defensive contribution and sees the game like an expert coach. By combining metadata from the defending team's formation with millions of historic events, the model understands that a centre-back is more likely to defend actions nearer their own goal than a midfielder, whose defending is spread across the pitch.

That's bait.

There's an interesting implication in blog though. When it comes to the section on centre-backs, players who make fewer defensive actions than the model expects aren't described as 'bad' or 'inefficient'. Instead, to take RB Leipzig's Castello Lukeba as an example, it suggests that "he relies more on his recovery pace and positional containment within Leipzig's structure than meeting and engaging with the play". Does this mean that even when adjusted for defensive responsibility we can't say 'higher number = better'? It doesn't seem unreasonable, but feels like a challenge to communicate.

'Not unreasonable but a challenge to communicate', must be a data scientist! – is the segue to the final note: recent job ads of the English Football Association. They've closed now, so I'll link to Dominic Jordan's post about them. He described them as follows:

The first role sits within the elite coaching development function [...] explicitly focused on the discovery, development and progression of English coaches. [...]

The second role is a more traditional but research-heavy data science and machine learning position. The emphasis here is on deep learning, tracking data, building novel metrics, player modelling and in-depth game models.

National football associations ramping up their data operations is fun and interesting. If you view FAs as their national teams, they have less to do than clubs, with only one major tournament a year (combining the senior men's and women's teams). But, of course, both have qualification processes, and then there are the age-group teams too. And then there's the other work of an FA, like the coach education that one of these jobs is aimed at. Their roles are both as competitors on multiple fronts, vehicles for player (and coach) development, and stewards of the national game's health.

A whole new world model

Sat, 30 May 2026 00:00:00 GMT

One of the most inspiring moments of the last few months was the Artemis II voyage around the moon. Space: cool, the astronauts: cool, a solar eclipse from the other side of the moon: very cool.

Also cool: using the moon itself as part of the return home. As folks who'd watched Apollo 13 already knew, the moon's gravity could be used in conjunction with Earth's own gravity to pull the rocket back down to the blue marble. That big floating orb of cheese could be used as a corrective, a steer. An extra force of gravity being used in conjunction with another.

If you’ve played with LLMs before, even just tweaking a system prompt, Artemis II’s ‘gravity slingshot’ trick might resonate as a metaphor. Working with language models, – particularly less powerful ones, and particularly on niche topics – you can feel the gravity of the model’s training pulling sentences back to its mass.

This isn't always a problem if what you're doing is just supplying a list of steps to follow or integrating up-to-date information, but it can be a real pain on some tasks. The companies creating the LLMs are certainly aware of it, aware of the need for something to help course-correct the gravitational pull of the model's linguistic weights. Enter: ‘world models’.

If you want an ‘explain it like I’m 5’ intro to world models, you could say that while LLMs understand lots of words, they only understand things like 'gravity' based on what they’ve read about it, and not the actual effects that you tend to learn by spending formative months as a toddler dropping things from your high chair. The Big Tech companies think that by injecting world models into the system, this will act as a moon to the LLM's Earth.

Being a dork who writes analytics blogs, I'd heard folks talk about this months ago, but earlier this week I had an apple-falling-on-the-head moment.

There’s a project I’ve been working on – which I'll talk a bit about in a moment – which relates to these concepts. The project brings together a few strands I’ve been interested in: language models, constructing things from event data, and how coaches can be better helped in bringing their ideas about the game to life.

I set about building an LLM agent that can pull in a coach’s game model, pull in information about available football data, and work some magic. 'Working some magic' is easy; 'working a little bit more than some magic, a little bit more than some of the time' is harder.

One of the problems, as it turns out, is that the language models I was experimenting with weren’t able to work out how to do complex things with event data just from first principles and some provider docs. (As I tend to, I was of course using the Statsbomb open dataset; my comment here is no shade on their docs, they’re better-suited to this than some competitor provider docs might have been)

So, not only did I have to add a rough version of a coach's game model to the system (which I wrote), I needed to add extra help to the system around how to work with event data. I did that, it helped things, and then, listening to a podcast where Big Tech’s world models came up, it hit me: that's what I'd been doing. I'd been adding a form of world model to the system. A coach/analyst-style 'game model' is just a world model for a specific type of world.

Here's the video:

One of the unexpectedly interesting parts of the project is that the Anthropic models have clearly been reading Inverting the Pyramid and Spielverlagerung. Tactical questions to the Claude web app get surprisingly convincing responses, terms like 'mid block' and 'pressing triggers' readily spring into its responses. In some areas, the web app would probably do better than my hobby project agent; although it obviously wouldn’t be able to call on the specific Statsbomb data it’s integrated with.

It should be said: LLM development is still a pretty terrible experience, and should turn any developer into a unit test-appreciator. Not only are the results non-deterministic, but if you’re cost-conscious (which, on a hobby project, I definitely am) then every test involves throwing Anthropic a non-deterministic fraction of a dollar. It used to be the case that the advice was to build with a powerful model, and then refine and compress efficiency into a smaller model once you had something working. I’m wondering now, as the trajectory of local-runnable models continues, whether the reverse would be true now: prototype with more limited models, then get a performance boost when you switch to a costly model from an LLM provider.

It’s partly for this reason that I don’t have evals for this system (but partly also because that’s an extra set of work that isn’t really needed). It’s been enough to just look at the results and scrutinise logs in Langsmith.

A special side-effect feature of this project has been a pay-off for the wonkish theory blogs written over the years, like this about phases and forms of control. It’s easier for agent tool purposes if the game model is semi-structured, and all that past thinking helped inform the structure. It only lightly affects the conversation about England’s press in 2023, but comes through more with questions like this:

Is this the "Claude AI-morim" that I riffed on back in January? No, obviously. Is it close? Ish.

One of the interesting things with agentic systems is the way that capability and testing requirements scale in completely different ways to 1) how they would in conventional systems 2) each other.

Imagine the user flows of traditional software like a set of rectangular corridors that you create. Each door in the corridor represents an action a user can take, and as you add more features, more corridors can be walked down. If you tested that system, you'd need more tests than features, because the corridor has more to it than just the doorways.

Dealing with an LLM agent is like if that same 'corridor as software feature' idea was switched so that corridors were no longer rectangles, but fractal shapes, like three-dimensional snowflakes. There will be more doors, and doors in better places and more interesting angles, but the surface area of the entire structure increases at a much higher rate than the number of doors.

This, after all, is why you still see so many laughable screenshots of Google's AI overviews. The sheer infinite variety of things people can type into a search box is more, by degrees of magnitude, than a non-deterministic system can deal with (without heavy guardrails).

Maybe a different type of world model, based around user intent and propensity to mischief, is something Big Tech needs to think about. Another gravitational steer for the LLMs, this time away from farce. As I'm sure the Artemis II crew said just before doing their lunar fly-by: 'what a world'.

Press 1 for rest defence. Then 2, 7, 4, then 1 again.

Sat, 16 May 2026 00:00:00 GMT

You have chosen to read this sentence: check box.

You have chosen to continue past the first line: check box.

You have chosen to subscribe? Check box pending...

Football data is hitting up against a new type of access problem. It used to be the case* that merely having access to the datapoints you were interested in was the hurdle. Nowadays you increasingly have that access, but the issue is finding what you want.

*To be totally accurate, there's still a chunk of professional football lacking good access; and a (diminishing) chunk of football problems lacking decent data to describe them.

Why this problem? First, the number of events being collected expanded. Then it expanded again, with opponents being bypassed and phases of play attached to everything. And now, every Tom, Dick, and Harry are packaging up off-ball runs. (And Peter, and various Pauls who don't have proper blog posts to re-find and link to).

Get Goalside never used to be a UI/UX blog. But the biz is insisting it become one.

You have laughed at weak joke? Check box pending...

There are two easy approaches to this kind of situation. One is – as you mighht have guessed from the conceit of this post – letting somebody find or create anything they want through a series of menus. This seems to be pretty popular.

It's probably popular in part because the other main option is scary. It's to be very opinionated about what you put in front of people from the start, leaving everything else behind a basic keyword search that you know only 5% of users will even look for. Opinions. Who'd have 'em.

Turns out I've actually written about this before, last September, positing a potential Third Way:

Manual curation naturally limits what is put in front of an individual, relies on smart experts doing the curating, and has a tendency of keeping people in the status quo. So are algorithms the way? Is that how people will make sure that the right dozen metrics, out of a pool of thousands, get put in front of people?

But this was only thinking about half of the equation. Given that football data is in such flux, and football is so fickle, you could just make like Meccano and let people create their own statistics. This doesn't solve the UI/UX problem, of course, just shifts it to a different part of the process.

As with many things, though, it's not like this is a problem limited to the football biz. Navigating phone or computer settings is frequently horrible. Toolbars of Microsoft and Adobe products are well-known for being packed with features that few people use but which are each used by enough people to keep them there.

You find comparisons to other consumer technologies tiresome? Check box.

To return to the well of analogies, we can always return to language. Data is a means of describing football just as the spoken word is; like how quaint and deprimentemente nublado are ways of describing England.

Google Translate, when you search for a word, not only gives you several translation options, but often gives you common synonyms for each of them in the input language underneath. For example, translating 'funny' to German gives you – flying in the face of English stereotypes – a wealth of options. The first is 'lustig' (which Google also relates to merry, cheerful, amusing), and then 'komisch' (comic, comical), 'witzig' (witty, humorous), 'spaßig' (droll, jocular).

Maybe this is an interface pattern to take inspiration from. 'You want line-breaking passes? You may also be interested in: progressive passes; Possession Value added; Buzzy Marketing Material.'

[Aside: at what point in a pass is a pass line-breaking? How deep past the line signifies the line being broken? If the pass delivery forces the recipient to remove the sting from the ball, during which time the line that was broken recovers, is that still a line-breaking pass? If defenders, anticipating the pass, pounce immediately on an isolated recipient, is that a line broken?]

You like line-breaking pass metrics? Check box.

Truthfully, this is all much more boring than the advanced modelling that Get Goalside used to write about. But so is data engineering, and everybody and their dog are hiring for football data engineers these days.

Via this post from Nicola Graham, I stumbled on this product-market fit diagram by Chris Peploe. It splits into four types of fit: Problem, Workflow, Proof, Ecosystem. Are you solving a problem; does it fit into how people work; is there proof of the efficacy; does the wider ecosystem support adopting and integrating the product.

We can talk Problem, Ecosystem, and Proof some other time (although: publish more research), but the UX of a product – whether a mass of filters or some other UI system – is gonna be part of that Workflow fit test.

How do you make something easy to find?

There's a deliberate typo in this post. How would you go about finding it?

Crabs, the everything app

Sat, 18 Apr 2026 00:00:00 GMT

If you're either Online or a biologist of some sort, you'll know what carcinisation is. Nature abhors a vacuum but loves to be a crab; a handful of different crustacean species all independently evolved to be more like them.

This kind of convergence on a 'good idea' happens all around us: slab touchscreen phones, caffeine, and (eventually) meatwall corners. Social media platforms that grow big enough eventually all add shopping, advertising, and 'shows'. White collar work platforms are converging on offering code, design, and documentation capabilities (Figma AI, Claude Design). Arguably, open-play football tactics converge to a mix of athletic defence and ball-respecting possession play for 'good squad' teams, and low-block defence and direct play for 'bad squad' teams.

The tech items in that list are interesting for being different versions of everythingification. Back when Elon Musk took over Twitter, he was mocked for his plan to turn it into 'X, The Everything App' – notably, it is still not an everything app, and the mocking was correct and just. But modern tech infrastructure lets technology do more and be more everything-y. Which will, one would expect, come for football tech too.

After all, a football club is one organisation, with criss-crossing of interests all over it. Data has a natural use with video for analysis; both have a natural use with marketing and social media; match data should be on the same system as training data; recruitment data could benefit from living alongside fitness and injury data of the first-team squad; training ground facilities management may as well be synced with fixture calendars and everything else logistical.

It's not like I'm saying anything ground-breaking. Teamworks is pitching itself as 'the operating system for sports, TM'. Infinite Athlete – still the closest thing to a permanent shirt sponsor that Chelsea have had since 2023 – was mocked for its plan to spend lots of money putting its logo in front of people while being oddly unclear about what they did – notably, et cetera. But 'operating system for sports, TM' does not seem far off what they should have been saying. Everythingification.

If this is to take root in football, where else might it happen? The combination of coding and design tools feels like a particularly interesting path to follow. Previously, in the tech world, the two were as separate as oil and water, Arsenal men and Spurs men. But pretty quickly, people realised that there was only so far that you could go with code-generation tools creating 'pretty' and 'good UI' by themselves. Equally, making good UI design files can be a painfully pernickety process, needing to create parallel universes of pictures for each button state: the kind of mostly-rote task that LLM-world is great at.

The thing that makes this all possible – we're coming back to football in a second – is that the design files are digital, and therefore, when you boil it down, merely representations of data structures. Match footage (and training footage for that matter) is not currently well-represented by data structures. Not to go all Silicon Valley about things, but this might be a blocker to progress.

There are a few techniques available for this. Tracking data isn't really a data structure, but it can be processed into a spec that someone creates. 'Traditional' video tagging by analysts is another means. Nowadays, the kind of event data that you get direct from data companies might even be good enough to do the job.

But what to use 'code plus designs' as a metaphor for? I think the closest thing would be to say that, in software, 'code' is what you are building, and therefore could represent the output of 'analysis'. The 'design' could be a game model, or general principles underlying the particular bit of analysis.

In the past, UI designs would be worked out and refined and then often handed off to a tech team to implement. Nowadays, powered by language models as an underlying technology, the 'process loop' between the two can be much quicker, tighter, improving both ideation and refinement. Maybe the same can be possible in football.

I don't know whether this is a good idea. I definitely don't have a TM for it. But data structures do seem to be the crabs of the digital world.

What's the worldview of your data?

Sun, 29 Mar 2026 00:00:00 GMT

Conflict of interest: My paycheck comes from a company that is pitched as being 'data provider agnostic'. Take an appropriate pinch of salt, or seasoning of your choice, when reading.

There's always a risk that publishing something after going to a conference will be interpreted as being about the conference. As proof that this piece predated Field of Play, here's a paragraph from the original draft:

Every event and stats data provider is implicitly selling you their worldview. Maybe that sounds dystopian to you, a straitjacket of the mind, but the alternative - doing your own event tagging - means that you need to come up with a complete worldview of your own. My own worldview?? I can't even come up with a coherent opinion on the Jorginho-Chappell Roan drama.

The world moves fast. The Google trends line for 'Jorginho' is now back at normal levels.

Referring to a data provider's collection as their 'worldview' puts an interesting spin on assessing their value. Some of them don't have (passable) phase of play data: does that fit with how you, and the job you're doing, see the game?

In certain parts, it's integral, but in others it might not be. There are parts of a football club where you could get probably get by with just expected goals and a couple of other statistics. Let the coaches analyse the games, then take a scan over xG, number possessions that enter the final third, maybe number of possessions starting in your defensive third that end in the opposition half. It'd cover chance creation, chance conversion, aspects of defensive efficacy – not an awful team-level review process.

But change your role, change your focus, and your view of the world will change. If you're focused on player development, you'll need much more granular data. But that doesn't just mean 'events': that a player blocked a cross is not particularly useful; you'd want to have data on any 'confrontation', whether a cross or a 'duel' was attempted or not, to help grade that facet of their game.

The idea that a data provider imposes their worldview on you is something that's been on my mind for a while. Partly because so much knowledge-sharing in the analytics world is about process, so it feels worth poking more at what actually matters in the data packages.

With this in mind, I put together a fun little game. Hit some buttons and find out what matters to you and what matters to the crowd. If you've got feedback, lemme know.

The ulterior motive here is obviously to hark back to a previous blog, 'What if we didn't care about passes?'. Hopefully we can establish once and for all that we could free up event collectors' time by being loosey-goosey with pass collection.

I'm curious, of course, how different the crowd's worldview will be from the 'worldview' of existing data providers. When companies like Statsbomb, Impect, Skillcorner burst onto the scene – giving off vibes of taking meaningful market share* – how much of that is purely because of the worldview of the data spec? How much is it because of associated toys – like platforms, interoperability –, how much of it is price, how much of it is 'grass is greener' syndrome, and how much of it is just a persuasive (or nicer) sales rep?

*I don't have actual numbers but, as always, am welcoming of more data

With the increase in broadcast tracking data, it feels like these questions will only become more relevant. Is the race among these companies going to be the race for a better worldview, for a dataset that lets people quantify their own worldview more easily, or for a 'good-enough' worldview that is easier to fit into Internal Processes?

I started this blog – all those many days ago when people were still mostly unaware of the blended family links between Jorginho and Jude Law – hoping to reach an answer to these questions. I didn't. My worldview is still in flux.

Amusingly, as I was putting the final touches of this together, I noticed Statsbomb recently updated their Python package to include their aerial duel rating system. 'Would you rather have duel win probability or a particular set of extensive, discrete tags' is exactly the kind of data availability trade-off I'm curious about and that went into the making of the football data feature ranking game.

Football is a simple game...

Sat, 28 Mar 2026 00:00:00 GMT

The saying, in England, is that football is a simple game: 22 players run around for 90 minutes and, in the end, the Germans always win. In the rest of the British Isles, I imagine the phrase is only slightly different: 'in the end, the English always lose'.

If you ask a certain group of nerds, of course, they have a different view: '22 players run around for 90 minutes and, in the end, rational thinking always loses'. In the broadest sense, this is the pitched battle that number-heads have been trying to have with football, stretching all the way back to the late '00s, Freakonomics era, like 2009's Soccernomics.

The introduction is titled 'Driving with a dashboard'; one of the early chapters, 'Gentlemen prefer blonds: how to avoid silly mistakes in the transfer market'. In my country, the book itself had a different name: 'Why England lose'. I hope, for their sales' sake, they kept the name in Scotland too.

This represents one of the two types of presentations at analytics conferences.

And, yes, after a lot of consideration (the ten minutes to write the opening three paragraphs) there are only two: 'here is a sophisticated use of data' and 'here is how to make organisations smarter'.

Take this week's Field of Play conference:

'Personal analysis: How elite players are gaining a tactical edge': Data use (although here 'data' means 'video')
'Designing for the decision, not the dashboard': Smarter organisation
'Unveiling true talent: Separating player skill from team context in football analytics': Data use
'The missing half of performance: the power of player grades in men’s and women’s football': Sales [Data use]
'How to be Hollywood smart with statistics: explaining the value of data and analytics to a Doubting Thomas': Smarter organisation
'Asking the right questions in women’s football': Smarter organisation
'Bridging analytics and tactics: turning benchmarks into on-pitch impact': Smarter organisation
'A model is only as good as it is understood': 20-minute stand-up set [Smarter organisation]
'Data Science meets NFL: Developing a winning project pipeline': Both, I think

One of the interesting things about this dichotomy is that on one hand you get work like 'can we identify good defending by running tracking data through a set of counterfactuals', and on the other hand you get talks like 'make peoples' lives easier and be a good hang'.

There is, though, a second set of nerds, who would sum up football in a different way. 'Football is a simple game: dynamical systems try to achieve a goal and, in the end, random variance has an annoyingly large impact'. (If you want a paper that has citations pointing to both pass completion models and work on starlings, here you go).

This would be the most accurate definition, on and off the pitch, even if it wouldn't pass the 'be a good hang' test. The perception of success for a club and its 'backroom' processes can hinge on a single match, can be disrupted by freak events, can be derailed by rogue individual decisions or mistakes.

'Trust the process' became a meme in the NBA, but one of the defining differences between sporting organisations isn't so much trust in a process, but the true understanding of what 'a process' is. A process, as in an infrastructure, rather than an assembly line. There's a fine line between wanting to reduce controllable uncertainty and wanting to deny the existence of uncontrollable uncertainty.

But, I think, one of the unspoken questions hanging over the 'smarter organisation' type of analytics talks is 'how do you know how good your process is?'. The 'sophisticated use of data' talks can at least point to a bunch of validation tests. Did the process work, or are positive results an unsustainable overperformance?

Maybe modern football needs a different maxim, to work for both the on- and off-pitch stuff. Football is a simple game: 22 players run around for 90 minutes and, in the end, some nerd is always asking to see the xG.

Friendly Neighbourhood Thanos

Sat, 07 Mar 2026 00:00:00 GMT

The Marvel Cinematic Universe is not as iconic in 2026 as it seemed it might become in 2016. But there is still one line that pops back to my mind every now and then: "Fine, I'll do it myself." [Thanos, the Big Bad, is revealed in a mid-credits scene to be fed up of sending minions to fill in the Infinity Gauntlet].

About two years ago, I wrote about the football data landscape via then-Chelsea-front-of-shirt-sponsors Infinite Athlete (Chelsea still, bizarrely, have no long-term front-of-shirt sponsor):

There are lots of different types of data that football organisations need. They come from all over. Some of it is even well-documented. But it’s a pain to traipse from market to market, picking up your event data groceries from one place and your physical data from another.

It is a pain. It's also objectively funny, in a Three Stooges kind of way, that football – the wealthy global sport, the peace prize-giving sport – still has problems with questions like "when was a pass made?". 'We need a fancy algorithm to match our data sources because tracking comes from cameras and events come from a guy with a keyboard.' Truly, the Beautiful Game.

Now, I'm no Adam Smith (the Invisible Hand that I'm most familiar with is Diego Maradona's) but this seems like a market failure. It should not be this difficult for the primary consumers of football data – clubs, etc. – to, y'know, consume football data.

I think that there are four reasons for the problem, which I'll cover as briefly as possible:

1) Football data is difficult to collect – player movement is important but has physical obstacles (either you need to be in-stadium or you'll have players off-screen in video footage); a large amount of on-ball events require high amounts of time and/or training to collect.

2) Football data companies are scared of losing business – because the data is hard to gather, you want to keep buyers locked into your services

3) It's hard to do more than one type of data well

4) Finally, the entire football industry is fragmented and fairly tech-youthful

But broadly speaking, football is sick of this. Some examples, in no particular order.

Example 1, Hugo Bordigoni of data company Skillcorner, recounting a story from 2019:

On our website contact form we had a submission from [then-Liverpool head of research] Ian Graham. He said: ‘It looks like you’re collecting tracking data, but I’m pretty sure you don’t have cameras installed in the stadium. If you are doing it from broadcast, we’re interested.’

Example 2, the association of European Leagues launching a Sports Data Value Chain project (snappy, Roman numerals-looking acronym of SDVC):

As sports data becomes increasingly central to commercial growth, fan engagement, and league operations, the SDVC project explores the opportunity to develop a centralised data platform that reduces fragmentation, standardises definitions and processes, and unlocks new opportunities for collaboration and monetisation across the ecosystem.

Example 3, a project to establish an IEEE standard format for football data:

This standard establishes a uniform and standardized format for providing digital data arising from football (soccer) matches. The standard includes data from the following sources: match sheet data, video data, event data, tracking data, match meta data, and player physical data.

Example 4, from our favourite technology accreditors and [something]-favourite tournament organisers FIFA, a challenge which is also shared to online community project SoccerNet:

To democratise access to this technology [skeletal tracking data], FIFA is launching the second edition of the Skeletal Tracking Light an Innovation Challenge to explore whether skeletal tracking can be achieved using a single broadcast camera — most commonly known as “Camera 1” — which features in all standard match broadcasts and accounts for approximately 70% of the footage shown during FIFA tournaments.

Through this challenge, FIFA invites the global computer vision and research community to develop methodologies capable of generating accurate skeletal tracking data from a single, dynamic broadcast camera.

Example 5, Sportec – German data company and DFL partner – touting automated event data from skeletal tracking data:

Upon successful completion of the proof-of-concept phase in early summer of 2025, the system now detects on-ball events – including passes, ball acceptance, shots as well as set-pieces such as throw-ins, corner kicks, kick-offs, goal kicks, shots on goal, free kicks and penalty kicks – with very high accuracy. The system entered its operational trial phase at the start of the 2025/26 season and is now being tested live, and optimised further.

Football, from a number of different angles – in fact, from almost every angle available – is saying "Fine, I'll do it myself."

There is one remaining angle on this.

Like Thanos, one can acquire power in a different way: by accumulating powerful objects to create an even more powerful whole. From a Get Goalside post, last October:

[W]hy does 'Monopoly' come to mind, in a month where data company Impect has been acquired by Catapult and Hudl have swept Athletic Data Innovations into the fold, its fifth acquisition of the year*?

Yeah, the other angle – the other solution to football data ecosystem problems like "it's hard to do one type of data well" – is merely to buy all the companies that deal with different types of data. Become the true supermarket of football data. (Aldi, Tesco, Waitrose: which are Hudl and Catapult most like?)

One could argue that this all comes back to Get Goalside's regular take: football clubs should not be technology companies. Sure, you've got your nation state-funded clubs, who may as well be tech incubators, but small, parochial clubs like, I dunno, Wrexham – they shouldn't be expected to piece together the broken Fabergé egg of football data themselves.

Maybe soon they won't have to. (I mean, by the time things sort themselves out, many clubs will have sorted it out by themselves to some extent anyway, because everyone is hiring data engineers at the moment).

Will this all be resolved by 2030? I dunno. In the MCU, Thanos wound up being beheaded by Thor. I don't know how far this metaphor extends.

What would Football 2.0 look like?

Sun, 01 Mar 2026 00:00:00 GMT

There are, as Benjamin Franklin said, only two certainties in life: death, and young'uns not having the attention span for long-form.

'Demise at the hands of an ambivalent generation' is an idea that keeps captivating people, even if it remains hard to see in practice. Folks have been worrying about 90-minute matches as long as I can remember and, if anything, (with attempts to quell time-wasting) matches are longer.

I guess that Doomsday cults do have a long lineage, although as part of a younger generation I'd prefer to see The End Of Days on a shorter time horizon.

This post has two non-footballing elements to thank for its existence: the Winter Olympics, and a question asked by host Richard Gillis on a cricketing Unofficial Partner podcast episode:

One of the questions for Ireland in the noughties was always 'how do they become a Test cricket nation?'. And I'm just wondering if that's still the goal, and whether or not that conversation has passed and Twenty20 is actually 'the thing'.

The Olympics is a little like a catch-all equivalent to T20 for other sports. There's a short-form, vertical video feeling to the Games; a collection of quick championships, generally in sports that people don't think about outside of the four-year cycle. Many sports want in. They wanna be in the Olympics feed. They wanna give people an introduction to their sport which lasts eight days rather than eight months of a year.

The cricket story of the past couple of decades, which Gillis is referencing, is how the sport managed to generate mainstream success with a format lasting four hours instead of five days. And also through India, and now non-Indian investors, really really loving the IPL (like, way more than other T20 competitions).

Football doesn't have either of these levers to play with. The matches are short. It already has a quadrennial festival. (The men's World Cup is surely the closest thing to a rival that the summer Olympic Games has).

What the football Doomsday heads really want - other than increased investment from an untapped global economy or two - is ice hockey.

Now, the rink design is not pretty, especially compared to the clean lines of a football pitch, but evolution only selects for aesthetics when the beauty has a function. Restarts take place with face-offs, offside is a matter of attacking players crossing a line before the puck, the 'crease' (patch of blue ice) protects goalies (a little) from being thwacked.

Then there's 'icing', an offence where a team clears the puck from their defensive zone and past the opposing goal line without it being touched. That brings play right back to the face-off spot in the defending team's zone. In football-speak, it seems to incentivise playing out from the back and high pressing.

Most of the 'invasion sports' that I sorta know (football, rugby, American football, ice hockey) have rules that compress the play in part of the field. Rugby disallows forwards passes; gridiron only allows one; football and ice hockey allow them but prevent players hanging out by the opposition's goal. Basketball's back-court rules have a similar effect, without anything being 'offside'. We'll return to that later.

If modern football feels a little dull it's not quite because the sport has become 'decompressed'. If anything, fitness and out-of-possession organisation has made it more compressed; gone are the days when teams would struggle to recover their shape after a turnover for entire minutes. But play is compressed into uninteresting areas. Tactical design, rather than game design, has pinned players into the middle third.

This is where we go back to basketball. The back-court rules disallow a team to return to their own half after entering the opposition's, but there's also a rule forcing them out of their own half within a time period. Netball, which actually forces players to occupy the full length of the court, creates the space for players (who aren't allowed to travel with the ball) to pass end-to-end freely.

Each sport, in their own way, tinkers with the rules to improve the conditions for action and drama.

Even, going back to the IPL, Twenty20 cricket. It took a long time for the modern version of the format to emerge; going via the 50-over one-day format, which itself tinkered with powerplays (periods where fielding rules are altered - particularly at the start of an innings, when the ball is newest and fastest, to encourage batting teams to take risks).

Experimenting with a sport's rules isn't easy, of course (although if we're looking for a grander purpose for the English League Cup, there we go). Fortunately, there are other vehicles for action and drama. And the Winter Olympics was full of it.

Before the Games even began, we had ski jump crotch chat, and things smoothly slid into Scandi infidelity chat, Minions copyright arguments, Lindsey Vonn's ACL tear and (we now know) near-leg-amputation-causing injuries. And it kept going: Nazgul the wolfdog, a more usual type of Olympics crotch chat (condom shortage), and the figure skating coach with no shortage of jackets.

A more sensitive organising body might clamp down on how much that potentially embarrassing stories like these are discussed. And yes, you wouldn't want most of these to become a pattern, but never in my life have I seen people talking about biathlon and figure skating so much.

If a sport isn't a part of a cultural fabric, people don't pay to watch it and advertise around it, and athletes can't train to be good at it. (Unfortunately, sport's success being so wrapped up in the cultural fabric butts up against parts of The Culture being hostile to people in the sport). This interwoven existence with the cultural fabric was touched on by Maggie Murphy, Aston Villa Women's managing director, speaking recently to The Athletic.

“[With the most regular kick-off slot being midday Sunday,] Then you’re asking talent [at pre-match activations] to perform at 10.30 in the morning ahead of a 12pm kick-off. Kids and their parents have Sunday morning football. We’re trying to work with the universities, but university students coming out here at 10.30 on a Sunday morning? Probably not. So who are we going for? Mates that want to get together for a beer? At 10:30 in the morning?"

And rugby and TikTok star Ilona Maher, on the Richer Lives podcast:

What I really love about the WNBA is these players have really stepped up who they are, to create a brand off the pitch [...] I get kind of annoyed with rugby. We're not a sport that's going to get attention just by somebody tuning in. We're not the NBA or NFL, we need to do more. [...] I wish more players were putting themselves out there. If you want the sport to grow we need that, because I cannot be the only one that people come to the games to see.

Arguably, Maher has the Gerard Piqué-backed King's League as evidence in her corner too. You would think that a venture with a World Cup winner at its centre, that gathered together famous streamers and famous contemporaries of Piqué, would be a smash hit. And don't get me wrong, the numbers seem fine. Its recent 'Kings World Cup Nations' streams drew 200k-400k on YouTube and Twitch individually (while also being streamed elsewhere) prior to the final, which was obviously much more watched and had a full stadium attendance. Its Spanish and Mexican 'domestic leagues' are reportedly profitable.

Kings League is aimed at being Of The Culture, where the culture in question is teens and 20-somethings who watch streamers on Twitch. That does mean that, outside of the showpiece events, it is not an in-person spectator sport. As a 'digital native' venture, it's not designed to be, although one wonders whether that's a face-saving decision as much as anything, despite its easy-mode levels of famous backing. The WSL's in-person attendances this season have been discussed and fretted over, but they would dwarf Kings League attendance. (To be very clear, not an apples-to-apples comparison, but comparing apples and oranges isn't totally unfair).

Culture is difficult to define. It might even be harder to quantify than defending. At least with defending, you don't have to wade through the fake data of bot armies. (Even if there are data providers whose reliability feels like that).

If anyone in football gets it, it's Arsenal. Partly, this is an excuse to link the Adidas Originals 1990 advert from 2020. But they're also the powerhouse leader in women's football attendances, despite relative lack of recent success. (Relative both to Chelsea's six league titles on the bounce, and to Arsenal's own dominance of English women's football in the '90s and '00s).

Arsenal are cool. Their on-pitch product is meatwall corners (men's team) and bizarre third-gear performances (women's team), and yet they are still quite cool.

Football is such a hegemonic sport that it doesn't need to be too worried about losing its status (and if it really cares about The Youth, it might wanna make it easier for them financially). But if it's really worried about being disrupted, there are so many tweaks you could make that are aimed at fun and action (instead of at auditing in-match admin and time-keeping).

'Death at the hands of an ambivalent generation' only happens if you let them be ambivalent. Maybe the tweaks to keep things entertaining are the taxes you pay to keep them engaged.

The 'Just About Managers'

Sun, 25 Jan 2026 00:00:00 GMT

'Gareth Southgate blogged on LinkedIn' is a real sentence. There's a lot of bad in the world - the resistance to it a light in the darkness itself - but every now and then the universe throws you a bone. Southgate even made some typos.

In 'Manager v Head Coach', he was commenting on the departures of Ruben Amorim, Enzo Maresca, and Xabi Alonso from their jobs.

Although each coach left under slightly different circumstances and for more than one reason, power struggles [...] were ultimately the root cause of the end of each tenure.

As Southgate says, the power of a 'manager' has been whittled down over the past couple of decades. It would now be uncommon (at the higher tiers of the game) for a club to not have a 'sporting director' figure working either alongside or above the 'head coach'.

I've previously wondered whether this has been one of the obstacles (among others) to 'analytics' being used in first-team, on-pitch matters:

‘[F]ootball people’ have accepted a lower level of control of the off-field stuff, retaining control of tactics and player-management (although some head coaches do gripe about the say-so of their medical departments). How might things be different if managers had demanded they keep control of transfers, and relinquished some on-field control to ‘directors of tactical methodology’ instead.

Now, neither me nor Southgate are saying that modern managers can do it all. As Get Goalside's new blogging rival says:

[I]n a classic example of Parkinson’s Law (where time saved by reducing responsibilities is quickly filled by something else), the modern Head Coach now finds themselves managing larger squads, bigger backroom teams, far greater analytical demands, and ever-increasing media and commercial obligations.

It's probably fair to say that, of those added burdens on a head coach, not many of them are fun. And the one that is fun, at least to the individuals in those jobs - 'greater analytical demands' - is, yes, the one that data analytics may 'threaten'.

'Threaten' is in quotes there because whether a coach is 'threatened' or is not an inevitability. It's an organisational management problem, really, although one similar to transfers and medical load management advice, which are not without conflict at many clubs.

It's easy to understand the frustration. Heck, as with many analytics themes, Moneyball is a primary colours guidebook/caricature: diktats coming from On High down to the manager, whose role is reduced to being a figurehead on the side-line. In Moneyball, of course, the control of Upper Management got the team to the play-offs [edit: not, as originally written, the World Series]. But if Management make decisions that don't pay off... well, there's already a term for that in football: losing the dressing room.

There is a world where analytics can be coach-focused more than club hierarchy-focused, although the advantage is with Hierarchy. It's the club whose budget pays for the software, the club who has to think about succession planning. [Related: "So, everyone's a decision-maker now", October 2023].

Managers - of the type of Alonso, Amorim, and Maresca - also have very different datasets-of-interest than the ones that modern providers deal with. Southgate again, on why his LinkedIn bio reads 'Leader, Manager, Coach':

The Leader operates in the public eye, engaging fans, setting the vision, shaping the culture, and championing and protecting the players and staff.

The Manager is ‘managing’ stakeholders, departments and people, maximising collaboration and making plans stay on track. Oh, and don’t forget the role of persuading 25 financially independent elite performers to put collective purpose ahead of their ego.

Finally, the Coach works on the pitch, setting the style of play, preparing tactics to exploit opponents, and developing the technical abilities of individual players.

Arguably, the best bit of data software for football 'head coaches' would be Claude Cowork. But, because of the football domain that managers have managed to protect - the 'coach' part of Southgate's trident - any software would need an intense boot camp in football terminology. Cynically, one could rip a set of coach training materials to tune the LLM-aided system.

This wouldn't necessarily help the Hierarchy-Head Coach relations though. Let's look back at something from Get Goalside's 'Lessons in genAI' post, written just over a year ago:

Currently, the clubs are the ones who buy software - but if part of what they’re hiring in a head coach is that coach’s methodology, it’s that coach’s ‘data’ that is the most important to access. [...]

And, of course, if we’re talking about a coach’s knowledge as ‘data’, there’s the data ownership question. If LLM-based systems are going to be relying on coaching knowledge, coaches have got to make sure they can take this with them and use it in future roles.

A few days ago, Microsoft CEO Satya Nadella said (perhaps with some self-interest): "If you're not able to embed the tacit knowledge of the firm in a set of weights in a model that you control, by definition you have no sovereignty. That means you're leaking enterprise value to some model somewhere."

That message echoes a point made by Twenty First Group in a podcast/event collaboration with the Unofficial Partner podcast. Essentially, 'your company knowledge is valuable data, and you will lose out if you don't think of it as data and make use of it'.

So now, imagine you're Ruben Amorim. You come into Manchester United, a club building their data capability. You have a very distinct style of play which has had success in one context, but that you don't have lots of experience in adapting to different circumstances. A rich but partial dataset.

What kind of 'data infrastructure' would best help you? For someone who keeps saying that football clubs should avoid being tech companies, I'm going to say something shocking, but maybe a nice, clean API from United's institutional knowledge data banks [no jokes please] and a Claude AI-morim would be a good route forward.

And, circling back to Gareth Southgate's blog, this could help a head coach be a true manager. If the current and historic knowledge of different departments is open to be tapped into by a well-crafted LLM-aided system, it could be easier for an Amorim or a Maresca to 'manage stakeholders' and 'maximise collaboration'.

This isn't just a power play. The friction around transfers feels a particularly frequent occurrence, and reporting suggests that disagreements on medical adjustments was a cause for contention with Maresca and Chelsea. But if a club is able to say 'hey, here's our research data on this topic, can you sit down with your CoachingBadgeGPT and see what ways it can align with your thinking', maybe that would feel less like Upper Management instruction and more like Cross-Management partnership.

I await Sir Gareth's further thoughts on the matter.

Square one

Wed, 21 Jan 2026 00:00:00 GMT

Alt title: "The publicly available data source is dead; long live the publicly available data source."

On Tuesday 20th, a lotta folks were saddened by the abrupt departure of advanced stats from FBref, one of - if not the - best public reference site for data. We raise a perfectly-measured glass in commemoration, and hope for a swift resolution in some form.

But it left a lot of people thinking about access to 1) data 2) smarts.

So this post will have three sections:

what we kind of know about football
where to go to find data
where to go to find insights

My opinions are my own and will be inevitably skewed by various factors. This post will avoid being long, because it's better for people to read a smaller amount of detail than not read a larger amount of detail.

To the long-time readers and in-football workers: let me know if and where I'm leading the newbies astray. For the newbies, I've been around long enough to say that my first football analytics conference was nine years ago.

State of play - football

The biggest thing that 'analytics' has produced is still - and possibly will always be - expected goals. But xG encapsulates so much of the rest of football data: its insights are less about discovering a Hidden Factor, and far more about the balance between Factor A, Factor B, and Factor C.

Expected goals' biggest learning isn't really "don't shoot from distance". It's more like "you're meaningfully underestimating how much harder it is to score from 25 yards, and from wide angles in the box". (That, naturally, leads you to shoot from distance less often).

There's probably another couple of features waiting to be neatly packaged into soundbites: sight of goal, and pressure on the shooter. Data provider Statsbomb have written about these before (sight of goal here; can't find the blog on pressure), and it's a near-certainty that others have done this work in private.

Again: these are factors that everyone knows makes some kind of difference. But knowing how much difference is information that could help you craft your tactics. For example, if shots from 25 yards are worth 0.02xG on average, does an absence of pressure raise this to 0.03xG, 0.05xG, or 0.1xG? That knowledge is the difference between deciding to consciously create those chances and deciding they're not worth the effort.

Elsewhere on the pitch/spreadsheet, things are similar. 'Everybody knows' that good teams have good athleticism and more of the ball, but people are still feeling out what the magnitudes are. What does it mean if a team can sprint 10% more than their opponents? What does it mean for a player to be safer on the ball when under pressure?

There's an increasing focus on mixing tactics with physical data: players focusing on sprint distance will soon be a thing of the past. It's more likely that coaches will start pinning up things like 'recovery sprints' in the dressing room after a match. The change is partly about what actually adds to player fatigue (frequent accelerations and decelerations can be more onerous than the distance at high speed), and partly about what's meaningful in a game. The interaction between physical output and time spent in- and out-of-possession, as well as time winning/drawing/losing, will be better explained too.

When it comes to possession, things seem more open-ended. The buzz of 'line breaks' is still around, but it risks being a false idol. There's little use in fizzing a line-break through to a teammate who can't control the pass or lay it off to anyone.

I think there are a few areas of fertile ground here: 1) the interaction between a pass and what happens next 2) body orientation 3) first touch. Watch players like Youri Tielemans and Yui Hasegawa (work your way back down the alphabet afterwards). Maybe the concept of 'usage' will get explored and refined (ported over from NBA, 'usage rate' currently refers to ending a possession through a shot or turnover).

Meanwhile pitch control models, using tracking data, are exactly what they sound like. They put a number on who has control of what part of the field and have already been used to aid player positioning/decision-making. They could become a great feedback loop for player development. That said, there's a difference between 'in the moment' decisions and 'full match' decisions. A player's assessment will change based on how tired they are/expect to be or what the feel of the game is.

Speaking of players, we've not mentioned transfers yet. Putting together a set of statistical profiles or filters is basically a norm nowadays. If nothing else, they help to focus deeper analysis of scouts/video analysts/background checks. Squad composition, or 'philosophy' more generally, is a surprisingly open question. Teams may 'want to play this way', without a clear sense of why. Meanwhile, some clubs take a strategic overview of the league or talent pool available and tailor their strategy around that.

'League adjustment' (whether a good performance in one league translates to another) is still an issue, but clubs will often have a feel for the relative quality of competitions. (There are large exceptions to this though, particularly when money is unevenly distributed). There's a statistical modelling approach to league adjustment, but some clubs are just smart in the way they focus their attention.

The largest area of study, though, will be about competitions outside Europe's elite men's football. Primarily, I'm referring to women's football, but this would equally apply to lower-tier leagues in major European countries or to the whole pyramid elsewhere. How much does set-piece effectiveness change depending on the technical skill of the takers? Is it easier to benefit from a high press in leagues where players may lack composure? What effect does it have on receivers of line-breaking passes if the pitches are low quality and the ball is bouncing?

Fortunately for fans and unfortunately for researchers, the landscape is always shifting, presenting slightly new problems. Because every team is always trying to counteract and outwit every other team.

State of play - accessing data

This is the thing which the FBref update has caused most concern about. Unfortunately, the state of play is a little bit 'it depends'. WhoScored has a good array of data - as stats and chalkboards -, including for the WSL, but not for the NWSL*. FotMob might be your best bet for team and player xG totals over a season, as well as a bunch of other stats, but with a bit less control over what you dig into. There seem to be a raft of score apps using data as a pitch to users nowadays, so it might be worth scouting around if you're curious. Both those highlighted services are decent for checking out players' season histories too.

*For NWSL, your best bet might be American Soccer Analysis (who also have stuff on MLS and USL).

If you're interested in trying analysis of your own, I do think there's something to be said for copying data into a spreadsheet by hand. Now, I know that sounds a bit 'when I was young we didn't have X and we did just fine', but think of it as how hand-writing notes aids memory retention. From personal experience, you will rarely know a set of teams or players more closely than when you're putting their stats into Excel week-by-week. If you get a community together, that can lighten the load. But make sure you focus the collection at least semi-smartly, and collect a little bit before committing to collecting a full season.

Then, if you're really eager to dive into 'analytics', there are some public datasets to explore. Statsbomb have released event data covering full league seasons to one-off cup finals. Impect released a season's worth of data. Skillcorner have released a set of data that covers tracking data and less traditional event data.

With all of these, I'd advise starting small on something you're curious about. Shot maps are always tempting, but I've always found them a little dissatisfying. I've tended to get more joy out of pass maps with a particular filter applied to them. Working with an unfamiliar set of data is like doing press-ups with one arm in a sling. Libraries like mplsoccer (Python), ggsoccer (R), d3-soccer (JS/D3) can help do some heavy lifting plotting events visually. Also check out Friends of Tracking.

State of play - accessing insight

To quote, well, myself, "You can think analytically without using data, and you can use data without thinking analytically."

It's easy to get carried away with the data itself, but the data is just a representation of football. So there are three things to understand: 1) football itself 2) the manner in which the data represents the football 3) the statistical techniques of working with data.

If you're already a mathematical person, you'll have a headstart on the third of these. The second is fairly easy to pick up with a bit of time and cross-referencing data with video footage. (A tip on that one: some of Statsbomb's open data is from FIFA World Cups, and FIFA have some World Cup games on YouTube).

The first is complicated by the fact that football itself is always evolving (although the general themes stay the same). Back in the mid-2010s, set-pieces were seen as a very inefficient method of chance creation; the recent focus on training routines has made that particular data-led analysis seem outdated.

As I say, the principles are generally consistent. If you're really interested, I'd recommend Spielverlagerung's Tactical Theory posts - starting with the oldest, because they're the broadest. The Athletic's 'How Football Works' series is quite good too.

My other advice would be to get a means of watching football matches (whether live or old games) on something where you can easily rewind by a couple of seconds. A really important part of my learning about football was spent watching clips, pausing, rewinding, watching again, rewinding, watching again, to try and work out why Player A did that instead of that.

In terms of data's interaction with football, I'd recommend the Hudl Performance Insight conference (previous editions at the Hudl Statsbomb channel). Also, Pysport's talks and output from the DTAI Sports Analytics Lab. Statsbomb's research on gender-aware modelling is also kind of a must-read. Previously-mentioned American Soccer Analysis are also a nice hub for some interesting work.

We're getting into the realm of research papers now, and if that's your thing then my starting point would be Jan Van Haaren's annual list. I will do one extra 'individual paper' recommendation, and that's this one on rest defence by Forcher et al. Rest defence is (or was, recently) a hot tactical topic, so this is an unusual case of data work being developed alongside tactical discussions at roughly the same pace. The citations/references are also good.

Post-script

Hope is not lost, even if data in tabular form is now harder to come by.

There's also this blog/newsletter, which you can subscribe to. Some representative lines from the latest post.

The idea of data staff being 'embedded' with a team's performance staff seems to be slowly growing.

A tracker in the ball and on boots (such as PlayerMaker; again, the BBC urge strikes) and you've got tracking data that can be used... anywhere?

I don't know what to do with this information other than file it under 'interesting'. I feel a similar way about La Liga putting out actual research papers.

The Grand Rewrite

Fri, 16 Jan 2026 00:00:00 GMT

If all the world's a stage, each day is a script rewrite. Some rewrites are easier than others, and – this is the segue – the same is true for data pipelines.

That's where we're going to start, in this rattle through of things that I've bookmarked on LinkedIn: Manchester United and London City Lionesses.

United are hiring three roles, all earmarked as 'senior' in the job titles (data engineer, data scientist, software engineer). There will be no jokes, referencing the men's team's current travails, none at all. If you want jokes, there's enough in the Get Goalside piece from 2022 when they were hiring for decision scientists.

There are a number of different avenues that you could go down with a data department at your disposal, let's list as many as possible:
- Identifying general undervalued (or overvalued) areas in the transfer market
- Looking for players who fit specific role-based requirements for the first team
[...]
- Optimal C-suite engineering (retain as few of your bosses as possible, as many as needed)

Meanwhile, WSL side London City (actually their ownership group, Kynisca) are hiring for an 'embedded' data scientist and data engineer, and who'll work on-site either in London or in Lyon.

The idea of data staff being 'embedded' with a team's performance staff seems to be slowly growing. There was a time, a few years ago, when people talked about the rise of the 'analyst-coach', a coach who got analysis and got numbers who'd help bridge the gap, and this is a similar idea just starting from the other side of the river.

Both (or all five) of the job ads are better written than job ads you'd see even a couple of years ago. That said, it's interesting - within the broader context of football data roles - that the London City data engineer role "will own and evolve the data foundations that support performance analysis, innovation, and decision-making across our multi-club football ecosystem", and asks for "2+ years of experience building data pipelines from APIs or external data sources".

I don't highlight this to put Kynisca in a bad light (partly because the company is named after Cynisca of Sparta, first woman to win at the Olympic Games, and I know the beginnings of an ancient curse when I see one). I only mention it because there are football clubs who will have hired people to produce their data pipelines with far less experience than Kynisca are aiming at. Those pipelines are gonna be in more of a need of a rewrite than most, at clubs who will likely be less familiar with the concept of tech debt than most.

In the grand marketplace of job roles, that probably leaves a gap (in 2-5 years' time) for experienced freelancers. This is another segue. Nick Meacham of SportsPro Media posted the other day about what feels like a spike in 'advisors' (in the sports industry more broadly):

It feels like every other day I see highly experienced leaders — people who’ve spent decades in sport, made mistakes, learned lessons, and delivered real success — struggling to land another full-time role at a similar level. [...] Instead, many are moving into advisory work, taking on projects for the very organisations or investors who won’t commit to them permanently.

The gig economification of the C-Suite?

Football data, at least, seems to be successfully swimming against the current. The United and Kynisca roles seem to be for the long haul, and the English FA recently had an ad up for a Head of Football Intelligence and Innovation.

As far as football data and technology is concerned, senior leadership roles are still very much being created for the first time. The wheel is not being re-made, it's being put together for the first time. Wheel re-inventions are happening though, and aren't necessarily such a bad thing, because PlayerData have collaborated with Mitre on a GPS-enabled football. As an English child raised on the BBC, the urge to say "other tracker-enabled footballs are available" is strong, but the only other one I'm aware of is the Adidas-Kinexon one.

Putting a chip inside the ball is the best football technology development since the Ronaldinho crossbar advert. A tracker in the ball and on boots (such as PlayerMaker; again, the BBC urge strikes) and you've got tracking data that can be used... anywhere? Any league in the world could have its own centralised (you'd need to coordinate between teams) tracking data deal. And maybe, down the line, automated offside and goal-line technology systems?

Now, maybe that's a naïve dream. But sometimes you gotta dream. There's two interesting, related datapoints from a part of the industry that's had to dream (another segue): women's football fans. This past week, an Unofficial Partner podcast interview with Nikki Doucet (CEO of WSL Football) and research from Even Sports (of the unofficial Fantasy WSL app) both indicated that English women's football has a lot of fans who follow multiple clubs.

In a way, this is unsurprising. Research and anecdotes tell us that many fans were energised by the England team; the England team has players from multiple teams; and we all know that young'uns follow players like old'uns follow clubs anyway. I was struck, though, by the degree of it. Doucet said that users of the new WSL website often select '3 or 4' clubs to follow; Even Sports' data - which, granted, comes from fantasy football players - had 46% of respondents watching 'as many games a week as possible'.

I don't know what to do with this information other than file it under 'interesting'. I feel a similar way about La Liga putting out actual research papers.

Another day, another blog, another rewrite. Keep rewriting.

Finding the bench to mark against

Mon, 05 Jan 2026 00:00:00 GMT

In this season of resolutions, one that has not yet failed for me is this: if you don't like a film/TV show/band, then look up what that artist likes. At best, you start appreciating aspects of something you didn't like; at the very least, you understand their choices. It's easier to judge something in context than in isolation.

Similarly, - go with the segue - if you read a research paper on running data, then at some point you'll get a line like: "This research could be used to design more match-relevant training sessions for X, Y, Z." Because designing the preparation for athletes depends on the context of what they actually do in-competition.

With that in mind, if you have some time off over January and want some juicy reading material, I'd recommend checking out this Google Scholar list (of works that cite the 2018 paper, 'Are Current Physical Match Performance Metrics in Elite Soccer Fit for Purpose or Is the Adoption of an Integrated Approach Needed?'). Save that Waterstones voucher for another time.

The notion of 'needing to know what players are required to do' is the same for 'event/on-ball' data too, and the results can be quite surprising. Recently, the Opta Analyst recently wrote about Martin Ødegaard and some of the areas he excels in. At time of writing, he had more line-breaking passes into the box than any other Arsenal player, and averaged more possession regains in the final third of any Premier League player. However, those figures were only around 2 per 90 minutes and 1 per 90 minutes: two skills that coaches (and fans) desperately want in their players, but which happen only a little more often than fans getting hitting in the face by a stray shot.

High-impact events in football are rare - that's not revelatory news, but the extent of it is more far-reaching than high-quality shots alone. Here's a selection of stats from the current Premier League season (only counting players who've played a decent amount of minutes, time of writing 01 January after 18-19 games):

Only two goalkeepers face 5 shots on target per game, on average
Zero players are taking four shots per game, on average
Only a little more than a dozen players are making three or more tackles per game, on average
Only three players are making two or more interceptions per game, on average
Only seven players carry the ball into the penalty area at least twice per game, on average

Name a consequential football action, and the chances are that you can count the tally for the league's third-best player on your fingers. Maybe that's just the nature of something being 'consequential', but it's a strange kind of situation where looking for a 'midfield destroyer who can break up the play' means someone who makes a particular action less than half a dozen times in a 90-minute game.

Part of the solution to this 'problem' is to give people context of when a number is big and when it is small. But is this, in a real sense, a 'solution'?

Players who make 'consequential' actions on a regular basis are unicorns. If you're looking for a striker who takes lots of shots, you might think of Erling Haaland... who takes a shot once every 20-25 minutes. And so, maybe the search should be for slightly-less-consequential but slightly-more-frequent actions, or traits, that players show.

Martin Ødegaard may 'only' average two line-breaking passes into the box per 90 minutes, but his control places the ball in the right spot for him to make improvised passes without being dispossessed, his weight of pass is astonishingly well-controlled, his frame of mind is towards reward rather than risk-avoidance. Some of these will show themselves in other statistics.

Using event data is often about working out proxies for valuable skills. There is, clearly, a literal value to players who make lots of passes into the final third or penalty area, but there's also implicit information about technical ability, vision, composure, et cetera. Generally speaking, the further you get from goal, the more that the value of implicit information will outweigh the value of the literal information.

At the moment, this post is skating pretty close to gate-keeping; making it seem like data is Aramaic, able to be deciphered only by a special, learnéd few. Yet the aim is the opposite - maybe data is Aramaic, but how can we bring Aramaic to the masses?

One of the most efficiently thorough lessons I learnt about football statistics came through typing stats into a spreadsheet: despite the fuss made by TV commentators, players who made the most clearances were generally on bad teams, and the best defenders did not tend to make many clearances. The reasoning is simple: bad teams come under the type of pressure that leads to clearances being necessary; good teams, where the good defenders play, do not. But how do you signal that to people, many of whom approaching stats for the first time, without requiring the (useful but tedious) time spent in Microsoft Excel?

It would be tempting to try and link stats to success-based outcomes like 'winning', but probably a mistake: there are far too many factors at play. And people have a tendency to see causation in correlation, no matter how many times you ask them politely not to.

But knowing that Brandon Flowers listened to Depeche Mode and Roy Orbison growing up makes The Killers' back catalogue make a lot more sense. There's something in that, I think. Maybe.

Hey: if you didn't like this blog, try and work out who Get Goalside's influences are. That'll make it better.

Cutting room floor

Bits of internet research I did but didn't end up putting in the main post.

The Pablo Picasso museum in Málaga is, from memory, quite good for putting Picasso's work, and his successors, into context of the art trends and global events that they were experiencing.

The difference between the first half of Hot Fuss and the majority of The Killers' other work always kinda surprises me. But knowing that he was big into Depeche Mode's Songs of Faith and Devotion as a teen, but that Oasis was a deliberate influence for Hot Fuss makes it make more sense.

UK school examinations body AQA have a document outlining their current policy/thoughts around comparative judgement as an alternative to rubric-based assessment. It can kinda be boiled down to 'we like it but it's not viable at scale right now'.

'Analytics history' - there is so much to write about and share

Thu, 18 Dec 2025 00:00:00 GMT

This piece has been re-written about four times.

At one point, it opened by using The Verge's fun tech-history podcast series, Version History, as an analogy. At another point, it was bookended by referencing 'The Dark Ages' and 'The Renaissance', talking about how the naming of time periods is as much about the era that the names were coined as it is about the periods of history themselves.

Instead, you get the eventual fruit of trial and error, a muddle when things don't fit neatly into place. That's fitting, really, because that's what the history of 'football analytics' is really like.

The popularly-told version tends to go a little like this:

A long time ago, in a galaxy far, far away (1950s Swindon), RAF accountant Charles Reep collected some data. He was opinionated and wrong, but somewhat influential. Later, Prozone and Opta were founded in England. Opta 'created' expected goals, and Liverpool used it to hire Klopp, sign Salah, and win the Champions League and Premier League. (Brighton and Brentford also used data, because their owners were into betting, but no-one there talks on the record).

The Get Goalside version - for example, in 'The path to now' - has tended to go like this:

We have football datavis dating back to 1920s Hungary, a generation before Charles Reep began collecting his data. He was overly critical of possession football, but had decent ideas, and his dataset was used to create an xG-lite model in a 1990s paper. A full generation later we got Twitter Analytics, xG popularity exploded, coinciding with its role in Liverpool hiring Klopp, signing Salah, etc etc. The nerds won, but history rolls on.

Neither version is good.

By an extremely strange coincidence, one of the issues with both versions traces back to the very same city held up as the Face of Analytics: Liverpool. But not the Liverpool of Klopp, Salah, not even the red side of Liverpool at all. We're talking Everton. And we're talking the 1970s.

Early in that decade, an agreement between Dr Vaughan Lancaster-Thomas and Toffees chairman John Moores allowed a (coincidentally) Everton fan Thomas Reilly to help monitor physical and running-based metrics of players. Reilly - by all accounts an athletic man himself - took physical and running measurements about Everton players that would nowadays be considered routine, but back then most certainly wasn't.

"[T]he statistics they gathered from pre-season were useful in a few ways," Joe Royle, Everton striker at the time, told The Athletic in 2020. "If you had a player who got injured during the season, did rehab and got back to what they thought was full fitness, they could say, ‘Well, this is what you were doing pre-season’ and compare the numbers.” This was the early '70s![1]

In the middle of the decade, Liverpool Polytechnic (the establishment that Vaughan-Thomas was connected with, later named Liverpool John Moores University) launched 'the first BSc (Hons) degree in sports science' [quoting from LJMU themselves].

By that time, Mohamed Salah's birth was still almost two decades away. Yet, coincidence again: this period - the late '60s to early '70s, the era of Bobby Charlton and Gerd Müller - was also the time that two statistical papers using Charles Reep's shot location and passing sequence data were published; co-authored by Bernard Benjamin (1968 and 1971) and Richard Pollard (1971).[2]

Those papers would create a ripple that is, at the very least, detectable to a medium-grade Google Scholar search. For example, there's debate about the statistical pattern that goal-scoring follows, with two seemingly unrelated pieces of work in the early '80s arguing for a different tack to the Reep-involved papers.[3]

Later that decade, now-Professor Thomas Reilly helped convene the first World Congress of Science and Football. It would be held every four years (the year following a men's World Cup), and would be a place for research and researchers of all football codes: association, rugby (both types), American, Aussie rules, Gaelic, futsal. The congress - and subsequent editions - would have a focus on various features, from the 'measure lactic acid' end of sports science, to what'd now be called 'event data' studies, to psychology, to kinematics. And even, in 1987's first edition, a coinciding with a burgeoning interest in computer-aided analysis systems.[4]

In the Get Goalside world of football analytics, there's always been a special place for the Forums put on by Opta [later Stats Perform, and later, really, superseded by StatsBomb, now part of Hudl]. The presenter alumni of the Forums would now count multiple heads of analytics departments in their number, not to mention others who work elsewhere in the game. (Including, I suppose, myself).

Besides providing 'outsiders' with the opportunity to rub shoulders with folks currently working at clubs, the Forums were also a hub for work and for connections of all kinds. That's probably, really, their biggest benefit. The Science and Football congresses seem to have had a similar effect, quarter of a century earlier. If you find an interesting football statistics paper from the '90s or 2000s, chances are that a paper from a Science and Football congress will be in its citations.

And there are interesting papers. To throw some examples from the sixth Science and Football congress (2007), there's 'Analysis of actions ending with shots at goal in the Women’s European Football Championship (England 2005)', by Józef Bergier, Andrzej Soroka and Tomasz Buraczewski. And: 'Match analyses of Australian international female soccer players using an athlete tracking device', by Adam Hewitt, Robert Withers, and Keith Lyons.[5] The year 2007 wasn't just ahead of the trend in terms of sports science on women's football, but it was early days for GPS tracking devices full stop.

There's a wealth of interesting stuff from Spanish researchers too. For example, 'Use of the polar coordinates technique to study interactions among professional soccer players' (2002), by Carlos Lago Peñas and M. Teresa Anguera Argilaga, which plots interactions between players in a 'passing sonar' style well before it became popular online. Or, 'Análisis de las posesiones de balón en fútbol: frecuencia, duración y transición' (2008) by Julen Castillano, looking at the number of possessions per game and their durations, and comparing them with studies stretching back to the '90s.

The Spanish aren't singled out for any greater reason than some of the work was easy to trace online through the citation trail. The contents of Science and Football congresses (easier to find online than the full books of papers, published after the events) reveal a range of other nationalities, judging by the imperfect metric of surnames and countries cited in paper titles. The Get Goalside assumption holds: there've always been folk who've wanted to apply some rigorous counting to football.[6]

Why, then, is this rich seam of statistical exploration into football not more well-known?

Partly, because studies were damn hard to do before modern companies were able to take the task of data collection off researchers' hands, and that naturally limited what could be explored.[7] Take the following passage, from a paper at 2007's sixth Science and Football congress, on turn demands in the Premier League by Bloomfield, Polman, and O'Donoghue:

The on-field activity of 55 FA Premier League soccer players was recorded from Sky Television’s PlayerCam facility for approximately 15 minutes each. The 15 minutes recorded was reduced to approximately 5 minutes per player by only including video sequences where the player was in possession of the ball, [et cetera...]

Much of the work (that I've been able to read) from the 'pre-Opta era' has a similar problem, and similar sections devoted to the mode of data collection. Often there are theoretical justifications for the choices behind the method. Say what you like about the specs of modern data collection companies, at least they give you a framework to start from. It may not be an ideal one - it may even be one that diverts attention from interesting areas of research - but in many cases it will have been better than starting from scratch.

For the history of analytics to be told in the mainstream, it needed a narrative, a cause-and-effect, a directional arc. That's part of the success of Moneyball; the book (moreso than the movie) tells a neat tale of data's entry into sport, from box scores to Bill James to his adherents to Billy Beane.

And the Liverpool story is the closest that football got to that. Against the odds (remembering that the club were in a tough spot in 2010), they won trophies with a manager tainted by a catastrophic season and talismanic players whose purchases were derided; they did so in part because of a faith in analytics, in expected goals; and between xG and Michael Edwards, the go-between for the data-heads and the manager, the story traces back to Opta (who popularised xG) and Prozone (where Edwards worked in the 2000s).

The line about history being written by the victors is only half-right. Sometimes history is what's actively remembered, sometimes it's shaped by what is deliberately suppressed, and sometimes by what's merely forgotten. Evolution follows a similar pattern; it led to opposable thumbs and complex brains, and also the platypus.

A building at Liverpool John Moores University is named after the late Professor Thomas Reilly. There's no significant reason why the history of football data became so synonymous with expected goals and not with the work of Reilly's precursors, peers, and intellectual descendants. The separation of 'sports science' history and 'analytics' history walks, talks, and sounds like a platypus.

Ending on 'platypus' - maybe this could've done with one more draft...

Footnotes

[1] The Athletic's article on Thomas Reilly's time at Everton, which inevitably has a certain word in the headline: 'How ‘private boffins’ helped Everton become sports science pioneers' - Kudos to The Athletic for the article.

[2] Charles Reep's data papers - 'Skill and Chance in Association Football' in 1968 (Benjamin, Reep); 'Skill and chance in ball games' in 1971 (Benjamin, Pollard, Reep). Like Reep, Benjamin had worked in the RAF, joining towards the end of the Second World War as a statistician, and would become president of the Royal Statistical Society (among many other things). Pollard has had a life worth reading about too.

[3] 1980s debate about whether goals follow a Poisson distribution - 'Is goal scoring a Poisson distribution?', 1981 (Colwell, Gillett) in The Mathematical Gazette; 'Modelling association football scores', 1982 (Maher) in Statistica Neerlandica. Pollard, who co-authored the '71 paper with Benjamin and Reep, responded to the discussions.

[4] The World Congresses on Science and Football - I'd been working on this post for a while before digging into these and ideally I'd spend a lot more time reading the proceedings from them. But I know a rabbit-hole when I see one, and unfortunately this would be a big one.

[5] Keith Lyons - This post owes a large debt to the late Keith Lyons, whose blog digitised and/or collected together a number of PDFs, including Reilly's 1975 PhD thesis. Other work is harder to come by. To take one example, an intriguing paper that Lyons references in a blog - 'Human Factors in Sports Systems: An empirical investigation of events in team games' (1983) is available online, but paywalled by Sage Journals. Working through the references of Reilly's PhD thesis, things quickly become harder to trace (though obviously, in both cases, access to an academic library would help). It would have been a lot harder for me to stumble on the lines of enquiry in this piece were it not for Lyons' commitment to recording his memories, and the work of his peers.

[6] The 'Get Goalside' theory of nerds being into football for as long as football has existed - It's hardly a take worth a victory lap: association football, as a sport, is a century-and-a-half old. It predates a dozen American states' admission to the Union; predates the unification of both Germany and Italy, countries that have won eight men's World Cups between them. It was codified closer to the time of Beethoven than the era of World Cups themselves.

[7] The difficulty of a decent dataset in the 'pre-Opta era' - It feels pretty reasonable to say that a significant factor in Charles Reep's footprint on history is the size of the dataset he developed over the years. It also seems reasonable to assume that this was helped by his contact with Wolves boss Stan Cullis in the early 1950s, shortly after his data collection started. It's hard to understate the effect that a show of belief from people in the profession can have on an 'outsider'. If you'd like a look yourself, Keith Lyons (see above) gathered together the data tables from the '68 Benjamin and Reep paper in a GitHub repository, stored as CSVs: https://github.com/2622NSW/Reep-and-Benjamin

"There is an opportunity to extend these stories and provide thick description of a pivotal moment of sport analytics in England. It requires a comprehensive, co-operative story-making effort. The outcome could be an inclusive and participatory account that is reflective and critical. [...] There is so much to write about and share."

-- Keith Lyons (1952-2020), writing in 2019

Are footballers artists or accountants?

Sun, 16 Nov 2025 00:00:00 GMT

The singularity is coming. And it could bring about an apocalypse of all we hold dear: football.

It doesn't seem impossible that at some point in the relatively near future*, we'll be able to say with a lot of certainty exactly what football players should have done in any circumstance. And because, when there's money on the line, people can be taught to do incredible things, we might get football players who will know how to respond in any given situation.

*Get Goalside's speculative fiction tends to find its way into reality a few years down the line; so by 'near future' let's say 2032.

This is how it could go: possession value models are reliable down to the slight changes of body orientation and acceleration, model features are tuned to the players' actual error bars, sums get run to draw up how often player X will face situations Y and Z in a match: the player duly rehearses variations of these in the simulator. Formula One drivers are now routinely within tenths of a second of each other across laps that last 70-90 seconds, a margin as small as the Max Verstappen fan club's British division; squint and a footballer's path of travel during a match is a series of lightly undulating laps.

Obviously, this would be awful. Or is it obvious? The bubbling rise of artificial intelligence is forcing a lot of questions about what the role of this or that in society is: teaching, art, friendship, copyright - the good things in life. Is a footballer's role anything more than simply providing the optimal on-ball value (OBV)

Out in the real world, Rosalía's LUX has been released to rave reviews - an album that features 13 languages, influenced by flamenco and hagiographies. There isn't a standard path to either making a work like that, or, more pertinently, to making artists who make work like that.

That said, every artist who threads the needle of artistic and commercial success has, yes, an adventurous artistic spirit, but also a sense of public appetite. They're not so crass as to use $$$ as a metric, but it's not like their art is experimentation for experimentation's sake. So, we don't need to say "oh, footballers aren't artists, they want to win", because 'art' and 'external success' can clearly co-exist.

One could argue that LUX will be a fantastic money-spinner, and that inventive players like Lionel Messi are fantastically OBValuable. But the choice of approach changes a number of things.

For one, the way you go about developing artists and developing accountants is different (even if 'creative accounting' is incredibly lucrative in itself).

And then there is the 'product' that fans tune in to see. A lesser man would say something dismissive here about 'football isn't chess', but chess isn't chess. Perfection in chess has been accomplished and, in a way, rejected: computers started beating grandmasters in the '80s, but people don't follow chessbots playing chessbots - they follow people.

Some reading this might be thinking 'nobody truly thinks that data and machine learning will fully replace human thought'. That's naïve. There are coaches who think their decision-making on the sidelines is better than the players on the pitch; there will inevitably be those who think the same about PitchControlGPT, when it comes along.

Maybe the notion of 'correct decisions', and even of 'better decisions', is wrong; a paved road for autocrats. When we talk about artists, we tend to talk about 'taste' instead, even though there are decisions which are pretty inarguably better than others (a notion even baked into western music terminology - a melody can be 'resolved', chords can be 'major' (triumphant, positive) or 'minor').

'Taste' points towards better decisions, while being a little less conceptually narrow than grading decisions quantitatively. Would it look different to coach a player towards better decisions versus coaching them to develop 'taste'?

And what does it look like, as a fan, to see a team aiming for 'better decisions' compared to one striving for 'good taste'?

Based on the reactions to the Premier League's set-piece focus this season, I think we know the answer to that one, at least.

Congratulations, you have been acquired

Sun, 26 Oct 2025 00:00:00 GMT

Disclaimer/disclosure: I work for a sports tech company, and am a bad board game player.

It may not be number one, but in the running for 'greatest ironies of our existence' has to be the fact that the inspiration for the board game Monopoly was as a playable critique of capitalism.

The premise of Elizabeth Magie's The Landlord's Game was, more or less, 'the extraction of wealth by the already-wealthy is bad'. 'What if', went Charles Darrow's spin on it, 'it's actually fun'. Monopoly was born.

And why does 'Monopoly' come to mind, in a month where data company Impect has been acquired by Catapult and Hudl have swept Athletic Data Innovations into the fold, its fifth acquisition of the year*? Last year's acquisitions added StatsBomb to their portfolio, bringing together a football data blended family that already included InStat and Wyscout (although neither of them as focused on their event data as event data as StatsBomb).

*(Sports Business Journal described the acquisition of Titan Sport as their 16th in total and the third of the year, and ADI as the 18th in total)

Don't send me to Jail for the comparison. Neither Catapult nor Hudl are monopolies of course, but Get Goalside has long been a chronicle of "huh, what's goin' on around here at the moment?". And what's going on at the moment is that football tech - and sports tech more broadly - is firmly in an Institutional Money era.

You see this in some of Catapult and Hudl's data provider buddies: there's Stats perform (Opta**, RunningBall), Sony (Hawk-Eye), and Electronic Arts (TRACAB).

**The Opta branding now incorporates the event and tracking data together.

Also at play is the amount of players (past and present) who are investing in companies and clubs themselves, either directly or via venture funds. The way some players talk about it, dressing rooms sound like MBA break-out sessions. Elite sport is one of the few areas where TV companies can still rely on hefty subscription fees (although even that is wavering), and one of the few industries minting millionaires upon millionaires.

You can't mention portfolio management and football players without mentioning multi-club models too. The modern age has an increasing amount of transfer rumours phrased like "Player has agreed deal with Multi-Club Ownership Group" (see: Julian Enciso).

This isn't just an "ah, capitalism" comparison. In its 'squad management' guises, football tech has long been in the 'asset management' game. What is the search for wonderkids on Football Manager if not an early introduction to the football tech industry and a forerunner for the transfer strategy of certain MCOs.

Really, the surprise is not so much that the football landscape is where it is now, but that a certain kind of nerd was ever considered an outsider in the first place. Somehow, The Man was punk. Not anymore.

So what, if anything, is punk nowadays?

The thing that comes closest might be the sub-industry going direct to players. Contract negotiation, personalised coaching and analysis, a hub to track your own progression - like some kind of forest plant, companies are sprouting up in some of the more fertile niches that larger trunks aren't making use of.

From a business development perspective this can have awkward challenges though. It's enough of a struggle to keep up with the different needs and quirks of individual clubs. Individual players with all of their different positions and skillsets multiplies that 30-fold. And, annoyingly, the most cost-effective use of football data - as a layer in the player scouting process - isn't something that players need to concern themselves with.

There's also the fact that Big Money is only a one part of town. Some players are millionaires, some players - in the precarious career of being an athlete - are able to set aside some of their income for extracurricular learning rather than savings. (And this might be why player agencies are making steps into data use).

If you want to focus on in-match data, it can also be extraordinarily expensive, and big companies are under little incentive to let other people create products that they could be creating (and owning the IP for) themselves.

But the giants aren't the only supermarkets in town. There are smaller companies that can be found with FIFA-accredited tracking data systems: BePro, Footovision, ReSpo.Vision [please make your name more conducive to a style guide], Skillcorner, and more.***

***There's no equivalent to the FIFA EPTS scheme, which seems a reasonably good accreditation system, for event data. In terms of FIFA-certificated optical and broadcast tracking data systems there is also EVS, primarily a TV production company; and Sportlogiq, who've renewed their EPTS status this year but are predominantly in the ice hockey world.

There's no real through-line of this post, other than 'something is happening'. Something generational - not in the sense of 'rare', but in the sense that the current landscape is the result of an accumulation of the past few decades.

In the mid-2000s, the record men's transfer fee had settled, unchanged since Real Madrid's purchase of Zinedine Zidane in 2001. It stayed there for eight years, a stability unheard of during the '90s. In 2009, Madrid broke the record it once again for Kaká and then for Cristiano Ronaldo in the same transfer window. Four years later they broke it for Gareth Bale. At a similar time across the eastern fringe of the Atlantic Ocean, the Premier League domestic TV rights almost doubled to £1.7billion for the start of the 2007/08 rights cycle, and then again to just over £3billion for the start of 2013/14.

Amidst that, in 2011, Gary Neville retired from his playing career. He may have been unusually business-minded among his cohort, but he was also only catching the beginnings of elite football's (latest) mega-boom. His final contract at Manchester United was £2.3million per year [around £45k per week, or £67k after inflation], a middling amount by the standards of today's Premier League, particularly at big clubs.

It's not exactly new for football to usher people past 'Go' from the working class to the landlord class, whatever Elizabeth Magie may have thought of that, but Neville is probably the best showcase of what has changed. In the old days, the English football tradition was for players to buy pubs as an investment for their future. He is developing Manchester skyscrapers and co-owning lower-league football clubs.

The number of Gary Neville-level wage-earners in football will have grown exponentially. I'm not sure that I - or the tech conglomerates who will presumably want to lock clubs into using their ecosystem and their ecosystem alone - will know what to do with that.

Time on the clock

Sun, 12 Oct 2025 00:00:00 GMT

There isn't enough public analytics done with the humble fan in mind.

Previously in Get Goalside, I argued that the focus with 'ball-in-play' time over the last few years is somewhere between benignly silly and a player welfare concern. Some coverage of 'BiP' has tried to spin it as a value-for-money issue (despite the fact that in-play time looks pretty similar to historical levels), but it's far from clear that 'match is live for longer' means 'match is good'.

In fact - with the memory of several cold, dull in-stadium experiences in mind - I might go so far to say that teams whose tactics end up with more in-play time actually make for worse matches.

[Ball-in-play Part One here and ball-in-play Part Two here]

One should check one's biases though.

The same data that was used for those previous posts can be used for just this question. There are various ways that you could measure 'excitement' in a match, as a starting point let's just look at the time spent in the final third. It's not a good proxy, but it's passable.

In the graph below, the platonic ideal is to be in the top-right.

This is every team in the Statsbomb publicly-available dataset who've played 15 games or more, which is why you're seeing Indian Super League and NWSL teams.

The bulk of this chart actually seems like much of a muchness, a fairly round clump with a few groups of outliers. In the bottom-right are the possession-focused bores; the top-left are top-half ISL teams from that season; and middle-top are a group of teams we'll touch in in a moment.

(Several of the labelled teams are partial: any labelled team-season that Lionel Messi was in only represents matches he played in; some NWSL and WSL teams are slightly-partial seasons due to video availability iirc)

One could argue that this is a minor confirmation of my theory: money doesn't buy you happiness and ball-in-play time doesn't buy you action. I think I'd want a broader dataset before I pat my assumptions on the back, though.

And maybe, in fairness, we should not blame teams like 2022/23 Paris Saint-Germain for being so dull.

The previous plot showed the time that either team spent in the attacking third in matches featuring those sides. So for that partial 22/23 PSG season, their games averaged about 12 minutes of PSG or their opponent in the attacking third. Indeed, it takes two to tango, and if we just split out the time that individual teams spent in the final third - as we do below - we see that PSG 22/23 actually spent a 'normal' amount of time in the final third.

Red line dotted indicates half of the match-average final third time being for the named team; above the line = more dominant teams, below the line = less dominant teams.

That clump from middle-top of the first image that I said we'd touch on later? Yeah - generally speaking, matches with those teams saw a lot of final third action because of their opponents. (Poor Yeovil Town in their WSL season, making their opponents look like Bayern Munich).

The middle-top of this second chart is a totally different type of team: the throttlers, the dominant* possession-heavy teams who simply did not let the ball into their own defensive area.

*And Goa, who finished 9th that season. You'll be unsurprised to learn that all five of their foreign player slots were taken by Spanish players. A quick scan of FBRef tells me that their opponents had the highest on-target rate of their shots, and the highest on-target shot conversion in league too. That's a story worth blogging about.

Unsurprisingly, Serie A teams make up the bulk of that 'bottom left quadrant' clump.

You could tinker with the definition of 'exciting time', most obviously to include time spent further back in the field during quick attacks, or time when a defence is in possession close to their goal under intense pressure. You could also throw shot stats into the mix here, look at xG created.

If you have the data [cough, Opta Analyst], you could spin this data up for current teams, and see what direction trends are going in. The logic is fairly simple: actions into the final third start the clock; turnovers, restarts, passes back outside the final third, or non-turnover actions by the non-possession team that clear the final third (e.g. blocked passes, clearances) all stop the clock.

If anyone needs some football-watching to pass the time, I guess you should go find 2018 NC Courage game film.

It's a grab bag

Mon, 06 Oct 2025 00:00:00 GMT

Sometimes there's a theme and sometimes there are just disparate thoughts.

It's the latter today, and in the grab bag are: the future of TV analysis, the future of football data job ads, the future of athlete health and technological development. Alright, maybe there's a theme after all.

Let's get the least shiny out of the way, because it's to do with football's big bureaucracies, FIFA and UEFA.

Leading from the front

Back earlier this year, FIFA cracked a champagne Docker container over its Innovation programme, relaunching it on the high seas of technology. (There's a 'VC' pun in there that I desperately wish I could've made work). I don't have an opinion on the value trade-offs of the scheme, but I think it's indicative of direction of travel; last month, four new areas were added to the list of things that FIFA are interested in:

audio transcription tool
basic (affordable, reliable) referee bodycams
professional (TV broadcast-grade) referee bodycams
'innovative' (and broadly accessible) talent development tools

The details about each category vary, but here's a snippet from the basic refcam one:

Recorded data, whether stored on the device, a personal device, or transmitted to a secure repository, must be protected using strong encryption aligned with recognised security standards. AES 256-bit encryption is recommended, but alternatives providing equivalent protection will be considered.

Secure storage solutions must include role-based access, audit logging, and safeguards to comply with relevant data protection legislation.

Solutions that use the referee’s personal devices (e.g. smartphone) as the recording device are highly encouraged.

That's quite a bit of specific detail!

Meanwhile, UEFA recently coordinated a consensus statement on best practices for athlete menstrual cycle tracking in football, including drawing together a view on where further research is needed. From the 'Background' section of the work:

However, despite increasing research interest, the evidence base regarding the physiological effects of the menstrual cycle on performance and injuries remains fragmented and incomplete and perpetuates misconceptions[...]

Most studies have focused on traditional endurance sports (eg, long-distance running, cycling and rowing), often excluding team sports such as football[...] Technical precision, tactical decision-making and intermittent high-intensity performance are key components of the women’s game in football. Understanding these specific demands in relation to the menstrual cycle would facilitate football-specific training and performance strategies and improve our broader understanding of player health in the women’s game.

It feels worth noting that not only did UEFA's medical unit put this together, but the focus is fully holistic, recognising not just physical impacts but the tactical/technical associations and athlete privacy/comfort/welfare as well.

As somebody who got into football analytics when it was a clear 'outsider' thing, it feels quite significant that these large organisations - with the resources and network that they have - are taking on these tasks. There's been some critique of FIFA particularly in recent years over their roles as both administrators and tournament organisers, and - while some of that is still relevant and valid - the use of the administrative capabilities for more than just calendar coordination is interesting and nice to see.

Seeing it through their eyes

As a rule, Get Goalside doesn't do shout-outs for individual companies, because it's far too easy to do someone's PR for them. That said...

This clip from the BBC's Match of the Day is the best use of VR in football that I, personally, have seen. (A far cry from Jamie Carragher gamely wearing a headset and controllers in-studio to try and talk the viewers through a clip). Part of this is that the player body positioning in the Match of the Day clip is so clear - a function of data and technology - and part is purely an editorial maturity. The tech is no different to choosing the best camera angle with 'real' cameras.

I'd argue that what makes the above clip work isn't that it "lets you see what Martín Zubimendi sees" (if I could see what Zubimendi sees on the pitch, I'd be a better player) it's that it's the best angle for the job - showing the time constraints Zubimendi had and the way that the relevant space was briefly available. It may be unsurprising that the camera angle that best shows this is the one from Zubimendi's angle, but it's the clarity for the viewer rather than Zubimendi's perspective that makes it valuable.

When the Saints go hiring in

Job ads for data gigs in football can still be pretty hit and miss in terms of 'how informative is this about the job' and 'how likely is it that the hiring team know what they want'. So it's interesting to see job ads like this six-month contract as an AI Engineer at Southampton, awash with fairly specific tasks and technologies that the role will be fulfilling and using.

This comes shortly after Newcastle United advertised for a Principle Scientist, whose 'Your Impact' section I'll copy and paste for future reference as much as anything:

We’re after someone who knows when to use a well-tuned statistical model and when to unleash deep learning. You’ll be comfortable working with large-scale time series data, building bespoke metrics, and integrating diverse models to create rich, layered analysis. If you’ve worked with football data providers like StatsBomb or Second Spectrum, even better.

You’ll need a Master’s or PhD in a quantitative field, strong Python skills, and extensive experience applying core machine learning libraries such as Scikit-learn, XGBoost, or LightGBM to structured data problems. The role may also involve developing and deploying deep learning models using frameworks like PyTorch or TensorFlow where appropriate. Experience with version control (Git) and applying software engineering best practices is essential. Familiarity with MLOps principles and integrating cloud-based data platforms like Snowflake and Azure is a plus.

There's nothing particular to add to this other than 'this is an interesting development'. But there are always Get Goalside bonus points for naming your data providers, a black-and-white marker (no pun intended) for the ebb and flow of popularity.

The bag holding the thoughts together

There's a phantom of a cohesive thread holding these three things together.

Obviously, they all relate to maturation of the football data/technology/analysis industry. The VR use in TV and the club job ads are cases of those spaces developing somewhat 'naturally' though (to an extent), while the FIFA/UEFA input is a case of 'stewards of the game' observing the landscape and saying 'these are places where we need to step in and support the progression ourselves'. In the case of the basic refcam strand of FIFA's Innovation Programme, it is also 'supported by the English FA and [IFAB]', like some kind of 'stewards of the game' Avengers. (Does that make Gianni Infantino Nick Fury, or is it Arsène Wenger, or someone else?)

If nothing else, this is a sign of a sport in the 21st century, embracing and innovating technology from multiple angles: not just from science-y wonks, not just from corporate efficiency-driven investors, not just from That One Club That Read Moneyball.

It's interesting.

If there's anything Quite Interesting that you've seen recently that hasn't been mentioned here, get in touch!

Spotify for 'successful line breaks'

Thu, 18 Sep 2025 00:00:00 GMT

There are a lot of disagreements about data, but the one thing that most companies seem to agree on is that you can never have too much of it.

You could say that we're living in an 'age of abundance' of football data - but only if you wanted to look horribly out-of-date in three years. There will undoubtedly be more and more datapoints added to our data packages as time goes by, which makes it all the more shocking when you see the amount of metrics that some companies currently advertise themselves having. Hundreds? Ha, smol babies: try the four figures.

Having so many to choose from is, in some ways, very nice. Metrics can be much more specific on a range of different criteria: phase of play, action type, area of pitch, relation to opposition structure. But abundance introduces its own problems: in this case, discoverability.

Think about music (or podcasts or TV). There are so many people creating so much of that stuff, that it can be hard to find things you like. (Or, if you're a creator, hard to put your work in front of an audience who might like it). Take that situation and abstract it out slightly, and you have a case of 'how do I find the things I want for my current circumstance'.

Spotify, YouTube, Netflix, and all their various competitors all do this with algorithms. But this isn't the only way, nor the original way, of doing this job. In the olden days, there were these things called radio stations and TV channels, which chose what to put in front of their audience. This is similar to the various bits of default software that companies - including the data providers - offer to their customers.

Now, you could listen to stuff that wasn't on the radio, but let's be honest, most people mainly just listened to what was on the radio. That might be because they're un-curious, it might be because the radio music is what they're familiar with, it might be because the radio station's choices were genuinely really good.

You could extend this metaphor even further, but I'll spare you that. Manual curation naturally limits what is put in front of an individual, relies on smart experts doing the curating, and has a tendency of keeping people in the status quo. So are algorithms the way? Is that how people will make sure that the right dozen metrics, out of a pool of thousands, get put in front of people? Is that the answer to the coming (or present) football data discovery problem?

There are, unsurprisingly, problems with this.

The first is that it isn't exactly an established best practice for data providers to create good-quality metadata about their statistics offerings. The second is that building an algorithm on usage data relies on a large amount of users. And giving someone a 'bad' recommendation in a professional football environment matters a lot more than if Netflix pushes a dud TV show in front of you.

You also get into a question about what the algorithm - or, more likely, algorithms plural - are meant to do. Think of music streaming, which does this more and better than Netflix or YouTube. Music streamers are well aware that people have different moods and situations, in which they'll want to listen to different things. But do you want to listen to 'Rock' or do you want to listen to 'Energy' (or 'dirty rock happy indie thursday evening')?

So, if you're logging onto your SoccerPlatform2027, will you want to look at 'Attack' or 'Final Phase Chance Creation' or 'sexy vibe xT aura farming'? And should the algorithm tilt towards 'things the team did well this weekend', 'things they could improve', or 'things most correlated to success'? (and how much would you trust the company producing the algorithm to accurately surface each of those things).

At a guess, an algo (or several algi) would be useful but would need to be smartly limited in scope and be an addition to curated features, rather than replacements. That's all for now, otherwise I'll tangent onto talking about the difficulties in dealing with the many slightly different terminologies and conceptual approaches that football practitioners have, like 17th-century variants of Protestantism. (Honestly, you give an inch on translating juego de posición into the common tongue...)

Play smarter, not better?

Sat, 30 Aug 2025 00:00:00 GMT

Long-time, no blog. But it's time to continue the old theme, 'the best football analytics publication is the Harvard Business Review'.

Watching some games lately, a thought occurred to me: when a manager is trying to implement a new system, is it better for them to play players who...

1) understand the requirements of the system in-the-moment, but who can't execute them fully, or...

2) can execute the requirements of the system, but who don't fully understand the requirements in-the-moment

Like many things, this isn't a static question. If you're very early in the journey of system implementation, number 1 makes more sense. That's because footballers - big shock - are not just individuals, but are members of a team. And so (in theory) it could help other first-team players to learn the system quicker if they're alongside someone who yells the correct things at them, even if the yell-er in question can't pass/run/tactical foul properly.

However, this raises a familiar question to an analytics blog/newsletter: how do you measure that?

'But Mark', you might be saying, 'why would one need to measure this? We can already identify players who've played in similar systems, and using that transfer strategy as the basis for a team's success never fails'.

Even taking that at face value, there's a chance that you find yourself in a scenario where either the system is fairly unique or the pool of available 'system-ready' players is just... uninspiring. In which case, someone learning the system role will be required.

And then, football is not just about transfers. Injury cover, introducing youth team players, manager hiring, the way that tactical considerations shift over time - this is always relevant to some degree.

The measurement would not exactly be easy (and, really, it's something that players and coaches are going to be better at than data wonks for a while). For one, differentiating between execution ability and execution choices is, even I will admit, possibly impossible. But why let that get in the way?

I suspect that the way to go would be to identify 'football-specific' situations rather than take a top-down modelling approach. For example: body orientation and distance from a marker are going to be important in various different roles. The circumstances from one pass reception to another will differ hugely, but it should be possible to model the likelihood of 'possession stickiness' in certain broadly similar scenarios, based mostly on these two factors (and accounting for the fact that in some circumstances the distance to a marker is out of the receiver's control).

Not all roles will need possession stickiness, though - some winger roles might call for the ability to roll instead, for example. A small-scale instance of this 'understanding vs execution' experiment, then, could be to model instances where a winger should roll inside vs prioritise possession, and combine that with the model of their execution of the initial reception. Maybe Player A makes the right choice more often than Player B, but doesn't have the passing chops when they move inside - maybe you decide that's fine because it's helping reinforce the positional choices of everyone else, or maybe you think Player B is more able to improve their decision-making than Player A is able to improve their passing.

Arguably, all this is nitpicking over players. The qualities of a player that make them adaptable to new systems and coaching points most likely shows up in their performance for their old team(s). There's usually a limit on one-trick ponies, and it usually gets found out pretty quickly in the cauldron of professional football development. But, to go back to the HBR well, the 'Peter principle' theory is that people get promoted until they reach the limit of their competency. That does mean that, sometimes, one-trick ponies get brought into a two-trick pony role.

Identifying whether a player is the former or the latter of those is important, but I guess to go back to the original question: if you end up with one-trick ponies as your options for a two-trick role, which trick should you prioritise?

Answering the hardest question in football

Tue, 05 Aug 2025 00:00:00 GMT

Ajax, Club Brugge, Hammarby IF, Forest Green Rovers, Toronto FC... if someone asked you to connect the dots here, would you know where to start?

Maybe it'd help if I added a few more: FC Midtjylland, Brentford, Brighton. Getting there? All of these are examples of why the hardest question in football isn't about players, isn't even about the pitch - it's about answering the damn (legitimate) query 'how do football clubs use data'.

Those eight names are all clubs who do stuff with data (or have done in the past) that is streets ahead of not just their immediate competition, but of significantly bigger clubs too. Yet there's very little that connects them on the surface. Sure, most are western European, but that's at least partly my awareness of public info*. Two share an owner, of course, and a third is also owned by a betting man, but none of the others fit that pattern.

*(In some cases that's press coverage, in some cases that's academic papers, in some cases it's job ads)

After a bit of thought, I think I've got a better answer than 'it depends' to the seemingly simple question 'how do football clubs use data?'.

Tough on Moneyball, tough on the causes of Moneyball

There are (at least) five factors in how a club uses data.

Two are covered by Moneyball: money and belief. The ol' tale of the Oakland As may have begun with a lack of dough, but it ends with the Boston Red Sox calling up Billy Beane and Paul DePodesta joining the LA Dodgers.

After those, there's aptitude. Converts to the Good Word of Expected Goals can go down very different paths depending on if they know what they're doing. Factor four is equally obvious: the length of time that a club's already been 'doing data' for.

And finally, business relationships. This can relate to money, but they can also be struck through the ambition to use data, or just by plain luck. Brentford, Midtjylland, and Brighton all had the connection of their owner's gambling company; Hammarby had a local academic; Toronto's ownership also owns the Toronto Raptors NBA team, so will have been courtside to that sport's analytics boom.

Ownership is also a factor for Toulouse FC, whose RedBird Capital majority owners had also, by the time they bought into the French club, invested in data consultancy Zelus Analytics. And you can't talk about French multi-club orgs without mentioning (almost literally) one-time employers of Savinho, Troyes AC, part of City Football Group, and the wider landscape of MCOs. Not every MCO has the same level of data set-up, and not every club within an MCO will have access to the same knowledge and tech, but it's not not a factor.

Cutting down the nuance

Five factors, though, is too many to talk about comfortably. At a push, they can be boiled down into two: resources and ideas. Money, time, and business relationships are all resources, and ideas covers ambition and aptitude (aiming for the stars when you can't build a rocket is still a bad idea even if it's admirable).

You can put all that into a grid, like this:

But even this can be simplified further.

There's an extent to which clubs with great ideas on low budgets will be landing on similar things as the clubs with more resources and less ambition.

With this in mind, even if clubs don't have much budget they can increase their resource in other ways, through strategic relationships or just by starting early (although, granted, that last one easier said than done than anything else). But even with a cap on resources, better ideas can punch up a weight class or two.

But how, then, are football clubs within these tiers using data?

The tiers

(This is the part of the newsletter that educated dissenters are obliged to get in touch about. It's (obviously) not going to be exhaustive, but it gives a feel of how data can be used)

Tier 5 (low resource, low ideas)

In many professional leagues, these clubs will still be using data of some form. But it'd be in ways like checking the goal tallies of prospective attacking transfers or clean sheets of defensive transfers; and probably collecting running and gym data but not doing much with it.

Tier 4 (Medium resource & low ideas/low resource & medium ideas)

What does 'medium resources' mean, in the grand scheme of global football? Eyeballing a couple of sources (clubelo, UEFA coefficients) just to be able to point to a ballpark, let's say teams like those in the upper reaches of Scottish, Croatian, Swedish, Czech men's top-flights. That does mean that 'low' resources has a very long tail, but that's sort of the way the industry cookie crumbles.

This tier probably has processes for feedback around scouting and match analysis, which will involve data in some form, but the processes might not be settled and with very limited data available. They'll probably have access to a data company's reports or platforms, but the ceiling on their ideas or resource stops them getting something better or more bespoke than that.

The low resource (medium ideas) clubs in this tier are probably upwardly mobile as long as money lasts, while the medium resource (low ideas) clubs are probably gonna be falling off if they see a drop in revenue.

Tier 3 (High resource & low ideas/medium-medium/low resource & high ideas)

By nature, this is the broadest tier, and therefore the most varied. All the clubs in this tier will have standard processes and reports, with a defined sense of how these should look. That sense will likely be driven either by industry knowledge absorbed by osmosis or (at the lower-resource clubs) a strong, 'first principles' set of ideas about what these should cover.

In this tier, the high-resource clubs can brute-force their way through a lot of things. For a long time, managers used data even if they professed not to: video analysts would tag and clip matches according to key moments or tactical concepts that the coach believed were important. Throwing a ton of video analysts at a problem can - for some things - get you the type of analysis that a more sophisticated data set-up could get you far quicker.

On top of that, through either money or relationships the higher-resource side will almost certainly have a bunch of 'normal' data on-hand. It may well be looked at, but that doesn't mean good analysis is done with it.

Clubs at this tier probably have a knowledge bank too, although they'll call it different things. A database of scout reports would be a form of knowledge bank, but low-resource clubs might enter this tier by building their bank with public research that they synthesis into internal documents. Even if you can't afford to do research in-house, that doesn't mean you can't benefit from research findings.

Tier 4 (High resource & medium ideas/medium resource & high ideas)

Now we're reaching the tasty part of the grid.

Clubs in this tier - even though they may look very different on the outside - will all have some kind of expertise in analytics. Reports will almost certainly have components which implicitly or explicitly touch on things like the coach's tactical game model, player advantage battles, the chain through which chances get created.

Whether through availability (high-resource) or smart prioritisation (medium-resource), clubs at this tier will be using tracking data for something. There are a bunch of competitions which offer tracking data to its competitors, which can feasibly be used for tactical analysis; while TV broadcast-based tracking data can be used for scouting.

Clubs at this tier will also probably be doing bits of their own research too, which links to the tracking data. It seems to be a not uncommon use-case of broadcast tracking data to use it for running data while scouting. It's always useful to know how statistics of a player might be altered by moving leagues or teams (the classic: do Eredivisie shot monsters keep taking lots of shots elsewhere), and this applies to physical data too. You can't just rely on Skillcorner blogs for your knowledge here, interesting as they are.

Squad-planning and player development will also have data involvement at this tier too. Clubs will be able to benchmark players against particular goals and against their peers. This can sometimes be done without a huge amount of resource (it can often come directly from a data provider's platform nowadays), but for player dev in particular you'd ideally have a long history of data.

Oh, and this tier will have a good data engineering set-up, which underpins everything else.

Tier 5 (the rich nerds)

idk, someone bug the [redacted], [redacted], and Liverpool offices.

But seriously.

This is the one tier that can feasibly be doing work with body pose/skeletal data. That doesn't mean that they definitely are, but if they're not I'd like a word with them - I mean, come on, why would you not be doing the cool stuff?

Clubs in this tier will probably have a substantial internal base of knowledge, drawn from:

internal research
public research
general insight of their club personnel (who, at this tier, are plentiful and/or experienced)

A bunch of that internal research will be on things that underpin the regular decisions football clubs make. (e.g. strength of leagues relative to cost of players; tactical trends; fatigue patterns in different in-game circumstances; possession value models).

The processes that the team has (including but not limited to their scout reports, match reports, transfer decision-making) will probably have elements that are backed up by real research. That could be the relationship between specific metrics and specific goals, or about how people absorb and interpret information.

There will likely be something that clubs in this tier do really, really, really well too, the fruits of a previous project. But maybe only a couple of things - being in this tier certainly doesn't mean that a club will be world-leading across the board. The thing with research is that outcomes are unknown, so it's a bit of a law of averages that you need to do a few projects to get just one that really shifts the needle.

But although we might expect there to be a heavy focus on tactics, that research could easily be done on other things. It feels like a safe bet to say that clubs in this tier will have some fancy recommendations about load management, given that 1) they'll be paying players a lot 2) it's easier to measure than tactics (monitoring a single player vs monitoring the interactions of many players). (It's also probably an easier analytics sell to managers who are already ceding control over transfers).

The reason for putting this together is that data usage is not like player quality: the ties between revenue and output are far looser. It's difficult, then, to say 'clubs in X league do this' or 'clubs with Y wealth do that'. It's still not particularly easy to say 'clubs in tier 3...', but it's truer to what the current landscape of football is like.

TL;DR - It depends.

Vocabulary for chaos

Thu, 17 Jul 2025 00:00:00 GMT

Every now and then, an American discovers a diagram of cricket fielding positions and has four minutes of joy. ("Fine leg? Silly point? Wicket keeper??"). I kind of think we need the same for uncontrolled football phases of play.

Football is one of the few major sports where the extremities that make humans unique - the dextrous, dextrous hands - aren't the thing in control. The sport's soul is an unpredictable bounce. So maybe there should be a clearer way of describing that.

Of course, there's a risk that trying to put names on these flux-y states deceives people even further into thinking there's order to them. But maybe it's worth trying.

A large part of why this comes to mind is a project I recently blew full-time on, where chaos was a significant part of it. There's a screen-recorded journey through it below, the main takeaways being that open-source packages like kloppy and databallpy are cool and that there are smarter ways of doing a 'low effort' project when your main focus is Wimbledon.

Some brief admin, in that I've moved newsletter service to Ghost, which shouldn't affect y'all but does mean that the main https://www.getgoalsideanalytics.com/ site is the place to go to re-read, subscribe, etc.

And, lastly, you can buy a PDF of the Get Goalside writing from the 2024/25 season - which helps cover web hosting costs and things like that - here:

Happy reading. Now time to work out what the 'silly mid-off' of football is...

Some reading, some vocabulary

Sat, 31 May 2025 00:00:00 GMT

If you’re following the news of the LLM world, you’ll be aware of an increasing number of protocols. The investors might be drawn in by the promise of god-like AGI, but the techies get drawn in by things that actually improve their lives.

A matter of days ago, a similar thing happened in the football data world. The Common Data Format could, like language model-powered agents, usher in a whole new world. It might not. But I’m gonna keep bringing things like this to peoples’ attention because, for one thing, it’s a recurrent theme of Get Goalside, and for the other, a common data format might make switching data provider easier, and switching data provider easier tests the established players. And either the giants are slain or they create better products - win, win.

However, the vocabulary of football data is still evolving and, more importantly, expanding. When was the last time you talked about entry speed and turn angle when talking about a player changing direction? Well, maybe you should (or, at least, read this paper on the subject). And I think there’s a lot more juice to squeeze out of the ball-carrying lemon too, in part because of the range in this piece from Lily Wood-Blake at Hudl.

Now, because I am approaching ‘old’ (30), I remember a decade ago when older football pundits had a little moan about the hype around ‘pressing’, because of course ‘closing down’ was par for the course in their days. They were, in some senses, correct - the German-Austrian pressing revolution was partly just a case of trends circling back around. In terms of terminology, the new lingo was a change of old vocab rather than extra vocab.

But I think as far as quantifying football in the present day goes, the vocabulary is going through a fairly rapid expansion. It’s only going to continue, and it’s gonna be here to stay. Which may test attempts like the Common Data Format, but is fantastic news for the writers.

On that note, you can still get a collected works of this season’s Get Goalside in a lovely PDF booklet. It’s a nice way of covering newsletter hosting costs - for £5 or your currency’s equivalent - because there are enough of you lovely subscribers to create that need.

Get Goalside 2024-2025

A collection from the Get Goalside football analytics blog over the 2024/25 season.

Happy reading.

Fishing for principles

Sat, 10 May 2025 00:00:00 GMT

In writing Get Goalside, I don’t usually like being personal - but today, today I’m going to open up about my least favourite fact in the entire world.

The most generous thing I can say about the ‘fact’ “there’s no such thing as a fish” - made famous by the British TV show QI and its spin-off podcast - is that it isn’t actively harmful. It’s not going to run over your child or poison your dog. It won’t force you to watch Mikel Arteta’s Arsenal. But, as much as the categorisation of aquatic life into a single group is zoologically problematic, by heck it’s tiresome to go ‘that thing which obviously exists, doesn’t’.

Which means that it’s with great sheepishness that I say: ‘defensive lines’, they do not exist.

Sorry, I can’t bear letting the Bit last this long - defensive lines clearly exist, it’s just that they’ll be more meaningfully understood if we poke at the definition a bit. [exhales] That feels better.

Poking at definitions is something I usually do with event data. For example, a few days ago I was looking through this year’s Get Goalside posts (for reasons), and re-read this line from the end of ‘Just run some more’:

“There’s another advantage that running data has. No-one has to argue about what a ‘duel’ is.”

Running data is, in a sense, elegant. It’s pure and simple - the movement of a person through space. Sure, we divide it up at increments of speed, but ultimately it’s a case of ‘fast’, ‘not as fast’, ‘double-speed fast’, ‘Micky van de Ven fast’. Lovely.

The way that event data works is obviously different, more similar to trying to fit art into genre boxes, but if you wanted to then you could approach it in a similar way to running. The actions which event data collects are usually a case of a player enacting a change in motion of the ball. Shots are fast (usually) strikes of the ball towards the goal; passes are somewhat fast strikes of the ball towards a teammate; ball-carrying is just a player making small strikes of the ball. (Duels are difficult because you need to get granular about limbs, but they’re still just a matter of opposing players, one of whom is (usually) the ball-carrier, coming into proximity).

And defensive lines, then?

If you’ve ever heard a manager pooh-pooh the idea of formations (‘telephone numbers’), you’ll know that tactics are in the eye of the beholder. [A sidenote: I would pray that everybody keeps in mind the love they hold for the ‘fluidity of football’ next time they ask for stats about ‘goals/shots from set-pieces’. Nobody knows when a set-piece becomes open-play. If God had wanted there to be clear delineation between phases, he’d have invented the NFL.]

The best paper that I’m aware of on this area of football, identifying formations and phases from data, is this one: ‘Putting team formations in association football into context’. And even here, I think that parts of it - like the notion of ‘fish’ - are convenient fictions.

“Figure 1 provides an example of the phases of play classification scheme developed by German Bundesliga analysts. In this scheme, open-play during a match revolves between periods of offense, transition to defense, defense, and transition to offense[…]”

With the utmost respect to all football coaches, if it was a data analyst proposing this rather than video analysts they’d be glared out of the room. The four-phase (plus set-pieces) split may be how coaches want football to be played, but it has to be coached very strongly before it even starts to look like it falls into this pattern. Y’know who I’d call as a witness if going to trial on this? Any coach who’s introduced ‘artificial transitions’ to their in-possession play.

All of these tactical notions are also too strongly tied to coaching practices to be usefully used as generic concepts. A classification of football shouldn’t break down just because Fernando Diniz walks in the room. Running data and event data classifications will still work if you’re looking at a match from 2025 or 1955; the tactical aspects should as well.

So how can you approach this?

If running is just the movement of a player and events are the interaction of player and ball movement, tactical features are- well, first, something that they’re not. They’re not the movement of groups or teams or units, because that just creates the question of defining the unit. Instead, in all concepts I’ve been able to think of, they could be said to be the coordination of movement of players.

Happily, this allows me to say ‘yes, there is such a thing as a fish’. A defensive line can still exist in this ‘coordination of movement’ way of defining things, because the main feature of a defensive line is that it’s coordinated in one dimension. A ‘press’ is a particular set of coordinated movements*. ‘Established possession’ is (simplifying slightly) the moment at which the movements of teams have become sufficiently coordinated. ‘Rest defence’ is a coordination of positioning in deep central area when a team has advanced possession.

*different types of press, or a disjointed press, could be flagged by identifying coordinations between a small number of out-of-possession players, whose collective movements are un-coordinated with the rest of their team.

If this was a grant application instead of a blog, I’d need to argue for the practical implications of this insight. It’s fortunate that it’s not (I’d have never slipped the Arsenal joke past the editors). But the applications have already started: the Opta Forum talk by Guillaume Hacques which I mentioned last post, focused on asymmetries in defensive blocks.

(It’s probably not a coincidence that the focus on asymmetries and coordinated movement was prompted by the submission question posed by Monaco’s current Head of Sporting Insights Vignesh Jayanth (in 2023, at Stade Rennais)).

I suspect that ‘phases of play’ can come down to gradations of stability (which in itself can be related to coordination of player movement) - which will have at least one extra gradation than ‘transition’ and ‘not transition’, the tactical equivalent of dividing player movement up into ‘running’ and ‘not running’. I suspect assignation of player marking stems from the same principles (and now I think about it, think I’ve seen work doing that before(?)). And therefore so will dis-marking.

Fish exist. My life is very slightly incrementally enriched by learning that categorising animals is hard. I hope this post has been minimally tiresome.

If anyone knows of research on line-breaking passes, could you send it my way? I was reading a piece which said that there wasn’t much research on it, and searching Google Scholar that appears to be correct, but surely such a strongly-pushed metric would have some research examining it….

Looking for stability

Thu, 24 Apr 2025 00:00:00 GMT

In true newsletter style, this will be a collection of thoughts. Mercifully, it’ll be short.

They’re both in the grand Get Goalside tradition of ‘the nature of football'. The first is a fourth-dimensional take on Rafael Benitez’s football tactical metaphor of the ‘short blanket’. The point is that, like a short blanket will always leave part of you cold, a defensive team will always leave part of the pitch exposed.

Football is a sport where it’s very difficult to score a goal, played for a length of time which is very long. What an absurdity to play a game for an hour and a half and, very possibly, not score a goal at all. This is the most popular sport in the world.

And so, there isn’t, and possibly never will be, a tactical blanket that a team can throw over a football match and cover it all. (Though of course, you can cover the match better or worse). The point is that in any tactical consideration, time is a factor. Not a revelation, but also not something that is talked about as much as all other tactical aspects.

Secondly, on a team level, is the idea of ‘stability’ in a team. This is a thought that stems from the perennial problem, ‘why do some teams game the progressive pass metrics?’. (Not ‘why do they choose to game them’, but ‘why do their stats look as if someone were gaming them’). At time of writing, three Premier League teams have averaged between eleven and twelve passes into the penalty area per game (FBRef) - Arsenal, Liverpool, Manchester City. By FBRef’s (Opta’s) numbers, Liverpool have created around ten expected goals more than City, and almost 20 more than Arsenal.

So, stability.

Teams attack and they defend. The ideal, for an attacking team, would be of being in a stable state and attacking an unstable defence. The issue which we see so often is that a team seeking stability will often allow the defensive team to find stability. Equally, a team that tries to attack while a defence is unstable will often be in an unstable state themselves. There are teams which enter the box when the opposition defence is essentially stable, which is usually still dangerous, but the ‘value gap’ in these ‘entry’ stats gets wider when you take a step back to the final third, or to ‘progressive’ metrics.

The pass stats are worthy attempts at proxying danger, but there’s a risk that teams trip themselves up by chasing them.

But what is stability? A couple of past Opta Pro Forum presentations have touched on this - Mladen Sormaz and Dan Nichols investigating shape-disrupting runs (2020) and Guillaume Hacques looking at destabilising a set defence’s symmetry (2023). [the biggest race in football analytics is between ‘solving football’ and ‘establishing a central, easily-findable location for the history of the Opta Pro Forum’].

My current sticking point is finding a ‘stability’ definition which works for counter-attacks. If a team has been caught out by a ball over the top and is running back to its goal, chasing an opposition attacker, in perfect synchronisation, it’s not exactly ‘stable’. You could just say ‘look at pitch control’, but I suspect that, when facing set defences, instability often precedes the changes that pitch control would be detecting. It could be as simple as ‘stable systems don’t move quickly’.

A side effect of dwelling on ‘stability’ is what it would do to the metrics that come out of it. You’d probably end up with something like ‘how many times did a team create instability from a stable defence’… but that would leave open the possibility that a team racks up numbers by (unintentionally) allowing the opposition defence to get stable lots of times. Aiming for a ‘high percentage of stable defences rendered instable’ is not even necessarily a good goal.

The one downside of this focus on stability? It gives all the physics PhDs in football even more to do. Football isn’t played on a Newton meter.

Building capacity and breaking lines

Thu, 03 Apr 2025 00:00:00 GMT

Over the years, I’ve met a lot of people working in football from around the globe. The nature of this industry is mobile, international, insecure, and ever-changing - my thoughts have been with those of you affected by the many life-altering circumstances that have been taking place around the world recently.

Every now and then, an idea just sticks with you. Most often it’s not so much the idea itself, but the way the idea is phrased. After all, people tend to echo the ideas that others have had before them, but it can sometimes be hard for a thought to take root until the right seed is sown.

‘Building capacity’ was one of these for me. It’s a phrase of long-time friend-of-the-newsletter Tiotal Football (of Absolute Unit and the Post-Script podcast). The phrase referred to the way that teams act in possession, a subtle difference to usual terms like ‘ball progression’ or ‘build-up’. We’ll get to the difference in a second.

First, the thought had to flower. It had latched, roots don’t make a fruit. (even if you can get some useful tubers). At some point, Tiotal also made a comment about cut-backs being so effective as goal-creation chances partly because they give the defending team a particular set of bad choices. It’s like the Rafael Benitez ‘short blanket’ metaphor for tactics in miniature.

‘Building capacity’ is shuffling things towards forcing these bad choices on your opponents. If cut-backs are the ‘fork’ - to play to the readerbase and use a bit of chess lingo - then building capacity is the move before the move to create the fork.

The idea of ‘build-up’, ‘ball progression’, and particularly ‘building through the thirds’ had always jarred slightly. It’s true that moving the ball closer to goal helps create chances, but how often do you see a team progress the ball through the thirds, establish possession high up-field and then… just be faced with a robust, structured defence. Ball progression is often relatively easy until you hit the point that the out-of-possession team is actually bothered about defending. Sometimes a mid block is more of a negotiation tactic than anything.

Building capacity will often mean that the ball moves forwards, but it ball progression doesn’t mean that the capacity has been built.

Neither, necessarily, does breaking a line.

I’ll grant you, breaking a line is better than breaking most things. It’s definitely way, way better to be a line-breaker than a heart-breaker, leg-breaker, strike-breaker. However, a pass that breaks a line is not necessarily ‘building capacity’. You could put together incredible highlight reels of players breaking the first line of an opposition block, which the opponent immediately recovers from, or breaking a midfield line to a teammate facing away from goal with no passing options.

A sidenote: a focus on breaking lines with passes de-emphasises the ability of ball-carrying to shake up an opponent. We need more centre-backs maraudering and gallivanting.

The problem with ‘building capacity’ is that it’s difficult to count. But culturally we seem to be swinging back towards valuing expertise instead of just wanting easy-to-grasp simplism, so maybe that’s not a deal-breaker.

Tracking data might come sweeping in here claiming to save the day. There are already companies like [no, I’m not giving free advertising] who have models calculating both the likelihood of pass completion and the pitch value of the potential receiver. You could imagine a step further than this, some way of factoring in immediate options that the receiver would have if they received a pass.

Could thinking about ‘building capacity’ (or ‘building bad options for the opponent’) help make modelling choices easier? Let’s take backwards passes (the anti-line-breaker). Analytics people have often argued that backwards passes aren’t necessarily ‘bad’, because they could be the pass which opens up a progressive pass. This fits in the framework of ‘building capacity’. But they can also be valuable for removing risk of a turnover, which is a ‘bad option’ that the opponent is trying to force on the in-possession team. Although the quality of options might still be assessed in a goals-derived metric, you’re dealing with a much larger bunch of ‘rewards’ than if you’re looking at goal-creation directly.

Sometimes the better option might not be taking a pawn, not even waiting for an opportunity to take the queen, but starting with a fork on the bishop and rook.

The path to now

Sun, 23 Mar 2025 00:00:00 GMT

In the beginning there was darkness.

And then… Then there was the light of a headtorch, illuminating the notepad of Royal Air Force accountant, Wing Commander Charles Reep. It is 1950. Like many of us, Reep is annoyed at a football team. Unlike many of us, he’s so annoyed that he’s marking down his own data.

Within a few years of beginning his data notation at Swindon Town, of all teams, Reep will be a minor celebrity. He will be credited with helping the success of Wolverhampton Wanderers, the most successful English team of that decade alongside only Manchester United. They won three league titles, two FA Cups, and their midweek friendlies against Continental rivals helped inspire the European Cup.

This is mid-century England, and it’s the first wave of football analytics.

And now, seventy-five years later, we have team cohesion of pressing chains, among other shiny things that data company Skillcorner is promising in their latest launch of metrics. This is quarter-century England - at least where this newsletter is being written - and it’s the latest wave of football analytics.

How did we get here?

As much as it would be nice and neat to begin football’s data history with Charles Reep, we must go back into older and fuzzier times. Newspapers from 1920s and 1930s Hungary show visualisations that would look modern in football coverage today.

A ‘match chart’ from 1922, from ‘Three Sportviz Inventions By a Hungarian Newspaper’, Nightingale

According to Attila Bátorfy, the writer of the linked article, the Hungarian newspaper Nemzeti Sport mentioned its creations being copied by outlets in Italy and Sweden. We know disappointingly little about this spread of football datavis. Historical terms are always relative to geography and the present day, and so for the moment Charles Reep remains our most secure starting point for ‘football analytics history’. But it seems very likely that as long as there has been football, there have been dorks making data out of it. Often it hasn’t been imagination that has held things back, but finance or technology.

Those charts (which, to reiterate, were published around the time Hungarian icon Ferenc Puskás was born) focused on themes we’d find familiar. Shot maps, for the match’s chances, and a momentum chart for which team was on top. These were also themes that Reep homed in on in his data. In 1997, a paper was published in the journal The Statistician using Reep’s data. The statistician Richard Pollard - who by that point had worked with Reep for more than a decade - put together what would nowadays be recognised as expected goals and possession value models. The match’s chances, and which team was on top.

Reep, though, was motivated specifically by what helped teams to win. This was a difference from the match charts of earlier newspapers. It also seems like he was motivated by a particular tactical viewpoint, a familiar-sounding one: that the modern style of short passing was overhyped.

From our modern perspective, this seems flawed to say the least. And Reep’s analysis of the data he collected, in articles that he wrote in newspapers and magazines, often seemed cherry-picked. However, some context about 1950s football is useful here.

Imagine the dying stages of a modern football match, when both teams can still get a result but everyone is very tired. The game becomes spread, formations become less clear, as players get caught between wave after counter-wave of attack, unable to get back ‘into position’. This - based off a bit of Footballia-watching - is what 1950s football was like all game.

Teams were simply less compact, which meant that passing was much more forwards, as it often can be in the sunset period of tense, modern matches. Teams on the ball weren’t facing a defensive block so much as a defensive lattice. In this context, sideways passing instead of forward movement would allow a defensive structure to form, something which our modern experiences simply take for granted.

Technical ability was also lower. If Reep had been around for the tactical evolution of the 2010s, he’d have probably liked the high counterpressing approaches coming out of Germany. If your 1950s-calibre players were going to be prone to miscontrols, why make them in midfield and not, after a longer ball, in the final third?

Although he was probably correct that there was efficiency to be squeezed, Reep’s public analysis was contested in newspapers at the time, and his professional involvement in the ‘50s and ‘60s didn’t lead to an ‘analytical revolution’. Successful, popular English teams still played ‘nice football’. Yet Reep’s data collection being affected by his opinions about football was not to be a one-off.

All data collection implicitly has an opinion. Way back to Nemzeti Sport, the data that was created was the overlap of what ‘mattered’ and what could feasibly be collected. Part of Reep’s skill was developing a shorthand system that could collect much more information than you would expect to be possible. (It may have helped that he was an accountant who joined the British Royal Air Force’s administration - numbers and regimented systems was his day job).

Those who took, and possibly bastardised, Reep’s research were inspired by contemporary trends themselves. Shortly after the 1966 World Cup triumph, results turned dismal for the England men’s team. They failed to qualify for all four major tournaments between 1972 and 1978, and had disappointing results at the following two that they actually qualified for. After the disappointing 1982 World Cup, an article in the Daily Express pointed to the beliefs of FA assistant director of coaching, Charles Hughes, as the way forward:

“Yet five years ago, and again before these finals, the FA had at their disposal information and analysis which could transform the future. […] [T]he logic of the Reep-Hughes analysis and statistics over 30 years, rejected by [England manager, Ron] Greenwood, is undeniable. Every four years we attempt to explain failure, we have a thousand arbitrary opinions. But Hughes has the facts.”

(Although the article says that Hughes learnt from Reep’s work, it names Reep as a ‘retired naval commander’, the wrong branch of military. It seems likely that the journalist knew Hughes’s opinions far better than Reep’s).

These facts: few goals come from long passing moves, lots of goals come from set-pieces or final third turnovers. Hughes and another figure the article mentions, Graham Taylor, got their chance a decade later. After a memorable fourth-place under Bobby Robson at Italia ‘90, Taylor got his chance; unfortunately, his England side won zero games at Euro ‘92, and failed to qualify for the 1994 World Cup. Few goals came from long passing moves; fewer came from a long-ball England team.

Having traveled the historical timeline all this way, it’s only a short jump to the founding of two industry-defining modern data companies. As has been the case with the characters we’ve seen so far, operations were influenced by circumstance.

Opta and Prozone (like Reep, both English) came into being around 1996-1997. Fresh off the back of a memorable semi-final for Terry Venables’ England at the ‘96 Euros, Opta would be collecting data for the Premier League. But they needed to fund their enterprise. A hungry media with pages and airtime to fill helped immensely. Prozone, meanwhile, found business viability through video analysis and player running load data.

Technology helped each company create a fully systematic data collection operation, where Reep had initially been limited by needing to not only be at the games himself but to process all of his shorthand into data by himself too. Yet by the mid-2010s, the technology was not enough. The context of these companies’ origins and the context of contemporary football was beginning to frustrate people.

The wave of Pep Guardiola imitators had made it clear that incisive passes needed to be identifed, while the pressing style emerging in Germany made the conventional collection of tackles and interceptions seem… small-fry.

A German company called Impect sought to address the first, with an emphasis on how many players passes ‘bypassed’, while the second was taken on by England-based (again) company StatsBomb and their ‘pressure events’.

Prozone, meanwhile, had apparently grown used to football clubs being uninterested in the details of their data. They “had been the gatekeepers of tracking data for many years and were loath to share it with anyone,” wrote Ian Graham, Liverpool’s Director of Research from 2012-2023, in How to Win the Premier League. “The data revolution in football could have happened years earlier if it wasn’t for Prozone’s protectionism.”

Little surprise, then, that Graham helped lobby for the Premier League to institute full access to tracking data in the early 2010s. Not only that, but a couple of years after league-wide sharing of data was finally agreed, in 2016, Liverpool Football Club welcomed a new tracking data company.

It was a relatively new company, and evidently neither side of the partnership wanted to wait around for innovation to evolve slowly. For once - unlike Charles Reep, Hughes and Taylor, Opta and Prozone - it was not English by birth. It was Parisian. And it was wanting to do something completely crazy.

Around the time that Opta and Prozone were being set up, the host for the 2002 World Cup was chosen. It would be the first-ever shared World Cup, between South Korea and Japan; the first-ever in Asia; and the first since the 1930s when a host, at the time of announcement, had never qualified for the competition before. (Japan would shortly afterwards qualify for the 1998 edition, their first appearance).

Whether caught up in this World Cup buzz or not, this period saw some interesting work by Japanese academics. Beginning in 1996, Tsuyoshi Taki and Jun-ichi Hasegawa published a series of research papers on the ‘dominant region’ of teams - areas of the pitch that one team or the other had hold over. Or, in modern terminology, ‘pitch control’.

The idea and calculations would still appear innovative twenty years later, but a problem at the time was getting the data itself. Work was published in academic journals for computer vision (a field where computers process images to detect objects, basically), but results were either small-scale or theoretical. Taki and Hasegawa even had to propose an in-stadium camera set-up that would work for their calculations. (“The telecasting image is not useful for analysis,” they wrote, “[…] because most scenes are intermittent and are focused on a player with the ball.”)

This in-stadium set-up was the kind of thing that Prozone was able to do, but still suffered - to a lesser degree - with what Taki and Hasegawa believed made TV footage unfeasible to work with. The issues of players being off-screen occurred on a smaller scale when they passed in front of each other, blocking one player or another from the view of the in-stadium cameras.

Technology and imagination, though, find a way. Among other things, TVs got bigger.

When Liverpool partnered with Parisian-based Skillcorner in the late 2010s (yes, we’re finally getting back to Skillcorner), the camera shots of football matches were both larger and crisper than they’d been in the late 1990s. Computer vision technology had also improved. And so too had understanding about how football teams moved.

Football coaches talk about ‘shape’ and ‘units’ a lot. Teams move in coordinated ways - in defence, particularly. This is hardly ground-breaking information, but it means that with enough processing power and prior data you can make smart guesses about where off-screen players are. And with a wider aspect-ratio of television, there tend to be fewer players off-screen, who are off-screen for a shorter time.

All this combination of technological advances allows companies to create more and/or better tracking data from relatively cheap, and very available, TV footage. For the first time in history, teams could scout with tracking data.

But what to scout? Would you simply use the ‘distance run’ figures, the thing that English coaches of the ‘00s seemed to like?

The fascinating tension in innovation is finding the balance point where imagination can be pushed just enough past technology. The latter often proves to be easier to improve than the former.

As it turns out, just like Impect and Statsbomb created unique selling points born out of frustration with the industry’s existing data, the world of ‘physical data’ stats has its sticking points too. At a similar point in time, the early 2010s, researchers were trying to find better measurement points when it came to player running too.

Separation between speed thresholds, ‘running’ vs ‘high-intensity runs’ vs ‘sprinting’, was fairly established, but still dissatisfying. In a particularly punchy 2015 research paper, subtitled ‘Shedding some light on the complexity’, we get this line: “Contemporary time–motion analysis of soccer still only offers a basic snapshot, and it is imperative that future research attempts to integrate multiple approaches to unravel the complexity of the game and its performance determinants.”

Three years later, one of the co-authors was involved in a paper whose title went even further. It was called “Are Current Physical Match Performance Metrics in Elite Soccer Fit for Purpose or Is the Adoption of an Integrated Approach Needed?”.

The floors of the 2010s data scene were littered with gauntlets being thrown.

No longer would it be acceptable to just focus on running at different speeds. Running - in fact, all movement on a football pitch - has a tactical component. That, ultimately, is the field in which Skillcorner’s ‘Game Intelligence’ data, particularly the recent out of possession launch, is pitching itself in.

Early event data captured on-ball events like tackles; later event data added direct pressures to the ball-carrier; the latest wave adds layers of variety about what effects that pressure has. There’s even a metric for how many times a defender turned their opponent backwards, something that Get Goalside imagined for an alternative football data universe several years ago.

It remains to be seen whether this fancy new data - like all data - is as reliable and insightful as it markets itself to be. The point at which Get Goalside tends to turn from enthusiasm to scepticism is - out of step with some media outlets - when someone starts flogging it. Thankfully, no-one is (yet) shilling a service based on the historical lineages of football data innovation.

To say that history repeats, or echoes, suggests that history itself is alive. Humanity repeats itself. People echo people that have come before them. It’s no surprise that association football is an echo of other ball sports from global history, because the bounce of a ball is a magical thing. It’s no surprise that the urge of a small group of people, to understand those bounces a little better, echoes too.

We know for sure that, a century ago, there were people tracking the key events of football matches. We know that, as England was still only five years removed from the Second World War, extensive and systematic data collection of match events was underway. We know that forerunners to modern tracking data applications are almost thirty years old.

Many things about the game have changed. For one, tactics no longer have the athletic limitations of the 1950s: research suggests that high-intensity running increased 30% in the Premier League between 2006 and 2013, and by a further 10% between 2014/15 and 2018/19. This increase in athleticism may well be an inspirational spur for some of the latest data innovations, or at least making the ground fertile for Skillcorner to put its roots in.

But many things are the same.

In late 1960, Reep wrote a pointed article about Tottenham Hotspur’s victory over Stan Cullis’ Wolves in the FA Cup, suggesting that their short passing would make their success short-lived (they ended up winning the double that season). In a response to Reep in the Hull Daily Mail local newspaper, their football correspondent closed his piece with the following:

“Football is a sophisticated game, and it will grow more so as the years go by. At the moment, possession and change of rhythm are the two principal weapons of the world’s best teams.

The long, forward pass still has its place in the armory of any successful team. But its very value lies in a team’s ability to use it as a surprise, not, as Wolves [Reep’s former associates] have done, as a monotonous, obvious weapon.”

Tactical variety, tactical debate, and arguments about data. This is mid-century England.

Notes

References/Further Reading

Attila Bátorfy, ‘Three Sportviz Inventions By a Hungarian Newspaper’, https://medium.com/nightingale/three-sportviz-inventions-by-a-hungarian-newspaper-b5c0df489d6c

On Charles Reep: https://www.bbc.co.uk/sport/football/61084931, https://www.bbc.co.uk/sport/football/61648608, https://www.wsc.co.uk/the-archive/grim-reep-/

Ian Graham, How To Win The Premier League

Rory Smith, Expected Goals

Richard Pollard & Charles Reep, ‘Measuring the effectiveness of playing styles’ (1997): https://www.researchgate.net/publication/227692321_Measuring_the_effectiveness_of_playing_strategies_at_soccer

Paul, Bradley, & Nassis, ‘Factors Affecting Match Running Performance of Elite Soccer Players: Shedding Some Light on the Complexity’ (2015): https://www.researchgate.net/publication/273071207_Factors_Affecting_Match_Running_Performance_of_Elite_Soccer_Players_Shedding_Some_Light_on_the_Complexity

Ade & Bradley, ‘Are Current Physical Match Performance Metrics in Elite Soccer Fit for Purpose or Is the Adoption of an Integrated Approach Needed?’ (2018): https://www.researchgate.net/publication/322277340_Are_Current_Physical_Match_Performance_Metrics_in_Elite_Soccer_Fit_for_Purpose_or_Is_the_Adoption_of_an_Integrated_Approach_Needed

Taki, Hasegawa, Fukumura, ‘Development of motion analysis system for quantitative evaluation of teamwork in soccer’ (1996): https://ieeexplore.ieee.org/document/560865

Taki, Hasegawa, ‘Dominant region: a basic feature for group motion analysis and its application to teamwork evaluation in soccer games’ (1998): https://www.spiedigitallibrary.org/conference-proceedings-of-spie/3641/1/Dominant-region--a-basic-feature-for-group-motion-analysis/10.1117/12.333797.short

Taki, Hasegawa, ‘Visualization of dominant region in team games and its application to teamwork analysis’ (2000): https://www.semanticscholar.org/paper/Visualization-of-dominant-region-in-team-games-and-Taki-Hasegawa/beff32a0a37a8d094a471067895cf420dd2e20de

Previous Get Goalside on the subject of Charles Reep and analytics history: https://www.getgoalsideanalytics.com/36315087-analytics-is-older-than-you-think/

Get Goalside, ‘What if we didn’t care about passes?’: https://www.getgoalsideanalytics.com/what-if-passes/

The wobbly chair of sports software

Mon, 10 Mar 2025 00:00:00 GMT

A few weeks ago, I made Sportscode. Hudl, watch out.

Well, actually, the point is that Hudl doesn’t need to watch out, because - alas - there is a difference between a working DIY video clipping bit of software and an actual product.

A regular theme of Get Goalside over the last year or so has been that football clubs probably shouldn’t all be tech companies. Now, part of that is because tech companies are increasingly susceptible to megalomania, but part of it is also practical advice.

The world of software engineers is currently split between those who love ‘generative AI’ coding tools and those who think they’re kinda trash. In the ‘love’ camp are people who like to move fast and break things. In the ‘trash’ camp are people who don’t like things being broken. (No but seriously; tools introducing bugs that are hidden because of a lack of oversight are basically the main complaint).

My version of Sportscode has a video player, custom-assignable hotkeys, an option to export to XML, and was created in about four hours (which, to be totally honest, feels slow). What it doesn’t have is any infrastructure that makes me trust it won’t crash if I try and code half a football match.

And yet, and yet, and yet.

The fact that someone with a passing grasp of React (the Javascript framework that it’s written in) can ‘manage’ a coding tool through the task of creating a fully working DIY app does change the landscape for football clubs.

I think the nature of these code generation* tools still means that you shouldn’t be using them to set up critical infrastructure, but developing, say, an quick, basic expansion to an existing set of regular reports for coaches; or a webapp to let players interact with their data? Seems more plausible.

The downside, the major downside, with any of this ‘small team employee-produced’ software is that, while they’re at the club, they’re the creator and the support desk and the bug-fixer. If and when they leave, you don’t care so much about the first of those roles, but you’ll probably still care about the other two. (That’s a large part of why you don’t want to build critical infrastructure this way too).

Another way that it changes the landscape is that what can be done with these tools will change depending on the user’s knowledge level. The tools probably become more powerful the more advanced the user is, for a similar reason as to why experienced line-managers with domain expertise can be useful. Part of managing is delegating, and part of managing is experience-sharing. ‘Ah, this sounds like a problem I’ve experienced/read about before, have you tried this?’ can be surprisingly powerful. People without coding experience - or without much of it, or without much football experience - can do the delegating with code generation tools, but they can’t do the experience-sharing.

Bundle that together, and the value of a fairly-experienced data-and-tech-person may have gone up by more than a fairly-entry-level one. And that’s quite interesting.

As a serial metaphor-er, I’ve been searching for the right comparison for the way that these code generation tools lead to changes. Basic as it may be, power tools of the ‘normal DIY’ kind are probably the best thing.

Imagine everyone had access to materials and power saws and belt sanders. Everyone on your street would, in theory, be able to build you tables and chairs for your business. You probably wouldn’t want them to, unless they could already build something sturdy. But there’d be uses for it. You might be able to redo some cupboards. They might be able to replace some faulty chairs in a pinch. Again: not critical infrastructure, things that have some common templates to follow.

What kind of DIY jobs do you consider doing yourself at home and what do you think is best to get a professional in for? Those same kinds of dividing lines seem applicable to the new age of DIY software in football clubs.

Hudl don’t need to watch out. Not yet, anyway.

*I am loathed to call everything ‘AI’. The major innovations have been around language models and their surrounding infrastructure. This doesn’t quite mesh with the term ‘generative AI’ which briefly had some time in the sun, because a lot of value comes from language interpretation as well. But the coding tools specifically tend to be useful because they produce code, therefore ‘code generation tools’. Let’s all remember that we don’t need to use the buzzwords that companies trying to show value to investors try to coin.

Who will win the processing war?

Thu, 27 Feb 2025 00:00:00 GMT

Football vs Transphobia: https://www.footballvhomophobia.com/fvt/

In the olden days, before even the Proper Football Men were denouncing spreadsheets, football had different rules. Depending on your viewpoint, it either had too many or none at all: yes, we’re talking pre-1863 codification.

Football was very popular even before it was The Beautiful Game™️ (sponsored by Kingdom Airlines), but everyone had their own way of doing things. This limited what you could do. It made it tough to play with other teams, and with so many interpretations some were bound to be worse than others. [a cheap crack at rugby could be made here]

And it’s with this in mind that we head to Japan, to quote a recent paper about analytics: “we propose […] a unified framework designed to streamline event annotation, data standardization, and various deep learning modeling for soccer analytics.” This is OpenSTARLab, coming from a group of Japanese researchers, the latest addition to a noble line of groups who want to make things easier.

As the paper references, they’re not the first. In 2019, a group of researchers from the Belgian university KU Leuven and Dutch company SciSports presented the Soccer Player Action Description Language, in a paper on their possession value model. The fellow Central European-born Kloppy Python package has a similar concept. The idea: while data providers stubbornly produce different types of event data, there are fundamental similarities that can be mapped into the same ‘language’.

An aside: there’s a tangent we could go down here, on a favourite Get Goalside topic of the internationalisation of analytics after a period of heavy English centrality. In fact, we’ll come back to it in a bit.

At one point in time, I even wondered whether FIFA would get in on the act. In 2021 they launched their FIFA Football Language. Arsène Wenger’s opening note uses the phrase “open-source”! Unlike the other frameworks, FIFA weren’t (yet?) trying to squeeze other data providers into their football language, but it formed the basis for their own data collection for FIFA tournaments.

Just like in the 1860s, the football community would quite like things to be simpler. Over the past few years, and for the next few years as well, a wave of football clubs will be going through their first major data provider switch. In many cases, this is because data providers have different offerings - more detailed and/or more shiny - but these differences don’t make up the majority of an event data spec.

In May last year, when writing about STILL-yet-to-be-replaced-as-Chelsea-front-of-shirt-sponsor Infinite Athlete, I asked a question. It was a question mainly about data engineering, but applies to the issues that OpenSTARLab is trying to solve too:

Which of the following is the more likely winner of the next three-to-five years?

- Interoperability between data providers becomes seamless on its own, allowing for integration of different data sources within a provider’s own product, or allowing for foolproof entity matching between any provider to use data in third-party applications like Tableau
- Organisations will turn to cloud providers like AWS for API integration and setting up data storage, either through some (semi-)automation (AI anyone??) or as an affordable managed service
- The above, but provided by domestic leagues or national FAs
- The scale of the task will have simply shrunk enough for clubs of all sizes to hire employees for the set-up and maintenance of data pipelines, and creation of internal tools
- None of the above, it’ll be as complex as always
- Something else

Maybe part of the ‘something else’ will be packages like kloppy and OpenSTARLab. If they become robust enough to use as standardisation systems while still offering the unique features of chosen data providers, a football club will only have to build a data system once*.

*(Well, as much as anyone only builds a data system once. The need to re-build and re-write would certainly reduce, and/or the need for them to develop their own abstractions).

This still doesn’t mean that it’s sensible for all football clubs to create their whole data engineering and software infrastructure themselves. It’s not. Smart (and sufficiently wealthy) football associations should be helping their clubs to establish a baseline standard, particularly if they’re a country in some sort of continental competition coefficient race. Like, I dunno, Belgium.

Unfortunately (from certain perspectives), this is unlikely to be a problem that really slaps people round the face for another few years. Not only are there many leagues where ‘using data’ means using software rather than the raw data itself, but therefore many leagues where it’s a competitive advantage to quietly build things yourself even if that means building inexpertly.

Now would, I suppose, be a good moment to declare the interest of working for a company that 1) deals with multiple event data providers 2) deals with interoperability between different data providers 3) has (smart, capable, witty) employees to pay and investors to create value for. Hopefully you, dear reader, trust that Get Goalside’s only biases are towards the entertainment of yourselves and, more importantly, of the writer.

(Also to the reduction in usage of the word ‘democratise’: democracy has enough on its plate without being dragged into sales pitches).

At some point, the boring parts of using data in football will get easier. What’s less clear is how.

Football got codified when a group of English elites decided to argue until they reached an agreement (which was then tweaked and then completely rewritten for clarity in the 1930s). Their agreement was picked up as the standard by the world. Who will end up writing the standard for football data collection and processing?

…

All of that was written about a week ago, and I didn’t get around to finishing it off. And then along came the Open-Source Avengers.

🌉🌁 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐛𝐫𝐢𝐝𝐠𝐞𝐬! We brought together contributors from different open source projects to discuss how we can align our work and improve interoperability in football analytics. Looking forward to the next steps. #FootballAnalytics #OpenSource #PySport
— PySport (@pysport.org) 2025-02-26T13:46:03.373Z

Wonderfully, not only does it circle straight back to the origin of this post, but goes right back to the internationalism point. Six nationalities are represented here, none of them English. England by no means has a monopoly on analytics history, but it did have a commercial tracking data provider in the 90s, it did supply the mainstream breakthrough for expected goals, it does have the club(s) commonly cited as the sport’s leaders in analytics.

But it’s very exciting seeing this cross-continent collaboration, and it’s very exciting seeing interesting analytics job ads from a widening set of nations. I would love to see more of them outside the historic nations plus America; I would love to see more of them in women’s football.

The ‘globalness’ of the global game has clearly pushed players and coaches to be better. It seems likely that it’ll help push the data side of things to be better too.

System stability

Mon, 03 Feb 2025 00:00:00 GMT

There’s an idea I’ve never been able to get my head around, called the ‘Lindy effect’, which (to crib from Wikipedia) “proposes the longer a period something has survived to exist or be used in the present, the longer its remaining life expectancy.” The idea is that if something has been around a long time, the end of it is usually a long way away.

And yet, all things (usually) come to an end.

So it’s strange to see a dominant global force, a force who’s been at the very top of the tree for a long time, floundering at the moment like Manchester City men’s team. Early diagnoses pointed to a single factor, Rodri’s injury, although there are, of course, surrounding factors. Maybe most system failures come down to one central factor, but the magnitude of their impact often seems to depend on how strong the rest of the system is.

Being in the tech side of football gives an interesting perspective on this: large (and/or experienced and conscientious) software companies use ‘chaos engineering’ to test their systems. (Microsoft Azure even has a product called Chaos Studio, which is how my colleagues refer to my presence in meetings). (That is a joke).

Of course, software is an easier type of system to test for failure than others. But you can see the value.

If nothing else, the chaos scenarios can force you to think through problems that you might unconsciously file in the box of ‘not going to happen’. What happens if your data provider suddenly changes the structure of their API; what happens if your Ballon d’Or nominated central midfielder ruptures their ACL.

These ‘what ifs’ might seem annoying, but many systems may seem stable not because of design but because of lack of testing. Rodri’s absence tests parts of the City system in ways, and with a frequency, that they simply hadn’t faced before.

This goes beyond the on-pitch action too. I’ve just finished reading Miguel Delaney’s recent book ‘States of Play: How sportswashing took over football’. In multiple cases, it seems like football’s governance structures were assumed to be reasonable until a moment in time where they were stretched further than it had been imagined they’d be stretched to. Regardless of your takes on those issues, it’s clear that systems (like around multi-club ownership) have been adapted to following new circumstances rather than adapting in anticipation of their potential arrival.

Obviously, the idea of scenario planning exists outside of software and football. But situations where chaos testing feels like it would’ve been helpful keep cropping up in the Premier League. Outside of Manchester City, there’s Manchester United who ended up with a marquee midfielder in Casemiro suddenly (though somewhat predictably) unable to cover the ground a central midfielder needs to cover. Then there’s Tottenham Hotspur, where an injury crisis has combined with a continued commitment to Ange Postecoglou’s pressing with fairly spectacular results.

Maybe this is where the biggest win for tracking data simulation could be. The idea, and practice, of using tracking data as a ‘prediction’ of player movements has been around a long time. Players and coaches don’t have a lot of time on the training ground. Perhaps, looking at their artfully constructed system on paper, the senate of Guardiola’s coaching staff thought that, yes, a combination of Bernardo Silva, Mateo Kovačić, and İlkay Gündoğan would able to cover for Rodri’s responsibilities. But maybe running that through tracking data simulations would have revealed vulnerabilities.

There are far more ‘chaos scenarios’ that you could run, too. Red cards springs to mind, but you could extend this to “what if our key player in X position is having a shocker” or “what if our press is horribly disjointed for no apparent reason”. “What if the front office sign 10 players and no-one knows each others’ movements”. ”What if our expensive new signing was actually a system player and also falls out with the manager”.

The idea of running simulations might seem fanciful to some (if so, why are you reading this, it’s the most Get Goalside idea going); but equally there’ll be some who’ll think that using it for chaos testing would be wasting it. This second group will be wondering whether you could use it to test different tactical set-ups before a match. After all, Formula One teams simulate different race strategies and conditions.

Personally, I would imagine that simulating football matches will be a lot less precise than simulating F1 races, and therefore the results are gonna be fuzzier. Do you trust the system to precisely determine the dynamics at play of using one first-team wide forward over the other? Maybe one day. But giving you the broad strokes of how the team might react to unlikely, but drastic, circumstances could be valuable without needing that precision.

Ultimately, as Manchester City show us, no system is perfect and no system’s robustness is set in stone. (Evolving circumstances, like recently-injured or ageing players, can be a slow boil of fragility). Each system will have trade-offs of goal maximisation and risk avoidance too. But systems shouldn’t necessarily be assumed to be stable just because they haven’t broken yet.

Possession adjusting: Part Two

Sat, 11 Jan 2025 00:00:00 GMT

When we’re young, we believe everything is possible. At a certain point we get old, tired, prone to saying “It’s complicated”.

The idea of adjusting a player’s defensive statistics to account for [whatever]… it’s complicated.

After last week’s investigation into ‘possession adjusting’, I’ve taken a look into ‘turnover adjusting’. A write-up with data tables is here, but the long and short is the same as the last one: defensive actions don’t seem linked to possession/turnover share.

Briefly, I should acknowledge an error I made in the code for the original study (now corrected and updated in the GitHub files for the project). It was a bad but ultimately insignificant one, as far as the results go. (Lesson: do basic data exploration steps throughout, particularly if you’re in an unfamiliar dev environment).

Here be numbers

A reminder of what these studies are: using Statsbomb’s open data for the 2015/16 ‘Big Five’ European men’s league seasons, running a simple correlation analysis of player defensive output (per 90 minutes, with different actions analysed separately rather than grouped) compared to a variable representing ‘possession’. Previously this was the team’s share of passes, this time it was the average number of possession sequences that happened per 90 minutes of game-time.

Some headline numbers from this latest bit of analysis, comparing to turnovers. Outside of the small group of Wing Backs, no combination of player position and defensive action had a correlation strength against the turnovers per 90 stronger than +-0.31. Only six out of a possible 45 combinations had a stronger correlation than +-0.2.

I was genuinely surprised by this. I thought there’d be a stronger link between how ‘turnover-y’ gametime was and how many defensive actions players made, given that defensive actions often are turnovers.

Even when performing the correlations for players league-by-league (which controlled a little for the fact that the Premier League had noticeably fewer turnovers than other leagues), there was little of any meaning to draw out.

So what now

Looking through the data in more detail shed some light on this.

La Liga’s correlations are really confusing, with negative correlations for Defensive Midfielders - meaning, a (very weak) relationship where a higher amount of turnovers was linked to a lower amount of defensive actions.

Why? Partly because Rayo Vallecano DMs had a really high amount of turnovers per 90 but weren’t hugely defensive active. Meanwhile players in the Spanish Big Two like Busquets, Mascherano, and Casemiro played in matches with far fewer turnovers but were much more defensively active.

And it was like that across the leagues and across positions. Shift to Premier League centre-backs (my favourite group) and at the low-turnover end you have the incredibly-active Nicolás Otamendí, alongside Otamendí aspirants like Ramiro Funes Mori and Laurent Koscielny , then at the high-turnover end you have the almost lethargic (in statistical approach) Scott Dann and Brede Hangeland.

No, so really now what

At a certain point we get old, tired, prone to low defensive output, and saying “It’s complicated”.

And there are so many complications. The play style of the match affects the statistical output, as do both the relative and absolute quality of the two teams playing it. A player’s role affects their statistical output, and can often be detected through it, but not always. A player performing their role badly may look like a different role entirely, and a player unable to perform their role may be because of them or their teammates or their manager.

Not only that, but weaker teams tend to defend with more players which affects how they can attack after turnovers which also affects how they can defend after losing the ball again. Sergio Busquets putting up ‘midfield destroyer(ish)’ tackle numbers, on a possession-heavy, low-turnover team is a clear sign that football is played on an odd, odd playing field.

Isn’t it exciting?

If you’re new to football data, you may well have missed the introduction of expected goals, the introduction of possession value models, the jump into physics PhDs and graph neural networks that tracking data has brought, and you may well be missing the first steps of body pose data (for something cool on that, see here). Yet there’s still no public consensus about how to take the tackle numbers of a player on one team and say ‘this is how best to compare them to a player on another’.

I still reckon that’s possible.

Possession is not nine-tenths

Wed, 01 Jan 2025 00:00:00 GMT

There are several motivations for this post, but one is to add to the public analysis conversation and show that (hopefully) interesting work doesn’t need to be complex. Small children often ask more interesting questions than adults.

This post is being written just after Christmas, a time of the year when the fridge is full of odds and ends that are all a little imperfect. Everyone will have their own favourite ‘leftover’ recipes, ways to squeeze the most out of the ingredients available.

Football data has an equivalent: defence. The full-bodied enormity of soccer’s contests for possession are carved and trimmed into duels, interceptions, clearances. Unlike expected goals - akin to a centre-piece cut of prime meat - they require ingenuity to work with.

Ideas about adjusting defensive stats began. Ingenious. But not all ingenuity is genius.

It’s now about ten years since I first ventured into ‘possession adjusting’ as a way of tinkering with defensive stats. (Others arrived at the idea independently before and since). I’m now pretty unsure about it.

This is partly a post about some data analysis work, but if you poke it hard enough it’s about the essence of the sport. Can you tweak an ingredient enough to make it a centre-piece on its own or are there insurmountable limits? What is and isn’t separable in football?

And how do you recognise ingenuity?

Data tables are in the eye of the beholder

Code for the project can be found here: it goes without saying that I would appreciate mistakes to be ruthlessly hunted down.

The theory behind ‘possession adjusting’ is simple. A team having possession of the ball (or not) affects what their players can do. So you adjust a player’s stats based on their team’s share of possession. In one sense, it’s the same theory as averaging a player’s stats per 90 minutes - that things need to be adjusted to make them justly comparable.

However, an update to work I did a few years ago has given me the same result as previously, one that doesn’t mesh well with possession adjusting.

We’ll see the data below, but there’s a lack of a clear link between possession share and player defensive actions output. This is not what the theory behind possession adjusting would expect.

Below are three tables that you can glance at, one for each of three positions: centre-back, full-back, and defensive midfielder. They’ll show three things, which were also features of the previous work:

The link between possession share and defensive actions is small if existent at all
The link varies by position
The link varies by defensive action type

The tables are ordered by the absolute strength of the correlation for each position. In none of them does the strongest correlation hit +-0.2. (A little note about the data follows them, as a reward for getting through the numbers).

Centre-backs

Defensive action type (per 90)	Correlation to possession share
Dribbled Past	0.144
Tackles	0.132
Interceptions	0.12
Clearances	-0.118
Pressures	0.104
Blocks	0.087

Full-back

Defensive action type (per 90)	Correlation to possession share
Clearances	-0.19
Blocks	-0.101
Interceptions	-0.071
Tackles	-0.067
Pressures	-0.066
Dribbled Past	-0.03

Defensive Midfielders

Defensive action type (per 90)	Correlation to possession share
Dribbled Past	0.099
Tackles	0.081
Interceptions	0.054
Blocks	0.037
Clearances	-0.034
Pressures	0.032

The data being used here is the Statsbomb open dataset for the 2015/16 seasons across the men’s ‘Big Five’ European leagues. (that is, the English Premier League, French Ligue 1, German Bundesliga, Italian Serie A, and Spanish La Liga). It’s accessed via the kloppy Python package. (I’m slightly wary of discrepancies this might cause compared to working directly with the event data, but I don’t think the trends would change).

The defensive action types are ones found in Statsbomb’s dataset, although the definitions are relatively common to event data. The player positions are also from the Statsbomb data, and only feature players who played 450+ minutes in those positions - the code for this project can be found here and a full table of results here.

In fairness to the concept of possession adjusting, those three tables aren’t the entire story.

There are higher correlation strengths among other positions, although none greater than +-0.35 in Attacking Midfielders, Center Forwards, or Wide Midfielders (and most not greater than +-0.25). The Winger position gives the greatest support for the possession adjusting principle (Wing Back correlations are stronger, but a tiny group):

Wingers

Defensive action type (per 90)	Correlation to possession share
Pressures	-0.386
Blocks	-0.365
Clearances	-0.355
Tackles	-0.339
Interceptions	-0.309
Dribbled Past	-0.297

There is more that can (and will, soon) be said about the gap between these figures and those for the other, more defensive, positions, but one thing is clear. This marks a blow against the principle of a uniform possession adjustment of defensive statistics.

More than meets the eye

There are a lot of interesting little avenues that can be found in this data. It seems curious, for example, that positions known most for their defending - central defenders, full-backs, and defensive midfielders - appear to be affected least by their team’s share of possession.

Defending is a team game though. It’s not unusual for ‘forwards’ on weaker teams to be more concerned by defending than attacking, while their counterparts on stronger teams are sometimes given a bit of a pass from defensive duties. Although I suspect that the Winger position correlations might look different if some lower possession 4-3-3s were classed as 4-5-1s, it seems true to my ‘viewing experience’ for possession share to have a more discernable effect in ‘forwards’ than ‘defenders’.

(An aside, although it’s covered in the code repo for this project - as Statsbomb’s event data has possession sequence IDs included with it, it’s fairly straightforward to identify when players were on-pitch and use the sequence IDs as a quick reference for events which happened while they were on-field).

This is where the fun begins

If you’re only interested in the data tables, you can safely close the email now. If you’re interested in the fine line between wonderment and futility in your numerical analysis, continue.

Let’s duck back to my theory that correlation strengths for Wingers are partly affected by the formation assigned by Statsbomb. My (unverified) assumption is that in a Statsbomb 4-3-3 formation, the wide attackers are designated as Wingers, and in some other formations - like a 4-5-1 - they’re designated as Wide Midfielders (whose correlation strengths looked much more like the Center Forwards, with a mean average correlation of the six defensive actions of -0.166 and -0.186 respectively - again, full results can be found here).

Assigning a formation to a team is notoriously difficult. It’s a subject I’ve written about before, a subject that even data-reticent teams complain to data providers about, and, as a result, an area which use of tracking data has tried to solve.

Although football punditry commonly talks about a team’s ‘in-possession’ and ‘out of possession’ shape, my conclusion in the post linked above (which I still stand by) is that the ‘formation’ a team is assigned by fans, media, and data providers is the one which they ‘see’ most often. It’s a shorthand for the roles that the players within it are playing.

When you come to the division between what the shorthand ‘4-3-3’ indicates and what the shorthand ‘4-5-1’ indicates, the difference is mainly in the defensive roles of the wide attack/midfield players.

If the switch between assigning the two formations was systematically done based on the ‘defensiveness’ of a team (for which possession share might be a plausible proxy), then this would just be a quirk of data. However, if teams are sometimes but not systematically assigned to ‘4-5-1’ or ‘4-3-3’ based on their possession share, then that might be a confounding factor in this particular type of data analysis.

(In fact, in the dataset, Wide Midfielders and Wingers in this study had the lowest (44.9%) and highest (53%) median possession shares respectively).

Tying the threads of the Get Goalside cinematic universe

As I said before, the concept of possession adjustment is sort of similar to the ubiquitous concept of ‘per 90’-ing statistics. We recognise that there are ways that data isn’t ‘comparable’, and we think we can address that.

No method is perfect though (even ‘per 90’-ing gets questioned).

Possession adjusting supposed that a team having the ball affected the defensive output of players. (In reality, many defensive actions are ends of possession sequences, and so wouldn’t be affected by the duration between these sequence endings).

You could head in the direction of a different sort of possession adjusting. Instead of adjusting based on the possession share, you could adjust based on the count of possessions/possession sequences. Given that defensive actions often end sequences, you might reasonably expect that players involved in ‘pinball’ matches will have higher figures than those without. (In fact, there was a slight indication of this in my earlier work, though I haven’t tried replicating that with this 2015/16 dataset yet).

Unfortunately, football throws another spanner in the stew. There is an obvious stylistic difference between teams who play ‘high turnover’ matches and those that don’t, and there is likely to be a quality difference between these teams too. Some previous Get Goalside work with Statsbomb open data has shown a link between ‘ball-in-play time’ and breaks in play, and a suggestion that high-possession (Guardiola-type) teams have more in-play time because of this link with breaks in play. The relative quality of teams (and possibly the absolute quality) appears to have an impact on ‘style’ whichever way you slice things.

I suspect that this is inevitable. I’ve previously tried to distil football down into teams competing over space control and ball control, and everything else springing from there. If a team has the quality to do so, it will naturally want to have more control of the ball, the only method of scoring. The hugely imbalanced value of space on a football pitch - extraordinarily heavily concentrated around either goal - steers the (weaker) defensive team’s strategy. With such a dynamically tilted field, is full comparability through adjustment of statistics possible?

So what now?

According to my own memory of my views (a trustworthy record if ever there was one), I have two longstanding opinions about data.

One is that a large part of the craft of data analysis is tuning an internal ‘fuzz-meter’, the sense of when things are ‘close enough’, ‘probably legit’, etc. I am wary of false precision.

The second - perhaps paradoxically - is that measuring discrete, specific skills is very valuable. It’s just that most of event data analysis doesn’t do that.

Take a statistic that is pretty clearly useful, expected goals. Reliably getting a large amount of xG isn’t ‘a skill’ as such; it’s an indication, or a symptom, of the various skills that a top striker has.

To go back to the (far worthier) subject of defending, there are a variety of skills that you’re looking for in a defensive player. It turns out that these skills aren’t necessarily observable purely through stats like tackles and interceptions. And it turns out that they’re probably not observable through possession-adjusted versions of these stats either.

Other types of data might let you measure those specific skills (whether that’s tracking data or other types of event data).

There are some in this world who just want to create single number models.

Maybe you can create steak out of scraps.

That would be ingenious.

Lessons in genAI

Mon, 09 Dec 2024 00:00:00 GMT

It’s a matter of time before someone releases an ‘AI assistant coach’. I imagine that, when they do, it will follow the common AI hype cycle:

product (or, more likely, product beta) released
interface is engaging enough that normies can use it for jokes
the ‘LinkedIn Apex’: declarations of a game-changer for the world
within two weeks, usage drops 80%

Despite this, generative AI systems are here to stay, in one form or another. Why? Because they’re getting quite good. As someone who spends a lot of time in code editors, I can attest to that.

And, attesting to it, there are some features that guide us to what football-based systems might be like when they inevitably (but not necessarily imminently) come.

How fast the road takes us, I don’t know. But I think there are three strands to the direction of travel: tools, context, and accessible content.

Tools for tools

There’s a small part of the generative AI industry who really believe, or want investors to believe, that large language models can become superintelligent on their own. But we don’t really need that - in either a practical or dystopian sense. “Agents being able to use software is how AI becomes more general,” Amjad Masad, the CEO of a very popular company called Replit, said recently.

These models, increasingly good at interacting with human languages and ‘non-human’ languages like code and APIs, can be hooks into things. And so your football LLM systems don’t need to be trained into being expert analysts. They just need to be good enough to use the tools at their disposal.

For example, some speculative fiction: “How well does this full-back defend against overlaps?” could be turned, by an LLM-underpinned tool, into a series of requests to other tools. Maybe you have the data for ‘overlaps faced’ already available, in which case gathering it is the first step. But maybe not, in which case a data science process can be kicked off (along lines of previous work). After that, another process, gathering data about how often corners, shots, and goals are conceded shortly after the defender comes up against an overlap. Another process might grab and edit clips.

(This feels like it’ll have implications for how different services can be used alongside each other, if ‘interfaces’ are going to be more geared to code scripts and API calls than human users. But that’s for another time).

The point is, a ‘generative AI world’ in football won’t need the LLM to be a football expert; the ecosystem of football technology will do a lot of the heavy lifting.

Context is king

Context is fun. This is objectively true, because mistaken context or innuendo makes up 90% of Shakespearean comedy. But it’s not fun if your LLM is missing it.

Code assistance tools are settling on ways to help this. For in-editor helpers, users can add files as context to their question - for example, you could ask it to write some data parsing functions based on the schema files you already have.

For football, the big and slightly hypey way this might be useful is context of terminology and game model. If we go back to the ‘overlap’ example from before, a club or coach might implicitly mean ‘overlaps in the final third’, calling similar movements elsewhere on the pitch something else. Or, a slightly more realistic example, terms for types of press or phases of in-possession play or player roles.

Beyond football, but perhaps in football more than elsewhere, I think LLM systems will live and die on understanding of context. It takes far longer to learn how to use a genAI helper if everything you ask it needs to include all relevant context. It’s like talking to a particularly stubborn child who’s decided to only follow instructions very literally. Or like working with someone in Quality Assurance.

But…

Content is king too

If context is important, that means the system needs to be able to access it. And that means that more of a coach’s/club’s work needs to be on a system that plays ball with an LLM-tool.

Think about training sessions. The majority of a coach’s work is done outside of a matchday, which has been the traditional bulk of data availability. They may well have a bank of reference video clips, but a lot of knowledge might come from conversations with the rest of the staff and exist in their heads, or on paper.

Think, as well, about coaches changing jobs. Currently, the clubs are the ones who buy software - but if part of what they’re hiring in a head coach is that coach’s methodology, it’s that coach’s ‘data’ that is the most important to access. Head coaches (at the top level where it can be afforded) often want to bring assistants along with them into a new role - would an LLM-based assistant be the same?

And, of course, if we’re talking about a coach’s knowledge as ‘data’, there’s the data ownership question. If LLM-based systems are going to be relying on coaching knowledge, coaches have got to make sure they can take this with them and use it in future roles. (The same is presumably already true for scouts, whose reports have long been logged into centralised knowledge banks).

The ‘pool of knowledge’ problem

Yes, LLMs spit out text that is incorrect. But it’s amazing that they work as well as they do, pretty reliable conversation-bots created by probability.

Of course, they’re more likely to get things wrong when the data isn’t there to produce ‘good’ probability estimates of the next word in a sentence. When coding, that’ll often happen when using a small-usage package, or non-mainstream language, or just a quite specific type of problem.

Fortunately, football is the biggest sport on the planet; and if you’re reading this newsletter in its original state, then your command of English will help avoid potential disadvantages of LLMs in other languages too. Purely keeping my British Isles locality in mind, it’d be interesting to know how good LLM systems are in Welsh, and Scots and Irish Gaelic.

All that being said, while I’ve made the point that it’s the football software ecosystem that’ll do most work, maybe this will be genuinely harder to achieve outside of the major languages and/or in more niche areas of the game. Will off-the-shelf LLMs fit into ecosystems where practitioners want to organise, say, periodisation and training microcycles in Japanese? (sidenote: I’m interested in how this would interact with things like the dialectic variation in Arabic too).

The end

You could probably go a long way without specialist software.

For text notes, tools like Notion have AI helpers which can search your notes and files elsewhere, like in Google Drive. The issues with domain knowledge of LLMs would come in here, but if you’re a low-budget club then you could probably get value out of using something like it as a repository for your coaching or scouting notes.

If you are the type of coach or analyst who collects video clips, and these clips are consistently labelled, an LLM-inflected search tool might help you find relevant clips easier, without you having to act as a folder-expert organiser. But maybe this will be even easier than that: given that tools can integrate with Google Drive, perhaps you could keep a your clips there alongside a spreadsheet of names, tags, and links (essentially metadata) and use that as the entry point for an LLM-searchbot.

Get Goalside has written about what we mean by the term ‘analytics’ before, and whether ‘football tech’ counts or is stretching the definition too far. But, regardless of your opinion on that, the end of that piece is still relevant here:

And the thing with Moneyball is that it wasn’t about data per se[…] What it’s about is questioning orthodoxies and finding edges, finding the most efficient way to get wins that you can.

So whaddya do when the Yankees are ‘doing analytics’ too? You’ve just gotta try and find another edge.

‘What we talk about when we talk about 'analytics'‘, March 2024

Maybe this is one of those edges.

Bias for bravery

Tue, 19 Nov 2024 00:00:00 GMT

A couple of months ago, noted Ralf Rangnick sceptic Michael Cox questioned whether data had made a dint on football’s on-pitch tactics.

“Analytics has become better at measuring which players do these things, and the things leading up to them, and increasingly things like off-the-ball runs too, but has it influenced teams to play any differently?”

‘Has the impact of analytics on modern football been overstated’, The Athletic

I found this interesting. As I wrote shortly afterwards, maybe that’s right and, what’s more, maybe there’s a clear direction that cold, calculating data would push things.

In fact, maybe it already has.

I had a theory - and if you’ve taken enough notice of this post’s title, you’ll know it too - and I started listing things that played into it. More of the list than I’d expected were already on-pitch trends.

Across the board, I suspect that data analysis would reveal more risk-taking is better. Ian Graham, Liverpool’s former director of research, tells an anecdote in his recent book, How To Win The Premier League, about Brentford owner Matthew Benham. I used it in the previous piece - it’s about advising Brentford to Always Be Attacking - but I didn’t use the best line from it:

“Benham’s reply was that there is so much risk aversion in football that you have to demand extreme behaviour to have any hope of getting a manager to be anywhere near as attack-minded as optimal.”

How to Win the Premier League

In other words, coaches are so affected by risk-avoidance that you need a caricature-level opinion just to drag their Overton window towards the optimal strategy. (Or so Benham believes).

After this had marinated in the back of my mind a little, my little grey brain cells made a link to goalkeeper cross-claiming. Analyst and data scientist John Harrison has previously published some work around shot prevention by claiming crosses: measuring how active and accurate claimers can save their team potential goals before shots even happen. And I remembered that David Raya (at data-savvy Arsenal) is particularly prolific in this department. (In fact, Harrison’s models have suggested that shot-prevention was a key part in Raya being an upgrade on Aaron Ramsdale’s performances in the men’s team).

Of course, we can’t talk ‘risk’ in goalkeepering without mentioning Manuel Neuer’s sweepering antics. Neuer didn’t invent keeper-sweepings by any means, but he’s a big reference point, and the position of goalkeepers in general play is a key point of analysis nowadays. Neuer was such an influence because of how successful he was.

Where else can teams play with more risk? High defensive lines, trying to keep the ball when under pressure, more take-ons, working the ball into better shooting locations, playing an attacking style even when leading, substituting as soon as necessary. (Any others?)

A lot of these are pretty familiar to modern football, even if analytics hasn’t driven take-up.

This doesn’t mean that these risk factors are a one-way street though. A push for high lines opens up a new risk-taking option for the in-possession teams: adventurous passes in behind the defensive line. But perhaps the highest line achievable is always the optimum.

At Amazon - a company even more successful than Matthew Benham’s Smartodds - they have a leadership principle called ‘bias for action’. The two beliefs which underpin this are that speed is extremely important to businesses, and that many decisions can be reversed if they turn out to be incorrect.

Risk avoidance serves an important purpose in business as in life as in football. But maybe the most effective path is to be poked a little bit more into the path of bravery.

Video is all you need

Thu, 31 Oct 2024 00:00:00 GMT

This is the age of unstructured data. There’s your pull-quote. In the real world, ‘unstructured data’ often means ‘words’, and part of the large language model hype is that they can provide an interface for this. In football, the more important world, the big mass of unstructured data would be video.

What we call ‘event data’ - the shots, the passes - is just a ‘structured data’ way of viewing matches. But the game tape - yes, plus wordy things like scout reports or coach analysis - is unstructured.

For all the benefits of structured data, it takes time to structure, into a system that you will inevitably hate when it’s just too late to change it. (something something data engineering, too). So why not just squeeze the juice straight out of the unstructured lemon?

Unstructured lemon juice like goalkeeper save technique from video, or goalkeeper save timing from video. (Goalkeepers don’t get enough attention, but are also on camera in just the right way for body pose detection). If outfielders are more your cup of tea, here’s a 2019 paper on body orientation from video.

If we’re being picky about metaphors, video and LLMs aren’t equivalents. Video and words aren’t even the direct equivalent. In football, tracking data is a form of unstructured data that has been around for a long time; working with the video directly is like working with audio. But the point remains the same: unstructured data.

However, to leverage this unstructured data you first need the game film. As Nancy Hensley pointed out at the recent Hudl Statsbomb conference, the lower level (in quality and existence) of women’s football coverage on TV affects not just fan engagement but data collection. In Belgium, they’re making a big play of installing cameras and getting data on everything that moves.

This harks back to 2022 Get Goalside, ‘How football competitions are their own competition’:

“Just like in the television industry, football leagues are now competing much more directly with their overseas equivalents. This is why La Liga (not just Real Madrid or Barcelona) are taking it upon themselves to complain about Paris Saint-Germain's and Manchester City's finances. It's also why they have their own analysis and visualisation tool, Mediacoach, which forms part of LaLiga Tech, which launched last September. [2024 ed: now called Sportian, and is part of the Belgian Pro League deal]. All a way of trying to make sure that theirs is the best product around.

On a slightly different scale, the relatively recently-formed Canadian Premier League has made a concerted effort to help the entire competition with its own CPL in-house analysts and expertise.”

As alluded to above, leagues centralising technical advancement is something that makes a lot of sense, not to dictate usage (which would likely stunt innovation) but to set a minimum standard. Although it does need to be a reasonable minimum standard. From Ian Graham’s How to Win the Premier League: “We also received tracking data for all UEFA games, but until 2021 UEFA did not exercise any quality control over it, so we could not trust it.”

(Related reading recommendation, The Formula by Joshua Robinson and Jonathan Clegg: “After nearly seven decades of [Formula One] teams fighting tooth and nail for every advantage […] Liberty presented them with a new reality. Instead of being rivals, these teams had to understand once and for all that they were all in business with each other.”)

The ‘pivot to unstructured data’ creates another interesting dynamic.

For a long time, access to data has been an issue for wannabe analysts or researchers. In their lifetime to date as a data provider, Statsbomb have been admirable in the amount they’ve made openly available. But if DIY collection from video takes off (not necessarily for tracking data - you could imagine someone taking this and deciding to create a shot-detection system), that would open interesting doors.

But to where?

Well, if video/tracking data is a rough equivalent of words and text analysis, maybe the recent use of generative AI can give some pointers.

There are two indisputably ‘successful’ use cases for genAI. One is coding assistants (for the more popular languages); another, though ‘successful’ is a loaded term, is art. Now, the development of lucrative tools built on scraped art, made by the profession who’ll be undercut by said tools, is the type of societal wrinkle you’d find in a dystopian novel. But these tools - think of Photoshop’s generative fill feature rather than entire artworks if it helps - produce convincing results. In the hands of artists, they produce art. In the hands of schmucks, they don’t. A lot of early genAI ‘art’ production was schmuckery.

These two use cases make sense when you think of how these kinds of generative AI work systems work. ‘Art’ does not follow the same rules of ‘factual accuracy’ that so much of the rest of the world does (photorealism, and adherence to specific styles, aside). Where else but art could Caravaggio and Kahlo, Rothko and Ruysch exist as greats. Certainly not business chatbots. Coding, meanwhile, has a much stricter sense of ‘accuracy’ but a far more limited ‘vocabulary’. LLMs work by predicting the probability of the next word; English has an estimated 170,000 words, coding languages will have far fewer (hats off to the Reddit user who asked this question a few years ago).

So, to football.

At some point in the past year, I heard someone with a lot of experience (on a podcast episode (I think) that I now can’t find) give a warning about tracking data. They said that it’s tempting to go after the gold mines of off-ball metrics, but that that could be a red herring. A marsh that one would sink into.

I’m embellishing slightly, but I think the tendency of work using tracking data to split player movement into ‘runs’ is telling. Some of it is to create physical metrics, some of it is around concepts like ‘running in behind’ or ‘overlapping runs’: clear concepts, clear vocabulary, turning unstructured data into well-understood structured datapoints.

But there’s so much that isn’t well understood, or well verbalised, in football. Leander Forcher recently released his PhD dissertation on ‘success factors in soccer defense’, much of which highlighted the lack of pre-existing work in defensive analysis. Lack of pre-existing work often means lack of clear understanding of terms. (‘what if we’d focused on different parts of the sport’ is a continuing theme of Get Goalside)

We’re in the age of unstructured data. We’re also in the toddling age of learning how to get the best use out of unstructured data.

The only thing that’s certain is that we’ll need more data engineers.

What's your research question?

Tue, 22 Oct 2024 00:00:00 GMT

Which is harder to do well, asking questions or answering them?

At some points a few months ago, I was speaking to someone who works with physical data. A few years ago, they’d have had to be selling stakeholders on the answers th data could give them. Now, everyone eats it up. But the person was saying ‘I say to people, do you know why you want this data? What do you actually want to use it for?’.

It’s a good question to ask.

Ultimately, though, there is only one question that anyone in sport has: “how can I win more?”. And so you need to play a little game of trade-offs, narrowing the broad focus onto something narrower, a tangible area, a specific factor.

This is why this year’s Hudl-Statsbomb conference (no more capitalised ‘B’ for ‘Bomb’) steered its research competition entrants towards ‘trade-offs’ as a theme. There was one on the value of a lesser-quality left-footed left-sided defender vs a better-quality right-footer; one on the value of booting it and pressing a throw-in instead of trying to play out of pressure. (The 2024 conference research papers can be found here)

It’s not just this year’s HudStatconf papers where the strength of the question shines through. Frequently, the thing that strikes me in analytics ‘research’ work is the clarity of the question. It’s there in some of my favourite research papers (who doesn’t have favourites), like fellow 2022 Sloan conference appearances ‘Learning from the pros: extracting professional goalkeeper technique from broadcast footage’ and ‘Beyond action value: a deep reinforcement learning framework for optimising player decisions in soccer’. It’s there in my favourite broad genre of work: projects which didn’t find what they set out to find, but which were conceived clearly enough that the journey was worth it and the direction of future investigations is clear.

It’s a craft.

For example: Pep Guardiola has referred to formation notation as ‘telephone numbers’, Emma Hayes has said discussion around formations is ‘archaic’ (around 9:20) - but we talk about ‘formations’ for a reason, as a convenient shorthand. So there will be some cases where splitting data by ‘formation’, or where methods of determining ‘formation’, will make more sense than others. (If you need in-the-moment formation/shape information, then tracking data or video analysis may be your only viable options, but build-up patterns might be gleaned from event data)

But this whole thing about ‘what is your actual question’ is not just true of research.

Three years ago to the day (as this is being written), I wrote an overview of where you might spend money set aside for ‘analytics’. Quite frankly, I’d forgotten I’d written it. Let’s have a look at the end conclusion:

If you're one of the elite [teams], it makes sense to get ahead of the game and get a department set up internally, on the condition that you retain that knowledge. There's no point in the exclusivity benefits of an internal department if you don't make sure you still have it if someone leaves.

However, most clubs are not yet at the point where they're hiring data people and allowing them large amounts of time for research projects. For the majority, it probably makes sense to choose data provider smartly as much as cost allows (and to be honest about whether the shiny things in the data will get used) but then to make strategic use of third-parties.

‘Where to spend your analytics money?’, October 2021

There are two biases I have here:

I work for a company you’d class as a ‘third party’, which one may want to make strategic use of (bosses and colleagues at Twenty3 Towers claim to read, so I better link the company website)
It feels nice when things you wrote years ago still hold up

That said, I want to highlight two specific parts of this extract. “There’s no point in the exclusivity benefits […] if you don’t make sure you still have it if someone leaves.”; apply this to your internally-created tools as well as report formats and research projects.

If you think players are the only things whose performance can drop off a cliff due to getting old, you might wanna google ‘tech debt’. Several months ago, I quoted Charlie Marshall of the European Clubs Association: “There are so, so many [clubs] and the vast majority of them are quite small businesses.” If you’re a small business — a community events business, really — do you want to also be a software company? Why?

(This is worth its own blog, but can probably be boiled down to 1) difficulties in combining services from different data sources 2) difficulties in wrangling external software to work for team game models 3) an employee’s time doesn’t appear as an additional cost on the balance sheet 4) you don’t have to wait on the external company’s timeline to update your internal platform 5) as Andy Warhol said, ‘in the future everyone will develop a scatterplot tool for 15 minutes’)

The other particularly important part of the extract is “strategic use” of third-parties. A couple of paragraphs later in that 2021 piece, I wrote:

Using these would allow you to flex your capacity for 'analytics' as needed, without hiring full-time. As a club, this can also enable you to build up an internal knowledge bank if you make sure that the third-parties work and/or findings get stored somewhere people will remember them.

On top of this, assuming that you're not their first customer, third-parties are also likely to have processes in place that mean you can skip some of the tricky 'training wheels' stage of setting up an analytics department

‘Where to spend your analytics money?’, October 2021

(I neglected to mention data engineering, and I apologise to the gods of cloud computing for this)

The problem with all that, of course, is that following this advice may mean hiring a sufficiently capable head of data to then spend money on a data provider as well as then spending further money on outside services. Maybe that won’t play well with bosses who expect some guy (often a guy) with a quantitative degree to get things going within a month or two.

But, look at Bayer Leverkusen. Thankfully not too tight-lipped in what they allow to be shared on LinkedIn, they’re on the roster of Kitman Labs, Catapult Matchtracker, MyGamePlan, and SportsDynamics, and that’s only the software platform partnership announcements I could easily find.

It may well be that Leverkusen’s eventual aim is for all of that tech to be internally created and managed (as one of the ‘Analytics in the US’ panellists at the Hudl-Statsbomb conference put forward as a general truth). It may well be that they’re not even using those platforms (which would be strange, but ‘pays for an unused subscription’ is hardly a novel situation).

But, while the best research papers ask a tight, well-framed question, the best clubs will be doing the same.

It's the incentives, stupid: when expected goals are the problem

Tue, 08 Oct 2024 00:00:00 GMT

A couple of weeks ago, I wondered whether analytics should have made football be played in a radically different way to how it currently is. A recent episode of the Double Pivot podcast reminded me of an oversight: football might not have a three-point line for goals (and an associated NBA-type change in shooting behaviour) but it does have a three-point rule for points.

The change was designed to improve the incentives for teams, after a period of doldrum-esque scoring in 1970s England. In the start of that decade, the English men’s top division saw four seasons in a row with an average goals per game of around 2.5 or under. This was a huge fall from a time in the late ‘50s and early ‘60s when the average could be north of 3.5 per game.

Goalscoring alone probably isn’t the best metric to judge the rule change on, but the top division’s scoring rate picked up before the change came into effect in England in 1981. It stabilised at around 2.6 per game, until a hop upwards in the 2010s. The 2023/24 season saw the first average of over 3.0 goals per game since the ‘60s.

But this isn’t about goals.

Tweet: "Had some really interesting conversations with people all around Europe about why long term projects don't get seen through in football.

Instability is rife. How many clubs have the same senior leadership / ownership for more than 5 years? Very, very few."

Tim Keech [@SBunching], 07 October 2024

The above is something of a riff on a theme:

Tweet: "Even with the rise in prominence of the Sporting Director role it is very rare to find people in decision making roles in football clubs who are confident enough to make investments with 3+ year waits before on pitch benefits are seen."

Tim Keech [@SBunching], 19 November 2023

Even at this high level - the level supposed to be about long-term thinking - it seems long-termism can be more theory than practice. Possibly because the incentives at play are job security at an individual level and, increasingly if not wholly, investment value at an ownership level. Regular readers of Get Goalside will identify a bias here, but ‘driving wedges between clubs and their long-standing communities removes an important incentive to generational long-term thinking’ does not sound an unreasonable hypothesis.

But this isn’t about club ownership per se either.

By coincidence, the past few days have thrown together these reminders of incentive structures with something else. On YouTube, a glimpse into Burnley’s data-gathering operation on their academy footballers, and on social media, a post about high-school American Football tracking data. Add these to the consumer products selling the dream of being scouted from afar, and soon we will be telling uncomprehending children of the days when we played sport and left behind no record other than the score. What do we do when there are no datapoints left to conquer? And what of the data already being coerced into databases?

There will, or should, of course, be data protection policies in place. But though Get Goalside is pretty pro-data, the push for player data rights is important. And it should extend to youth football. The power of data is often in its quantity, which leaves individuals - in successful data-gathering operations - as mere specks in the whole. Valuable as part of an aggregate, inconsequential on their own terms. There are football ownership models where groups will buy clubs, or large stakes in them, as a means to gaining a foothold in that continent’s ‘talent development market’ - how passé. A scouting network is such a twentieth century concept. Cycles of age group teams entering one’s data warehouses, training talent identification and development plan algorithms: that’s an investable future.

As the saying goes, football is the most important of the unimportant things. The latter half of this may be a good thing.

By coincidence, modern football coaching mirrors the now-famous line from James Clear’s book Atomic Habits: “you do not rise to the level of your goals, you fall to the level of your systems.” In football more broadly, what systems are those to fall back to? Do we trust the incentive structures in place?

Nick De Marco KC, barrister (on Twitter, responding to a hearing on Premier League rules around ‘associated party transactions’): “All that I can say is we are living in the most exciting time for sports law. I have never myself been one to celebrate the greater commercialisation and therefore legalisation of sport and its regulation, but it is a real fact of life and economic activity, such that this tendency for greater scrutiny of sports regulation is inevitable.”

Positional Play and manager metrics

Mon, 30 Sep 2024 00:00:00 GMT

Everyone knows the problems and the buzzwords. The theory is this: Teams are difficult to organise; individual magic is great but usually creates inefficiencies; a structure, aligned with meaningful goals, speeds up a team to being good.

Am I describing Positional Play or Agile project management?

For those unfamiliar with the term:

Agile is a collection of software development practices that built on existing ideas, grouped under an umbrella term, with courses and how-to books springing up for this now-a-proper-noun methodology; it is sometimes criticised as a management fad that is applied in cases it needn’t be.

Positional Play is a collection of football tactical practices that built on existing ideas, grouped under an umbrella term, with courses and how-to books springing up for this now-a-proper-noun methodology; it is sometimes criticised as a management fad that is applied in cases it needn’t be.

Good, we’re all caught up.

Now, don’t close your inbox if you’re not in software. Everywhere that there’s management, there’ll be theories and snags where implementation hits the road. Government, supermarkets, building sites, military theory. (Google ‘Sun Tzu business book’, you could start a library).

My point is that it’s easier to talk about football tactical approaches when you think about them in this way. The terms are more like genres than rubrics. Or schools of thought, within which there can be disagreements and room for leeway.

There’s an ‘analytics’ point we’re strolling towards, this isn’t completely a newsletter of amateur HBR-ing. First, though, we need to visit Munich.

Vincent Kompany, the current Bayern men’s team coach, has an unusual management career path. From Anderlecht to Burnley, with a promotion and relegation, and then to Germany’s dominant club. The style of play at Burnley translated well to their position as top team in the English Championship, very badly to their position as one of the worst teams in the Premier League, but, crucially, quite well to mega-teams like Bayern.

If we go back to the business management metaphor, what works for multinational corporations will be different, to some extent, to best practice at a neighbourhood Tesco Express. Bayern didn’t look at Kompany’s Premier League relegation as an abject failure, they’ll have seen it as a datapoint in his (sigh) management philosophy.

An advantage that football has over the rest of the business world is that a relatively wide range of manager metrics are easily obtainable. While they won’t be able to tell you for sure how good a manager is, they can at least narrow down the pool of talent to ones who ‘do Agile’, if that were important to you.

If you’re in a boardroom, that is worth paying attention to. For everyone outside, it’s the least interesting thing about football managers.

Here’s the actual interesting thing. Kompany joined Burnley and oversaw them be comprehensively promoted and comprehensively relegated. From Kompany’s perspective, you can see this as a chance to practice his Nasdaq C-suite methodology somewhere that the C-suite hirers were definitely watching, unlike Belgium. But from Burnley’s perspective, you do wonder a little more about the value in hiring a McKinsey grad for the mill town Sainsbury’s.

This is the interesting part. The amount of teams wanting to transition between playing styles is almost certainly quite large, almost certainly a difficult task, and almost certainly a very valuable one. Recent Championship-to-Premier League success stories tend to have something interesting tactically about them, either in their initial season or in their later evolution: Brighton, Leeds, Brentford, Bournemouth.

I wrote recently that I think ‘Possession Play’ football is more or less what the best teams inevitably play due to the nature of the sport. The theory is that football is about space and ball control in comparison to your opponents: teams with better players choose a style that maximises their controlling talents; teams with worse players choose a style that minimises the differences or disrupts opponent control. Better tactics or coaching can eke out edges in this battle, moving you up the table, and allowing the team to bring in better players who bring a more advantageous balance to the space/ball-control match-ups.

Who are the managers who are best at this, and does that show up in the data?

(There is a sort of ironic point to be made here too that, at a time when statistical analysis for head coach hires is taking off, there’s a trend for hiring managers in their 30s. The assumption is presumably that they’re both closer to tactical innovation and have room to grow, the latter being a similar logic to snapping up 21-year-old players.)

This brings us to a final point.

In business, it’s recognised that people who are good at scaling a company may not be the best at running it once it’s big. Not only that, but people who are good at scaling may not want to keep running things anyway. (admittedly ‘Neither Moving Fast Nor Breaking Things’ does sound like a bit of a buzzkill). Football doesn’t quite have this same type of appreciation, beyond occasional relegation-firefighter specialists. Managers who can switch up a team’s style but aren’t great at coaching ‘elite’ tactics are seen as limited, both in their aptitude and their career prospects.

If that doesn’t change, it feels like we’ll only get more teams turning their promotion-brand football straight into their relegation-brand football, because it’s the best way for the coach to get a decent job next time.

A note on the Positional Play-to-Agile metaphor

Given that Agile is aimed at being, y’know, agile, it strikes me that the more literal metaphor would be Agile and ‘Relationism’, where the heavily process-dependent ‘Positional Play’ would be the kind of documentation-driven software practices that Agile was aimed at replacing. The metaphor is less of a 1:1 comparison, but still works on a ‘widespread methodology’ level. And now, the frog is truly dead.

Why is the impact of football analytics in question?

Sun, 15 Sep 2024 00:00:00 GMT

The Athletic’s Michael Cox prompted online posts this week with this article: ‘Has the impact of analytics on modern football been overstated?’. Writers don’t write headlines, of course, so as way of summary here’s the penultimate paragraph:

Ultimately, it depends how you consider the word ‘football’. If you take ‘football’ to mean the football industry — financial transactions, contract negotiations and clubs as businesses — then analytics has definitely changed football. If you take ‘football’ to mean the actual game — 90 minutes, 11 against 11, how should we play in order to win? — then, 3,000-odd pages later [a reference to books on the subject Cox has recently read or re-read], we’re still waiting for a convincing (or fully disclosed) account of analytics’ impact.

It’s a reasonable point. The impact of expected goals on shot distances is the most visible thing for analytics to hang its hat on, and even that is contested (numbers of long-rangers were already falling before xG became widespread, even in online circles).

Shouldn’t this be clearer by now?

Here are some possible reasons why it might not be. Feel free to skim.

As well as defining ‘football’, we might want to think about what we mean by ‘analytics’, which Get Goalside has, of course, done.

Football is ruthlessly efficient

Baseball and basketball are held up as the examples of ‘analytics’ in modern sport. Really, though, we mean MLB and the NBA. The NFL too, if we’re talking fourth-down decisions. The major leagues are plural, but in their own sports they’re singular.

Maybe the closed shop (with very limited international pollination) lets stale ideas circulate, or allows a culture which suppresses new ones. Maybe the global nature of football lets experiments play out somewhere, with the successful ones later appearing on the global stage.

It may not be closed- or open-shop per se. Football is also played by far more people, a larger marketplace for ideas to appear in, and has a fairly unique interplay between club and international football. While national playing styles might once have been a barrier to new ideas, nowadays it seems national FAs will take any approach that’s working. That, in turn, filters down to their coach education programmes.

The old-school saying is that only the results matter. Maybe 21st century football has a large enough and diverse enough sample size for that to be correct.

Other sports are uniquely inefficient

The three-point line is not wildly difficult to understand. In cricket, the largest influence of data has (I think?) been in Twenty20 strategy - a new format of the sport in which there was a lower body of ‘establishment’ knowledge.

Maybe we shouldn’t be expecting analytics to make huge on-field impacts after all.

Football should actually be much different

These are theories, so they’re allowed to wildly contradict each other.

In basketball, the ‘analytics’ change was to be more brave in shooting threes instead of long two-pointers. In baseball, strategy has shifted to walks and home runs. In NFL, the push is for fewer punts on fourth downs. The insight in Twenty20 cricket was around the value of boundaries and the trade-off waste of wickets in hand at the end of an innings. There’s a theme here: aggressiveness.

This is actually mentioned in Ian Graham’s book How to Win the Premier League too, in an anecdote featuring Brentford owner and Smartodds founder Matthew Benham: “Benham told me that his instructions to the Brentford manager were to attack, regardless of the opponent and the situation in the game. In minute one, Brentford must attack. In minute 90, leading one goal to nil, and down to 10 men, Brentford must attack.”

It’s a refrain that analytics-watchers will be familiar with. That teams sit back on leads too much, that managers don’t make substitutes early enough. It’s worth remembering that the disapproval of long shots is not about a conservatism around shooting, but about the magnitude of difference between a thirty-yard and a fifteen-yard attempt.

So, maybe football should be much more aggressive - or, maybe more semantically correct, less conservative - in its outlook.

The data isn’t better than (most) coaches… yet

Apart from the (contested) xG influence on shooting behaviour, transfers are probably the biggest evidence of analytics’ influence in football. “Data/analytics have had a HUGE impact on recruitment. Literally everything about the job and process is different now,” Ted Knutson tweeted on the subject of Cox’s article.

Scouting is also the area of football with the worst ‘human-to-required observations’ ratio. A Dean Oliver line that’s quoted in How to Win the Premier League: “Your eyes see the game much better than the numbers. But the numbers see all the games. And that’s a big deal!”.

Maybe the available data had an edge on scouting operations* that it didn’t on first-team coaching. (*Or, possibly, an edge it had over managers who tried to have the last say on transfers).

‘The data’ is changing though. Maybe the increasing availability of tracking and biomechanic data (and the tools people have for them getting better) will lead to more of an on-pitch impact.

It’s also worth noting that ‘football people’ have accepted a lower level of control of the off-field stuff, retaining control of tactics and player-management (although some head coaches do gripe about the say-so of their medical departments). How might things be different if managers had demanded they keep control of transfers, and relinquished some on-field control to ‘directors of tactical methodology’ instead.

The smart people are quiet (or, vice versa)

To go back to Ted Knutson’s tweet: “Data has had a moderate impact on style of play and game models. I don't want to go into huge detail here because a lot is still amazing IP we and others have developed and I get grumpy DMs when I talk too much.”

This is something that Cox’s article touches on as well. The thing with this one is that its argument is visible to fans, but a case of ‘these changes are attributed to [some other factor], but actually analytics is responsible’.

Semi-related, a thought experiment: how different would perception of analytics’s impact at Liverpool Football Club (men’s team) be if the press coverage had been different? Their data-involved ‘transfer committee’ had a unique level of negative attention in the traditional English press, and they’re also the only club I can remember to have a New York Times magazine article written about them.

Mikel Arteta’s season-end finishes at Arsenal were 8th, 8th, and 5th before finishing as Premier League runners-up in 2022/23 and 2023/24: what would this conversation be like if a long-read on their analytics set-up had dropped in 2022?

We’re looking in the wrong place

This would fit alongside some of the other possible options. Football might be uniquely efficient at its top level, where many people focus their time. But there’s a world outside the richest leagues in western Europe, and maybe the impact is larger there.

Speaking of geography, an aside (is it even an aside?). In an open-air sport like football, ‘the right way’ to play will look different from place to place. Would the heavy-pressing football that is so tied to Germany and (through Red Bull) Austria ever have emerged in Spain, land of the siesta? Might the state of English pitches that we see in videos from the 1980s be a factor in the long-ball football that took root?

A different spin on ‘looking in the wrong place’ is, to use a singular example, the absence of new André Villas-Boas-es. The Portuguese manager, once hired by Chelsea as a sort of heir to José Mourinho, is the earliest ‘data story’ I can remember. Shortly after he was sacked by Tottenham Hotspur in December 2013, Michael Caley wrote, using a rudimentary expected goals model:

“You can see here again AVB's predilection for long shots. Even taking over a club which had generally not focused on emphasizing shot quality, he still cut the club's percentage on shots from the danger zone from 5% below league average to 20% below.”

‘Statistics for managers: a continuum of AVB’

Maybe you can chalk up the (lack of) André Villas-Boas legacy, a coach who made his name with an unbeaten season as Porto manager, as an example of football’s efficiency. But the absence of AVB imitators, conscious or unconscious, is probably down to the acceptance of expected goals too.

Analytics types have long said that their biggest impact in transfers is just saying ‘no’ to bad ideas. Maybe that’s the case for on-field matters too.

Just run some more

Sun, 08 Sep 2024 00:00:00 GMT

It's only 'a sprint' if it's from the Prozone region of Leeds; otherwise it's just sparkling high-intensity runs

Fashions come and go and, inexplicably, 90s aesthetics are back and running stats are a hot topic.

Well, maybe it is explicable.

Football has had regular running data since the 90s, but a few things might’ve held back its usage and usefulness. One of them is ball tracking - harder to do than player tracking but really helps to know what’s going on on the pitch. Another is arguably imagination (see ‘Are Current Physical Match Performance Metrics in Elite Soccer Fit for Purpose or Is the Adoption of an Integrated Approach Needed?’ (2018, Bradley and Ade))

Once you have that, it’s easier to chop the numbers up by phase, and when you can chop the numbers up, they get more interesting and useful. Split them by position and tactical phase, and you get a very quick, very nuanced picture of a player’s physical demands on matchday. As I often do, I’ll point to the FIFA Training Centre as an example of this kind of work.

That detailed matchday information also lets you work backwards. Backwards from matchday: to training, to recovery from injury. MLS team Chicago Fire are using French company Footovision to help with this. Other systems will, one presumes, be available. (Perhaps like fellow French-born company, Skillcorner).

Platforms have power. It’s not just Parisian-based companies springing up over the last few years: over the last couple I’ve noticed a few research papers using data from the Kitman Labs platform (like this one on hamstring injuries). My venturing into the sports science realm doesn’t even reach ‘amateur’ status, but I’m familiar with the ‘n=8’ sample size problem. This paper, though, is able to call on data from 36 team-seasons in elite men’s football.

Get Goalside throw-back to September 2022 after an analytics conference:

‘[A]llow me to suggest some taglines for the company's marketing team to use in their post-conference content: […] 'StatsBomb Conference 2022: All analytics is web apps!'

Would now read ‘all analytics is SaaS’.

Software (usually) means scale. Loop back to the FIFA article I shared earlier: that was about women’s football, data from the 2023 World Cup. As the sport tries to catch-up on its historic lack of support and active hindering of women’s football, some things scale better than others. Training infrastructure and coaches can only be shared so much. Computer vision data collection scales better than event data collection. The insight into how to split tracking data into tactical phases, to benchmark players against their peers, barely needs to scale at all. Once it’s there, it’s there.

And the global game being the global game, there are geographic implications. Check out the countries that viewers were tuning into this computer vision for tracking data Q&A stream from - there’s Turkey, Uzbekistan, India, Congo, Uruguay. I won’t repeat my previous line about ‘why not make your own tracking data instead of buying it’ (also from that September ‘22 post above), but a rudimentary system may still have value. Even if you work with large error bars, maybe you could take inspiration from the literature, gather non-quality assured data for the Uzbek second tier, and be able to benchmark attacking full-back requirements against their peers.

So, yeah, running data is cool again. Ditch the 2010s slimfit jeans, dig out the colourful maximalism.

There’s another advantage that running data has. No-one has to argue about what a ‘duel’ is.

The four quadrants of football

Sat, 24 Aug 2024 00:00:00 GMT

This blog won't be about Ian Graham's book, How to Win the Premier League. I might write about it at some point when I've finished reading it, but really y'all should just read it yourselves, it’s good.

However, reading it reminded me I haven't written anything fun in a while. Let's rectify that.

Going back through the five years of Get Goalside, there was something I forgot I'd thought about. It was a piece theorising why the best teams throughout history tend to play a possession-focused game and, equally, why weaker teams tend towards deep blocks and long balls.

Here's the core idea:

"'what makes a football team 'good'?' is easily answered (though much harder implemented): being better at controlling the ball and better at controlling the space. [...] From here, from this basic framework, you can break down football archetypes into the 2x2 quadrants that high-or-low space and ball control form."

I want to poke at this. I suspect it'll seem like kindergarten tactics to some readers, but we've all got to go through kindergarten once.

Let's try filling in this 2x2(x2) matrix (each segment needs to cover in- and out-of-possession sides of things).

High space control, high ball control
In possession: Guardiola-ball (for want of a better term)
Out of possession: aggressive, zone/option-oriented press

High space control, low ball control
In possession: [unsure - defensive possession with low-technical players?]
Out of possession: effective block, but little direct pressure on the ball carrier

Low space control, high ball control
In possession: Vertical ground passes or ball-carrying (Red Bull-ball? for want of a better term)
Out of possession: aggressive ball- and/or player-oriented press (Bielsa’s Leeds)

Low space control, low ball control
In possession: hoof ball
Out of possession: low pressure on ball-carrier and porous in the middle

Trying to fill in these archetypes, it becomes more clear what each space and ball control is being distiled into. In possession, ball control is obvious, while space control feels more about the amount of controlled passing options available. If you think about times when a team feels like it's losing control despite retaining possession, it's when they're passing around a press/block, where each pass is the only one available, often going backwards.

Out of possession, space control (intuitively) would be about denying these options. People talk about compactness, but the compactness isn't exactly an aim in itself, it's a method of achieving an aim. Compactness for compactness’ sake is silly. The ultimate aim is to prevent progression through the most dangerous area of the pitch. Blocking passing lanes and being close to potential receivers, that's out of possession space control.

'Ball control' when out of possession, meanwhile, might sound weird, but strong duel ability would count, as would applying pressure to ball carriers. Maybe tenuously, you could count the pressure on a ball-carrier as an extension of the duel, or a preamble, but it's not exactly controversial to say that pressure on the ball-carrier affects control of the ball.

Simple stuff so far. But you can start to use this as building blocks for more complex things.

You should always try and maximise your advantages relative to your opponent. If your team has elite ball control in possession, that makes your in-possession plan obvious. But if your ball control in possession is only moderate and your opponent's out of possession ball control is great, maybe you change tack. You can also play in a certain style to affect how the opponent plays in response: playing a low ball-control style in possession often lowers the ball control the opponent is able to have while in possession too.

So far, sounds like basic match planning, but these balances will flow and flux moment to moment in-game. None more so than in the ‘transition’ phases. Ultimately, these moments are not so separate to settled possesson as the popular distinction might make them seem; you could just phrase them as moments where space and/or ball control is up for grabs. (This also makes terminology around ‘manufactured’ transitions easier, when teams try and coax a press to then rush through - they’re just trying to reduce the space control of the out of possesion team).

Frameworks aren’t inherently good, but they can be a useful bedrock or scaffolding or starting point for further ideas. For example, you can start to imagine how you’d use this space-ball control matrix to build out a suite of KPIs, geared specifically towards identifying space control or ball control.

Unfortunately, there are some potential flaws in this particular framework.

On a team-wide level, 'ball control' should probably include the ability to offload the ball if under intense pressure, which requires teammate options, which is also more or less how I've defined 'space control' in possession. Diniz-ball seems like a useful test case here, if positionism discourse is willing to stoop to the level of kindergarteners. Are the interchanges between players in close spaces about achieving high space control, high ball control, or both?

Secondly, is there an inherent advantage that space control has over ball control, or vice-versa? If you have to choose between optimising one or the other, is there a clear answer which it should be? If there is, does that affect the usefulness of the matrix?

I suspect that there are more. Are there holes in this you can find? Or, in the other direction, are there advantages that I've downplayed too much or have missed completely?

Assuming that this theory tracks, that means that football is:

a game of space control and ball control, in and out of possession…
…where the aim is to score more and concede fewer than the opponent (read the recent 'understanding AI' piece for that differentiation)

And everything goes from there.

Unclear at this stage how it leads to Premier League titles. Maybe that’ll be in the second half of the book.

A fifth birthday

Mon, 29 Jul 2024 00:00:00 GMT

Get Goalside is five years old. I was aware that it pre-dated the pandemic, but hadn’t put together its age until going back through the archive after publishing the latest post (understanding machine learning through the analogy of football).

So I went back and reviewed it all.

Inevitably, more of it than I’d like was… meh. Not bad, just a waste of precious words. For those who’ve been around a while, thank you for not unsubscribing. However, thankfully a lot of it was quite good, or at least had interesting ideas.

Because people are always newly arriving (welcome), and I’d forgotten a lot of this anyway, here’s a thematic wrap-up of Get Goalside’s greatest hits.

Defensive stats

Get Goalside actually started as a defensive analysis newsletter/blog, but in April 2019 I published ‘Possession adjusting: an essay’. I would eventually look at some proper data on the topic, but two thoughts from it stick out even now:

“Adjusting defensive stats isn’t about seeing which are the ‘good’ defenders, but if you’re using the defensive stats to determine a player’s role, then you need to isolate that role as much as possible. In other words, you have three things that can affect a player’s stats. Their execution, their tactical role, and things outside their control like the quality of their or their opponents team and, from that, how much of the ball either side sees.”

And

“We don’t adjust attacking stats like shots or expected goals just because a striker plays for a good team.”

In public work, this latter sentence is still pretty accurate. The Athletic published a piece about adjusting attacking stats in 2022, but even adjusting per 90 minutes isn’t totally mainstream. This won’t be the case behind closed, professional doors. Coincidentally, Tim Keech of MRKT Insights (who are also celebrating a fifth birthday) noted the other day that Anthony Gordon’s time at Everton popped positively for them when adjusting his stats. However, to distil the point of these two thoughts more succinctly, it’s a good message: what in the raw data needs contextualising, and why?

A couple of years later, I came back with some graphs, some Hamilton references, and a prompt:

“I present these not as definitive takeaways to apply to your own work, but to show [that] whether possession adjusting makes sense depends on both the stat and the position of the player.”

‘What is football’

The defensive stats focus led to rumination on the nature of football more broadly.

It really hit its stride in 2022, but there are a couple of good lines from earlier pieces as well:

“To me, football is too much of an inter-connected sport for things to be as simple as coming down to the strongest link or weakest link. […] Maybe instead of the theory argument being strong link vs weak link it should be strong selection vs weak selection, or maybe strong unit vs weak unit.”

— ‘Is football a ‘weak link’ sport?’, September 2020

“As [Seth] Partnow says about shooting in basketball [in his book, The Midrange Theory], if we were starting from scratch I don't think we'd have the statistical landscape for defending in football that we have now. In fact, as he also teases in a footnote, our lack of conceptual understanding of defending probably holds back how we choose to collect data on it.

If defending is all about space, why are the defensive statistics so much about how a player affects the ball?”

— ‘Do we know football well enough to have good defensive stats?’, December 2021

A mixture of Partnow’s book, The Midrange Theory, and Twitter user @TiotalFootball are probably largely to blame for 2022, where I really hammered on the theme…

‘What’s in the way of analytics solving football?’ (partly the ball, partly the fallibility and slow learning rates of humans); ‘The Theory of Everything (in football)’ (split things into better conceptual categories); ‘Fear and fatigue in analytics modelling’ (maybe these would affect pitch control/possession value models); ‘What if we didn’t care about passes’ (and what stats we might have collected instead); ‘Ball control, space control, and why good teams play Pep-ball’ (with a 2×2 framework which also explains high pressing and deep blocks); ‘What is midfield for?’ (time-wasting).

Good collection of ideas there.

Conference-watch

The best conferences are collections of people and ideas, interrupted by sales pitches. The worst conferences are sales pitches, interrupted by ideas. As a result, post-conference pieces tend to weave half-threads together.

The post-conference pieces and their threads:

2021 Opta Pro Forum and StatsBomb ‘Evolve’ event (halfway house of event and tracking data; data providers competing on modelling); 2021 NESSIS and StatsBomb conferences (‘We know what players do, but we don't know how they do it’ — Vosse de Boode); 2022 StatsBomb conference (‘Insight is one thing, productising insight is another’); 2023 Opta Pro Forum (asking good questions); 2023 StatsBomb conference (‘everyone’s a decision-maker now’).

The best thing about these conferences (most of which I attended in person) wasn’t what I wrote about them, it was the conversations over various beverages. And one truly fantastic free StatsBomb pen.

‘Research’ work

At times I’ve even looked at real data (outside of the defensive data mentioned earlier).

In 2019, I wrote something that’s still a favourite of mine: looking at how good people think ‘good’ is and how that might affect the ratings they give to various players. The following year, I looked at the percentage of minutes Englishmen made up in the Premier League (amid Brexit rules controversy). Would really like to see what updated data on that looks like.

Two really fun pieces based on StatsBomb’s freely-available data: ‘How would you play against the Invincibles’ and trying to find a similar dribbler to Messi. In more of a data science vein, in 2023 I had a go at adding player position labels to their 360 data freezeframes too.

Finally, and more seriously, I went back to the StatsBomb free data to question the rationale behind increased added time on matches. The follow-up piece questioning the reasoning also delved into what actually drives match length. Hopefully we’ve seen the back of that fad.

Silly ideas

But I guess I can’t throw too much shade - this newsletter has been a home for lots of silly ideas in the past.

Get Goalside’s is an intentional, whimsical silly though. The silliest things you can do with advanced tech (2019); stealing ideas from other sports (2021); the list of things Manchester United could task their new data scientists with doing (2022 - “Optimal C-suite engineering (retain as few of your bosses as possible, as many as needed)”); silly decision-making that data scientists could do in-game (2022).

There was no whimsy in 2023.

Analytics history

Analytics is older than you (might) think. Although even the recent stuff is now old enough to be the subject of books and podcasts. And what do we mean when we talk about ‘analytics’ anyway?

History is a particular pre-occupation. Partly because Get Goalside, in some ways, is a record of analytics history. Then, building on that thought, it’s partly because I’m aware of the holes in my knowledge (and memory). Here’s a paragraph from the second piece linked above, a post partially about the book Expected Goals by Rory Smith, in which Chris Anderson heavily features:

“While editing this post I leafed through my copy of The Numbers Game. Although Anderson, who co-wrote it, had been a blogger, he was kind of in the 'early' section that feels slightly separate from what came later; before the famous Opta expected goals blog by Sam Green, before the StatsBomb blog took off in a big way. The closing chapter of the book features forecasts, one of which is that ‘Geometry – space, vectors, triangles and dynamic lattices – will be the focus of many analytical advances’. How smart would I have looked in the 'early analytics Twitter' era of blogging if I’d just repeated that over and over again?”

The point being: if we’re building ideas, we should, as much as possible, use what’s come before as our foundation.

And then in a slightly different vein, from the same piece:

“[M]aybe Moneyball’s Anglosphere legacy is larger than we thought, but, for example, what if any legacy does Nemzeti Sport’s early twentieth century data visualisations have in Hungary?”

History is as much about who is forgotten as who is remembered.

Special guests

Finally…

The best newsletters, like the best conferences, are a collection of people and ideas. Get Goalside has been lucky enough to have a bunch of smart people pop up and form the basis of the following pieces:

‘Can you teach tactics in a lockdown?’, ‘Bringing advanced data to the public’, Get Goalside #100, ‘What do we actually know about football’, and ‘Understand football and you’ll understand AI’.

I’ll end with something that Javier Fernández (ex-Barcelona, current-Zelus Analytics) says in the special 100th issue of Get Goalside:

“This sport has an incredible opportunity of becoming even more popular and even more enjoyable. Organizations will benefit immensely if they share more data; analysts need to prepare more and better use the data. We will all grow and enjoy more. xG is great. But football is not simple. Don't settle.”

Thank you everyone for reading.

Understand football and you'll understand AI

Tue, 23 Jul 2024 00:00:00 GMT

The famous Bill Shankly line is that football is a simple game complicated by idiots. AI is a simple concept complicated by almost everybody. What happens when the two come together?

Most versions of a 'football and AI' article would be about the who-s and the what-s of this collision, and there'll be a bit of that here, don't worry. But what's really important are the how-s and the why-s.

If you're looking for 'X team is the leader in artificial intelligence', you'll be disappointed by this piece; but if you want to understand how AI itself works, and why the basic mechanics of it are a lot like football, you're in luck.

What's in a name?

To begin with, let's leave the term 'AI' at the door. It's more of a brand name than anything at the moment, but maybe that's always been the case. Since the 1970s there have been variations on the following quip flying around: “As soon as it works, no one calls it AI any more.”

Instead, we'll talk about 'machine learning'. It's a more literal term, involving no philosophy about what 'intelligence' really is. You train a footballer to master a skill; you train a machine learning model to complete a particular task.

Simple. Bill Shankly may well approve.

Goals and goals

There will be one and only one piece of hair-splitting in this article. It's this: on the face of it, football is a sport about scoring more than your opponent; but to be specific there are two parts to this, scoring more and conceding fewer. If you don't believe that there's a difference, ask a fan of Tottenham Hotspur men's team.

Whole schools of tactical thought, from Ange Postecoglou to Jose Mourinho, are based on the idea that scoring more or conceding fewer is either more important or achievable than the other. Being specific about objectives is important.

For example, you can imagine how a TacticalTheoryBot might reach a different conclusion if the ultimate objective is 'avoid conceding' or 'score quickly'. If the latter, you might end up with something like route one football. If the former, avoiding conceding, you might end up with something like 'defensive possession', a more modern term for an old concept of killing a game off by playing keep-ball.

However, whether your objective leads to you lumping it long or passing it short, both footballers and machine learning modellers have to deal with the very low-scoring nature of football. In both cases, they need some kind of signal that they're learning correctly and whether what they're doing is good or bad.

In some cases it might work just to switch 'goals' for 'shots'. But in some, you've got to change your whole mindset.

Learning styles

We've said already that you 'train' a machine learning model, the same word we use for footballers. Like with humans, there are different methods of teaching and learning too. (As a tangent, I'd love to hear from education specialists on what they think about 'learning styles').

In machine learning, there are some broad categories that different techniques can fall into. There's supervised learning, where each bit of training has a label attached, like a tick or a cross; there's unsupervised learning, which is more about pattern-finding or grouping; and there's reinforcement learning, which is a bit more like human trial and error.

For some examples: creating an expected goals model would be a case of supervised learning, with the model being trained by 'goal or no goal' info for each shot. Examples of unsupervised learning would be anything on 'clustering', like this or this from The Athletic. However, for other parts of football, reinforcement learning can have its advantages.

Dr Pegah Rahimian is a football data scientist at Twelve Football and post-doctoral researcher at Uppsala University (PhD in Football Data Science). One of the research papers she's worked on (a Get Goalside favourite) used reinforcement learning to evaluate choices that teams make in spells of possession. She explains: “We cannot immediately see if, for instance, a backward pass on the left wing in a particular game state is good or bad in terms of increasing the chance of scoring at the end of a possession. However, we can design some short-term rewards for that pass which depends on our modelling approach.”

“For example, whether a pass was successful or a turnover, the expected threat value of the pass, et cetera. Having these short-term rewards and sequential nature of the game, we can model it as a Markov Decision Process [a mathematical framework] and use reinforcement learning to get an optimal action.”

As a general concept, reinforcement learning is similar to human trial and error. The model gets nudged towards better outcomes, but without needing a definitive 'right' and 'wrong' answer. When training footballers, you can't give them a list of situations and a 'right' and 'wrong' thing to do. But you can set them training drills, maybe focused on specific areas of the game, where a points system can give them some indication of what was better or worse.

There's probably even a machine learning version of a rondo.

But if there is, it'll involve data, not footballs. Because, depressing as it may sound, there are no sports in the world of a machine learning model. Only maths.

Knowing the game

“Data science algorithms in general are just sets of rules,” says Maia Trower, a football analytics consultant and PhD student in data science at the university of Edinburgh. “Computers allow us to implement complicated algorithms on huge amounts of data to get solutions to problems that weren't previously possible.” David Sumpter, professor of applied mathematics at Uppsala University and co-founder of Twelve football, said something very similar when speaking for this newsletter: "Machine learning is mostly statistics and fitting curves to data."

To see what they mean, let's take an example many people will have come across at school, the 'line of best fit' to a set of data. The maths here is quite simple: you draw a line, then add up how far each datapoint is from it; then you draw another line and do the same sums and see if it's closer. It's possible to do that as a human over and over again until you find the line of best fit, but it's a task that computers can do in a flash.

The 'maths' behind the algorithms isn't always as simple as this, but the point remains: the skill of doing machine learning isn't all about fancy equations. When grumpy football coaches or pundits say 'you have to know the game' they do sometimes have a point, and the same is very true with machine learning.

"Many algorithms come with a set of assumptions and caveats, and results or properties that are true if those assumptions are met," Trower explains. Although the algorithms might just be 'a set of rules', "someone has to develop the rules, and someone also has to prove that they work and how well they work." Which is where the professors and PhDs come in.

This is why they make you study maths at school, really. Someone's gotta grow up and understand the maths that underpins all of the data science in our lives, even if most of us will only be consumers of the tech that makes use of it.

One thing school doesn't force us to learn, though, is football theory. That's a problem for football data science. The best machine learning in football will always be done by the groups of people who understand both areas. When speaking with David Sumpter for this piece, he noted that “lots of things we observe in football are not in the data […] So I would describe the decision of the practitioner who works only from data as throwing away everything we know about football before we start!” (a complexity which he writes about further here).

Knowledge, it turns out, is important.

ChatGPT can make mistakes. Check important info.

(header taken from message displayed when beginning a fresh ChatGPT session).

Although this post has hopefully been readable so far, you really have to be applauded for going through so many facets of machine learning - the goal-setting, the ways of teaching a model whether they're on the right track, the value of maths - without even getting a mention of ChatGPT yet.

It's a genuinely fascinating tool. 'Tool' is an important word though, because behind 'ChatGPT' is the 'GPT' model itself (a set of initials we'll explain soon). Like all machine learning models, we can apply the previous sections of this post to it.

The goal of a GPT model - part of a wider family of 'large language models' (LLMs) - is basically to predict the most relevant next word in a sentence. “Essentially, LLMs are advanced probabilistic auto-complete systems,” as Dr Ryan Beal, CEO & co-founder of SentientSports, puts it.

When humans speak, we usually have some kind of thought in mind, even just a vague one, that we then put into words. The large language models are putting one word in front of the next based purely on what has just been said. This is where 'hallucination' comes from: it's a prediction bad enough to stick out like a sore thumb (or, more worryingly, not to stick out like a sore thumb even when it's wrong).

This prediction of the next word is also how they're trained, and explains why so much data is needed by the companies which create them. A model which is built on the relationships between words will always need to be able to examine those relationships.

“Beyond technical advancements, significant progress [in LLMs] has been driven by substantial financial investments in training these models,” says Beal. “Training GPT-4, for instance, reportedly cost around $100 million. Thus, the combination of innovative architectures and large-scale financial investments has been instrumental in achieving the current capabilities of models like ChatGPT.”

There may well be some financial parallels to football. I prefer not to speak.

But yes: alongside the heft of money bags and illegal data scraping there are genuine technical expertise in these models - the 'GPT' in 'ChatGPT'.

GPT stands for 'Generative Pre-trained Transformer'. A 'transformer' in this case is a type of machine learning architecture, a very sophisticated 'set of rules' for running data through a model. As with football tactics, these approaches have evolved over time, with the big leaps forward usually having pre-cursors that the older heads would recognise well.

To go back to the 'machine learning model as player education' metaphor, an 'architecture' of a model is like the intricacies of a pressing strategy and the teaching methods used to get a player to do it right.

“Transformers have been pivotal in analysing and generating text by understanding the contextual relationships between words,” Beal says, “enhancing tasks like translation, summarisation, and text generation through self-attention mechanisms that process sequences simultaneously.” Self-attention. Sequences. We don't even need to return to the 'training players' metaphor to apply this to football.

Connecting the lines

The choices about which machine learning approach and architecture you use touch on the end goals, touch on the type of 'mathematical problem' it might be similar to, touch on the type of data that you have. They can even extend to how you arrange that data. Fantastically, football is a perfect sport for understanding this.

'Tracking data' gives the positions of the players and the ball in a way that looks like the top-down pitch map views on Football Manager or FIFA. You can view this data as an image, or you can view it as connections between points.

"If you look at tracking data frames, you can look at the players as one graph," explains Amod Sahasrabudhe, a machine learning engineer at Gemini Sports Analytics. A 'graph' in this case is another mathematical term; a collection of 'nodes' (think of circles) joined together by 'edges' (think of lines). "If the image representation allows the model to look at a 2D representation (of the position of the players and ball), the graph allows you to encode additional information like velocity of the player, acceleration, distance from each other, et cetera during the play."

Isn't that fun? It's like a real Transformer - the Optimus Prime kind, not the machine learning architecture - where the same thing can be configured differently for different uses. Some information will be able to be picked up by both types of data lay-out, but others might be easier to work with in one form or the other. “For example, if there are two nodes in your graph that are connected, you could pass in the distance from each node along with the angle created with each other,” Sahasrabudhe says. “You can also represent the velocity difference, acceleration difference which I think an image based representation might not be able to interpret.”

Let's take a brief pause.

In a second we're going to drop 'neural networks' in, and we're not going to break stride when we get there. The pause that we're taking is not because we need a run up, but because we don't need one. Life is full of complex things that everyone grasps the gist of (gravity, evaporation, Spursiness). There's no reason at all why machine learning shouldn't be one of them.

Just like how training drills can teach footballers to improve their technical or tactical play, training a machine learning model involves repeating a task until it starts achieving the outcome you're after. Picking that outcome might influence the approach you take. The approach will involve a machine learning 'architecture', much like deciding on the details of a training session. A neural network is a broad category of architecture (based on ideas about how the brain works with lots of 'nodes' which interact with each other). The nodes in a neural network do maths, which isn't necessarily complicated when looking at one single node, but becomes incredibly powerful when scaled up, and lets the model learn complicated, non-linear relationships. Modern computing power lets us do that to a huge degree.

And 'graph neural networks' are a version of neural network where the data going into the model is arranged as a 'graph', a concept that basically looks like a football passing network. These 'GNNs' are also almost everywhere in the modern world.

"Graph Neural Networks excel in tasks involving complex relational data, such as social network analysis and recommendation systems, finding patterns that traditional neural networks might miss," says Dr Ryan Beal. And after barely 2500 words of this post you'll understand a summary of a research paper Beal co-authored, which used graph neural networks to help estimate player locations purely from event data (a type of data which usually only has player locations for the person making each 'event', like a pass or shot).

Fun, huh. And we're almost at the end.

Being specific about objectives, choosing a data format, choosing a machine learning framework. It's all in the service of achieving something. Here's Maia Trower again, on what a data scientist might be thinking about when coming to a data science or machine learning task:

“When choosing an approach, I'd say the three things I'd think about are: One - what is the exact problem or question, and how will I evaluate whether or not I've solved or answered it? Two - what data do I have access to? Three - what models do I already know about that might be appropriate?”

Before the recent wave of generative AI buzz, data scientists were already applying these questions to techniques like regression models and neural networks. Now there's a new type of tool in the arsenal.

The right tool for the job

Talk to a smart mathematician or data scientist and they'll tell you a lot of smart stuff about maths and statistics. But talk to a really smart one, and they'll tell you when all of that won't be necessary.

Towards the start of this post, we talked about 'defensive possession'. Sometimes the obvious objective (have the ball = score a goal) isn't the real objective (avoid conceding). Similarly, creating the most sophisticated machine learning model might not be the true aim. It'll usually be to help a coach or player improve something. Those two things might sound like the same thing, but they can be as different as Angeball and Mourinhoball.

The fact that something is a 'tool' is important. It's important in ChatGPT, where OpenAI clearly built in a lot of safety rail-type features. Well-known examples are around its refusal to use slurs or give out bomb-making recipes. Guardrails could also help the 'hallucination' problem, but won't be enough.

Remember when we said that humans turn thoughts into words but ChatGPT just lays them one after another based on probability? A technique called 'retrieval augmented generation' (RAG) adds a little bit of actual structured 'thought' in and around the language models.

To re-phrase the syllable-heavy term, the process augments (improves) the generative language model by retrieving some information (from a database).

This way, new information doesn't need to enter into the large language model's training data, where it would need to muscle past out-of-date information in the melee of probabilities.

When speaking with Dr Ryan Beal for this newsletter, he pointed out that the GPT3.5 model underpinning ChatGPT still thought (at the time) that Harry Kane played for Tottenham (he'd moved to Bayern Munich several months earlier). For the language model to talk about Kane being at the right club, it would need to be re-trained with all its data. A retrieval augmented generation system, on the other hand, would fetch the data and then use the language model like an interface or conduit for the updated information.

David Sumpter has an interesting framing of this kind of language model usage. “When we use LLMs, we don't use them to model the world,” he says. “We use well-defined statistical models and mechanistic models. Instead we use the LLMs to communicate about our models.”

“You should think of it like a visualisation, like a shot map. A visualisation does not tell you the single truth, but is a reasonable representation of the data. We use the LLM in the same way, to get a reasonable representation of the data.”

The who-s and what-s

Some of the who-s of machine learning in football are, of course, the contributors to this piece. There are others too, and the biggest English clubs are still steadily building up departments for it.

Money doesn't guarantee good work, though. Ultimately, the real who-s and what-s will simply follow on from the how-s and why-s. The best work will be done by those combining the best domain knowledge (how) with the best questions (why).

Plus a little bit of creativity. “Skill without imagination is barren,” wrote Walter Isaacson about Leonardo da Vinci.

Skill and imagination. That's basically all there is to it.

Many thanks to Dr Ryan Beal, Dr Pegah Rahimian, Amod Sahasrabudhe, David Sumpter, and Maia Trower for their help in writing this piece

Notes

Engineering the (super)market

Tue, 14 May 2024 00:00:00 GMT

An aside before starting: I’m still working on what I hope to be some really interesting pieces on ‘AI’ in football, new data sources in football, and data in women’s football. If that sounds like your domain, or you have other topics you think Get Goalside should cover, please get in touch, to this email or getgoalside.newsletter@gmail.com

They’re taking a while to put together; the piece that follows is looser than I’d ideally like these newsletters to be but I hope there are parts of it that are useful to people.

If you watch the Premier League or WSL, you’ll have heard of Infinite Athlete. Which means it’s £40million well-spent to get on the front of Chelsea’s shirts.

The company say they want to be the ‘operating system for sports’: in a City AM interview, CEO Charlie Ebersol references Intel; another piece of promo uses the Google Maps API as a comparison. Both have bugged me, because I can’t quite get a grasp on what they mean, but to Infinite Athlete’s Infinite Credit, they’ve given me an excuse to think about metaphors for data.

Because I am, at best, two updates an update away from being completely superseded by an OpenAI model, I find metaphors useful as a sort of categorisation test. If you don’t quite have your head around something, you can toss up a bunch of metaphors and say “is it more like this or that” and work out where you are.

Although I don’t quite understand Infinite Athlete’s chosen metaphors (or whether comparing yourself to both Intel and Google Maps works), trying to come up with one that I can wrap my head around is interesting. And it’s hit upon one that feels like it has a wider usefulness.

I think that what Infinite Athlete are getting at is that they see sports data as a resource (like Intel did with electrical current or Google did with ‘where roads are’) which they can refine in some way (like Intel does through microprocessors or Google Maps does with data and API maintenance) to facilitate other things being created. They want to sync different data sources, all meshing nicely in one API. They want to sell you the convenience of your data in one place.

Like a supermarket?

It’s not a perfect analogy, but it provides a surprisingly good framework for football data as a wider landscape.

There are lots of different types of data that football organisations need. They come from all over. Some of it is even well-documented. But it’s a pain to traipse from market to market, picking up your event data groceries from one place and your physical data from another. And tracking data— you know that you should eat your greens, but frankly the fruit stall is up a hill and you don’t know what to do with pomegranates anyway. Your father never used tracking data and he turned out all right, scurvy or no scurvy.

Data companies, like specialist grocers, know that most customers will prefer either refined versions of the raw ingredients or some kind of ready-to-go product. Pre-sliced bread from the baker, pies from the butcher, stats portal from the data provider.

Sometimes there will even be crossover between grocers who’ve come to agreements between themselves. Deli meats and bread at the cheesemonger’s; Sportscode-ready data export options; a specific, strategic set of entity ID matching.

But wouldn’t it be easier if everything was just in one place?

It would. But that would also probably require a much different relationship between data collection companies and the people buying the data (or data products). The centralising party would become the main focus and contact point for the customer, and it’d probably benefit the ‘supermarket’ if customers could mix-and-match their purchases more easily. Like how the Apple ecosystem works, the pain of switching is a small reason for staying put. (And I should mention, data collection costs are high; you can understand why companies want to hang onto their customer base).

Something vaguely similar happens at football organisations beyond the pitch as well. The commercial operations also have a lot of data from a variety of sources to deal with: membership, tickets, merchandise, any other types of fan engagement data. Although the ‘football’ and ‘business’ datasets don’t need to mix with each other, the problems in both are similar.

Like with the footballing data, the cost of putting everything in the same place, let alone analysing it, is prohibitively high for many clubs. It was a point made multiple times on a recent episode of the Unofficial Partner podcast. “There are so, so many [clubs] and the vast majority of them are quite small businesses,” said Charlie Marshall, managing director of the European Clubs Association (ECA). “There is no economic sense whatsoever in hundreds and hundreds of small businesses all trying to build their own quite sophisticated data strategies, and the technology systems that support those data strategies.” [episode timestamp, ~1:01:00].

The discussion was about business data, and although the participants in that panel drew a line between that area and what happens on the pitch, the same is kind of true there too. It doesn’t make sense for every team to be doing very similar, boring things to build out their football data strategy. As some leagues have recognised: La Liga developed Mediacoach; the Premier League offered a combined feed of Stats Perform event data and Second Spectrum tracking data.

Those are quite wealthy leagues though, and even their solutions are a distance from a fully-fledged Sainsburys or Big Asda. Who’s helping all these clubs do their weekly shop?

You have Infinite Athlete, who have supermarket aspirations, but it remains to be seen whether they’ll be successful. You have data providers who, as discussed, often have their own product range alongside their basic ingredients. And now, increasingly, you have the big cloud services providers.

The cloud providers don’t appear to have direct supermarket aspirations, but they have all sorts of add-ons that can help manage and analyse the data once you’ve hooked it up. (Like a high street?). AWS has a bunch of promo about the sports organisations who use it. Both Databricks and Snowflake have similar content marketing you can read. Oracle, another cloud services provider, sponsor and service Red Bull’s Formula One team but don’t hold that against them too much.

The twin paths of sports’ increasing interest in data and sports’ continued rise as a business proposition has caught Big Data’s eye. With that comes Big Investment and Big Start-Up (Cerberus or the dragon meme, depending on your viewpoint).

However, the space that these new companies can exist in - as ‘third parties’ outside of football organisations or data providers - will differ depending on the landscape of Big Data Engineering. (Infinite Athlete, to briefly return to them, exists to try and be the solution to the Big Data Engineering problem, which itself still relies on a certain formation of the wider engineering landscape).

A question for Get Goalside readers: Which of the following is the more likely winner of the next three-to-five years?

Interoperability between data providers becomes seamless on its own, allowing for integration of different data sources within a provider’s own product, or allowing for foolproof entity matching between any provider to use data in third-party applications like Tableau
Organisations will turn to cloud providers like AWS for API integration and setting up data storage, either through some (semi-)automation (AI anyone??) or as an affordable managed service
The above, but provided by domestic leagues or national FAs
The scale of the task will have simply shrunk enough for clubs of all sizes to hire employees for the set-up and maintenance of data pipelines, and creation of internal tools
None of the above, it’ll be as complex as always
Something else

If data isn’t easy to access across providers and across ‘users’ (organisations or the outside products or services they might be interested in), then either data engineering will be part of the product/service or there’ll be annoying integration friction (or both). This has echoes in a related business area:

“Looking back and assessing where we are, there’s a clear winner in Enterprise AI. Bolt-on, good-enough ML, sitting besides data storage and data processing has proven to be the option the market wants and everything else trails by a wide margin.

Standalone AI/ML-focused offerings plunge teams into an endless cycle of POCs and procurement woes, where our best technical minds spend their days not innovating, but rather, navigating customer problems and bolstering operational capacity.”

— ‘The Hidden Cost of AI as a Service’ blog post

Convenience is, well, convenient. You could put a price on it. Infinite Athlete are banking on that price being something worth paying (hopefully enough to make back that £40million they spent in shirt sponsorship).

But it also matters because an API is harder to deal with than potatoes. You can parboil some spuds, heat some oil in a pan, add some garlic and herbs and make a delicious side dish. Try doing anything as efficiently and easily delightful with a football data API as good roast potatoes. Football data has a steep learning curve and arguably an even steeper implementation curve.

Data gets talked up in regards to scouting as a big time-saver, but there are an awful lot of game model-related things that coaches and analysts code up themselves which you could almost certainly automate with tracking data. But that’s out of reach to implement for almost everybody.

I go back to what Charlie Marshall said again: “There are so, so many [clubs] and the vast majority of them are quite small businesses.” All of these small businesses are having to work out how to use data, but many will not be able to work out how to use it (on their own) unless they’re able to actually use it. But their ability to use it is limited by either engineering costs where no provider-produced product exists, or the confines of the provider-produced product where it does.

Whichever way you look at the situation, it’s both understandable and slightly weird.

What we talk about when we talk about 'analytics'

Wed, 27 Mar 2024 00:00:00 GMT

Moneyball

Every movement has important dates. Football analytics has a few: 2003, the year Moneyball was published; 2014, the year of the first Opta Pro Forum, an event which has a stacked array of former presenters; October 2015, publication of the ‘air-conditioned offices’ article in the Daily Mail.

If you’re unfamiliar with this particular part of analytics lore, the basics are worth knowing. Brendan Rodgers had just been sacked by Liverpool, amid rumours of discontent with a ‘transfer committee’ at the club infiltrated by nerds. Two paragraphs in the middle of the piece encapsulate the whole:

The increasing influence of analysts, young men who have no experience of scouting or recruiting players, has meant the end of the road for good football men such as Mel Johnson. […]

Instead a new breed sits in air-conditioned offices, cutting up videos from matches all over the world and burying their heads in the stats. [Michael] Edwards, along with his vast team of analysts, constantly monitors the opposition, providing detail about playing positions, style, routines, set-pieces and other important matchday information.

This is what analytics was seen as at the time. On one hand, a cosy obsession with spreadsheets that was an opposition, or even a threat, to traditional ‘football men’ and their hard-earned expertise; but on another hand, not just a new hobby but a potential career.

The tradition-vs-nerds framing was always a suspect one, even if people did clash, and now it’s almost totally fallen by the wayside. So what do, or should, we talk about now when we talk about analytics?

Laptop analysts

Let’s briefly return to the 2015 article:

They [Edwards and his ‘vast team’ of analysts] profile players based on their last 10-20 appearances, gathering information and helping Rodgers build a presentation for his players before matches that was usually a maximum of 10 pages on each team. It is a useful, but far from infallible, tool.

For an article that was so derided, even these three paragraphs give more information about day-to-day work than you’d expect. And it’s not exactly outdated nine years on.

Opponents are monitored for playing styles and patterns (set-pieces still a key point of interest) and information drawn into reports for coaches. Teams’ reports will differ based on the ‘vastness’ of the team of analysts, their own style of play, their priorities in how to approach matches, how coaches digest information, and the tools and data available.

This is just one part of what we’ve always called ‘analytics’, with models like expected goals being another part. For those new to modelling, a 101 summary would be that statistical techniques are used to find the value of various features, and the resulting ‘model’ is tested (and re-tested) on a new set of data to see how it stacks up.

There is a degree of human decision-making, although ‘subjective’ might be a slightly misleading term for it. A data scientist does choose what to put in a model, but they can (if they want) just throw everything at the wall and see what sticks, before discarding what doesn’t. The choice of modelling techniques, deciding how to approach the problem at hand, may also be important.

But in 2024, the reach and scale of data is even broader.

More data, more data

Nine years on from the ‘air-conditioned offices’ article, it’s both technologically and culturally much easier to have data around. Inevitably, there are also more data companies around than ever, and more data technology companies around than ever.

Part of the reason for this is the cultural change, of course, and part is a broad technological advancement. Computer vision improvements have helped to collect data more easily, and methodological improvements have helped with both the collection and analysis (particularly with tracking data, on both counts). ‘Line goes up’ financial factors mean everyone sees an investment opportunity too: even if not data-specific, a timely example is the recent announcement of JP Morgan’s sports-specific investment team.

However, this preponderance means that "we’re kind of at a point in the evolution of the industry where people know enough to be dangerous." Those are the words of Sarah Rudd — formerly of Arsenal, now co-founder and CTO of consultancy src ftbl — on a recent panel at the MIT Sloan Sports Analytics conference. "People can get their hands on data and information that’s outside what’s being curated within the club, so you have to go through that education process with a lot of people."

In some cases, that might not even be outside sources of data. On a recent episode of the Winning with Data podcast, Parma’s chief performance and analytics officer Mathieu Lacome described a perhaps unexpected way that clubs can badly use data. "You [can] have a club that starts to buy everything, every single source of data, every single piece of software without really understanding what you can do with it. That ends up with overspending and very low usage of the technology to do something valuable for the club."

This is the important distinction to make, that ‘analytics’ isn’t simply having data around. Think of it like the ‘shot on an iPhone’ adverts compared to what most peoples’ phone videos look like: just because everyone has a high-quality camera in their pockets doesn’t mean that everyone is ‘doing cinematography’.

The death of ‘analytics’

"This is a book about analytics." opens The Midrange Theory (a book about basketball rather than football). "I hate analytics."

The author, Seth Partnow (who was previously director of research at the Milwaukee Bucks), continues, "The word has become hopelessly poisoned, reduced, confused, and misapplied." What analytics is is written about a few paragraphs later:

Analytics exist at the intersections of math, statistics, and computer science. However, those are merely the tools rather than the field itself […] Rather, it is a mode of thought seeking to reduce the impact of the cognitive biases we all suffer from. In a world wrought with imperfect information and uncertain outcomes, it is about putting oneself in a position to be less wrong.

This is the thing. You can think analytically without using data, and you can use data without thinking analytically. ‘Analytics’ is where both come together, not with the aim of being perfectly correct, but to be more right, more often.

The old ‘reports and models’ understanding of ‘analytics’ as a term does fit neatly into this. Reports help coaches get up to speed quicker, with a consistent methodology, providing a useful context, perhaps challenge, to their own opinions. The models seek to represent the sport more accurately, more succinctly, or more efficiently than the human eye and brain alone can.

Why are we saying that things have moved further than just ‘reports and models’ then?

When describing the human involvement in data modelling earlier, interpreting the results was left out very deliberately. How you interpret a model’s results is clearly important to whether or not you’re ‘being analytical’, but an excellently trained and validated model could still be used badly wielded. Part of the battle is avoiding that.

So, is ‘how you package your reports and models’, ‘analytics’?

‘Please buy my product’

This is where Get Goalside might be partly guilty of stretching the term ‘analytics’ to breaking point. Because if the tech wrapper around the numbers is ‘analytics’, then where do we end? "Not only is ‘understanding football’ an important skill for ‘analytics people’," GG wrote in the previous newsletter, "increasingly so is ‘understanding business’ and ‘understanding management’."

reader, sobbing: You can’t just point at everything and call it ‘analytics’! Get Goalside, pointing at employee-empowering policies like flexible working, high wages, and professional development budgets: Analytics.

But seriously. To take from the Midrange Theory quote, anything directly connected to maths or statistics is probably ‘analytics’. Anything else is probably in a category of ‘analytics implementation’. You can’t really ‘do analytics’: you can do an analytics project, and you can implement analytics. Both can be done badly or well.

Both categories have also seen changes since that article in 2015, although covering those fully would take, like, a book. Possibly Ian Graham’s upcoming one. More briefly, and less Premier League-winningly, let’s try summarising each in a paragraph.

The raw data (of all types) going into reports and models is of better quality in most cases. Possession value models (like expected threat or on-ball value) are now fairly widespread. There will have been far more research at the top end using tracking data, which lets you investigate space and team structure, and pitch control models are knocking about but not exactly common. Other research projects will have been embarked upon, and if an organisation started doing projects on career progression topics back then, they’ll now have a strong longitudinal dataset on those intriguing problems.

For implementation, there are simply far more people able (and being asked) to do the implementing. The packages that come with the data itself, from providers (or from/via competition organisers), have moved on, in some cases even merging datasets to make them easier to work with. This merging of datasets can also be done in third-party tools, but clubs are also building out more tech capabilities of their own. As Sarah Rudd alluded to, it can be sort of like the Apple ecosystem: you want to keep peoples’ data consumption streamlined and in-house (although for the organisation’s benefit, rather than your own profit).

As an example, Manchester City — sorry, City Football Group — are currently hiring for a Machine Learning Engineer who will, among other things, "research & build machine-learning models to identify and interpret complex patterns and structures in data and use this information to understand how teams play." Liverpool have their famed department; Arsenal have theirs too, whose lower public spotlight presumably rests on silverware and media access. Some might say this is the real prize at stake in this season’s Premier League title race.

Little England

If Get Goalside has a fatal flaw it is being tragically English.

For a while, this was forgivable; even understandable. It was on the English-speaking blogs that expected goals entered the lexicon. (The English phrase, and English abbreviation ‘xG’, still gets used in other languages). North Americans featured heavily, but the data was all English (from Opta), with access to it often coming via English stat reference websites WhoScored and Squawka, and the app StatsZone which was associated with English magazine FourFourTwo at the time.

It’s not that Opta was the only data company around, even in England (as former employees of Prozone have previously been quick to write in about) but it was at the front of the public conversation, helped significantly by the annual Opta Pro Forum analytics conference.

Fast-forward to the present, and the week in which Opta’s latest Forum took place also saw the fourth edition of the Sports Data Forum in Seville and the third edition of the DFL’s SportsInnovation trade fair, in Düsseldorf. At the latter, the DFL announced a continuation/expansion of their partnership with Amazon Web Services (AWS), as their official generative AI provider.

On a similar theme, Major League Soccer launched a sports tech start-up incubator scheme — MLS Innovation Lab — at the start of the year (reminiscent of NBA’s Launchpad program and Tennis Australia’s AO Startups, to name just two).

AO Startups is possibly a better comparison to the DFL and MLS’s schemes than NBA Launchpad is. In those three non-basketball cases, the support for innovative ideas and companies benefits the sport (and sport) as a whole, but also aims to keep their event(s) ahead of the competition. MLS may be a closed shop itself, but it can’t regulate its outside competition (although it sometimes tries its luck).

The embrace of data and ‘analytics’ by league bodies doesn’t have to mean startups though. Current Bournemouth manager Andoni Iraola recently praised Mediacoach, a data platform developed by/in collaboration with LaLiga. This was a quote that Sportian — the new name for what was previously called LaLiga Tech — took note of, adding to their tweet "Is it any wonder LaLiga is producing the world’s top coaches?". Back in Germany, the DFB announced in September that the 3. Liga and Frauen Bundesliga would be brought into their partnership with data company Sportec, which also includes a match analysis hub. It’s not exactly Apple but it’s a tech ecosystem; one wonders whether the NewCo for English women’s professional football could take inspiration from some of all this.

And leagues aren’t the only entities giving a boost to start-ups. Arsenal, Barcelona, and Real Madrid (among others) all have or have had schemes of some sort, the latter currently in the beginnings of an Asia-specific accelerator programme. Clubs also come to types of agreements with young companies in other ways — Liverpool (them again) reportedly collaborated with tracking data company SkillCorner to help improve their product for a year prior to an official partnership. (SkillCorner are also French; more internationalism).

The point is, the modern football data industry is (proportionally) less ‘show me your math slide’ and more ‘move fast and break things’.

‘Moving fast and breaking things’-ball

You thought we were going to get to the end of this without mentioning Moneyball again?

In 2015, football was still searching for its version of the Oakland As. Then things got confusing, quickly. Leicester City won the Premier League and people tried to sell FC Midtjylland as football’s ‘Moneyball’ and neither really hit the spot. Liverpool won the Premier League and Champions League with the same Roberto Firmino that the Daily Mail article had maligned, but Liverpool were also too rich and storied to be ‘Moneyball’.

And now, nine years on, the three (rich) teams vying for the Premier League title may have the strongest data outfits in football. Manchester United are in on the act too. There’s still room to punch above one’s weight, but it’s mainly at the margins: promotion, Premier League midtable, mainland Europe. You could try winning a prestigious American tournament like the US Open Cup, if that’s your thing. (Sorry, being tragically English again).

The point is overdone for effect. The underdogs have less room for shocks if the big dogs are using data well. Moneyball (movie) has the line "if we try to play like the Yankees in here, we’ll lose to the Yankees out there" — whaddya do when the Yankees are taking inspiration from Moneyball?

The thing with ‘Moneyball’ is that, like ‘analytics’, it was always flattened and compressed until it fit into a buzzword. What we call ‘analytics’ is basically using data and maths in an analytical way. You’d be hard-pressed to find a reason why it couldn’t simply be folded into the umbrella of ‘sports science’; after all, rigour and evidence and testing is part of the scientific process.

And the thing with Moneyball is that it wasn’t about data per se. Get Goalside has argued in the past that the book is actually largely just about Billy Beane. What it’s about is questioning orthodoxies and finding edges, finding the most efficient way to get wins that you can.

So whaddya do when the Yankees are ‘doing analytics’ too? You’ve just gotta try and find another edge.

What do we actually know about football?

Thu, 29 Feb 2024 00:00:00 GMT

Let’s take stock of where we are.

Expected goals has been on Match of the Day since 2017, two years after Arsène Wenger made waves by referencing it in a press conference. Between those two events, we got ‘pitch control’, a way of looking at a football pitch like a multi-dimensional tug of war between all 22 players on the pitch.

In 2018 and 2019, the public (or the public who avidly read papers from the MIT Sloan Sports Analytics Conference) got glimpses at exciting possession value models. In 2021, Opta - one of the largest data companies around - brought out their own Possession Value model. Later that year, Statsbomb (another data company) did the same.

So why, three years later, is a football analytics newsletter even asking the question ‘what do we actually know about football?’?

Because Pep Guardiola doesn’t use his substitutes.

Undersubscribed

When competitions like the Premier League were thinking about making the covid-induced ‘five substitutions allowed’ rule permanent, there was a lot of worry about rich teams. Would teams like Manchester City simply blow poorer opponents away by bringing on even more talent? How would we know, when there are games when Guardiola makes literally no substitutions?

Like a lot of things Guardiola does (apart from, perhaps, his clothing choices) the substitution-aversion seems to be catching on locally. The Premier League has consistently given substitutes less game-time than other major European men’s leagues but, unlike the pre-pandemic years, its now on its own in the extent of this. As Michael Caley notes in his newsletter, Expecting Goals:

Premier League teams only manage to get a fifth substitute into the match about one-third of the time under the new rules. In one match in ten they don’t even get around to the third substitute, while such a choice would be vanishingly rare in Serie A or La Liga.

This seems counterintuitive, right? That the richest football league in the world will, in one match in ten, not even use half of the substitutions available to them?

But then again… “I think there is a whole different strategy component to substitutions that no one has figured out yet,” Lucy Rowland, who’s worked at San Jose Earthquakes and Canada Soccer, told Get Goalside over email. “What time frame would be the most disruptive to your opponent for you to sub? Or is there a clear trigger in their tempo that will make you want to sub to stop that tempo. Do you sub all 5 players at once to give them a chance to find cohesion?”

Caley’s analysis has found that subs play more minutes when teams are losing (£). This fits an “if it ain’t broke don’t fix it” viewpoint, but still leaves Guardiola - who often seems to avoid subs when his team is struggling - as an odd exception.

It’s not beyond the realms of possibility that there’s something the Catalan intuitively understands that the rest of us don’t. Maybe he’s thinking along the same track as Rowland, that making substitutes when things haven’t yet clicked might actually disrupt the rhythm he’s worked in training to cultivate. But how do we find that out?

Subscribe to the newsletter

Vibe check

As we’ve seen, a simple question around substitutes opens other doors too, to things like tempo, team cohesion, momentum.

Those are three things that you could, if you wanted to, file as different varieties of ‘vibes’. Although analytics people tend to scoff at the emphasis put on ‘the intangibles’, that doesn’t mean they’re afraid to try and investigate them. There’s a paper here from 2020 on quantifying and predicting team cohesion. And there was a paper from the first Statsbomb conference in 2019 - whose authors may be familiar to Get Goalside readers - that tried looking at tempo, among other things, when breaking down a set defensive block. If running data is your jam, consider the following line from a 2014 paper on the physical output of substitutes, on further work that could be done:

It could be possible that substitutes are covering more relative distance [than starters] late in the game, but these attacking runs are less effective as the fatigued players on their team cannot respond to their runs.

Fascinating, but dastardly difficult, questions to get to the bottom of.

Tempo in particular is something that has been brought to the forefront of Get Goalside’s mind by the arrival of Roberto De Zerbi in the Premier League. It’s hard to watch his defenders stand with their studs on the ball and not think about it. But now consider statistical modelling that needs an outcome to aim at, usually selected as goals or shots, an ideal outcome that is very far from where De Zerbi’s players dawdle on the ball. How do you fit his strategy into a modelling technique that only counts a high-xG chance within ~10 seconds as ‘success’?

And, more broadly, isn’t it strange that in a sport that averages only three goals per game, teams spend so long… waiting?

This is probably not a problem of football players and managers spectacularly missing the point of their own sport. It seems much more likely that it’s a problem of analysing it.

Statsbomb CTO Thom Lawrence has previously talked about how “different players at different parts of the field have access to slightly different rewards.” Various types of modelling have repeatedly found that the ‘real estate value’ of areas of the pitch shoots up very quickly near each goal - perhaps best displayed in Karun Singh’s fantastic blog post, which coined the term Expected Threat.

Now, if the value of moving forwards in midfield is close to nothing, you might think that the logical thing to do would be to get the ball into the penalty area as quickly and frequently as possible. But that can’t be right. For one, because English football and analytics icon Charles Reep basically proposed that fifty years ago and was dismissed by modernity. For another, it’s not how the best teams in the world currently play or, really, have ever played. Something must be missing.

Perhaps the approach of this paper on modelling in-possession decision-making, from 2022, is a way to go. It uses different aims for different phases of play rather than just focusing on whether a goal might be scored or conceded in the next X seconds, as many possession-value models do. At the very least, it seems a little more like how football’s elite coaches think about the sport.

What we do know

It’s about time we talk about some things that football analytics does well.

“There are so many confounding factors in a team’s success that it’s really difficult to come up with an optimal way to play,” says Sarah Rudd, former vice-president of analytics at StatDNA and Arsenal and co-founder of analytics consultancy src ftbl. “What we can do, however, is keep chipping away at the factors of success – how do we create advantageous situations on the pitch and how do we avoid them.”

“I think the areas of soccer that are pretty well understood are penalty-taking, passing patterns and passing progression in build-up,” Rowland said. (And it seems quite significant that, after heat maps and varieties of radar charts, passing networks are probably the most popular data visualisation - there’s something meaningful that data captures there).

A line that football coach and founder of Spielverlagerung Academy, Martin Rafelt, said to Get Goalside about coaching struck a chord here: “More generally you could just say that things which are repeatable and consistent are well understood.”

“You want consistent effort, running, immediate transition, you always want to control space, always want to put pressure on the opponent, stop them from progressing into the centre and always want to not lose the ball, have clean passing and back passing options, always want to be well protected against counter-attacks.”

And in a wonderful point of synergy, we can bring pitch control back into this. Professor and co-founder of Twelve Football David Sumpter has previously spoken about using it to develop a strategy to protect against counter-attacks at Swedish club Hammarby, alongside the coaching staff.

It makes sense that repeatable actions and control are areas where coaching and analytics seem to align. Data people love a good sample size, or fear its absence, so anything repetitive feels like dry land (and is probably more automatable in reporting). Control, meanwhile, follows a simpler logic, or produces simpler probability curves, than chaos.

Improvements in the type of data available helps with this too. “Data sources such as tracking data, and more specifically broadcast tracking data, are really changing those types of discussions,” says Rudd. “Lots of things that couldn’t be measured in the early days are now being measured in a way that people feel confident in.”

It all adds to mean that “one thing that football analytics is really good at these days is accurately profiling a player,” she says. When a single video feed is able to provide the basis for event-based statistics, and physical and spatial metrics through tracking data, you can see why.

Subscribe to the newsletter

The long path of history

It’s worth taking a moment to reflect on how much things have changed in the sport. “Nowadays you can walk into just about any club in the world and people will be at least familiar with xG,” Rudd says. “That wasn’t the case when I got started [in 2012 at StatDNA]. I don’t think there’s a debate any more about if analytics can be useful in football, but there’s still a debate about how useful and in which applications.”

The simple fact that practitioners like Rudd and Rowland have established careers is a factor in the development of this knowledge too, and the sense of institutional knowledge around tracking data is growing as well. If you leaf through past papers from the MIT Sloan Analytics Conference or the (gone but not forgotten) Barça Innovation Hub Analytics Summit, you get work on automatically identifying formations, off-ball run categorisation, space occupation gain, and automatic corner kick categorisation.

None of those things necessarily help ‘understand’ the sport better in the way that expected goals models forced a reassessment on shooting locations, but they’re the bedrock of future analysis. And are already being turned into data products.

To phrase that all a little bit differently, even if we don’t know that much extra about football, we can definitely describe it better.

That said, while describing things better is a good thing, introducing new data as a means of achieving it creates its own questions. “In the early days we were really limited in what predictive models we could build,” Rudd says, “because we just didn’t have enough historical data, so everything was very descriptive. That’s changed for some data sources, but every time there is a new data source, you enter that cycle of starting off descriptively while you wait for the data to accumulate to build something more sophisticated.”

Where data has been consistent, the passage of time is a great thing. “That's the other thing about doing this work now,“ said Caley, of his newsletter Expecting Goals. “We've got nearly 15 years of statistics that have been collected [and available publicly] and way more leagues [than before].”

And bigger sample sizes matter because sometimes the data is messy for reasons outside of collection techniques and randomness. “If you've got thousands and thousands of nineties [minutes played by a player are often grouped into ‘nineties’ as a consistent denominator for analysis], some of those players are getting better and some of those players getting worse,” Caley continued, “some of those players are in a good mood, some are in a bad mood, but all of those things wash out over large enough samples such that you can start to make clearer claims [about trends].”

Rejecting the premise of the newsletter

Data being consistently collected, though, is not consistent throughout football. The richer leagues will tend to have the best and most-immediate coverage, and are the ones where data companies might consider delving back through the video archives to collect extra, historic seasons of data (if, of course, video archives exist, which has its own inequalities).

That goes for geography but also, unsurprisingly, for gender. “If you’re looking at a player who’s 18 and you want to project who they’re going to be when they’re 24,” Arielle Dror, director of analytics at NWSL team Bay FC, recently told The Athletic, “I mean, it’s hard even with the amount of data we have on the men’s side, but it’s certainly not possible with the data that we have on the women’s side.” Part of this, as Rudd pointed out to Get Goalside, is that women’s leagues often have fewer teams and fewer games, meaning that even when data is being collected, the sample sizes are usually smaller.

This topic obviously extends to data modelling. Statsbomb have previously presented on their work investigating gender-specific or ‘gender-aware’ models (the former being when only data from men’s/women’s football is included in the model; the latter when a model uses all data but has an indicator of which it came from). The upshot is that their gender-aware models performed better - which seems obviously better for analysts in women’s football than using the analytics equivalent of shrunk-down male football boots of xG trained on the men’s game.

However, even this raises a further question, about where and why you’d draw these lines. Would you use an expected goals model largely trained on adult men’s football for youth men’s football? (Here’s a recent paper looking at shooting in different levels of German men’s football that touches on this very subject). If differences between the men’s and women’s game stem from sex-based height differences, for example, would men’s football in shorter parts of the world also benefit from context-aware models; if the differences stem from gendered differences in funding, does that have implications in similar global circumstances?

Any headline containing the word ‘we’ is making an assumption about who ‘we’ is; but maybe ‘what do we actually know about football’ was also making big assumptions about what ‘football’ is.

Bored-room

There are some, of course, who still say that football isn’t a game played on spreadsheets. And maybe they’re right. Balance sheets are where it’s at. (What’s the old adage… Offence wins games, money wins championships? The most important data scientist is your accountant; less p-hacking, more gr€€n-hacking; etc).

Analytics people have long put their value in these terms, to be fair. There’s the old line about ‘I could save you millions just by sitting in the corner and vetoing bad transfers’. But perhaps wider squad composition is the next step.

“I think some of the areas we really struggle with are what is the optimal tactical approach for a club given the league, their budget, etc, and how does that impact squad-building,” Rudd says. Coming full circle to the impact of substitutes, “that also leads into roster composition,” Rowland says. “If research shows that subs are more or less important than we originally thought, then how we spend resources on getting depth players will surely change.”

But let’s broaden the scope again, beyond just playing staff. Analytics consultancies are moving more and more into the realm of advising senior, exec-level decision-makers. Even below the C-suite, analytics departments are more widespread and growing in size. So not only is ‘understanding football’ an important skill for ‘analytics people’, increasingly so is ‘understanding business’ and ‘understanding management’.

Managing people and projects is a real and underrated skill, and not only are analytics departments new, but they’re often adding several new information streams to existing organisations. Sarah Rudd has a positive spin on this: “A nice by-product of all of this is that it really forces you to think about what that process looks like.

“When there’s just one source of information (i.e. scouting reports) your process doesn’t need to be too defined or rigid, but as soon as you add more information you need to have something set out to explicitly state ‘when will I use this information versus that one, how will I weight these, what if they disagree?’”

But on the flip side, “Without something in place it just turns into chaos.”

A contrived bow

Get Goalside once wrote that you can measure the progress of ‘analytics’ by listening to what analytics people are griping about. Conference papers, to be fair, can be used as a barometer too.

This year’s MIT Sloan Sports Analytics Conference has three soccer papers in the final of its research paper competition. They are about:

So, they’re about how we generate more high-quality data; about long-term data; and about high-level strategic approaches. That’s not because on-pitch matters are solved, but they’re just… not what the public work is like at the moment, a big departure from the 2015-2020 period.

What do we actually know about football? More than we did, for sure.

But what do we know about what we know about football? Maybe less.

And what does Get Goalside know about what we know about football? TBD.

Thank you all for reading. Get Goalside will be back with another edition soon but please get in touch with any comments, clarifications, criticism, complaints, sales pitches, gossip, chit-chat, questions, or ideas at getgoalside[dot]newsletter[at]gmail[dot]com.

So... everyone's a decision-maker now

Sat, 14 Oct 2023 00:00:00 GMT

It’s sort of become an annual tradition at Get Goalside to write about how the analytics industry is converging around one specific point. First it was pressure data. Then it was hybrid data (a mix of event data and tracking data). Now, well, it’s not a type of data at all.

A brief bit of context first. This newsletter is written off the back of attending the Training Ground Guru and StatsBomb conferences, which tend to have shiny, exciting speakers and therefore produce thoughts about the shiny, exciting side of the industry. And sometimes, like in 2021 when I wrote about Stats Perform having pressure data, I’m a little ahead of the data company’s own product announcement (‘Opta Vision’ splashed pressure intensity in 2022, which then got written about in the hybrid data newsletter).

Better to be seven months early than late. Send me an email if you want an accidental peek into any other company’s roadmap.

So back to the newsletter: everyone’s a decision-maker now.

I don't say that just because StatsBomb conference’s keynote talk, from Zelus Analytics’ Luke Bornn, was ‘cognitive biases as Taylor Swift lyrics’. (Get Goalside would like to make clear its firm support of tenuous and extended metaphors). Let’s head over to three of the most recent data-centred consultancies to start operating in the sport, and see what they say about themselves in the opening line of their website copy. Bold emphasis is mine:

SRC FTBL, whose Ravi Ramineni also spoke (about decision-making) at the StatsBomb conference: “SRC is a football consultancy and data provider. Our mission is to use data and analytics to help clubs make optimal decisions.”

Ludonautics, the Michael Edwards and Ian Graham venture aiming to be the best spin-off from a Liverpool institution since Wings: “Ludonautics is a sports advisory business dedicated to helping sporting organisations improve their decision making capabilities through access to insightful statistical analysis.”

Prospect Sport, who recently hired former-RB Leipzig and Stats Perform data scientist Tom Worville as their Head of Football: “Prospect are revolutionising decision-making in sport with the help of artificial intelligence.”

I’ll know who to call next time I don’t know what to have for tea.

(Just so you don't think this is some kind of selection bias, Analytics FC, a much earlier entrant into the space, calls itself a "football analytics specialist with proven experience in delivering data-driven solutions". My own paymasters, Twenty3, keep it short and sweet with "We maximise the potential of sports data." Even Zelus Analytics care not about your decisions, they're just "building the world’s best sports intelligence platform.")

As with any convergence, there are multiple reasons behind it.

To a large extent, analytics has always been about making better decisions (only very occasionally about helping players execute better). And analysts have always talked about this too, going on about the path of data usage going from descriptive, to predictive, to prescriptive. First you can use data to describe what happened, then to predict what will happen, and finally to advise on what you should do in pursuit of a certain aim.

On top of that, there’s the other thing that analytics practitioners have always said: it’s no use having data if it doesn’t get listened to. That’s both an internal thing and an external thing, whether getting a manager to pay attention to your graphs or getting someone to buy your products. In either case, getting the buy-in of the people who hire, fire, and sign off budgets can save you a lot of hassle and wasted time. And I'd guess they're very receptive to phrases like 'decision-making'.

Finally, it's also quite simply a story about how established football analytics is. Everybody named so far in this article has close to (or more than) a decade’s experience around professional sports in a data or data-adjacent role. If we include Bornn’s research papers on NBA analytics while an assistant professor at Harvard (which we should, because one was about defending), all of their pro sports careers started while current Arsenal men's manager Mikel Arteta was still a professional footballer. These are people who've worked for a long time, in several cases been at teams who've won (multiple) major trophies, and therefore have the heft to go right to where decisions get made.

So, there’s all of that. Decision-making. Not ‘using data better’ or even applying specific sets of expertise like ‘helping sign the best players’. Just decision-making. "Drop the The; it's cleaner".

Football will probably be better for it. Not least because clubs often simultaneously need good decisions while being scared of anything but the usual ones. Lee Mooney of MUD Analytics (formerly of Manchester City) said, during a Training Ground Guru conference panel, that clubs are a bit like nuclear power stations: people are scared to tinker with them in case things go horribly wrong. The point of good decision-making, I suppose, is to work out how close the reactor actually is to blowing up, reduce any actual risk, and then improve the low-danger areas in the meantime.

(It may also be why English clubs that embrace data tend either not to be at the same risk of catastrophic nuclear collapse (Man City in one sense, Brentford and Brighton in a different one) or have already gone through the process of blowing up (Liverpool, whose data journey began after a debt crisis-induced ownership change in 2010; Wigan, who are currently working with Prospect, after their own ownership problems saw them docked points for not paying staff on time under the old regime; Arsenal, who bought StatDNA months after the third successive May date of St Totteringham’s Day, in 2012, and then hired the topically data-inclined Sven Mislintat months after the first season without a St Totteringham’s Day since 1995, in 2017; Manchester United, who announced the hiring of their Director of Data Science a few months after Manchester City reached their first European Cup final in 2021))

Anyway.

Subscribe to the newsletter and demystify football from your email inbox

While ‘analytics as decisions’ is a fun way to understand the business landscape, it also made me think about something more fundamental. It's a framing which I think explains a lot about the course analytics has gone through so far, and may give a better idea of how it will progress from here.

Is ‘analytics’ tech, or is it an information resource?

It’s been easy to think of it as the former for most of the ‘boom’ since the mid-2010s, evolving in public consciousness alongside social media, grandparents getting smart phones, and AI. But while there’s tech involved, analytics obviously fits better alongside other ‘movements’ of information and ideas.

The internet, computers, television: none of these took off because people understood them. In fact, their most rapid growth happened only once they could provide value without people understanding how they work. There's a reason ordinary people don't know what Linux is.

Now think about ideas like cognitive biases, microaggressions, issues with capitalism: you have to understand them to get value from them. Although the general population might not have as sophisticated an understanding as experts, a big driver in them taking off in the last decade has been a mix of pop-sci books and Twitter. In other, less medium-specific terms: because enough people got exposed to enough of the information to understand it.

This framing is meaningful because if analytics was a technology, then simply giving users better interfaces would be the revolution. And is that what happens? No: people don’t get good at smartly applying data to football problems because they have a login to software with possession value models. They get good because they hire a decision-making consulta- no, wait. They get good because enough of the concepts get enough of an explanation that they can understand the information and apply it.

Does this matter - like actually matter - in the grand scheme of things? Not really. Only as much as you care about the application of football analytics beyond your ability to win more games or earn a living. But it does mean that “why aren’t more clubs using data well” isn’t going to be solved by better tech alone, or by the existence of a handful of smart people in the industry.

There's an elephant in the room here. The title of this newsletter is clearly false. Everyone is obviously not a decision-maker now.

Most people are, and will continue to be, far, far from the room where it happens, falling into one of three categories.

Two are simple: those who feed into a decision-making process earlier in the chain, and those who don’t. Analysts who create reports which get read by senior coaches will be in the first group; analysts who create reports which get a cursory glance at best will be in the second. We know where fans tend to find themselves. Not everyone can feed into every decision-making process, football is full of people giving miles to gain an inch or being told their input is listened to when it’s merely being recorded.

The third and final group are the facilitators. Data engineers, report-creation tools, data providers themselves. On this, the StatsBomb conference – being a branded event – had some news: that StatsBomb will be doing football tracking data for themselves in the near future. That means that after basic event data, 360 data, and stop-motion gridiron data, the company are filling up their football data Pokédex fast (someone told me afterwards in a shady corner of the pub that if you press up, down, left, left, right, up, down while StatsBomb IQ loads then you get Gaelic football stats too).

An amusing quirk of facilitators is that a rapid rise in their number may end up producing 'two steps forward, one step back' situations. New data sources need to be examined and studied for their peculiarities; new APIs may not be well supported; new products may not have all of the features that were demo-ed. To take a point from Ramineni’s talk, decisions compound, for better or worse. Rejigging your work processes to fit in a different product or service is likely to be one of those compounding choices.

How long will it be until the average club uses the same core set of data for, say, five years in a row? (Will that be a new marketing point: 'you know us, you have a track record of data with us, you know how to navigate our bugs'?). It may not even be within the club's control, depending on what league-wide deals are in place. And even then, even if a club is happy with their current provider(s) and able to stay with them, the aspects of that data they use every week will probably change. New models, new stats, new events.

A lot of new decisions to make.

If only there were someone who could help you out with those…

“The wedding was charming if a little gauche / There’s only so far new money goes,” –Taylor Swift, warning against assuming new wealthy club owners will solve all problems.

"What's hard about art is getting any good – and then getting any better. What's hard is solving problems with infinite solutions, and your finite brain." – Elisa Gabbert, 'Why Write'.

“I had not the hardihood to suggest to Dora’s father that possibly we might even improve the world a little, if we got up early in the morning, and took off our coats to the work.” – Charles Dickens, David Copperfield.

Purpose of innovation; process of innovation

Mon, 09 Oct 2023 00:00:00 GMT

In the words of the musical heavyweights Scouting for Girls, "There's a little bit of Christmas in the air."* Not, like, literally. But the dawning of this morning brings the Training Ground Guru Live conference, with a day of sporting directors and analytics today on Monday, and the StatsBomb conference tomorrow on Tuesday (papers from the StatsBomb research competition are available here). All that so soon after the sweet, sweet farce of a replay operator uttering the words "the on-field decision was offside".

*Keep it bookmarked for December

How lucky that on the eve of those conferences, we (some of us) also got a brand new camera angle to watch football in. On the UK's Sky Sports Football channel, viewers got to see an in-development 'Game Mode' camera angle. If 'like in FIFA/EA FC' means anything to you: like that. If not, the camera was on a track the length of the sideline following the play, and pivoting slightly for better angles/to reduce the distance it needed to travel.

Now, I'm a sucker for these things, but the near-total lack of action in the game left me plenty of time to dwell on 'why'. Is it there to be like video games for the people whose interaction with football is mostly through their PlayStation; is it there because there's something genuinely benficial in that type of angle which everyone can enjoy; or maybe it's to be like weirdly-flavoured Oreos, where the strategy is to create brand awareness through peculiar, temporary novelties.

Alas, a tangent: is Gerard Piqué's retirement venture, the indoor football Kings/Queens League, a pumpkin-spice Oreo? Should FIFA (or UEFA) incorporate it to keep people coming back to plain old, milk-flavour 90-minute matches?

Let's wrestle this back on track.

Innovation comes in many forms and, like a Slinky, rarely arrives front-to-back in a regular timeline. The explosion in 'AI stuff' in the past year wouldn't have happened if OpenAI hadn't put a very user-friendly interface around their GPT model. That in turn (if various podcasts about the subject I tuned into are to be believed) wouldn't have happened without That One Paper From Google, 'Attention is all you need', which proposed a new type of learning architecture that now goes into all the language models.

On a much smaller scale (though less potentially-dystopian) I recently had three awkward projects at work move along by leaps and bounds within the same week, as if they were proverbial London buses. Progress and invention isn't linear, it leaps, like Archimedes.

Innovation may not be able to be controlled, but it can be nurtured. There are various methods but a main ones tends to be getting more people able to work on things, preferably with some degree of interaction.

That's often been difficult in football analytics, with the differing ways that people 'see' the game - not just in terms of opinion but in the actual brain processing. The way that I see a match unfolding won't be the same as the coaches in the dug-outs. For me at least, a part of that stems from the data itself being a framework, a shorthand for breaking up complicated patterns. I'd be surprised if the same wasn't true for other 'data people' too, whether that framework be 'pass event'-based or 'possession value model'-based or 'high-intensity run'-based.

There's a quote from Juanmo Lillo in an old interview with Sid Lowe that I like (but sadly can't find the link for right now): "Qualities go together. Naturally, you see the passes you can make; if you can't make them, you stop seeing them." In more ways than one, the passes are the data.

As data people we've often talked - or been told - about how we have to learn to 'speak football'. This is true, undeniably, both because it's helpful and is also more fun than hyperparameter tuning. And yet, in our defence, if 'football' wants to get the most out of 'data' then the opposite has to happen too.

I've been working on a side project which I've just got to a 'yeah, I'm done with this' stage, a 'build your own metrics' webapp using event and tracking data (from Metrica Sports' open repo). Coaches and analysts watch too much video than is healthy for them, but they can also be very picky (and hard to pin down) about their definitions. So the idea was: why not let them create their own?

Of course, you'd never leave an analyst without any video, which is why the prototype gathers all the sequences that fall within their definition and lets them watch. (With tracking data, of course, although you could link this to 'real life' video if you had it).

In a perfect world, there would be a number of benefits to this. One would be analysts and coaches (potentially) getting more time to do things other than watching football (though that might just be a pipe dream). Another, I think, would be that by offering a suite of options for definitions, it matters less if the data scientists/engineers/developers get things exactly right. It could also reduce the breadth of definition, if data people are called upon to add a component of a concept rather than one in its entirety.

And then there's one about innovation too. If you help bridge the gap between the two groups, you increase the number of people whose minds are bubbling away on the same problems. You not only understand the framework that the other party is seeing things through, and the way that may be shaping their viewpoint, but you get to understand the opportunities it brings too.

There's a second half to my Lillo quote, a passage or two along in the interview: "A guy who knows he can place a ball 75 metres, opens up his perspective to do that."

The link to the 'metrics creator' is here, with a screen recording walkthrough is here. The code can be found through my GitHub, here.

Time to give ball-in-play measures a kick into touch?

Fri, 18 Aug 2023 00:00:00 GMT

In Sunday school, as a child, we were taught that God loves all His creation in all its diversity, from the lowliest frog to the highliest mountain goat, whether they be big or small, beautiful or not. But FIFA is not God.

There's a diversity of ball-in-play times in football, and football authorities do not seem to love it. As we talked about in the first newsletter on ball-in-play time, matches appear to drift towards an average of 54-56 minutes, within each average is a pretty wide range of ball-in-play values. Many of the sample competitions that we looked at last time had matches with around 45 minutes of in-play time (sometimes called 'effective time'), stretching all the way up to 65. Take a look below: every dot is a single match.

[Data, of course, from StatsBomb's open dataset]

What causes these differences though? Why do some matches have a lot more ball-in-play time than others? Why is it apparently perfectly normal for the in-competition range to be ten minutes or more?

That's the thing I've been looking at, and I, simply put, have a stupidly lowly frog-level thing to say about it.

If the ball doesn't go out of play, the in-play time is higher.

That's what it seems to boil down to. Lemme take you through it.

Subscribe to the newsletter and demystify football in your inbox

In, out, in, out, shake it all about

After the first newsletter, the first thing I wanted to look at was the ball-in-play time for different teams alongside some 'stylistic' statistics. It was kinda boringly predictable though: as a trend, high-possession teams with lots of short passes were involved in matches which, on average, had a higher amount of ball-in-play time.

This was fun and all, but felt like distracting from the point. So let's go right back to basics.

The charts below have four statistics which are compared with the ball-in-play time of all matches I ran my code on. They are:

Short passes: the percentage of passes which were 15 metres or shorter
Ground passes: the percentage of passes which were along the ground
Live passes: the percentage of passes which were 'live' (so, not goal kicks, free kicks, etc)
Restarts per minute of ball-in-play time

[Data, of course, from StatsBomb's open dataset]

There doesn't look like much of a correlation between a match having lots of short passes (top left) and its ball-in-play time, across these samples of data anyway. (That might just be about the particular cut-off point of 'short'). There's more of a link when focusing on passes along the ground (top right), although still with quite a wide range.

That brings us to the two charts on the bottom row, '% of passes live' and 'restarts per in-play minute'. In a way, they both represent the same type of data: how many passes are in live play compared to being a dead-ball restart. I read them both as saying 'more restarts/stoppages means less in-play time'. After all, this makes sense: if the ball doesn't go out of play then the 'in-play' timer just keeps ticking up and up and up, even if nothing much is really happening.

Let's look at this in its purest state then, with the total number of restarts for each match compared to the in-play time. The link is a little fuzzier than some of the previous charts, but it's still clear:

[Data, of course, from StatsBomb's open dataset]

This restart correlation might even explain the link between ground pass proportion and in-play time, although my intuition is that any meaning in the link is actually about high passes:

restarts like free-kicks and goal kicks and throw-ins tend to be high passes
high passes are less controlled and therefore probably more likely to bounce out of play (possibly also having longer stoppages as teams collectively move up or down the pitch to follow the long path of the ball).

The link between number of restarts and in-play time is always going to be a little fuzzy. The stoppages between the restarts can vary in length, meaning two matches can differ in in-play time even with the same number of stops, and the length of added time to make up for excessiveness of breaks will also vary. But, yeah - this is a pretty plausible and stupidly simple link.

Subscribe to the newsletter and demystify football in your inbox

What this means...

We've done the data, now we're going to do the semantics.

There's a way of describing football that says that if the ball is in play it's 'live' and if it's not, or if the play hasn't been restarted yet, it's 'dead'. I re-watched half an hour of the 2018 men's World Cup 3rd-place match between Belgium and England to check that data and, believe me, in-play football can be dead.

If you think I've been banging away at this, Twenty First Group have also called the added time push 'a sledgehammer looking for a nail'. And Omar Chaudhuri there reaches a similar conclusion that I do:

"The single biggest driver of effective playing time is the on-field technical quality of play – not time wasting, as is commonly believed. This is why the Premier League has 15% more ball in play time than League Two, and the Champions League 5% more than the Premier League. Better players and teams keep the ball in play."

I'm inclined to believe that there's a tactical choice element as well, but the tactical choices to play ground pass-based, keep-ball football only become unlocked when technical, spatial, and pitch-quality levels are good. (That said, high-quality pressing may still disrupt this).

I don't think it's too far over the line to say that a continued push to increase ball-in-play time beyond levels we have historically seen in football risks casting certain styles of play as illegitimate. If you want a guaranteed 60 minutes of the ball in play (something that appears to be more uncommon than common historically in football, even at the highest level) then what you do not want is the ball to go out of play.

A smarter approach - although an investigation that may be difficult with data taken from TV broadcasts - would be to look at the average time that certain stoppages usually take. There'll be a range for throw-ins, a range for long goal kicks, short goal kicks, corners, goals. There may be important effects for high-physical intensity matches. If the length of time of goal-kick stoppages is increasing compared to matches of similar intensity and exertion from five years ago, then we can talk. But ball-in-play time is a clumsy, flawed - and possibly tactically geo-engineering - measure to use.

Epilogue

Hidden amongst this is a convincing argument that I think organisations like FIFA or the Premier League could be making. While the work I've done hasn't established any causal links, it seems fair to assume that better technical, tactical, and pitch quality could lead to more control-seeking in-possession play, which could in turn lead to fewer stoppages. Seen this way, you could argue that although present-day ball-in-play time isn't that different to historic figures, you would expect it to be higher based on increased quality across the board. In this argument, it's not that the line is going down, it's that it should be going up but isn't.

I still think there's a hole in that. Defences are better technically, tactically, and physically now too. Football is always a combative sport. We're already seeing how an increased focus on pressing strategies, varieties, trigger moves, and mid-block organisation is nudging Premier League teams into embracing The Long Ball, albeit in a much more controlled form than two or three decades ago.

There's one place where I do empathise with footballing authorities though: tactical injury time-outs. Player welfare means you can't just ignore it when players go down injured to disrupt their opponent's rhythm and let the coach give out some instructions, but it's also a joke at this point (literally, in WSL-watching circles). This - more than time-wasting at goal-kicks, say, which already has clear punishments referees can issue - is an area that probably needs addressing and will be awkward to do so.

I hope to make the rest of the code for this piece available before the end of September. The code for the first newsletter is already available in this GitHub repo. In the meantime I hope that the ridiculous aspects of this new push die down and that maybe some footballing authorities publish their own data, preferably over different timespans and different levels of football.

Remember: football is a lot like life, everybody needs to rest at some point.

Notes

Mea culpa, and data checks

Doing further research for this piece I got even more curious about what my code said were very long ball-in-play times for the 2018 men's World Cup. I re-watched the 3rd-place match between Belgium and England (the highest figure in the data produced by my code), taking notes on breaks. It was not a scintillating watch. It also didn't seem to stack up to the numbers I'd produced.

When I checked the event data event-by-event, I realised something I should have gone actively looking for beforehand: events missing when the TV broadcast was showing replays. This was a particular flaw in my methodology - a goal missing a kick-off due to replay meant I never went back to look for the duration of the stoppage (and wouldn't have got an accurate reading anyway).

I watched part of the 2019 World Cup final alongside the event data too to check StatsBomb's more recent data version, which it turns out every sample other than the 2018 World Cup one was collected under. Their documentation online indicates that they changed a bunch of stuff between these versions, and between these checks, previous checks, and the fact the 2018 average values were such a clear outlier in the previous newsletter, I am willing to trust the legitimacy of the rest of the data. But I've chucked out the 2018 World Cup stuff.

Stop the clock

Mon, 07 Aug 2023 00:00:00 GMT

Edit - 18 August 2023: A previous version of this post used data from the 2018 World Cup. This data has since been suspected of being suboptimal for the purposes of this study. Where the data is used, it is asterisked and figures blanked.

I have terrible, terrible news for the administrators of football. Y'all are pebbleheads. Either that, or some code of mine is wrong, in which case pebblehead status is mine.

If you've been watching football, or just permanently have Sky Sports News on, you'll know that footballing authorities want to CRACK DOWN on time-wasting, a SCOURGE on the game. Viewers of Premier League matches, it is reported with Helen Lovejoy-ian despair, see a shade under 55 minutes of in-play action in an average match. (Although, as Opta's The Analyst site pointed out in January, this 'nearly-55 minutes' figure is higher than a swathe of major European men's leagues).

The thing is, though, nobody actually knows whether that is out of the ordinary. Ordinary people do not sit in the stands or on their sofa with a stopwatch. But this is a subject where data is handy, because with event data at your disposal you can add things up pretty easily. And thanks to StatsBomb and their open data, we have quite a trove of it. Here's a set of samples from their data and the average ball-in-play time (code repository with calculation is here).

Premier League 2015/16 season: 55 minutes, ~30 seconds
La Liga 2015/16 season: 54 minutes, ~10 seconds
Arsenal men's team Invincibles season, 2003/04: ~55 minutes
StatsBomb Icons sample (21* games from the careers of Diego Maradona, Johann Cruyff, or Pelé): 55 minutes, ~10 seconds

(*See note at end of post)

And here's some more!

Women's Super League 2020/21 season: 54 minutes, ~20 seconds
Indian Super League 2021/22 season: 49 minutes, ~45 seconds
FIFA World Cup 2019 (matches that finished in normal time): 54 minutes, ~50 seconds
FIFA World Cup 2018 (matches that finished in normal time): ~XX minutes*
FIFA World Cup 2022 (matches that finished in normal time): 59 minutes, ~40 seconds

One of those is a major outlier, the 2018 World Cup (edit 18 August 2023 - for more on why it was an outlier, see the follow-up post on further investigation of the data). It, as well as the ISL season, are also outliers in another way: in most samples, the ball-in-play time is between 56% and 58% of the total match time; for the ISL season it was 50%. (There's a note on methodology at the end of the post)

So in this data analysis, including matches from two decades ago, we've got a central band around 54-56 minutes of ball-in-play time and 56-58% of match in-play time, and an outlier in either direction. Going by The Analyst's article from January 2023, much of the top European men's leagues fell into or close to this central band too. They generally averaged 53-55 minutes ball-in-play time and 54-57% of total match time: a touch less than the analysis on StatsBomb data samples, but not much.

The push for added added time hasn't just had pushback from silly newsletter writers, but from players too. As this piece is being edited on Monday 7th August, Raphaël Varane has tweeted concern about it:

From the managers and players, we have shared our concerns for many years now that there are too many games, the schedule is overcrowded, and it's at a dangerous level for players physical and mental well-being.

Despite our previous feedbacks, they have now recommended for next season: longer games, more intensity, and less emotions to be shown by players

The results of both my and The Analyst's data analysis should be serious food for thought for 60-minute stop-clock advocates. A 60-minute stop-clock system would be like every match running to around 105 minutes without it, or like normal time plus a half of extra-time (albeit without stoppage time) week-in, week-out, in cups and every league match.

(If you are still committed to the stop-clock then a more sensible figure would be 50 minutes - lower than most current ball-in-play times but probably allowing for more sustained and sustainable high intensity than larger amounts of time).

There is, though, a high degree of variability in ball-in-play times. In The Analyst's piece, the Scottish Premiership had an in-play time and ratio close to the Indian Super League figure, around 50 minutes. English coverage at the start of this season has reported that League Two matches had an average of 48 minutes. Meanwhile, teams like Manchester City tend to have higher level of ball-in-play time than others**. Similarly, in the StatsBomb data, an analysis of 37 LaLiga matches from Barcelona's 2011/12 season - part of the dataset of Lionel Messi's league career - found that their matches had an average of over 58 minutes in-play time, with an in-play percentage of 62%. Not far off a figure FIFA would nod at.

**The Analyst has 2022/23 Premier League team-level figures in this piece from May, but I am disappointed that they succumbed to framing this in terms of time-wasting.

The link between Guardiola teams and in-play time is probably both a tactical one and a quality one. The Analyst's January article noted that the Eredivisie's in-play time was around 57 minutes; the Dutch league isn't higher-quality than the English Premier League, but it does have a certain threshold talent level as well as a very particular stylistic history. It's not clear to me, though, that trying to push English League Two or Scottish Premiership or Indian Super League football towards an hour of in-play time is desirable for anyone. Not everyone can play Dutch/Guardiola football, not everybody wants to (long ball football is sometimes the optimal strategy!), and not everyone wants to watch it be done badly.

Most importantly, there's a further reason why the City and Barcelona figures are not high purely because of talent. The figures for Arsenal's Invincibles season and the StatsBomb Icons sample, you'll remember, were around 55 minutes. Yet the coverage of in-play time makes it seem like current-day football has reached a nadir. It makes it sound like we need to go back to the good old days before time-wasting was robbing fans of ball-on-pitch action. But it seems - from this albeit limited data analysis - that the current day has as much ball-in-play time as the good old days. (The 48 minutes figure for English League Two is probably less than ideal but, the question remains, what is standard for fourth-tier in-play time?)

The funniest part of it all is that, despite FIFA's efforts in the 2022 World Cup, the percentage of the match time that was in-play was still below 59%. They got close to an hour of in-play time not through anything smart but by brute force, introducing several minutes of extra dead time in every game at the same time.

Why push for more in-play time, which I have not yet seen evidence for having ever existed, when the in-play action has increased in intensity while barely decreasing in duration? Research has found that high-intensity running increased by between a quarter and a third, depending on outfielder position, between 2006 and 2013 in the Premier League ('Evolution of match performance parameters for various playing positions in the English Premier League', 2015). More recent research by data company Skillcorner has found that, over the past five years, physical demands are still rising. Raphaël Varane notes in his tweet that "we have shared our concerns for many years now that there are too many games, the schedule is overcrowded, and it's a dangerous level for players physical and mental well-being." FIFA has recently made a big push into data (noted in Get Goalside last January), but it doesn't seem like the data is in their favour on this.

Let's stop this nonsense. Players may well be harder, better, faster, stronger but so are their matches, without any ball-in-play time changes. Fans are not being short-changed with less in-play action, and if football authorities are so worried about that they would be better to focus on ticket prices and transport availability. And every underdog victory that you have ever seen and savoured has involved time-wasting (many Guardiola victories also include time-wasting too, they just happen to do it with sterile possession).

Referees do not add on time for every stoppage, never have done, and never should. Football is a simple game: 22 players run around a pitch for 90 minutes and, in the end, only 54 of them were with the ball in play. 'Twas ever thus.

Subscribe to the newsletter and demystify football from your inbox

Methodology note

The code for this project can be found here. The approach I chose was to look for 'restart events' (e.g. kick-offs, corners) and look back at the previous event's timestamp, after having removed events which could occur during breaks of play (e.g. substitutions). This seemed easier than looking for events that ended in-play sequences, which depend on the referees and can be coded up in a variety of different ways by data providers. For total match time I used 'Half End' events.

For the StatsBomb Icons sample, I found that some games had less than 90 minutes of match time. This was because the old video sources sometimes skipped moments (probably out-of-play time). As such, I have no real reason to doubt the 55 minute in-play time figure, but I suspect the analysis undercounts out-of-play time slightly, even for the 21 matches I used whose match time total was higher than 90 minutes.

Although I've taken care with the code, and checked for anomolous results at stages throughout the process, it's possible that there's a mistake somewhere which would affect the analysis. Even if that is the case, I think several points in this post still stand. Based purely on the comparison to The Analyst's in-play time for the 2022 World Cup, my analysis seems more likely to be over-counting ball-in-play time than under-counting.

Notes

Coda

The footballing authorities taking note of The Guild of Time-Wasters seems unlikely, but if Gianni Infantino does happen to be reading I have one suggestion, one concession: Stop the clock for 'on-field' VAR reviews.

As someone who works with football data, I'm loathed to suggest a change to the way the clock works, but look at the data. The fact that the 2022 World Cup increased in-play time by a much more significant extent than the in-play ratio means there must also have been an increase of dead time. That almost certainly comes from VAR reviews.

Now, you can't stop the clock at every VAR review, because the point of video assistant referees is that they're always reviewing stuff. The rationale for stopping time would become fuzzy too: if you pause for five seconds while the ref double-checks something with the video room, why not pause for the two minutes it takes someone to get treatment after a clash of heads? That way madness, and a hard stop-clock, lies.

An 'on-field'/at-screen video review differs from these other events in one key way, which is the match official leaving the field of play. They even already give a clear signal before and after a monitor review, segmenting it from the rest of the match. These stoppages are also more frequent than the only other type of minutes-long break in play, serious injuries (although even here, a stretcher needing to be brought on could be a signifier).

There are bonuses here too. Games clocks regularly ticking over to 100 minutes is clearly stupid, making it look like the sport doesn't know how to organise its own matches. It screws with player minutes-played metrics. And it's just plain weird for fans to get to 85 minutes and not feel a sense of urgency that the game is coming to a close. Stopping the clock for monitor reviews might even improve general VAR use too, because refereeing teams might feel an impetus to keep things snappy unless something is serious enough for a stopped-clock review.

And there we have it, football's in-game bureaucracy problem is solved.

Subscribe to the newsletter and demystify football from your inbox

When PIF comes to shove

Fri, 28 Jul 2023 00:00:00 GMT

Hi there, it's been a while. With the second half of the calendar year now underway, let's take a look back at the turn-of-the-year piece and what Get Goalside drew out as the themes to watch for 2023:

Football analytics jobs conceived of as an entire ecosystem (not just clubs)
Set pieces
Skeletal/body pose data
Data on the sidelines
Global game, global analysis
Ethics in analytics

Events of the last couple of months have sure made that last one seem relevant, huh. What with FIFPRO's commitment to working on a centralised player data hub to better allow players to access their data.

And then elsewhere, with 'global' in mind, is the Saudi Pro League (SPL). The big news on this front has, of course, been about the players and the transfer fees and the seriousness or not of prior support for the LGBT community. It's turning out to be quite the summer on that front. But if the plan for the SPL is to become one of the best leagues in the world, and not merely what some people still call the league that Celtic and Rangers are in, the money will have to go further than big-name coaches and players. That will include the backroom infrastructure of a club.

A football team is an iceberg, and this iceberg, now more than ever before, includes data and analytics. In fact, the strength of analytics staff at these teams may well be a barometer to watch to gauge how seriously this project is being taken. Training facilities and physios are things which players (the local ones and the very expensive foreign ones) interact with every day, and are what they rely on so that their bodies don't break down. The same isn't the case for analytics. Kylian Mbappé may be put off by a sub-standard gym; he's probably not gonna be put off by an inefficiently optimised database schema.

Of course, some of you, many of you, may well be reading this and expecting the 'should they go' question to crop up. First though, it's worth breaking down why people are asking it. As with many things, there are multiple reasons behind peoples' feelings, and without recognising this we risk conflating one thing with another. There are, I think, four broad categories of reasons that people have opposition or conflict about all this:

1) Saudi Arabian laws: A broad category that could be split further into things like LGBT rights, women's rights, democracy and free speech, 'foreign policy', and profiting from oil. For the moment we'll just keep them together as a broad 'moral belief clash with the regime'.

2) The sheer amounts of money: While the sudden influx of cash may help some clubs plug some short-term financial holes, football has a tendency to spend beyond its means to pursue the wealthiest rivals. On a slightly different level, rapid increases in fees and wages will always seem a little ostentatious, particularly given that footballers are treated as heroes in a way that most other multi-millionaires are not (people may admire the CEOs who take home millions in pay each year, but kids don't wear their names on the back of their shirts).

3) The fact that the money comes from the nation state: national investment funds aren't just some rich bloke burning cash and, although politics is tied to everything, it becomes a lot easier for football to be made a geopolitical bargaining chip when the state itself is so tightly involved. This is also the thing that particularly amplifies category 1: these aren't just jobs which are situated in a country with certain laws, and go a long step beyond merely being helpfully facilitated by that government.

4) Don't take our thing: look, we've got to be honest that this is part of it for some people, to varying degrees. Part of this is that European football may get weaker and it's convenient as a football fan to have the best players in one place, and part of it is because - despite six World Cup appearances - Saudi Arabia isn't thought of as a 'footballing nation' by most Europeans. This too will have several factors, partly kinds of prejudice, but the lack of Saudi Arabian players playing in Europe will also be a significant factor (Pierre-Emerick Aubameyang will undoubtedly have boosted the perception of Gabonese football among Europeans over the last decade).

Most peoples' reactions will draw from all of these four in various different proportions, and it would actually be surprising if they drew from one alone. Some people probably don't care much about 1 but care a lot about 3. Some may care about 1 but not 2 but a little of 4, etc.

It's worth making the point that some people within the industry may care about issue number 1 particularly because, unlike (what we assume about) high-profile male footballers and managers, it may be their lives which Saudi Arabia's treatment of the LGBT community and women affect. Because of the lack of publicly out gay or bisexual male footballers/coaches (itself a symptom of cultural homophobia, even if not legalised) the LGBT rights issue has mainly been a question of abstract solidarity. For some people, as far as potential job prospects there go, it's going to be a question of their very tangible life experiences.

On the theme of solidarity, it is somewhat interesting that this is taking place at the same time as industrial action is such a prominent topic in (at least) England and the United States of America. The US writer and actor strikes in particular are very well communicated (in terms of their grievances, what strike action means, and what support is available for people in that industry) and well-supported by high-profile figures in the industry who don't personally need the strike to protect their working conditions in the same way as less-prominent figures.

Get Goalside is one of the few areas of public industry chitter-chatter so take this as the opening of discussion for the non-player/coach side of the SPL story. I'm interested in what you think. If you've been forwarded this email/post rather than coming to it direct, Get Goalside can be found at getgoalside.newsletter@gmail.com. Also on Threads and Instagram, at getgoalside.analytics.

Have a good day.

Is it better to be a leftie?

Fri, 09 Jun 2023 00:00:00 GMT

Rafael Nadal hits a forehand with his left hand and signs his name with his right. How different would the honours list at Roland Garros look if his uncle Toni hadn't decided to train Rafa to hit with his 'weaker' hand as a child.*
*(in fact-checking I've discovered this is an urban myth and Rafael simply played the left-handed forehand better once he was old enough to start hitting balls with one hand)

Being left-handed in tennis, like cricket and baseball, is a bit like being above a certain height threshold in basketball - a rare physical trait that gives some in-game advantages.

I had an idle wondering recently whether the same might be true in football. The amount of articles written about the value of left-footed passing angles seems to have risen sharply in the past few years, a trajectory not seen since Q4 2020 Bitcoin. If you were the coach of a young kid, would it be valuable to train them as a left-footer? If you were at an academy, would you specifically look out for them, even if other parts of their game don't match up to their peers?

And then I realised I could actually probably look into this with data, and was therefore compelled to.

First, the question has to be refined into something there's data for, and I landed on two easy ways to approach it. One is simply whether there are more left-footers playing football than exist in the general population; the other is whether they're smaller on average (worse is not at all easy to judge) than their right-footed peers.

On the first count, – well, actually, let's just take a moment so that you can guess what the rate of left-footedness in the general population is.

Just take a punt.

Did you guess close to the approximate figure of around one in ten who are left-handed?

Estimates that I found online ranged from around that level (https://www.psychologytoday.com/gb/blog/the-asymmetric-brain/202009/large-study-reveals-how-many-people-are-left-footed) right up to 24% who are 'non-right-footed' (https://www.nature.com/articles/s41598-020-71478-w).

For the footballer data, I gathered from FBref (so, Opta data) players who:

Are an outfielder
Had played 1000+ more minutes
In any of the Big Five men's European leagues
In any single season in the past five years (i.e. a minimum bar of 1000 minutes in one season, not 200 in five)

Just over 27% of these (2800 total players) were left-footed, so it does seem like being a leftie is more common in elite professional men's football than it is out in the wild. Just. But let's count that as Count One: ✅.

On Count Two, 'can left-footed players get away with being scrawny', things were less clear. The ~2000 right-footers have a mean average height of 181.8cm and the ~740 left-footers come in at 181.1cm. The presence of a marginal difference is there in weight too, righties a touch heavier than lefties (maybe because of the height) at a mean average of 74.2kg to 73.5kg.

Now, this is a nerdy newsletter so we're going to have a nerdy note about this visualisation. I had to reduce the smoothing on this KDE plot because the default in the plotting software smoothed away the fact that the modal height for righties in the sample is 180cm. Do we think it's interesting that there's a peak, going against the general trend, bang on 180cm? Who can say. Who. Can. Say.

The weights also have clusters of values, although I'm not really sure what pattern they follow (not on round kg numbers but around every 5kg, which is about 11 lbs). The distribution, at the same level of smoothing, is very spiky.

So, Count Two: ❓. The weights don't interest me that much anyway - it made sense to take a look given that the data was there, but you can put on mass in a way you can't with height - but what if even these slight differences are 'fake' ones?

The positions where being left-footed seems to be the strongest tactical advantage are at left-back and inverted right-sided forward. It seems likely that the relationship between the value of height/weight vs footedness varies by position, and neither of these are positions where height matters a lot (unlike, say, centre-back). Maybe this is a confounding factor when we're looking at all outfielders together.

(I did consider digging into this position-specific question, but didn't feel confident in categorising the position strings that FBref gives players).

So, I'm sorry to the Uncle Tonis-of-myth of the world: if you want to give your young relatives the best chance of being an elite footballer it may not help them that much if you persuade them to take up being left-footed. Depends on your philosophy on marginal gains. But it was fun to take a quick look at.

Data notes postscript

I excluded goalkeepers on this because I think they're a distinct enough skillset from outfielders, where height and hand-coordination is so much greater a factor than footedness
I get that FBref/Opta data might be fakey (all the fellas are six foot eh), but I figure that it'll be fake-y in the same ways for both types of footedness
I imagine that this will play out differently for different levels of football - it seems plausible to me that 1) there'll always be a minimum floor of lefties given that the disadvantages of a right-footer playing at left-back are so great 2) the advantage of being a leftie in build-up play matters more at levels where build-up play passing angles matter more

The only rule is it has to be fun

Sun, 16 Apr 2023 00:00:00 GMT

We're going to talk about television. We're going to talk about data. And we're going to ask what the definition of 'fun' is. It must be the Get Goalside newsletter.

Slime and shot speed

Ah, to be American. Able to watch Premier League 3pm kick-offs on TV, understand jokes about New Jersey, and watch the first appearance of the 'Data Zone' streaming option on NBC's Peacock.

Image form NBC press release article

Liverpool's comeback against Arsenal at Anfield was the first of four matches in April to get this option on NBC's streaming platform, which combines the match footage (on a slight delay to most live feeds) with event and tracking data.

Being in England, I didn't get to see it, but this isn't the first time that sports broadcasts have been 'augmented' like this. It's just the latest in a growing line. Over this side of the Atlantic, BT Sport's app won an industry innovation award of some sort in 2021 for its 'Hype Mode' option, which had some of the features that you can see on the Data Zone screenshot (ball speed on shots, 2D tracking data pitch map) as well as some energetic animation.

If you're less fussed about numbers but like cartoons, the NHL has you covered this year. Partnering with Disney, it's done some stuff to take the tracking data of the players and pucks and animate it in a style that I assume I'd recognise if I was a couple of decades younger.

(This follows a proud American lineage that brought you the NFL's Nickelodeon coverage, although as far as I know that one is just graphics work and presentation, no data).

Is this trend more to do with the progression of tracking data, US money and techieness, or continuing fears about how young audiences engage in traditional sports? I dunno, but it's probably a bit of all three.

The latter two speak for themselves, but this being a data newsletter it's worth touching on the tracking data point a little. Event data can be engaging and visceral (check out the intro to this video on some La Liga graphics), but it's much easier and arguably requires less imagination with tracking data, and it opens up more doors. There's more data, and to paraphrase the old IBM line, no-one ever got fired by telling people running stats.

And yet. Do these implementations in football know who, or what, they're for?

Now, in fairness to the people putting these things together: 1) they're treading new ground 2) even the people running football don't tend to know who, or what, they're for. (It is quite amusing that FIFA bigwig-types have pivoted from the idea that matches should be shorter, bc attention spans, to focusing on the dummy spectre of time-wasting and the idea that matches should be longer. In a sideways sense they're correct: my attention span for their time-wasting is quite short).

When you start thinking about this question in one area it starts to crop up everywhere. Who is the Data Zone for? Whose viewing experience is it aimed at improving? Who, other than industry professionals, was 'Hype Mode' for? Who, FIFA, is going to be more attracted to football matches by there being more added time at the end of matches?

I'm not really asking football rightsholders for something as clear-focused as a slime cannon-wielding, Patrick Star-starring kids show NFL broadcast. I am, more than anything, asking the question because I want to wonder out loud about the answer.

Supermarket segmentation

The latest episode of the Unofficial Partner podcast couldn't have come at a more convenient moment in time for this newsletter. On the subject of what sport can learn about customer data from other industries, guest Claire Kelly (formerly of Sainsbury's) was asked how the supermarket chain thinks about their customers.

"One thing I'd say as a starting point is they certainly don't segment by demographics or gender like perhaps we do in the sports industry," Kelly explained. After giving a couple of different examples of shopper profiles, she continued: "And so Sainsbury's would segment on needs and motivations and then you'd layer profiles around that when it moved into marketing and targeting strategy."

So let's play armchair marketer about armchair fans for a moment. What different profiles are there? (If you want some slightly less off-the-cuff fan segmentation, the European Club Association split fans into six groups in a report here, although that was about fans in general, not solely TV viewers).

I've come up with a few:

Background-noisers: people who tend to do other things while the football is on. This could be reading, cleaning, working, gaming, flicking through TikTok.
Home-stadiumers: people who are watching the match primarily as a substitute for being at the stadium. They'll usually be a fan of one of the teams, and be invested in the action.
TV-eventers: people who watch the match more for social or 'cultural' reasons, because it's the thing that friends will be talking about. The 90-minutes of football is more like watching a film or HBO episode for them, although they may still be invested in one of the teams.
Football-studiers: A small subset, but one who'll be overrepresented enough in Get Goalside's readership to deserve a mention. People who watch the game for some kind of educational or academic purpose. Probably some overlap with a hipster/nerdy niche of TV-eventers.

Unless you're going to offer a customisable experience or multiple streaming options (the latter of which is happening a little more (see also: the Manningcast)), the broadcasting experience needs to cater across these different groups. This is probably why half-time and full-time punditry is the way it is: as much as the football-studiers may hate it, they talk about the talking points to serve the TV-eventers and the background-noisers, and to give the TV-stadiumers all the slo-mo replays they need to be angry about whatever they're inevitably angry about.

Let's flick back to the Data Zone screenshot and see which of these profiles it might be appealing to. There are, I'd say, five features on the image: a player name, the speed of their shot, team running stats, 'top players' running stats, and a 2D overhead pitch map. There's also one feature-by-elimination kinda thing: the reduced pitch size.

Image form NBC press release article

Does any of this help the background-noisers? Yes: the name tag, and possibly the pitch map. Does anything appeal to the TV-stadiumers? The shot speed might make the viewing experience feel a little more visceral I think, particularly if it can visually differentiate between a screaming fast shot and a particularly tame one. But the surrounding stats take away space from the pitch, which might be a worse experience for them.

For TV-eventers and football-studiers, the choice of stats is where the success would live or die, I think. TV-eventers probably care more about the actual action than the sideshow, so taking away pitch space feels like it would be worse for them. Maybe it could be rescued if the stats bring out narratives that everyone's going to be talking about. Football-studiers will want to see the pitch for different reasons, but the stuff around the edges might be useful to them if it's high-quality.

The aim has to be being 'entertaining'. Or, to use another word, 'fun'. But what is 'fun'?

Subscribing to and supporting Get Goalside, that's what

The Oxford English Dictionary defines 'fun' as...

'Fun' is important, but hard to capture, and there's often a tension between people steeped in the traditions of a sport and people who, well, don't find the traditions that fun. Cricket's Twenty20 format is an obvious example. For a footballing one, the recent storm in a teacup around the WSL official Twitter account's tweet of a Barbie meme is a timely example (the Barbie movie meme-train was joined by the official Women's Super League account, uncertainty about the suitability of comparing women's footballers to Barbies was had by some, tweet was deleted, Discourse ensued).

I liked Flo Lloyd-Hughes' tweet about it:

Too busy to properly engage in the Barbie discourse but I will say, when we think about making women’s football culturally relevant, engaging in an internet meme and riding on the coat tails of a huge Hollywood film that has cultural cache isn’t the worst idea in the world imo - @FloydTweet [tweet]

As much as we might like to think so, people don't necessarily follow sports because of the intrinsic qualities that it has. In part it's about whose friends are into it. For example, people like making memes and if some fire memes (see, Get Goalside can be 'fun') are made around a sport then it helps people engage.

(Related notes: Football Beyond Borders recently released a report, approaching a year on from England's Euros triumph, that suggests inner-city teenage girls in England still aren't that engaged in women's football. This seems all the more concerning considering that eight of the Women's Super League's 12 teams come from three cities (London, Manchester, Liverpool). The latest Unofficial Partner podcast episode also featured Finn Bradshaw who spoke (among other things) about his time at Cricket Australia where initial assumptions around marketing women's cricket were that the fans would mainly be women, but that the first people buying tickets were male cricket fans who had daughters. Observationally, I think this applies to women's football too; but to what extent does the daughter whose dad buys them a ticket when they're seven ends up buying their own ticket when they're 17).

Clearly, though, 'fun' is different for different people. However, there must be some commonalities which can be applied across groups. For example, in a more general sporting sense, people have different levels of capacity for enjoying frenetic action and random outcomes, but sports have a tendency to try and balance themselves somewhere between two poles.

A level of comfort or togetherness is probably a common component of fun, thinking of why warm-up acts exist in live shows and why football chants rarely take off unless a critical mass of chanters is reached. Being good at something usually helps make it fun too (or, at least, being bad is often one of the things that puts a roadblock on fun), although I'm not sure what 'being good' at watching sport means.

Let's think about how this applies to the made-up TV viewer segments. For background-noisers and TV-eventers to be familiar with what's going on they might need to be reminded who people are and what the stakes are. It brings them together with the slightly more knowledgeable fans that are watching, and catches them up on things they might have missed.

Think about a sport that's gotten a lot of hype recently, in Formula One. Their on-screen graphics show the current driver order and rotate through different sets of stakes quite regularly: gap between drivers, fastest lap, number of pit stops. It doesn't make use of the fancy tracking data but honestly, if we're adding new graphics to TV broadcasts then the league table should probably be one of them (but keep a special design for traditional 'As Things Stand' moments pls).

Use a momentum chart or xG timeline or shot map or something. Probably a momentum chart. I reckon it would appeal to TV-stadiumers, who want the emotional reinforcement of the momentum being behind/against their team; it serves as a talking point for TV-eventers; it helps background-noisers get caught up on what the match has been like while their focus has been elsewhere.

As for the stats themselves, could there be some use in comparing a player's match figure to their season average? Or to other players in the league? The TV-stadiumer would probably get value from knowing more about one of their players; it'd serve as a benchmark for more casual TV-eventers or background-noisers; depending on the context, football-studiers might get something from it too.

Glow pucks

If you've found this interesting then I thank you for reading this far. I dunno what I'm talking about but it's interesting.

One day these things will likely settle down and a new visual language of broadcasts will be established. I kinda suspect that what gets added to main TV broadcasts will be quite minimal and uninvasive (look at how little space current scorebugs take up and how unobtrusive graphic-generated ads are) but maybe the tech will move towards more customisable options. Who knows.

For the moment, here's to a true Hall-of-Famer piece of on-broadcast data graphics (which, it turns out, arrived at us via hockey pucks).

A Pandora's Box of football media stats use

Wed, 12 Apr 2023 00:00:00 GMT

It's 2009 and you've just stepped out of the cinema when a version of yourself from 2023 appears in front of you. You exchange exclamations, pleasantries, a copy of Grays Sports Almanac, and then get onto the movie you've just seen.

The 2023-you reveals that, just like the original, Avatar 2 has smashed box office records. "Wow," 2009-you says, "and do you still have to wear those glasses for the 3D?" "Oh, no," says the 2023-you. Your 2009 mind is blown: "Wow! Technology must have gotten really good then!". "Uh..." 2023-you errs, "no, no-one really makes 3D movies anymore."

It would be very exciting to go back to summer 2014 and tell the readers of Colin Trainor's StatsBomb post that, almost a decade later, that metric, as well as expected goals, would be getting mainstream usage. 'The standard of media and punditry must be really high', they might think. And yet...

A recent post by Casey Evans for Analytics FC ended with the following paragraph:

The data is there for the mainstream media to pull information from and there is also a massive pool of writers and analysts willing to interpret it for them. The question now is whether football media will take the next step needed to bring their coverage up to date.

The point Evans makes is one that, on the whole, I'd agree with - that despite some increase in data availability and a big increase in data consciousness, it doesn't seem to be well-used by the media industry. (There are exceptions of course - I think Sky Sports, with their online output and Monday Night Football in particular, have done pretty good work - as Evans also points out, but on the whole things are slightly more 'as they were' than you might have expected).

I think that there are three main reasons for this, which each have their own 'solutions'.

Stats are new
Football is hard
[insert 'el problema es el capitalismo' meme]

All three are kind of interlinked. Stats being new means that people in media haven't had time to 'become fluent' with them, but a different reason why people in media haven't had that time is because of what jobs look like (produce as much as possible to house ads) and what editors look like (they don't).

But football being a difficult, pesky, dynamic sport means that it's hard to pinpoint statistics that really cut-through the noise. Expected goals does a great job at what it does; everything else is messier than a Double Pivot Salacious Gossip podcast. Evolution will take its course and better metrics will rise, but, like we said, all these football stats are still pretty new. And with the money in football analytics pointing to proprietary information (mostly inside clubs), media mostly gets the leftovers in the fridge.

(Apart from media companies who purchase the services of Twenty3 and their Toolbox of goodies, the finest employer of the writer of this newsletter that there ever was. Although seriously. They also do pro-side focused work too. Get in touch with them.)

Darwinian forces are starting to have more of an impact though. Expected goals timelines and momentum charts are beginning to become pretty commonplace, neat visual ways of capturing quite a lot of information.

Liverpool 2-2 Arsenal

What. A. Game.

A combined xG of 5.4 - third-most in a Premier League game this season - with Liverpool accounting for 4.0 of that, almost twice as much as Arsenal have faced in any other league game all season.

How big will that point be for the Gunners? pic.twitter.com/49ilOE1roI
— Opta Analyst (@OptaAnalyst) April 9, 2023

Talking strictly stats, here's a list of things I'd consider throwing into a basic toolkit:

xG
High turnovers
Counterpressures (if available)
10+ pass sequences (spells of possession with 10 or more passes; ten is a bit of an arbitrary marker, but it makes you go 'huh!')
Fast-break final third entries (this is one I've made up on the spot but I conceptually prefer it to 'direct speed' that is used in some places)
PPDA
Crosses and cut-backs (both the volume and the relation between them would be interesting)
The passes -> final third passes -> shots chain

I think that those ten or so stats would cover most of what you want to look at on a team level. They cover a range of tactical approaches and are a good combination of effective at explaining something and quite easy to picture.

Here are some things I'd put in a booster pack:

% of goal kicks short and % of opposition goal kicks short (the former is simple, the latter can be a quick implied judgement on a team's high press when you look at the average for the season)
% of possession sequences starting in the defensive third that reach 5+ passes AND/OR % of possession sequences starting in their own half that reach 5+ passes without attempting to enter the opposition half (I've only used the first of these, but think they're both intriguing looks at how a team approaches build-up in a fairly simple way)
Post-shot xG
Expected pass completion and how it compares between two teams in different parts of the pitch/different situations (I think this is most interesting for players though)

There are some stats that haven't made either list not through oversight (although some might be oversight) but because I, personally, find stats easier to work with if I can see them. I know what a spell of possession with ten or more passes looks like; I know what a fast-break into the final third looks like; I know what a high turnover looks like. Like band and brand names, almost anything can be a household name with enough push, but being tangible certainly helps.

Players would have different toolkits to teams. Part of this is because some of these stats (e.g. PPDA, sequence-based stats) are literally, in the way they're constructed, team statistics. But final third passes is a great example of a different reason: it's pretty useful to know how much a team spends in the attacking third, particularly compared to their opponent; it's not that useful to know how much time a player spends there.

Inevitably, we arrive now at "what is football" theory, though I will keep it mercifully brief.

I assume that part of the difference in 3D execution a decade-and-a-half ago came from experience and time, but part from just better ideas about how and why it should be used. The same is true with football data - knowing what you want to measure helps you get to better metrics.

If we want football media to make better use of data, that's where it needs to start.

'What do you mean?': GPTs and post-Opta Forum thoughts

Thu, 23 Mar 2023 00:00:00 GMT

"With what shall I fetch it..."
"In a bucket, dear Henry, dear Henry, dear Henry / In a bucket, dear Henry, dear Henry, in a bucket."
"But there's a hole in my bucket-"

A centuries-old children's song there, telling the story of Henry's battle to get a helpful answer out of ChatGPT.

AI inevitably got an outing at this week's Opta Forum, the annual event held by one of football's major data providers. Chief Scientist Patrick Lucey opened the day by weaving neatly from talking about the GPT news du jour to how machine learning applies to Opta's data and products, all tied together with a neat overarching point about how Open AI's models actually function.

These things, and ones created by Open AI's competitors, are large language models (LLMs), although with things like DALL-E GPT4's image recognition they're not really limited to words anymore. Sport though, Lucey put forward, might not be an easy place to slot these LLMs in. These things might be 'language' models, but they're generalists, and sport has a very specific language of its own. Even human coaches sometimes have to learn the new vocabulary of a club and staff when they arrive at a new job.

Chat GPT Prompt: "Write half a paragraph about how, thankfully, the data provider Opta is here to save the day with their suite of data and product development, in a lightly but friendly needling tone"
Chat GPT Output: "Well, well, well, it looks like we have a hero in the data world! Opta, the trusty data provider, has arrived to save us all from drowning in a sea of information overload. With their impressive suite of data and product development, they're like a beacon of light shining through the fog of confusion. Thank goodness for Opta!"

Ok, that's the last time I let ChatGPT write any copy. That was significantly worse than I was expecting.

It's a little funny that LLMs are entering public consciousness at this current point in football data's evolution. I think you could argue that both the large language models and tracking data are at the early stages of exponential curves in usage. Versions of them have been around for years, but now the tech is usable enough and available enough, with enough people experienced in them, for take-up to rocket.

As well as this, both offer a tantalising possibility of opening up a whole new world to people. Search engines once did this, but then search engine optimisation (SEO) utterly borked Google; not only does GPT offer to bring the world back to your fingertips, but if you don't understand any of it you can ask it to explain in more detail, or simpler detail, or in the style of a pirate. (Actually, maybe be careful about pirate styles).

Tracking data, when paired with a good event data source, does the same thing for football analytics. You no longer have to stick to on-ball events - the equivalent, perhaps, of going direct to websites you know in a browser - instead you can go "hey, is our young centre-forward lurking in the central defender's blind spot as much as we've been telling them to?". And get an answer. (With a lot of data engineering).

But with the promise of so much comes a lot of potholes.

Hey, if you've got this far I bet you're enjoying this. Subscribe to Get Goalside if you haven't already

I've been working on a project that I hope Get Goalside will see soon, using our little Chat-3PO as a helper on some areas I don't know a lot about. There are times when it can be extremely helpful. But there can be times when, like the hole in the bucket song, you go round and round with it always seeming to be on the verge of getting the point, without ever actually getting there. Part of this was because of the limitations of LLMs, but part of it was just me losing sight of what I was actually trying to do.

Perhaps counterintuitively, in the world of seemingly limitless information, knowing when something is 'good enough' becomes a skill, because it's one way to stay focused on the real problem at hand.

The stat 'progressive passes' is a good example of this, I think. Definitions vary, but they generally revolve around a certain distance to goal being achieved. The current Premier League leaders, out of players who've played 10 or more games, are Oleksandr Zinchenko, Kevin de Bruyne, Rodri, Kyle Walker, and Thiago Alcantara (stats from FBref). Five undeniably good footballers. But you might notice that at least three of them play in deeper areas of the field, on teams who are dominant in possession. Maybe they're playing against defences who are set up in a way that makes it easier to achieve these passes. What you could do is tinker with the definition so much that you try and capture all of the possibilities to iron out potential issues and weird tactical interference... or you could accept the imperfection and just keep in mind that it might do 70% of what you want it to but at least it's quick and understandable.

This is why one of the research presentations, from Guillaume Hacques, appealed to me a lot. The title was 'Destabilising a Set Defence: Identification of Symmetry-Breaking Collective Movements', but you could boil it down to the difference in the direction that the two teams are travelling in. Combine all the player locations together and you have a team's centre of mass; if one centre of mass is moving to one wing while the other is moving to the opposite, there's something destabilising happening somewhere. There'll probably be some refinement that you could do, I'm sure, but it seemed right in that sweet spot in terms of bang for your buck and focus on the question at hand.

If you enjoy Get Goalside, support the newsletter for a mere child's handful of £ per month (£2-£6)

Part of the reason why events like the Forum, and the wider analytics community, are so useful is surfacing ideas like these. Because they're hard to come up with. Football's a complex, and always subtly evolving, game and human interpretation is too. Even the companies producing the data don't always (some might say tend not to) know how best to use the data they're producing. StatsBomb*, for one, have been very open about this in their releases of new datasets, their excitement at what the data could hold, the potential that might lay within it, all part of the buzz of the announcements.

*Or '[redacted competitor]', if you prefer.

Let's compare tracking data to proto-Ultron one more time. Compared to the old world, it feels like you can ask ChatGPT and tracking data anything you want and be able to get plausible answers out of it. I should know: I have spent far, far too long getting plausible (but ultimately unhelpful) answers out of ChatGPT.

As far as my project was concerned, the most useful thing that OpenAI could have added to the tool was something that, every now and then, might say "let's talk about what you're actually trying to do here". (Although this would've also relied on the LLM having the capacity for digesting information).

I suspect that the same is going to be true with a bunch of tracking data work. To an extent, this is always the question that data people have had to ask, and it's always been up to data people in the professional game to do that questioning and parsing of language and intent.

But I also suspect that the level of questioning is also going to grow too. There's more data available of course, but nowadays, and in future, there's going to be more questioning from coaches. As players grow up with data and become more empowered/encouraged to look at it, their queries might be in the mix too.

We're going to see that good problem-solving isn't just about the data or the tech. It's about recognising at what point 'dear Liza' should've just lent Henry a knife.**

Thanks for reading. You can subscribe to Get Goalside or become a supporter through the button below

**Henry's third problem is that the straw, which Liza suggested he mend his bucket with, is too long. All his subsequent problems are about sharpening his own knife.

How confident should Arsenal fans feel?

Sun, 05 Mar 2023 00:00:00 GMT

You can now follow Get Goalside on Instagram at getgoalside.analytics

Arsenal fans, how confident are you feeling about the Premier League?

The state of play right now is like this: Mikel Arteta's team have a lead of five points over Manchester City, and each team has 12 games left to play. Both still have games against Liverpool, Chelsea, and a strong-performing Brighton to come, as well as a huge clash against each other. Both are still in their European competitions.

We'll all have our own gut feelings about how things might go, but maybe you'd want to check some kind of statistical model to see if it matches up with what you think. You are, after all, the kind of person to read an analytics newsletter. So you go to US pollster celebrities FiveThirtyEight, who at time of writing give Arsenal a 54% chance and put City on 44%. Then, for a second opinion, you check Opta. They put things at 50%-49%. Hmm. And for a third opinion, maybe the Euro Club Index: 51%-47%.

Note: it doesn't look like any of these three had updated since the Liverpool-Manchester United match at time of writing on Sunday evening, a result which will have reduced United's slim title chances and marginally increased Arsenal and City's.

One of these models makes the contest look like it's on a knife-edge. One gives Arsenal a bit of breathing space. The other is Goldilocks-like right in the middle. What gives?

There are, as far as I can tell, two basic parts of this modelling: determining how strong each team is, and simulating what happens when two teams of strength X and strength Y play each other. Differences can come in both parts.

All three of the modellers mentioned give Manchester City a stronger rating than Arsenal. I think this basically matches with what most fans think too, that City have a better squad and better depth available, that Arsenal could drop points but might not have enough time to drop five more than City do. The way that they rate the teams is different though, and this looks like it could explain part of the difference in the predictions.

FiveThirtyEight's system (which gives the 54%-44% race) starts each season afresh. They base a team's rating two-thirds on what the team's rating was at the end of the previous season, and one-third on their Transfermarkt squad valuation.

On the other hand, neither Opta (50%-49%) nor Euro Club Index (51%-47%) seem to mark the end and beginning of seasons in such a way. The ECI's methodology notes that more recent results get a higher weighting in the team ratings, but things work on a more continuous basis for the two of them.

I mention this because FiveThirtyEight's system give Arsenal the second-best rating in the Premier League, whereas both Opta and the Euro Club Index put Arsenal fourth-best. Maybe there's something else in the modelling going on, but given that the Gunners last two seasons' saw them finish eighth and fifth, it seems possible that FiveThirtyEight's preseason resets have helped Arsenal climb the rankings quicker than in the other two rating systems.

(This also means that if you wanted to game FiveThirtyEight's system then hacking into Transfermarkt would probably do it. Why you'd want to do that, I don't know. Maybe if you wanted your models to look better than the season prediction that everyone on my Twitter timeline uses).

On top of this, each system updates their ratings slightly differently after every game played, with FiveThirtyEight again more different than the other two.

They all share a similarity: teams get a boost if they perform better than expected in a game, and lose points if they lose a game they should win. However, FiveThirtyEight's ratings use expected goals as a feature while Opta and Euro Club Index's systems are (according to their methodology pages) purely results based. This means that a team could win a game but lose points in FiveThirtyEight's system, if their underlying performance was well below expectation, whereas they'd still gain points in the other two. Maybe this also helped Arsenal whizz up the ratings, if their underlying numbers were more impressive than their results have even been.

Arsenal fans, please put two fingers to the side of your necks. How are your pulses? Better or worse so far? Let's give City fans something to worry about.

For the third time, FiveThirtyEight has a distinctive factor, and this one I can say for sure I'm a fan of. On the webpage, you can hop back to previous points in time and previous seasons with very handy dropdown menus. Last season, Manchester City opened the year with a FiveThirtyEight rating of 92.1 and ended it on 93.5. This season, they opened with a rating of 92.3 and currently sit on 90.6. Number gone down.

This doesn't necessarily mean that FiveThirtyEight is saying 'City are worse than they were at the start of the year'. It could mean that it's taken a while for their rating to accurately reflect City's quality. Either way, number done down. Which means that, even though City's rating is still higher than Arsenal's rating, the gap has been getting smaller and smaller all campaign, and it could get smaller still.

The thing with these predictions is that, like the table itself, they're going to change with each passing gameweek. On the evening of the first of April, things could feel very different. Arsenal will have played Fulham, Crystal Palace, and Leeds; City will have played Crystal Palace and Liverpool. They won't have played a third team in that timespan because of their FA Cup quarter-final on 18 March, a factor that I doubt any of these predictions will be factoring in: other priorities.

Arsenal only have two competitions to think about, the league and the Europa League, and it's pretty clear which of the two they would care most about. City, for now at least, have three, and if they had to only pick one, I bet they'd rather win the Champions League. The ratings that each of these models assign to each team are singular - as far as I can tell from the methodology pages, they don't try and factor in potential squad rotation, which might result in lower-quality teams playing matches (to be fair, this would be a very difficult task to do).

What we see with these models are Arsenal being given an advantage, with a range of one percentage point to ten. But the fact that City still have other distractions means that I, personally, would lean towards the more Arsenal-favourable side of that range.

How confident should Arsenal fans feel about the Premier League? Cautiously. Depending what model you ask.

Any opinions on models or other prediction models of interest? Get in touch at getgoalside.newsletter@gmail.com, on Twitter at @get_goalside, or Instagram at getgoalside.analytics

If you like and appreciate the unique work of Get Goalside, you can become a supporter for just £2 a month.

High-fat data for low(er)-fat costs

Wed, 15 Feb 2023 00:00:00 GMT

Science can do many things, and soon it may be able to make low-fat chocolate that doesn't taste of bland disappointment. The outside of a piece is the most important for the texture, and humans get fooled pretty easily about taste-related things. So feasibly you could coat low-fat fauxcolate with high-fat chocolate and not be able to tell the difference. Even if you could tell the difference, the drop in fat and calories and stuff might be worth it. Who knows.

It's taken me a while to get around to playing with StatsBomb 360 data. There's been some interesting stuff that people have done with it, like identifying line-breaking passes or analysing on-ball positioning. The company's senior analyst Jaymes Monte looked at space that receivers got the ball in earlier this year. Many of the research papers at the 2022 StatsBomb conference featured explorations of this newly-available type of data. But I'd never touched it. What kind of analytics newsletterer does that make me, for shame.

Then I stopped thinking about individual stats and thought about team shape. The shape itself seemed to present a problem: the StatsBomb 360 frames don't give you identifiable player information, just the team, whether they're the player making the action, and whether they're the goalkeeper. Beyond that a left-back looks like a centre-forward. Now I had a challenge: can you identify player positions from the freeze-frames in a way that lets you look at team shape.

Before I go further, I'd have found this obstructively difficult if it wasn't for StatsBomb's statsbombpy Python package and the mplsoccer package. They're the real MVPs.

I'd also have struggled were it not for remembering Laurie Shaw and Mark Glickman's paper from a few years ago, 'Dynamic analysis of team strategy in professional football'. That piece of work was about determining team formations from tracking data, identifying changes in formation, differentiating between attacking and defensive formations.

Slight difference: they had the positions of all 22 players. In the match I chose to work with, I had an average of 13.

However, Shaw and Glickman wanted to designate formations, but I was happy to use StatsBomb's and just try and assign players within that. I only looked at the off-ball team, for two reasons: 1) in my experience a team's out-of-possession shape is often closer to the noted formation than their in-possession shape 2) it's the out-of-possession shape I was interested in anyway.

Let me copy and paste wholesale the part of the paper I found most useful:

The final spatial distribution of the outfield players is determined by the following algorithm: first, we set the centroid of the formation to be the position of the player in the densest part of the team, as determined by the average distance to the third-nearest neighbour. We then identify the relative position of his nearest neighbour, the relative position of that player’s nearest neighbour (ignoring any player already considered in the process) and so on, until the positions of all players in the team have been determined.

Or as a tl;dr, as Shaw and Glickman write in the following paragraph: "[T]he location of a player in a formation is dictated solely by his position relative to his neighbouring teammates."

There are probably a raft of smart, technical ways to do this. But I'm doing this in my free time so we're getting the cheap and easy way. First: Make some fake data for a formation (goalkeeper at (1,4); right-back at (3, 1); RCB at (3, 3); etc). Second: code a long iteration sequence that compares the angle of a 360 frame datapoint to all the teammate datapoints, and compares those angles to the corresponding angles of all the possible options in the fake formation data. The former is easier to understand than the latter.

For example: imagine a point floating near the halfway line on the right flank, with another six teammates in shot. That's six angles. Let's go to the right winger in our fake data. Compare the first of the six angles to the ten angles that the fake data winger has to its teammates. Look for the smallest difference in degrees. What you have there is the hypothetical "if this 360 frame datapoint was the right-winger, the angle is closest to this second datapoint being the defensive midfielder". Repeat that for all of the possible combinations, choosing the smallest difference in angles each time.

The results, it's fair to say, aren't perfect. Here's a particularly bad example, the out-of-possession team in blue and defending the goal on the right (where the goalkeeper* is). England, the defending team, are playing a 4-2-3-1.

NB: Player, position, and team name come direct from the data

*Because StatsBomb do note whether a player is a goalkeeper or not, we can be sure that we get these right, as well as avoiding falsely labelling an outfielder as a goalkeeper.

And another example (this time with Germany defending in a 4-3-3):

I didn't include any information other than the defending team's locations in these calculations, so there's no knowledge that the England player on the ball is their centre-back Leah Williamson. That makes it somewhat unlikely that the blue dot closest to her is the defensive midfielder, as the algorithm has labelled it.

However, it's not all bad. Germany are back defending in this next example, still in their 4-2-3-1. I don't quite think it's labelled every point correctly*, but it looks a lot more plausible.

*I think that what's been labelled as the LCM should be the left-winger and that what's been labelled as the DM should be the left-sided centre-back, with the RCM and DM shuffled to the right, out of the frame's viewable area.

I suspect that this could be improved if you focused just on passes and took the pass and ball receipt together, as this probably increases the viewable area. On the other hand, that would mean you have to deal with 'tracking' datapoints across two frames. Shrug.

If you take the average of the frames though, things don't look too awful. (This time the defending team are defending the goal on the right, for no reason other than inconsistency in plotting between my code files).

There's a problem with the goalkeeper, mind: because they're in so few frames we only see them very close to goal. It's also possible that because we know which datapoints they are/aren't, they're not pulled towards the rest of the bunch by false labelling. I also think the full-backs are a little weird, but don't have a strong theory as to why the weirdness occurs.

Still, you can segment the data a little in interesting ways. Here are the two teams when their opponents had the ball in the final third:

And out-of-possession positions for when the opponent were making on-ball actions in their defensive third:

The goalkeeper and full-back weirdness persists through each of these segments to my eye (not least because the goalkeeper isn't recorded at all for these high defensive positions). In these latest two images, the full-backs have a tendency to be deeper than the centre-backs as well as being wider than the wingers - I suspect that's a function of camera angling only catching centre-backs when they're in very advanced positions and them being absent otherwise. Maybe a future step in this process could be estimating out-of-frame positions.

Anyway, this doesn't seem bad for some quick work on a single game. Maybe things would look more robust if the averages were over a couple of matches' worth of data.

However, part of me also wonders whether the plausible-ness of these averages might be a bad thing. If it looks 'kinda right' might that give a false sense of security in trusting it? I don't have tracking data to hand to compare this to, and maybe it's totally wrong! It looks close enough to fool you though, and one's biggest fear should always be unleashing a plausible idiot ChatGPT-alike onto the world.

If that's the case, at least it's wrong enough to appear not-too-useful. Out of possession average positions might be useful for getting a sense of compactness, or maybe whether a winger is allowed to stay high rather than tracking back, but the fact that there's clear labelling error going on in this means you can forget that. The only insight I feel vaguely safe drawing from these images is that Germany probably defended a little higher than England. (A quick check of the data shows me that they did indeed have more pressures in the final third than the winners of the Lionesses, so looks like that is correct).

All of this becomes moot if StatsBomb add individual player ID to their 360 data, although that hasn't as of yet been announced. This might be slightly flawed, it might not be hugely useful, but then neither is low-fat fauxcolate coated in high-fat chocolate shells. But if you're able to do them right, maybe they could be low-key game-changers, for a certain audience at least.

Get Goalside is just about the only place you can get this kind of work. If you like it, you can become an official supporter for just £2 a month.

Thanks again to the StatsBomb's statsbombpy Python package, the mplsoccer package, and 'Dynamic analysis of team strategy in professional football' paper, they're all worth checking out. Each helped make this a relative breeze. StatsBomb's open data repository can be found here.

The European Premier League

Thu, 12 Jan 2023 00:00:00 GMT

I should start this post by saying two things:

1) I think that the strength of the English men's football pyramid is one of the best parts of the nation's game, a history of community institutions that can do tremendous good in their local areas

2) I could count the number of Football League matches I've seen in the last few years, live or on TV, on one hand, while I've watched at least a dozen matches in the top men's divisions of other European countries this season. (I'd like to blame EFL's design aesthetics for that but that's probably not entirely fair).

These two things both being true doesn't feel right, and I was interested in why this is the case for me. And what better way to investigate one's feelings than with data analysis.

A lot gets made about promotion and relegation between the Premier League and Championship, as well as the gap between the relegated teams who receive parachute payments and other EFL sides. Maybe this is true, but in terms of promotion and relegation, the Premier League has never been easy to stay in.

By my count, four out of five seasons from 2002/03 to 2006/07 saw two of three newly-promoted teams go straight back down - the big step up in quality is nothing new. The year when Fulham, Blackburn, and Bolton came up and each stayed in the top-flight for over a decade (2000/01) was a huge exception.

As well as that, there aren't really more 'yo-yo clubs' in recent years than there were at the start of the century. There are two teams who've been promoted three times in the past ten Championship seasons (not including the current one of course): Norwich and Fulham. In the first ten seasons of the 21st century, starting with 1999/2000, there were also two: Birmingham City and West Bromwich Albion. Here's another stat: in that early set of ten seasons there were 23 different teams who earned promotion to the top-flight and in the most recent ten seasons it was 21.

[Please don't @ me about when the 21st century started. This is a newsletter which might attract people who would @ me. Please don't @ me about calendar systems. But do subscribe if you haven't already].

Despite these similarities in pro/rel, I still feel less of a connection to the Championship and the leagues below it than I used to when I was younger.

Part of this might be TV availability. When I was younger, the Premier League was on Sky Sports, with a series of second fiddle channels struggling to get off the ground. BT Sports' success in becoming a second sports subscription for elite football (first by sustaining the rights to big Premier League games, then getting the Champions League) has meant more access to European football, as those leagues are ones that BT fill their airwaves with. It's now just a lot easier (by paid means as well as murkier unpaid means) to watch the league games of the big, shiny European teams.

However, there's something else, something that I wanted to check with data. The pool of players in the Premier League nowadays owe less of their careers to the English Football League than they used to. At the same time, they owe more of it to the rest of the 'Big Five' European leagues.

If you like the work Get Goalside does, you can sign up as a paid supporter for £2 per month

What follows might not be the absolute best methodology, but it's a methodology - for the particularly data-minded, I'll include a link to my code at the end of the post.

We'll take all players who've played 450 minutes in the Premier League this season (up to the FA Cup third round break), the equivalent of five games, and call them 'active players'. We don't care about the others, and we particularly don't want ones who might be ageing squad-filler whose lengthy careers might skew this analysis. Using FBref data we can get the figures for all the minutes these 'active players' have played throughout their careers, in all the different leagues. (We'll discount the minutes for 2022/23, because it's basically all going to be Premier League).

Ok, now for the results.

For 2022/23's active players, the EFL (including League One and League Two) accounts for 21.4% of their total prior senior career league minutes. The rest of Europe's Big Five leagues (La Liga, Ligue 1, Serie A, Bundesliga) account for 19.3%. Pretty even, but with the EFL just on top. Most of the rest of the career minutes came in the Premier League itself.

However, I have a hunch that newly-promoted teams will skew this. They've just come from the Championship, so most of their players will have played some significant time there, and they'll have had less time to adjust their squads for the Premier League. There are different ways you could try and account for this but the easiest way is simply to cut the players on newly-promoted teams out of this analysis.

If you take the active players of the other 17 teams, the balance shifts heavily: the EFL's share drops to 17.1% while the Big Five's nudges up a little to 20%. The modern Premier League maybe owes more to Europe than the rest of the English pyramid.

We should do a comparison though. There's a chance it's been like this for a while, or not too different to it at least. I checked the same thing for 2017/18, using 1000 minutes as the cut-off for 'active players' this time, seeing as it's a full season so more players would've had the chance to reach the 450-minute mark I used for 2022/23. (When calculating the proportions of career minutes, I also didn't include the 2017/18 season - I'm basically looking at the careers of active players up to that point in time).

Before I saw the results I was fully expecting to need to go back even further to find anything significant. I was already dreading having to do the same work for the 2012/13 season just to have something to write about. But it turns out I didn't need to. Because the results in 2017/18 were just that noticeable.

For 2017/18's active players, the EFL accounted 23.6% of career minutes. Meanwhile the rest of the Big Five only accounted for just 16.4%. Comparing like with like - all 20 teams in 2017/18 to all 20 teams in 2022/23 - the EFL's career minutes share has gone from 23.6% to 21.4% over the course of 5 years while the Big Five's has gone from 16.4% to 19.3%, the gap narrowing by five percentage points. It's not a revolution, but it's still a pretty noticeable change.

We saw a moment ago that the presence or absence of newly-promoted teams made a difference to things though. Let's put the same figures together while ignoring the newly-promoted sides. The EFL's career minutes share goes from 20.4% for 2017/18's 'active players' to 17.1% for the 'active players' of the 2022/23 season. The rest of the Big Five leagues' minutes share goes from 17% to 20% in that five years. The gap was narrower to begin with but it completely flips!

The same pattern is true if you just look at the number of players who've played in the EFL and elsewhere in the Big Five as well. There are more 'active' Premier League players who've played in Europe's elite now compared to five years ago, and there are fewer who've played in the EFL.

The 'tl;dr' of it all: whichever way you cut it, the Premier League is noticeably more reliant on Europe's Big Five than it was five years ago, and less reliant on the EFL.

Get Goalside is just about the only place you can get this kind of work. If you like it, you can become an official supporter for just £2 a month.

You could talk a lot about why this has happened (I assume it's mostly the money flowing into the Premier League, maybe a change in scouting and transfer practices) but I'm less interested in exploring that here. This piece has been gestating for a month or more now and it's hard not to see the present footballing world through this lens when it's on your mind. Here's three examples:

Item One. Of the six Premier League managers hired mid-season (so far), one has been an interim manager made permanent (Gary O'Neil - Bournemouth), one a fellow Premier League manager (Graham Potter - Chelsea), one was a Football League manager (Nathan Jones - Southampton), while the other three came from Europe (Roberto De Zerbi, Unai Emery, Julen Lopetegui - Brighton, Aston Villa, Wolves).

Item Two. Leeds United's academy manager said to the Training Ground Guru website recently that he thinks English football should reconsider B teams. Part of this is about loan rules in England, where Premier League teams can't do loan dealings outside of a transfer window while EFL clubs can, but part of the whole B team dynamic is about Premier League teams having an outsized degree of power and squad strength in comparison to the rest of the nation's pyramid. As several people said when I tweeted about this, maybe players would get more first-team football if they weren't hoarded at Premier League academies in the first place.

Item Three. The Athletic expanded their football coverage for La Liga with seven new hires this past week, including two club-specific writers each for Real Madrid and Barcelona. Grace Robertson quote-tweeted the news saying (I think completely correctly): "Felt like this was a no brainer for a while. There has to be a bigger audience for English language coverage of Real and Barça than most British clubs.". To be clear, the tweet doesn't match exactly with what this newsletter has been talking about, as 'English language coverage' includes the US and everywhere else, but it's a similar subject; nowadays, even for an English audience, would there be more money for The Athletic in Spanish duopoly coverage or Championship play-off race coverage? What direction is that balance travelling in?

There's something, an idea, I'm grasping at here that I can't quite hit. It's a somewhat popular rhetorical thing to say at the moment that the 'Super League' already exists, it's just the Premier League. And I guess this isn't just because England's top-flight is where so much money is and so many top managers are - and consequently, through both, so many top players - but because the influence of the rest of the English football pyramid that the league is still supposedly attached to is waning and waning. And so if, like me, you watch the Premier League because it's the top men's league of your country, why would you feel more of an attachment to a Championship game than a top-half Big Five league game? It's the latter set of teams who are the Premier League's true peers, and increasingly the latter set of teams who are providing the Premier League with a player pool.

(It's worth noting that this isn't entirely the same as the 'are English players getting opportunities' discussion, given that English players can play domestically or abroad at any stage of their adult career, although it's naturally intertwined).

I can't completely blame this pattern of changing playing histories for my tendency to not watch much Championship football. Some of it is the design aesthetics (that logo, those fonts!). Most of it is that I just don't have the headspace. But I'm reassured that the data seems to back up a gut feeling I'd had: if I tuned in, not only would I be unlikely to watch a team that'll spend time in the Premier League (although the odds haven't changed that much there), I'm increasingly less likely to be watching players who'll spend time there in future too. That's not the only, certainly not the main, reason to watch the EFL of course, but highlights the divide.

It'll be interesting to see what this looks like in five years' time. A heads up for whoever does that work: we'll have to contend with the added factor of Brexit in our analysis then as well.

Thanks for reading. Please get in touch with any comments or questions, either on Twitter or at getgoalside.newsletter@gmail.com.

The code used for this newsletter is available here

Goodbye 2022, Hello 2023

Sat, 31 Dec 2022 00:00:00 GMT

Hello to you, dear reader, and hello to the oncoming 2023.

2022 was a year. England won the women's Euros, that was fun. Real Madrid pulled off some incredible Real Madrid-ing in the men's Champions League. Get Goalside reached 100 posts.

Back at the start of the year, I sent out some thoughts on what 2022 might bring in the world of analytics. It was hit and miss: definitely a year where data engineering got more of a spotlight, not one where clubs started using analytics as meaningless buzzwords.

Instead of predictions then, here are some general 'things on my mind going into 2023'.

Football analytics jobs conceived of as an entire ecosystem

Jobs around football data have always been more than just clubs. That said, 'how do I get a job in football analytics' tends to implicitly be asking about club football. It's then up to the person answering to point out other things.

I don't know if this'll ever completely change, but I think the wider football data ecosystem is broadening out, and that might mean people aren't quite so focussed on club jobs. You have jobs at data providers (of all types); multi-club organisations, where you're not tied to a specific team; player agency or management businesses; national associations; media outlets; consultancies; product-building companies.

There are far more clubs than these non-club entities, but all of the above sectors are likely to grow, at least a little bit, in terms of either number of hires or number of companies.

Set-pieces

This probably should have been a 2022 thing. Set-piece design is definitely mainstream now (in the Premier League at least).

An early draft of a thought: Will set-piece design change our idea of aerial ability like cut-backs and xG changed the conversation around 'finishing ability'?

But a warning: I already watch NFL, let's not turn this football into set-play-ball too, ok?

Skeletal data

We're now in a world where not just tracking data (Football Manager 2D mode) but also skeletal data (Football Manager 3D mode) is becoming a bit of a standard in some competitions too. You'll have seen it in TV broadcasts for offside calls, including what I guess you could deem a booty call in the World Cup final.

We're a long way from it being widely used, I think, partly because time is limited at clubs, there's not much data out there for 'the community' to have a go with, and the file sizes must be ginormous with accompanying increases in skills required.

If you're interested in this area though, Driveline Baseball's OpenBiomechanics project might be worth checking out. I'm not sure how much of the insights across the two sports will be transferable, or how similar the data is to football skeletal data, but where anything in the same ballpark will be available.

I'm aware that this skews heavily towards analytics at the top end of the game - let me know what it's like at other levels if you work elsewhere!

Sideline data

I forget where and how much this is allowed in various competitions, but it seems like The Next Thing.

A lot of other things in analytics have been 'data people going into football', but I imagine that this would be more 'football people picking up data'. This is partly because I think in-match analysis requires more domain knowledge than other areas of work. Elsewhere, a large part of data's use is in distilling information that is too widespread and multitudinous for a person to grasp; in-game there's less that needs distilling, more that needs expert interpretation.

Maybe they could put some of these suggestions into action.

Global game, global analysis

Three pieces of media towards the end of 2022 focused my mind on how England/US-centric a lot of analytics history (or the analytics history that is told in English) is. There might be a money-plus-Moneyball reason for this being the case, but my current feeling is that things are going to be different moving forwards.

In some ways, I think the pandemic accelerated this: it may be correlation and not causation, but having to hold analytics conferences virtually seemed to expand options and the geographic pool of presenters.

But as well as that, and StatsBomb releasing a free Indian Super League dataset, there are concerted efforts in non-Anglo countries that will probably bear fruit in a noticeable way soon. Spain has LaLiga tech, the make-up and quantity of analysts at Ligue 1 clubs intrigues me (Skill Corner, a fairly established company now, is French, as is Footovision, one of only two FIFA-accredited broadcast tracking data systems).

This is an area I'm lacking a lot of knowledge on so please get in touch if you know stuff.

Ethics in analytics

Analytics often has the objective of optimising things and making them efficient, but it's worth considering what is being optimised for and what is considered expendable in the pursuit of that optimisation.

One method of optimising an allocation of resources is the multi-club ownership (MCO) model. Big fashion trend, and also some of the most interesting-sounding analytics work out there. Hey, I'm not MCO-ist, some of my best friends are MCOs. Maybe multi-club ownership isn't intrinsically bad. Is consolidating ownership of football clubs in the hands of a smaller group of owners worse than whatever system we had before? What are the xSocialGood levels?

(A related thing that I found to be an amusing coincidence, although multi-club ownership is not the cause of what happened, the recent Melbourne derby that got postponed because of crowd trouble was also a clash between two MCOs, City Football Group and 777 Partners (reporting suggests 777 only have a minority stake in the Victory unlike CFG who outright own Melbourne City))

Finally, for some work on something else I found myself reading about Leeds United's measurement of player weight and body fat levels under Marcelo Bielsa. [Content note on the next five paragraphs for eating disorders]

I'd like to pair a quote from an article about it from The Athletic, with this introduction to a piece on the Women in Football website:

“We were weighed every day and sent to fat club. You got on the scales daily and someone would just say you needed to go and do more sessions or not.” – Claire Rafferty on how football fuelled her disordered eating.

Now the Athletic quote:

The reason it works, as one employee at Elland Road put it, is that there is nowhere to hide. Bielsa has his squad weighed and monitored for body fat-levels every morning. They do daily skinfold tests and the results expose the smallest of excesses (one player was panicking on his return from holiday after putting on a quarter of a kilogramme in Dubai).

I don't know to what degree Leeds' methods, and those like it, cause or exacerbate eating disorders or disordered eating. One study cited in a PFA article on eating disorders found that, in their sample of Norwegian elite athletes, eating disorders were three times more common than in the general population. However, I think it's one example of where data and the pursuit of it could be more mindful of its effects on the individuals being monitored.

I hope that 2023 brings good things to you all, and a great women's World Cup in June/July. Get in touch at getgoalside.newsletter@gmail.com if you have any thoughts on my topics here, any things I might have missed off, or just to say hello.

Happy New Year

Thank you for reading - if you enjoy the work of Get Goalside you can become a paying supporter to help keep the show on the road

What if you let loose a data scientist on in-match decisions?

Fri, 23 Dec 2022 00:00:00 GMT

Each year that Get Goalside has been running when Christmas has rolled around I've sent out newsletters that try to max out the silly while also being interesting.

In 2019 it was a broad one, 'The silliest things you can do with advanced football tech'. Last year it was broad in a different way, 'Pick a sport, any sport, and take ideas from it'. So this year we're gonna go specific: what are the stupidest in-game changes you could make with very advanced use of data.

Time-wasting

You could go about this two ways. The first is to work out when time-wasting will help you; the mental model we all have probably works fine, but there might be some times when the data surprises you. The big factors are going to be score, proximity to the end of the game, and a little bit about how well each team is playing.

However, the more fun way to use data for this would be to help change your style of play to make time-wasting easier. Are your long balls going straight through to the goalkeeper? Is the opposition left-back slower to take throw-ins than the opposition right-back? Would it be better to clear a cross up-field or smash it out of play into the crowd? Instead of building a model whose ultimate aim is goal-scoring, build a model whose ultimate aim is wasting the most time.

Data can make the art of time-wasting even more elegant.

Designated tactical foulers

There are teams who definitely seem to rotate 'tactical foul' duties, and it's not like it's a complicated system: "I just fouled someone, can you foul next time buddy?". But teams are getting the hang of data nowadays, so why not throw some numbers at the calculation?

If, as a manager, your team is being countered against, you could definitely use tracking data (which tracks every player as they move around) to see which of your own players is close to the counters. Could you use this information to put them in better positions to stop the attacks in the first place? Sure! But you could also pair that with some kind of 'likelihood of getting booked' model and, like, tell a full-back to invert and the central midfielder to stay wide instead because one of them has a history of more convincing cynical fouls or something.

Fouling more

On that theme...

All of the 3D images we've been getting of offsides in the Champions League and World Cup this season are produced with skeletal data. Lots of high-quality cameras all around the stadium have their images run through sophisticated software to know where every player's ankle joints are.

If you had some NASA-grade data people, I'm sure you could build some kind of system that highlights when a given amount of contact between two players would usually be called as a foul. With this, you might be able to say "hey, the data says this ref isn't calling fouls - or more specifically isn't calling this type of foul - we can afford to be a bit more aggressive."

I feel bad for referees now, but that would be a fun use of smart people.

Football Index Insider Trading

This is the one thing that makes me sad that the Football Index bubble burst.

If you're bored of building statistical models that tell you which players increase or decrease the chances of winning (who isn't!) why not pair that with "whose fake gameified stock value will increase the most if we win this game/start involving them in the game more/designate them as the penalty taker".

It would be morally wrong for football clubs/coaches to have been trading Football Index-type player stocks anyway, but if you're gonna be shady then you may as well do it right.

Sprinkler settings at half-time

At the end of the day, everything on a football pitch is a matter of physics. Newtonian forces and crap like that. Let's say you're a team who like playing passes in behind the opposition defence, and in the first half they've been bouncing too high for the forwards to bring under control. That's physics: the interaction of the ball and its speed with the firmness and dryness of the grass.

So maybe you could run those passes through one of your fancy physics models but you change some of the variables based on the pitch surface moisture level. Does the bounce deaden a little if the grass is wetter? How much pitch moisture would be too much before the passes just skid through to the goalkeeper? Adjust the sprinklers at half-time accordingly.

Would it make a meaningful difference? I highly doubt it. Would the story being leaked cause a great couple of days for fans to talk about? Absolutely.

Create leading survey questions to fool your web3 NFT-token-holding members into thinking they're taking part in a genuine democratic process

It's not something that really happens at the moment, but it's conceivable that, at some point, a bunch of crypto/web3-type clubs are going to try and give their members the option of weighing in on in-game decisions.

You may, like certain conservative-leaning political parties in certain countries of geopolitical significance, decide that too much democracy is a dangerous thing. In which case, why not work out what you think you want to happen, and then get some kind of Natural Language Processing system to write the survey questions and answers in such a way that it primes the voters to vote in a certain way.

Would it work? Probably not enough for it to be reliable but just enough to get you banned from being involved in running a football club ever again.

Conduct the crowd

I haven't yet read any research about the effect of geeing up the crowd on how a team plays, and in fairness it would be difficult to get the data, but it's the kind of research I think I want to read most.

Imagine your winger's low on confidence. They're not having a good game. You don't know how to change things tactically. But wait: the decibelometer indicates that the section of crowd that the winger is playing near are quieter than usual by two standard deviations!

You send word out that that needs to change. Luckily your intern has done a research project on home season ticket holders and their atmosphere network effects. You know the exact seat numbers most likely to start and sustain a chant. You contact your crowd management agent in their section and get them to start a loud and positive discussion of the winger and how they need a boost. The chant starts. The winger's head is held high. Three minutes later they go on a mazy, full-pitch dribble and score.

In another life you'd be overthrowing governments with this set of skills and support staff. But here, the only leaders you're deposing are your rivals at the top of the table.

Prompt: Write an article about potential AI use in football...

Sun, 11 Dec 2022 00:00:00 GMT

This article was not produced by an AI. Questionable that it was even produced by an I.

The Jetsons have a lot to answer for.

Of all the space-related culture that came out of the 1960s, the everyday life regular family Jetsons, who just happen to have a flying car and live in weird little towers in the sky, stuck. People say "where's my flying car" when they're dissatisfied by some new tech in large part because of The Jetsons and it's idyllic but boringly hopeful vision of the future.

You've probably seen some screenshots or articles about OpenAI's ChatGPT by now. You may, although this is a little less likely, have seen something about FIFA's Stadium Experience augmented reality feature of their app. It feels like we're on the doorstep of The Future, and there are a lot more companies in and around football a mere stone's throw away from the porch.

Get Goalside will aim to have some fuller pieces on these in due course, but for now let's get something out quick and easy on what some of these things could mean for the near future of football analytics. Call it a beta version, it's what tech would do.

AI

For simplicity and brevity, by 'AI' I'm going to stick to things that OpenAI list as examples you can use their tool for. There are too many to list them all but they include: Q&A, grammar correction, natural language (i.e. human language) to code (Python, R, SQL), summarising text, simplifying text.

If we were being cruel to ChatGPT, the thing that's caused the recent hubbub, we might say that it's a very successful example of Artificial Inintelligence. The Verge went for the term "fluent bullshit". It's read a lot of books and 'speaks well', but it's not too hard to trip it up (which might be reassuring?).

In my post for Get Goalside paid supporters [join for £2 or £6 a month] I managed to, without trying, get it to stumble into making up a citation. When asked to do some maths which required some interpretation, it gave an answer that was the exact worrying mix of surface-level plausible and badly incorrect.

Analytics staff who have to explain concepts to all of their colleagues might have been hoping that an AI chatbot could be used as a kind of FAQ machine. But alas, it'd probably do a bad job; or, worse, a good enough job that its mistakes go undetected.

The bauble of OpenAI that seems more immediately usable would be the set of 'natural language to code' features. Football is a long way from the large analytics departments of baseball, for example, and in many clubs a head of analytics department might be the department. At the very least, analytics staff often have to do their own data engineering, which is a little like asking your head chef to wire up all the cookers. Not that it's beneath them, they're just liable to get themselves killed.

There's been a lot of (justifiable) talk about the threat of AI 'art' to the creative economy, but I think the balance is different when it comes to coding. The old phrase goes 'writing is re-writing': nowhere is that less true than the cheap artistic work at the bottom of the corporate pile that might sustain a lot of artists, but nowhere is it more true than writing production-quality code. I don't know whether I'd trust an unedited webapp built from prompts by OpenAI, but it might save a few hours getting the first draft done.

AR/VR

If the problem with AI chatbots is that they're too smooth for their own good, the problem with VR and AR is the same problem as ever, which is that they're clunky as hell.

As intriguing as FIFA's AR Stadium Experience looks, I'm not hugely sure why fans would want to view stats projected onto the pitch via their phone screen. On the VR side, there are a few systems which aim at tactical analysis and/or reaction training, which have their uses but still sorta seem like they're finding their legs.

I mean that literally too. A favourite of mine (which, to be transparent, is also one that I've tried out personally) is from a company called Rezzil, for the pure fact that they've always hooked up users' feet rather than relying on a headset and those hand controller things. If I may be so bold, involving peoples' legs not only seems pretty useful in football, but absolutely essential.

Both AR and VR feel, to me at a distance, like they're just waiting for augmented reality glasses to get significantly better than they are now. Some current tech lets people see stats and other pop-ups, but it's a little way off Tony Stark's glasses in Marvel movies. I imagine that something like those - which can overlay more complex shapes than rectangles over moving objects - would liven up positioning and set-piece sessions considerably: project tactical pitch lines, opposition mannequins, or assigned run routes onto the grass through players' glasses as they stand on the training pitch.

Off-field, I'm kind of surprised that I didn't see a VR-plus-NFT ticketing scheme during the recent NFT bubble (if you have seen one, please let me know). In the half-year window when every footballer and their cartoon monkey were promoting some kind of collection it just seemed like this was an obvious step to come.

ML

Technically, separating machine learning out might not make sense, but it's different enough for me to justify it. It's a broad term that would technically include expected goals models, and is a step in the AI process, but machine learning itself doesn't necessarily aspire to pass the Turing test. The machines just keep on a-learning, and two things in the past couple of weeks were a great demonstration of where and what.

The first was at Training Ground Guru's Big Data webinar in Jonny Whitmore's talk showing some Stats Perform metrics. The company have demonstrated the underlying models before - expected pass completion, expected threat, expected pass target - but Whitmore broke them down into some really interesting statistics to look (principally) at some central midfielders.

Things like: how often did they make 'safe' passes, how often did they optimise likely completion with threat added, how often were they the target of teammates passes when they were an open option. (As with many things, Messi figured highly on that last one).

While expected goals was both a useful stat in itself and a building block with which to create others, this talk demonstrated how the same will be true of newer, tracking data-reliant models.

The second thing comes from Metrica Sports, who shared news and a dataset from the automatic tagging feature in their software. Their product is a video analysis tool, and this option tags up some basic datapoints when users upload game footage - no outside data source needed. I can't speak for the accuracy or consistency, but the demo example shows how a user could load in a video and let the tool take them to all of the moments when a team enters the final third from the left flank with multiple players in support.

It might not happen for a few years, but it's now not that hard to imagine a world where a lone data employee at a medium-to-small club uses software like the above to get immediate data from video, maybe for scouting purposes outside of their existing data coverage, and an AI bot to write the first draft of a database and webapp to store and present it. Or the staff at a larger club with multiple staff members use it to free themselves up for other tasks.

The 'Will AI kill us all' section

There are understandable worries about what might happen if some of these technologies are simply let loose. Will some precarious jobs just vanish? What happens to the first rung(s) on the ladder if the work usually done there can be done by AI instead?

It's worth saying that, even before talking about their impact, merely their creation is an ethics class in action. Models trained on as much data that can be found on the internet is kind of like the hypothetical we all waved away finally becoming reality: 'You know if you put all your stuff on the internet, it's there for everyone and can be used by anyone'. Lensa is the latest big focus for this, DALL-E has received scrutiny too. Keep an eye out for DeepMind's script-writing Dramatron to make headlines at some point.

If "hoovering up artists' output and creating products which can put them out of an income" isn't enough, there are other serious issues. Lensa has been reported to whiten the skin of people of colour and produce sexualised images of users, even when uploading childhood photos. Dramatron-produced scripts (The Decoder relays) featured some sexism. ChatGPT has some pretty bad biases that can be drawn out too. Although OpenAI have clearly tried to put in some guardrails to their chatbot, users found ways around them incredibly quickly.

This stuff, as far as I can remember from seeing re-runs, never happened in The Jetsons (although we never saw how they tested those flying cars). But whatever skeletons might lie in that society's closet, more than anything the Jetsons family made the future seem pleasantly mundane; they have a lot to answer for for that.

Get Goalside would love to hear from you if you have any thoughts or clarifications. Get in touch at getgoalside.newsletter@gmail.com

Everything you need to know about defensive statistics (well, the basics)

Sat, 03 Dec 2022 00:00:00 GMT

Stats have taken off, TV coverage is using expected goals, but what about defenders? As we'll see, defensive statistics are a little more nuanced, a little more sophisticated, but this article will cover all you need to know about the basics.

What stats are good for judging defenders?

Can you tell which defenders are executing their roles better than others?

What kind of defensive stats are about roles and what are about performance?

How is duel win rate affected by the role of the player?

What defensive stats should I look at then?

Why do data companies collect things differently?

So, what stats are good for judging defenders?

Let's start with a metaphor. Imagine you're a first-time parent and you keep track of everything you and your partner do to make sure you're even better parents for any future kids. After 18 months you look and there's one category your stats are way higher than your partner's for: catching the baby when they fall off things.

Does that make you a better parent or a worse parent?

Better: you catch more? Worse: the baby falls more? What's the point of the 'catch the baby' parenting analogy?

It's like a lot of defensive stats: is a higher number better? It's better than conceding goals, for sure, but is it high because something else is wrong somewhere?

And with statistics like goals and shots conceded, a single defender rarely has have full responsibility, or even most of the responsibility.

US sports are able to get around this a little with the amount of substitutes and rotation between games, but with such short seasons, few subs, and relatively stable line-ups, football doesn't work in a way that makes 'better/worse than replacement' stats easy either.

Can you tell which defenders are doing the roles they've been given by the manager better than others?

Kind of, but you need to know what roles they've been given to do that. Analysts in clubs will simply know that, of course, but on the outside we have to work it out from watching games or estimate it from looking at the stats in a slightly different way.

What kind of defensive stats are about roles and what are about performance?

Generally, the number of defensive actions and where on the pitch the player makes them are indicators of role. That doesn't mean that they're not linked to how well the player's fulfilling that role, just that big differences are likely to be about role rather than quality.

Things like tackle win rate or aerial duel win rate are more performance-based - but those can be affected by role too.

How is duel win rate affected by the role of the player? Surely it's just about how many they win?

Some defensive players will be encouraged to go for higher-risk tackles. As an example, a player who is asked to mop up behind a high press might be asked to attempt high-risk, high-reward duels.

More generally, previous work by Get Goalside has pointed out that tackle success rate seems to be lower in the middle of the pitch than near the sidelines. (Lighter colour on the visualisation below = lower success rate; smaller circle = fewer attempts in that location).

What defensive stats should I look at then?

Generally, high numbers of things like duels, tackles, interceptions will indicate who's an 'all-action' type of player and who isn't. If you want to look at tackle/duel success rates, it's best to compare players to others who play in the same position and attempt a similar number.
(e.g. compare a central midfielder who attempts 5 tackles a game to other central midfielders who attempt around 5 tackles per game, not to central midfielders who attempt 2 tackles per game - they're probably playing different roles)

Clearances usually tell you whether a player's team is under pressure a lot - the highest number of them happen inside the box in games where the opposition is swinging crosses in. Meanwhile recoveries are usually partly affected by the team's approach to possession - this is because some definitions of a 'recovery' require the team to have kept the ball for a certain length of time after the defender recovers it. A team who always try and play on the floor will be more likely to meet that threshold than a team who tend to launch it long.

Those are some general rules of thumb, but the way that data companies collect defensive stats varies a lot more than the way they collect passing or shooting stats.

Lol. Why do data companies collect things differently?

Apart from cross-shots, shots are easy to define without being controversial. Passes usually are too. But players 'duelling' or 'attempting to tackle' or simply standing in the way and getting lucky is often a bit more fuzzy and random. Shots and passes is usually higher up the list of priorities for attention too.

There also seems to be more cultural differences. In England, 'tackle' means a standing tackle or a sliding tackle; in parts of Continental Europe, what gets translated as 'tackling' usually means slide tackling only. Certain countries are more keen on using the term 'duel' than 'tackle'.

And then, a lot of defensive work happens where the ball isn't. It's not just tackles, but filling space to prevent passes and tracking runs. When data is collected by people watching video of televised games, this type of thing is harder to (cost-effectively) take notice of.

What is midfield for?

Sun, 13 Nov 2022 00:00:00 GMT

Have you ever, and you have no reason to have done, stopped to think about the story of the word 'Eureka's origin? There was a bath, there was Archimedes, there was an insight into measuring the weight and density of objects...

I hadn't, not since I was eight. I just looked it up, and it's really disappointing. It just means "I've found it" [source: Wikipedia]. For whatever reason, the story of Archimedes made it into our language but presumably ancient Greeks were saying that all the time. 'Where did I put my tunic' 'Do you want me to help look for it?' 'Never mind, Eureka'.

All of this is to say that I think I've had a good idea. Alas, the story of its origin is unlikely to get passed down from generation to generation, culture to culture; there was no catchphrase, no baths. There was barely an audible 'oh!'.

As we all know, the point of football is to win. The way you do that is scoring more goals than the opposition. If you want to do some linguistic algebra you could rearrange that formula slightly: it's about scoring more than you concede. I think that most of the fundamental definitions of the sport that you see usually stop there, but there's a key thing they miss.

You have 90 minutes in which to do this.

Now, I know that that will not seem like Archimedean insight, but bear with me for a couple of paragraphs while I try and displace some intellectual water.

The first draft of this newsletter began life the day before the men's Twenty20 cricket World Cup final, a sporting format that offers one of the clearest effects of 'time' as a game mechanic in sport. For those whose lives aren't blessed by wickets and yorkers, teams (of eleven players, like in football) have to split their talents between batting and bowling. In T20s, no bowler can bowl more than four overs (a set of six balls), meaning that, for most of the twenty overs a team is batting for, the specialist batters have half a mind on not getting out. With a team needing at least five bowlers, if the specialist batters get out then the non-specialists have to see out the remaining time.

But then, usually with about two overs of the twenty to go, the calculus flips. Magic happens. There's so little time left to bat that there's no point being cautious, and the risk of swinging for the fences begins to be outweighed by the risk of not swinging for the fences. There's a bit of a parallel of this in football, where teams will go for 'low percentage' balls into the box late in the game, but this tangent is all to bring us to a feature of football without a parallel in T20 cricket. When teams put their foot on the ball.

You don't see this often and there's a strong chance that if you do it's a Pep Guardiola team doing it. Defenders or goalkeepers will just stand there with the ball at their feet, not through lack of options or indecision, not even necessarily to draw out an opposition press, but just because. Because they've realised something: they simply don't need to do anything.

It's radical. It's like The Matrix, seeing the world for what it truly is. We're only 34 minutes into the game, why rush? Let's sit down, make some tea, watch some T20 highlights.

It's also fundamentally different to a sport that otherwise has a lot of similarities to football, and a lot of analytics overlap: basketball. Two sports each with one ball, a goal at both ends, and fluid in-play action. But basketball (the NBA at least, don't ask me about FIBA rules) has a shot clock and backcourt rules, aimed at getting in-possession teams into their opponent's half quickly, and keeping them there.

(As an aside, I think that the biggest single rule-change you could do to increase the 'entertainment' of football matches isn't cracking down on time-wasting or diving at all, but adapting one of those basketball rules. Maybe backcourt violations but for the defensive third or something)

Watching one of those Guardiola games was my Eureka. The world seems to stop. You wonder why anybody bothers to move the ball up-field quickly at all. There's still ten minutes to half-time, you could watch the last three minutes of the (50-over) men's 2019 Cricket World Cup highlights three times over (if your life isn't already blessed by cricket, this would be a fantastic and if bewildering introduction).

Anyway. The pause of Guardiola's players on the ball in inconsequential areas links to one of my favourite phrases in football analytics. In a 2019 talk at the inaugural StatsBomb analytics conference, Thom Lawrence used the phrase 'trough of meh' to describe how there's a big area in midfield that is far removed from the risk or reward opportunities close to goal that statistical models pick up on.

The slide. Link to the relevant part of Thom Lawrence's talk here.

The fun thing about the trough of meh, beyond pure linguistic frivolity, is that we know the name is only accurate when talking about models. We know that midfield is important in football (fifty million Guardiola brain cells can't be wrong), we just haven't found the numbers for it yet.

It's time to talk about time.

Most possession value models (they do what they sound like they do) base the impact of a particular event on how it changes the chance of scoring or conceding in the next X seconds or phases of play. This makes sense if you're interested in the impact of the action - it'd be silly to credit a pass twenty minutes ago as playing a part in a goal that was just scored. But it does sorta assume that football gameplay incentives are the same as basketball's. And football isn't basketball, it's a completely different sport.

Or, it sometimes is. The point made by Lawrence's 'trough of meh' slide is that your chances of scoring a goal increase very steeply as you approach the posts, with the same thing (as a risk of conceding) happening at the other end. What if this intensity of value in certain parts of the pitch, and the tantalising possibility of getting to it when close by, essentially eliminates the any clock-based incentive when in that part of the field? Although there could be 45 minutes' worth of unbroken play possible, if the ball is near your goal then you want it gone now. Similarly, if you're a step away from generating a, say, 0.5xG chance then it'd be silly to say 'well, we have an hour left'.

Neither of these sentiments is true for midfield though; it doesn't have the implicit 'shot clock'-like qualities that the areas near the goals have. If you want a cricketing metaphor - and why wouldn't you - the parts of the pitch close to goal are a T20; the midfield is a Test match.

In the hope that each repetition of this point will be more finessed than the last, I think this is why possession value models struggle with the midfield. The models' in-built time limitations happen to align with the realistic definitional aim of the game at either extremity of the pitch, but don't align with the the realistic definitional aim of the game in the middle.

Eureka.

Sadly, what is still to be eureked is what is valuable in midfield.

I think it would be fair to say that, as many current coaches appear to believe, control of the ball is a significant aim, and the middle of the pitch is a better place to do that than hear your own goal. The fact that your opponent will usually have the same belief means that maintaining control in itself will be a genuine challenge.

I also suspect that part of the reason why teams don't simply spend the ten minutes before half-time passing the ball near the centre-circle is psychological. You'll quite often see teams get a little bit lax when they're 'just' recycling the ball, and mental errors creeping in can often let the opposition in on goal as well. Perhaps, despite the 90-minute shot clock, there's an 'internal' limit to how long teams can keep the ball in midfield.

If all this is true then part of the value in midfield could be something like 'maintaining the capacity to attack', rather than necessarily 'to attack'. This aligns with a phrase of @TiotalFootball's which has stuck with me since I read it, of 'building capacity'.

it's about moving around as a team, and moving the ball around as a team to build the capacity to successfully move the ball into the penalty area before your opponent does
— Tiotal Football (@TiotalFootball) September 13, 2021

I prefer this as a concept to the common 'building through the thirds' which, to me, feels too sequential, too much like a step of Lego instructions. 'First build the base. Turn the base around; you've reached the middle third. Now build the Millennium Falcon walls; you've now reached the final third. Now build the Falcon's roof; attach it. You've now scored a goal'.

I don't know how you measure this though - if the aim of midfield is to build and maintain a capacity to attack, a potential, then what numbers do you attach to that?

In the hopes of helping, I have what is probably an incorrect but possibly useful oversimplification: is midfield possession simply about wasting the time that your opponent can score in while retaining the potential to score yourself?

Is the point of midfield possession primarily just... timewasting?

If it is, I'm sure Archimedes would have enjoyed the modern, midfield-possession era of football. Wasting time with the slim chance of divine inspiration, it sounds just like his bathtime.

Notes

The main body of this newsletter captures a pretty complete, neatly-packaged set of thoughts, but I wanted to collect some stragglers here which would've disrupted the train of thought if I'd included them above.

The first is that I have a nagging memory of seeing some kind of possession value with a time or score effects feature factored into it before. There are so many different sources touching on this that I didn't find it in some quick checks. It might be a fake memory, but I want to acknowledge that this idea might seem new to me purely through forgetting or being ignorant of other work.

Another important point is that, even if this 'theory of midfield' works for modern-day elite football, I feel like there's a chance that it - and certainly implimentation of it - could be too focused on that level of the game. At other levels, and in other times past and future, the short-term goal value of areas of the field might be different, or the steepness of the value curve might change.

Footballing differences applies not just to chance creation but to maintaining possession as well. That thought I threw out there about a psychological limit to the length of time teams can hold onto the ball but change, making long passages of keep-ball more likely.

I think that this area of thought marries up with the concept of 'defensive possession', maybe even 'rest defence' too, where your structure in-possession has to be considering how you'll be defending when you eventually inevitably lose the ball. They're both tactical theoretical approaches that consider possession as more than merely a means to an end of scoring a goal.

I don't know a lot about player fitness but I also suspect that the slower tempo that midfield possession offers is part of what allows footballers to keep performing for the whole match. That seems like something that would make modelling very difficult: what if a midfielder is effective because they can help control possession for long enough for their attackers to rest, mentally and physically, for a little bit?

Finally, it's difficult to credit the things that have been a background influence to an idea but, as well as the things mentioned in the main body, there are some things that deserve a mention.

One is Van Haaren, Rahimian, Abzhanova, and Toka's paper on a reinforcement learning model for player decision-making, which used different reward functions for different phases of play. Another is Ted Knutson's [once tweeted/podcasted, echoed in the book Net Gains by Euan Dewar] opinion that set-piece efficacy stats were so bad because teams didn't train for them; that the sample was essentially misleading. And another is an article that I think I read a few years ago about the factor of time in T20 innings, possibly this one by Jarrod Kimber, arguing that teams were leaving runs on the table by being too cautious of getting out.

Final final note: I started drafting this the day before the T20 final. Ended it just after the final ended. Congrats England. You justified the amount of cricket written into this football analytics newsletter.

Manchester United says 'hello world' to data science

Sat, 05 Nov 2022 00:00:00 GMT

"Sometimes you have a noisy neighbour. You cannot do anything about that." – Sir Alex Ferguson (2009), speaking about newly Abu Dhabi group-owned Manchester City.

"Sometimes you have a noisy neighbour. You cannot do anything about that." – City Football Group's Data Insights Slack channel (2022), speaking about newly 'hiring for data scientists' Manchester United. Probably. Maybe.

Manchester United are hiring data scientists. Yep, the big story of the day on this corner of the internet is a super-wealthy entity (belatedly) putting in place sensible processes that'll benefit the organisation and help it to thrive amongst rivals. So there's hope for Twitter yet.

But, seriously though, what does this mean for Manchester United? What does it mean for a club with means (and now the inclination) to build out an analytics department in the year 2022?

Many blue moons ago this newsletter wrote about City Football Group's own data department expansion. It was a jaunt through the job adverts they posted, things like "developing our Narrow AI platform" and "advancing our computer vision, simulation & reinforcement learning environments". Sadly, Manchester United's advert, which is here, is not so exciting.

They're hiring for machine learning scientists and decision scientists. To quote the ad:

"ML [machine learning] Scientists will work primarily on human-out-of-the-loop algorithms applying Data Science techniques to augment or summarise data at scale to solve significant, long-term problems. Decision Scientists will work primarily to improve human-in-the-loop operational processes, collaborating closely with our football experts across the club to integrate Data Science into their daily work. We welcome applications from candidates who have ambition to work in either area."

(For posterity and for people starting out in and around the industry, I'll post the person specification at the end of this post)

Beyond that, there aren't a lot of juicy specifics. However, as the Training Ground Guru site quoted, United's director of data science and the person doing this hiring, Dominic Jordan, said earlier this year that "the club is very much looking to be dominant in this space."

So, two years on from one Manchester team rapidly expanding their data operation, with other teams across Europe improving theirs in the meantime, let's consider the question: what would it mean to be dominant in football data science?

Let's assume that 'dominant' doesn't mean United's algorithms putting Liverpool's algorithms in headlocks and giving them a noogie. (That's for after-hours at analytics conferences). There are a number of different avenues that you could go down with a data department at your disposal, let's list as many as possible:

Identifying general undervalued (or overvalued) areas in the transfer market
Looking for players who fit specific role-based requirements for the first team
Researching and monitoring effects of match and training load on player fitness
Patching into Twitter's API to find Pep Guardiola's burner account
Developing 'fundamental' models, like expected possession value, and running experiments to try and work out advantageous tactical approaches
Goalkeepers. Nobody outside of about three people knows anything about goalkeepers.
How to turn as many academy kids into first-team players or valuable sales as possible
Optimal financial engineering (pay the players as little as possible, as much as needed)
Optimal C-suite engineering (retain as few of your bosses as possible, as many as needed)
Run and analyse medical/fitness experiments using the women's team and academy - so little research has been done on elite female athletes, building a large knowledge base could be a big win. 'Just' getting ACL injury probability down to male footballer rates would be neat.
Advanced research on new data sources, like 'skeletal' data
Find the chess.com usernames of all your rivals' data employees and grind their Elo into the dust, torpedo-ing their productivity as they spend their time memorising endgame strategy
Building out scouting KPIs or filters or models based on (video footage-based) tracking data, in addition to the ones that presumably exist based on stats or event data
Work with coaches to create lightning-fast, user-friendly apps to access in-game analytics-based strategy
Subscribe to Get Goalside
Use your chess.com username list to organise a Magic: The Gathering tournament. Invite Pep Guardiola. I'm sure he'd have a blast.

A small-scale interesting thing with Manchester United when contrasted with City Football Group is that United is 'just' one club. If you're a data scientist on the Trafford side of town you won't need to care whether free-flowing attack is a path to success in the Australian A-League; if you work for the light-blue team, who also own Melbourne City, you might do.

That's a different set of incentives, a different way that resources can be distributed. Frankly, with that long a list and a full starting XI's worth of clubs to be involved in, it's a wonder that CFG's department get anything done at all. Although you'd hope that, in the spirit of institutional transparency, they've already crossed Guardiola's burner off the list.

But, when you're building out a department, what do you focus on first? What are the first projects that get done? It's not like United will be starting from nothing: there have been data professionals working on the football side for a while; there's a Data Operations department that's referenced in the job ad, who presumably mean you don't have to do a lot of infrastructure work; the director of data science himself has been in the role for a number of months (albeit a number of months in which time the club changed men's team manager).

You'd probably want to be hiring some people with the capabilities of doing all the kinds of things that the Get Goalside newsletter gets excited by, even if you don't plan on doing it immediately. Alternatively, maybe that's who and what you'd focus on most. Maybe your interpretation of "dominant in this space" is being the first(?) club to squeeze precious lemonade out of skeletal data lemons.

This would be fun for two reasons. One, learning is fun. But even moreso: Two, getting one over on your rivals.

At the 2021 StatsBomb conference, Liverpool FC's director of research said, as cited by the Liverpool Echo, "Recruitment is the most important application of analytics." The fancy skeletal data is unlikely to be useful for that, but - like with pitching, batting, or bowling mechanics in baseball and cricket - it could be a treasure trove for technique improvement. Ajax used something like it for goalkeeper technique analysis. Use the fun, new, shiny data! Prove Liverpool wrong! What was it Sir Alex Ferguson said about that club and knocking off of perches? 👀

But perhaps you wouldn't want to go in that direction. Analytics types tend towards the genteel.

Perhaps, instead, you want to focus more on the here and now. Or, if you really want to make sure the men's team nails the next couple of transfer windows, the here, hopefully-not-here-for-long, and almost-now. A different focus, a different set of priorities.

You could get deep into predictions about how transfers might adjust to new leagues (i.e. the Premier League) and how certain positions (i.e. centre-forwards and central midfielders) might interact with current tactics. You could go deep into the holes in Chelsea, Arsenal, and Man City's women's teams to try and secure United's first Women's Super League title. The job ad does say that decision scientists will be "collaborating closely with our football experts across the club to integrate Data Science into their daily work."

I don't have any real insight into this. It's a chilly, rainy evening and I figured that this was a fun starting point to talk about the variety of things that new departments can do. For a professional's viewpoint, Inter Miami's director of analytics Sam Gregory spoke at this year's StatsBomb conference about building an analytics department. Coincidentally, Inter Miami's coach is ex-United player Phil Neville; Liverpool are used as a hypothetical example club in Gregory's presentation. Perch. Perched on.

We'll see what road United decide to go down. Well, let's be honest, we probably won't see. We'll probably be able to see who gets hired, and guess at what direction that means things are pointing, but the track record for clubs talking about what they do or it being reported on isn't strong.

In fact, this is a final way that you could be dominant in football data science. Publish a load of work. Tell us what you're doing. United's football heritage already involves long ball dossiers in press conferences (strangely, Louis van Gaal's hair-pulling quote has lasted longer in history's memory). Why stop there though? Erik ten Hag and Marc Skinner could deflect difficult questions by pointing to Appendix B of the latest research paper release. That'd show CFG and Liverpool, eh.

As promised...

'The Person' section of Manchester United's data scientist job advert is below (a link to the job ad is here, closing date is 18 November). Football analytics-related job adverts tend to be either pretty fuzzy or pretty specific and demanding; this is probably the most similar I've seen to a 'regular' data scientist job posting. Given that a lot of people early in their careers are interested in this kind of thing, this might be a useful list to take a look at even if you don't plan on applying: working towards it would probably serve you well in a lot of workplaces.

While learning is part of a Data Scientist’s day job at Manchester United, we would love you to be able to bring:

Excellent mathematical and statistical knowledge, gained from a degree in a quantitative discipline or equivalent courses, or demonstrable practical equivalent [Get Goalside's emphasis]

Excellent Python skills

Excellent general Data Science skills covering development of practical analytics applications to enhance established processes, data story-telling and KPI development

Experience in applying data science techniques to answer sports-related questions

Good understanding of software engineering principles, including test-driven development, CI/CD and version control.

In an ideal world you would also have:

Excellent SQL or similar data manipulation skills

Understanding of Bayesian and causal inference

Knowledge of cloud infrastructure and experience of working with data at scale

Experience of conducting code reviews

Experience collaborating with third parties, including academia, to solve problems collaboratively

Thanks for reading.

The analytics revolution is history now

Sat, 29 Oct 2022 00:00:00 GMT

There’s only three ways to tell history.

There’s ‘events, dear boy, events’ history, a focus on ‘important’ figures and/or ‘important’ events. There’s history of ideas, whether that be political or economic or societal. And there’s history of ‘the people’, those who might not have led armies or made medical breakthroughs, but who existed and had the same richness of life in large, large, often anonymous, numbers.

In the span of the past few months, three works which could broadly be described as ‘football analytics history’ have dropped into our laps. And although they don’t fit into those categories perfectly, they’re conveniently close. This post will be part-review, part-Get Goalsideian amble through the countryside of the topic, because I realised while drafting it that I’m simply too close to the subject matter to write a straight ‘review’. (For something more like that, I recommend reading Grace Robertson’s piece on the two books we’ll be discussing).

Expected Goals* by Rory Smith is in the first of those three types of history: a broadly linear telling of ‘modern’ football analytics, starting with English company Prozone in the 1990s. Along the way it hits characters that will mostly, but not necessarily all, be familiar to analytics followers: turn of the century Opta, Sam Allardyce’s Bolton, Decision Technology, StatDNA, the titular metric expected goals, FC Midtjylland/Brentford, Liverpool. Don’t worry, Charles Reep gets a mention too.

*(subtitle: The story of how data conquered football and changed the game forever)

Net Gains* by Ryan O’Hanlon is far less linear, and while it still focuses on individuals it’s far more focused on the now. The now isn’t necessarily the ‘present’, but the minds and the intellectual frisson that is shaping and has shaped recent football analytics. It also seems to be more interested in the subject of its subjects too: trying to understand football as a sport. There’s a passage late in the book where O’Hanlon recaps what the reader has learnt from the book as if it were being applied to an imaginary club, but the introductions of both works clearly shows the different tacks the authors take on this general topic.

*(subtitle: Inside the Beautiful Game’s analytics revolution)

Expected Goals’ prologue opens with a data collector in Manila before talking about the way clubs’ attitudes to data have, often quite quietly, changed. It’s about the companies and the boardrooms. Net Gains’ introduction is more of a personal biography. It’s a contention of mine that Moneyball is not a book about baseball analytics but a book about Billy Beane, and one could similarly say the introduction to Net Gains isn’t an introduction to a football analytics book, but an 11-page (ebook, font-size dependent) dedication to a father and a childhood.

The books aim at different things. Expected Goals is ‘how did club football get to this point of analytics adoption’; Net Gains is ‘how come we don’t understand football, and who are the people trying to work it out’.

John Muller and (the pseudonymous) @TiotalFootball’s Post Script podcast, meanwhile, is intentionally in the ‘the people’ category of history, focusing on a specific subset: bloggers. Its characters of interest overlap with Expected Goals and Net Gains only in as much as they happen to have been people who wrote about football stats online. Sarah Rudd (former StatDNA and Arsenal) and Chris Anderson (a key figure in Expected Goals) formed episode one; Howard Hamilton (whose appearance in the books is only a passing mention in Net Gains), episode two; Ian Graham (Liverpool) the bulk of episode three.

The project spawned from Tiotal’s attempt to draw the history of analytics and tactics blogging together, as the two areas had sometimes been seen separately and in opposition to each other, but grew from there. Of the three works in this newsletter, it’s the closest to a work of historical study: the primary sources are old blog posts, largely uncovered via painstaking use of internet archive services; Muller and Tiotal are the historians, placing them within a wider context, drawing out points of interest, and interpreting thoughts of the source’s authors.

Cards on the table, I really enjoyed Net Gains, which aligns so closely to my particular interests that I’m featured, very briefly, in the chapter about Charles Reep. I enjoyed Expected Goals less, but would probably recommend it to friends or family members to understand what I write about. The Post Script podcast is so far up my alley it’s on my doorstep, but is probably a podcast about analytics bloggers for analytics bloggers.

What I think is so interesting about them, and the fact they’ve come out around the same time, is that they complement each other in very interesting ways.

“And then, you and I know [that] coaches and front-office executives, sporting directors, are some of the most online motherfuckers on Earth. And they read this shit, and they always have.” – Tiotal Football, Post Script podcast (episode 1).

Not only do the three works start off in different ‘types’ of history-telling, they have markedly different energies to them too.

Talking about Expected Goals on a ‘Totally Football Show’ podcast appearance shortly after its publication, Smith gave an interesting sorta-mission statement of the book: “Behind the scenes there have been lots of people who’ve done a lot of stuff to change football, when football didn’t really want to be changed, and I think they have changed it far more than they recognise.”

The reason why I find this interesting enough to quote is because much of the book is focused on people who, if not failed, certainly didn’t achieve what they set out to. The clearest example of this is the running thread following Chris Anderson, which is the focus of alternating chapters.

Anderson starts out as a professor at Cornell University (one of my favourite lines in the book: “Moneyball was the perfect light read for a behavioural economist”) and then starts blogging. He attends analytics conferences, co-writes a book (The Numbers Game, edited by Smith), and tries to implement an analytics revolution at a club, any club, through increasingly (though necessarily) ambitious means. The ‘Chris Anderson’ chapter titles feel telling: Trojan Horse; Proof Point; How (Not) to Buy a Football Club; Alien.

And then there’s the comparison between Expected Goals and Net Gains when covering Anderson’s Post Script episode co-focus Sarah Rudd. Both books feature interviews with her, and though there are varying degrees of positivity in her section of Net Gains, there’s nothing close to these lines from Expected Goals: “Still, though, Rudd is a little rueful at what might have been. ‘We were not as successful as we could have been, given how far ahead we were.”.

This could be read as a pessimistic way to approach football analytics history, but in light of Smith’s words on the ‘Totally Football Show’ I think it reads as quite supportive of the characters that Expected Goals features. They were people who were inspired by Moneyball but couldn’t get Moneyball to work (“He [Anderson] could not become Billy Beane in the West Midlands” is a line in a later chapter). I’m not sure that the book really shows how they changed football, or even particularly proves that they did, but it seems to want to show that their work, Anderson’s in particular, wasn’t in vain.

“Marc Andreessen[…]has a dictum that, in his business, ‘being early is the same as being wrong’. The timing of technology is as vital to its success as the substance of it: the world has to be ready to accept and embrace your idea.” – Expected Goals (chapter 3)

What this perhaps quite British sensibility of Expected Goals also does is something that I think any ‘history’ of analytics leading up to the present moment needs to: engage with the question of ‘why now?’ and, by implication, ‘why not then?’. Were the ideas not right, the implementation not right, the conditions not right? Could StatDNA’s work under Wenger, for example, had the same impact as Liverpool’s current research department appears to have done if circumstances in the late 00s have been different?

I don’t think any of the works set out to answer this, but it permeates them. The early episodes of Post Script talk about struggles of the bloggers to find data; Expected Goals details the evolution of data collection, eventually moving past the need to physically drive discs of footage around England. It opens, as previously mentioned, with Ashley Flores, part-time pro-footballer and part-time data collector for Impect, the German company that popularised a player-bypassing metric ‘packing’.

Although the pay is, Smith writes, “good, way above Filipino minimum wage”, you assume that it’s far lower than would be required in Germany, which’ll be why the collection takes place in the Philippines (or, for companies like StatDNA, Laos or Cambodia; for StatsBomb, Egypt). Imagine what that would have required in Charles Reep’s day of the 1950s. Meanwhile, key characters of Net Gains like Luke Bornn and Paul Power have done fantastic work with tracking data, something which has only been made possible, on a mass scale certainly, with technological advances this century.

As someone whose interest in analytics tends to be towards ‘what’s next’, this ‘why now’ questions sticks out to me particularly. The characters in these works are clearly very smart, and building on the work of others, but, for example, if Prozone have had tracking data since the 1990s why did it take until the mid-to-late 2010s for pitch control models to emerge as (apparently) new ideas, and can we learn anything from that which might improve further development and implementation?

What can we learn from these histories that helps to shape a better future?

“All interpretations made by a scientist are hypotheses, and all hypotheses are tentative” – evolutionary biologist Ernst Mayr, quoted in Net Gains (chapter 6)

It might be because the book isn’t aimed at me, but I was mildly disappointed that the main beats of Expected Goals, outside a couple of the Anderson chapters, felt so familiar. Net Gains speaks to and about a slightly less well-known, but more Online, set of people. However, it feels like a smart and non-obvious choice to feature Paul Power in a fairly prominent role, someone who was doing advanced work with tracking data early on and who, to my knowledge, hasn’t been profiled or done a bunch of interviews before.

However, all three of these works have a very Anglosphere-centric focus. Maybe it’s a shared language thing, maybe it’s who is willing to talk, maybe it’s that these are the only people whose work has mattered. Some of these works elicited the following more than others, but it’s unnerving to me to suspect that there are notable gaps in my analytics history knowledge and to consume works which looks at that suspicion and goes ‘nah, I think you’re good’.

The only non-British, Irish, or North American contributions featured, that I can recall, are Impect and a mention of the Israeli scientists who created SportVU, the tracking data system which helped Luke Bornn’s entry into sports analytics (through the NBA, rather than through football). Maybe these are the only notable stories to tell, maybe Moneyball’s Anglosphere legacy is larger than we thought, but, for example, what if any legacy does Nemzeti Sport’s early twentieth century data visualisations have in Hungary? Skillcorner, one of the companies making waves in broadcast footage tracking data (where Paul Power now works) are French – is there an analytics history to explore there?

It seems worth mentioning at this point that Expected Goals and, more thematically similarly, Net Gains join Christoph Biermann’s book Football Hackers, originally published in Germany in 2018, on the ‘good analytics movement history books’ bookshelf. The Post Script podcast ‘limiting’ its remit specifically to ‘analytics blogging’ is a smart one in this context, particularly as a side-project; it sets out an explicitly smaller area to cover.

“The sheer quantity of brain power that hurled itself voluntarily and quixotically into the search for new baseball knowledge was either exhilarating or depressing, depending on how you felt about baseball.” – Moneyball (chapter 4)

Let me bring in Grace Robertson’s review of Expected Goals and Net Gains again here, because there’s something she writes at the end of it that crystalises something that was half-formed in the back of my mind.

“Let’s say that in ten years’ time, every club incorporates analytics into all their decisions, while using proprietary models far more advanced than anything we can see publicly. At that point, none of us will be able to know what good decision making and good strategy looks like. We’ll have come full circle [from the pre-blogging era] and we will understand football even less than we ever did before, because the astrophysicists have figured out all the things we will never know.”

I’m a little less worried, for a reason that O’Hanlon, and Luke Bornn, nod to at one point: “’People who are analysts have very clear incentives to say everyone should be using data because they want to grow their space’, he [Bornn] said. ‘They want to sell more product, they want to make themselves more hireable’.” Few things make you look more valuable than making advanced analytics sound simple in a popular book, or explaining a smart thing you’ve been doing (albeit that you were doing a couple of years ago, perhaps).

On this theme of secrecy and what is made public, I find it admirable that, in this passage on the incentive structures of analysts and companies to big up their achievements, O’Hanlon acknowledges these apply to the writers too. This newsletter has incentives too, although uncertain ones.

I sometimes worry, though, that we (by 'we' I probably just mean 'I') rely too much on the blogosphere in covering and identifying analytics. It doesn’t feel feasible to assume that, even in its heyday, it represented the full nuance of analytical knowledge and application. (That said, blogs should make a return. Blogs are good).

While editing this post I leafed through my copy of The Numbers Game. Although Anderson, who co-wrote it, had been a blogger, he was kind of in the 'early' section that feels slightly separate from what came later; before the famous Opta expected goals blog by Sam Green, before the StatsBomb blog took off in a big way. The closing chapter of the book features forecasts, one of which is that ‘Geometry – space, vectors, triangles and dynamic lattices – will be the focus of many analytical advances’. How smart would I have looked in the 'early analytics Twitter' era of blogging if I’d just repeated that over and over again?

Robertson’s right though; both the development of analytics knowledge and the telling of analytics history is impacted by the incentives that practitioners have in keeping their insights and edges (and screw-ups) secret. This isn’t like ‘regular’ science and technology either, where there are specialist reporters covering the field. There are no analytics beat reporters. (Regular journalists will find analytics staff as useful sources, but it seems doubtful that they’ll be as interested in the 1s and 0s as someone who’s analytics-specific).

“Still less well-known, at 65, than many far less influential managers, Bielsa is something like Velvet Underground of soccer coaches: Not many people buy his records, but everyone who does subs in an attacking midfielder.” – Brian Philipps, ‘Marcelo Bielsa and Leeds United Form a Perfect Union, The Ringer (2020)

If you’ve stuck with this post this far, well done. We’re nearly home.

There’s a chapter in Net Gains that I found unexpectedly touching. Granted, it’s the one that O’Hanlon contacted me about to contribute to, the one on Charles Reep, me having written a Get Goalside newsletter about him. If I’d had to bet on it though, I wouldn’t have guessed that book chapter about a mid-century accountant would be emotive.

In it, O’Hanlon speaks to Richard Pollard, who was a friend and collaborator of Reep’s in the later years of his life. “’He [Reep] was made fun of more than anything else,’ Pollard said. ‘And a lot of coaches used to downgrade him all the time. Weird, weird old retired wing commander with a hat and a pencil and paper.’” [original italics]

Let’s be honest, if you were stood on an English terrace watching an evening game in the 1950s or 60s and a middle-aged man a couple of rows in front of you put on a miner’s helmet, torch on, and got a notebook out… you’d have thought he was a little weird; you may well have made fun of him too, loudly or quietly. Not a thing to be proud of, but often happens.

In the hours after Robertson’s piece came out, ‘analytics twitter’, or fringes of it, came alight. People who’d been around a decade or more shared names they thought had been overlooked, people whose contribution they thought deserved recognition. Many had, at the time, been shunned or made fun of – if not by coaches this time then certainly by others online, even, in some cases, members of the media.

It turns out that the 1997 paper that my old post on Reep opened with, which featured a diagram that looks a lot like modern expected goals probability charts, wasn’t co-written by him at all. I’d sort of suspected as much, given that he was in his 90s by that time, but I still hadn’t known for sure.

“Reep,” O’Hanlon writes, of the man who’d spent matches scribbling notes and producing data on the game he loved and was fascinated by, “had nothing to do with the production of the paper; Pollard just included his name as a co-author as a tribute to his mentor.”

If you want a ‘history’ of analytics then we’re probably still too close to the turning point to properly recognise its shape. That doesn’t mean we shouldn’t try, but probably makes it more likely that things will end up feeling unsatisfactory.

Given this, the stories we choose to tell tend to be personal in some way, and I think all three works show this in their own ways. What interests us? What inspires us? Whose tale do we want to commemorate? Not so different to Pollard: whose name would we quietly add, in a kind of tributory fraudulence, as a co-author to our own research paper?

If you owned a big club, would you buy Ajax?

Tue, 11 Oct 2022 00:00:00 GMT

Some Get Goalside newsletters are very informative and verge on instructional; some are just weird blogs. This is one of the latter.

With the recent death of Hilary Mantel I’m reminded of one of my favourite things about reading her Wolf Hall trilogy: characters can be two months removed from momentous events, they might even anticipate those events happening, but the way Mantel writes captures how it can seem just as likely to them that Henry VIII will force through a divorce as he will get bored with Anne Boleyn and distract himself with a war against France instead.

We can all see that the trend of multi-club ownership structures is only going in one direction. A recent article in The Athletic says that nearly 33 per cent of teams in Europe’s ‘Big Five’ men’s leagues are part of such set-ups. As the number grows, so do the types of approaches. The ‘Red Bull model’ is different to the City Football Group (CFG) model, which are different to the Pacific Media Group model (investment in Barnsley, Oostende, Nancy, Den Bosch and others), which are different to the ‘rich American happens to own 5-20% stakes in various clubs’ model.

We don’t know how this is all going to end up. We can see the ways that the pieces could fall, but not how they will. “But [Cardinal] Wolsey’s becoming uneasy,” Mantel writes from the perspective of a Thomas Cromwell sat half a decade before Henry VIII’s eventual marriage annulment, “waiting for Boleyn’s daughter[…]to please the king. If she would do this, the king would take an easier view of life.” She didn’t. And he didn’t.

There was a time when Anne Boleyn was just a mistress. A time when the Premier League was just an idea discussed in expensive restaurants. A time before free transfers, a time before uncapped wages, a time before baseball farm systems.

If, in the 1520s, you were trying to map out how the next decade of England’s political and religious landscape would shape out, you’d have simply been wrong. So let us avoid being wrong in the same way; instead of trying to map the future out, let’s focus in on one very specific possibility to see where it might lead and what, if anything, we can learn from it.

Let’s buy Ajax.

Ajax, of Amsterdam, are one of the few teams to have won the European Cup and one of the few to have a club icon who is also a flat-out legend of the entire sport. The club has a definable culture and successful academy, and in the past four seasons, they’ve received, according to Transfermarkt, close to €600mn in transfer fees. I, like many people, don’t know what the Netherlands’ natural resources are, but Amsterdam’s is footballers.

And Ajax are good exporters. “It’s a steppingstone team,” Arco Gnocchi, host of the “Pak Schaal” Ajax podcast, told the New York Times recently. “That can be difficult to accept, but if we’re a steppingstone team, at least we’re the best steppingstone team.”

This may be because, despite their performances and heritage, Ajax are not one of the financial big dogs. Deloitte’s Money League report lists the top 30 teams by revenue; Ajax weren’t one of them on the latest edition. As coincidence would have it, 30 is just about the number of teams that each of the big three US major leagues have, leagues which Florentino Pérez and friends so clearly look to as inspiration.

So, money no object, if you were on Pérez’s little rolodex of financial ambition within that top 30, you’d do it, right? Buy the best team on the outside of the money list? (If, a large and inconvenient if, you had the sway with the various people who you’d have to persuade to sell).

Now, if you did buy them, or at least a majority stake, the benefits wouldn’t be as simple as a direct Ajax-to-Super Club pipeline. For that, we have Red Bull as precedent.

They made for an eye-catching test case when both their Leipzig and Salzburg wings got into UEFA competitions at the same time in 2017. That summer though, the governing body ruled that the two clubs were sufficiently separate: "no individual or legal entity had a decisive influence over more than one club". Sure enough, the Salzburg-to-Leipzig tunnel slowed to one transfer per season* rather than the three or four per year that it’d been from 2014 up to that point.

*It’s not quite clockwork but it’s not far off. From Transfermarkt again: Konrad Laimer in 2017/18, Amadou Haidara in 2018/19, Hannes Wolf in 2019/20, Dominik Szoboszlai and Hee-Chan Hwang in 2020/21, offset by nobody in 2021/22 (although Caden Clark was purchased from the New York branch). Nobody’s arrived in 2022/23 – yet – but Benjamin Sesko is lined up already for 2023/24. Including next year that’s seven seasons and six transfers, and we still have this coming January window to go.

So, if you’re a landed gentry club, you’d have to limit your transfer activity. A summer like Manchester United’s – bringing over Lisandro Martínez and Antony along with manager Erik ten Hag – would be out of bounds. You’d have to make do with Matthijs de Ligt one year, persuade Frenkie de Jong to hang around until the next, that kind of thing.

However, even if you can’t bring in multiple players at a time, that doesn’t mean that your extra hand in the transfer market is wasted. You have a possible landing spot for visa stuff, you have potential to nudge players who don’t fancy sticking around away from your worse rivals, you have a club to use as makeweights in other deals. On that third point, here’s an extract from the previously linked Athletic piece:

“In the summer transfer window just ended, CFG waited until Southampton had sanctioned Oriol Romeu’s departure to Girona, a club which it has a 47 per cent stake, before Manchester City sent Samuel Edozie and Juan Larios to St Mary’s. Southampton also loaned goalkeeper Mateusz Lis to CFG club Troyes.”

Neat. Mind, you probably wouldn’t be able to be as obvious as that if your junior partner was Ajax instead of Girona or Troyes, but you get the idea.

While we’re talking technicalities, Red Bull’s UEFA decision isn’t the only prominent example of separate-but-related entities. Towards the end of a much wider ranging recent interview on the Training Ground Guru podcast, chairman of Toulouse FC, Damien Comolli, was asked whether he had any interaction with the data people at Liverpool and Milan, who both now have ownership links to the French side. Understandably, Comolli sought to clarify the situation:

“Toulouse is owned by a different part of Red Bird which is Red Bird FC. So Red Bird FC owns Toulouse[…]and Zelus Analytics [a data consultancy who’ve been mentioned in a previous Get Goalside newsletter]. AC Milan is new[…]and the involvement in Liverpool – actually not in Liverpool, with Fenway Sports Group – is with Red Bird Capital Partners, not with Red Bird FC.”

To repeat that, because I needed to go over it a few times: Toulouse FC are owned by Red Bird FC, a company of Red Bird Capital Partners; Liverpool FC are owned by Fenway Sports Group, invested in by Red Bird Capital Partners (reportedly a ten per cent stake, although at one point they were linked to an attempt to get a larger stake through a Special Purpose Acquisition Company, unhelpfully for keeping track of names called RedBall); AC Milan are directly owned by Red Bird Capital Partners. Red Bird Capital Partners also bought a 30-40 per cent stake in Wasserman Media Group in 2021, a big ol’ company that has a sports agency wing with a ‘global football (soccer)’ sub-wing.

To be clear, this is neither a set of ownership stakes in Liverpool and Milan like the Red Bull clubs have, nor an agency-club relationship as appears to be the case at Wolverhampton Wanderers. Just an impressive business portfolio with lots of mentions of and links to the colour red. (Very glad I never had to care about the RedBird-backed RedBall in talks to buy a stake in the Red Sox’s parent company).

Apologies, we’re getting away from Ajax.

As we said, if you bought them – and, we should stress, big if – you could probably get away with a transfer per season from them, maintaining the club as the successful development ground that it currently is. You don’t even need to call yourself part of the same group (having everybody dress up the same and re-branding all the club badges is a bit passé).

Like all businesses, being larger can also bring with it some economies of scale that you could make use of. City Football Group – as the parent group – has its own data insights team, for example. Other models may not have a centralised department but workers at the individual clubs could have a ready-made network of quasi-colleagues to chat things over with. This newsletter has an eye firmly on analytics but it could be useful for coaches to discuss tactics and methodology in a safe-ish environment too.

As luck would have it, bringing Ajax’s analytics staff into your gold-SQLed fold would be a pretty neat thing to be able to do. They’re the type of club that do things like eye-tracking experiments and goalkeeper stance studies. Imagine the impact that smart, advanced studies could have not just on a first-team squad but on player development, and not just player development at one club, but at multiple clubs. (This is probably why CFG have their own centralised insights team).

At this point things are starting to sound like an investment pitch, which is fitting because the type of people who listen to investment pitches make up a large section of the people buying stakes in football clubs.

There is a form of advertising in some of the club ownership groups, most obviously Red Bull and, of a different sort, City Football Group and Qatar Sports Investments. But for many it seems to be just a fun piece of investment portfolio management, with a dazzling array of price-entry points, and almost every club with a ready-made customer base to build on. It’s like if you could invest in Coronation Street. Or dearly departed Neighbours.

Football clubs might not be money-spinners on a year-to-year basis, but between American business on the one hand taking to soccer like Ted Lasso takes to nurturing fellow men with deep-seated emotional issues, and on the other hand certain oil states recognising both the business side and the geopolitical avenues* that a globally-popular sport offers, the feeling seems to be that club ownership stakes are assets whose appreciation can only be rivalled by male-attracted fans’ appreciation for Jack Grealish’s shorts size choices.

*In fairness to oil states, it’s not like ‘leisure activities as soft power’ is anything new, nor even getting a political-avenue boost from football; the magnitude of it happening on a nation-state scale, and the signal that sends to the rest of the market, is new though.

Bringing this back to Amsterdam, buying a large enough stake in Ajax wouldn’t be cheap, particularly considering that a lot of it is owned by a member’s organisation who’d probably need a lot of convincing. Even if you were very wealthy, you’d need to share the cost. Maybe you’d get a big investment group with star-power names, like LeBron James (Liverpool again, in a sense, with a bizarrely timely recent news article on his and FSG’s links) or like, well, anyone who has an ownership stake in Angel City FC.

And this is where, even within the structure of a self-awaredly out-there thought experiment, problems arise.

If you own one of the mega clubs, you may already be sharing the cost. And you’re going to try and buy another one, which is a rung below the top step of the financial ladder? You might get the money together but would that be an investment coalition that’d be able to stick together?

Not only that, but buying Ajax might just be flying too close to the sun of regulation. Laws can change, and often do when presented with new circumstances. The two Red Bull clubs are one thing – they were ranked 32nd and 47th best teams in Europe by clubelo in summer 2017 (even now they’re ‘only’ 25th and 35th). Owning Ajax would likely mean owning two in the top twenty.

And finally, to dwell a little on that big ‘if’ we’ve noted throughout this, there doesn’t seem much on the table for Ajax (besides, if you have it, boatloads of money). They’ve won three league titles in a row (and were joint-top when the 2019/20 season was abandoned because of covid). While two of these were only by a handful of points, strengthening their domestic position would come at the cost of an implicit ceiling on how far they’d be expected to go in the Champions League, which they reached the semi-finals of in 2018/19. ‘This might be selling your soul but we’ll turn you into Eredivisie Bayern’ isn’t a hugely attractive pitch.

However.

If you, the club-owning wheel-reinventor, are still interested in an institution with history, a strong academy and player development pedigree, a decent analytics set-up, and, crucially, a lower clubelo ranking than Ajax’s current 13th... Would Benfica work?

We’ll hand-wave away the whole issue of winning over shareholders and fans again and I’ll point you to the lower chance of being hit by new regulation, the fact the club hasn’t won a league title in three years while they haven’t ever reached the Champions League semi-finals (they’ve got their own in-build ceiling already!), they still could have provided you with a signing-per-season list of João Félix, Rúben Dias, and Darwin Núñez.

Who would do it? Perhaps Liverpool/FSG, with their Darwin connection. It’d be a bit tough for Chelsea given that their own Portuguese link, José Mourinho, has history with Porto but maybe they could style it out. Real Madrid might make for some interesting Iberian politics hot takes, while Barcelona would probably be waiting til they have enough saved up for Ajax. Hmm…

I wrote a draft of this over the weekend and, having got to this point, decided it was all pretty terrible. For a start, even as a thought experiment, it goes into some fairly disrespectful territory towards the clubs involved. And for what? The multi-club model stress test of ‘what if a big club buys Ajax’ just ended up seeming that little bit too far-fetched, even if it did let me touch on a lot of other interesting multi-club stuff.

But then Monday came around.

QSI, the Qatari acronym that own Paris Saint-Germain, announced a 21 per cent stake in SC Braga (QSI’s portfolio also includes KAS Eupen, because if you don’t have at least a sizeable stake in a Belgian club are you even in the club ownership game). Braga, the Portuguese team who’ve finished in the top four of the Primeira Liga every year since 2017/18. Who may not have produced a Frenkie de Jong or a Rúben Dias recently, but did sell Francisco Trincão to Barcelona.

It may not be outright ownership or even a majority stake, but… The “let’s play a stupid game” draft had pivoted off Ajax to Benfica and ended up about 360km and 0.66 points per game away from where the PSG owners actually plumped for.

“But some people suggest that she is bargaining with the king; some say that she wants to be the new wife; which is laughable, Wolsey says, but then the king is infatuated, so perhaps he doesn’t demur, not to her face.

He has drawn the cardinal’s attention to the emerald ring Lady Anne now wears, and has told him the provenance and the price. The cardinal looked shocked.”

Postscript

Ajax being the subject of this newsletter has an added significance, in that they tried to be the head of their own multi-club-type model before it was cool. The following is cribbed from Wikipedia, although I was already aware of some of it.

In 1999 they acquired 51% of one club in South Africa and one in Ghana; in 2000 owned a majority stake in a Belgian club (always Belgium). In 2003 they sold the Ghanaian and Belgian stakes and pivoted to America before that folded in 2007. They opened academies in Greece starting in 2011, a scheme that dissolved in 2016; and in 2020 they finally sold the stake in 'Ajax Cape Town'.

Those types of markets are the ones that a lot of ‘rung below CFG/FSG/QSI’ level players are operating on. There’s a lot of Belgian ownership purchasing of course, a fair amount of lower-than-elite French, German, and Italian purchasing, smatterings in Central Europe. The ‘African academy’ idea has been taken up successfully by FC Nordsjælland, owned by the Right to Dream Group since 2016, and Aston Villa’s owners recently announced investment in an Egyptian and a Senegalese academy.

I wish I had a good-enough sense of where this ‘lower’ level of things is floating towards to do a version of this newsletter about them, but I don’t. I’d probably end up selling Belgium’s public utilities to Athletic Club.

That level of things is probably the more important to have a handle on, though less eye-catching, simply because it captures a lot more clubs. I’ve mentioned Belgium a lot, but, like, what if two clubs decided to use their Belgian subsidiaries as a proxy war between them? What happens if the Belgian government or populace turn away from football, the talent pool dries up, and suddenly everyone decides they’ve got to drop their investment in the clubs there? More sensitively, what happens if this pattern plays out in an area with a larger wealth gap to western Europe?

I dunno. In some ways, it probably depends what happens at the top end, but maybe it wouldn't.

I was trying to find another Wolf Hall quote to end this on that fit in with this postscript. Maybe something about laws, given that FIFA may have to write new ones. But Hilary Mantel just writes good lines and I liked this unrelated one.

“No ruler in the history of the world has ever been able to afford a war. They're not affordable things. No prince ever says, 'This is my budget, so this is the kind of war I can have.”

Ball control, space control, and why good teams play Pep-ball

Sat, 01 Oct 2022 00:00:00 GMT

Manchester City in the late 2010s and early 2020s. Barcelona and Spain straddling 2010. Sacchi's Milan. The Dutch of the 1970s. Hungary in the '50s. Scotland, in the late 19th century*. Every generation has its iconic teams characterised by their possession of the ball.

*(From 'Inverting the Pyramid': "That success [holding a heavily-favoured England to a draw in the first international football match in 1872] may have confirmed the notion of passing as superior to dribbling - north of the border at least - but it could never have worked had passing not been part of the game in Scotland almost from the start.")

As much as trends come in and out of fashion, I'd wager that the teams that win throughout football's history have, almost without fail, had more possession of the ball than their opponents. I think there's a solid footballing reason for this, and I think it has profound implications for statistical analysis.

I will try, where possible, to throw in a zinger to lighten the tone.

Football is an invasion sport - categorisationally, rather than method of global adoption - meaning that teams try to score points via entering the opposition's 'territory'. But it's also one with a ball.

We have a lot of these. The same is true of basketball, netball, all of the other offshoots of football (Australian, American, rugbies league and union), but the balance between 'space control' and 'ball control' is different in each one. Take basketball. Both conceptually and literally it's a much more 'ball-handly' sport than football, but space control is still important on offence and defence. Maybe the split is 70/30 between the importance of ball control vs space control, whereas football is closer to a 50/50 split.

If you break the sport down like this, then 'what makes a football team 'good'?' is easily answered (though much harder implemented): being better at controlling the ball and better at controlling the space.

Hey, is there a well-known approach to playing football that focuses on these two things?

Like any good little analytics newsletter, we now cross over to Moneyball to quote (Brad Pitt's) Billy Beane: "If we try to play like the Yankees in here [the pre-draft meetings], we will lose to the Yankees out there [the field]." Instead of back office approaches though, think tactics. If the other team is better at controlling the ball and space than you, either in or out of possession, then why try and control them in the same way? You won't! You will lose to the Yankees out there.

So what do you do? You make the game less about controlling the ball, less about controlling the space. Shake things up. This isn't ingrained Englishness, this is elemental philosophy, this is practically Aristotelian: get bloody stuck into them.

From here, from this basic framework, you can break down football archetypes into the 2x2 quadrants that high-or-low space and ball control form. The quadrants are less about absolute quality, and less about how much a coach values each thing, more about how the talent of their squad matches up against the opposition.

What we now call 'positional play' is a high-space control, high-ball control approach. Quick ball-carrying is high-ball control, but only has a low dependency on space control. Long-ball football has a low reliance on both ball control and space control. I'm not really sure if there's a style of play that's low-ball control but high-space control, but it's probably what clumsy La Masia graduates end up playing.

This can work on the defensive side too, although we might need to clarify our meanings a bit. By 'ball control', that could be direct 'ball actions' like tackling (or, in attack, dribbling and passing) but also close-quarters 1-v-1 stuff. Meanwhile 'space control' can be affected by fitness as well as tactical understanding.

So, a hunkering deep block is a low-ball control, low-space control approach (reducing both the frequency of duel situations and the amount of space aimed to be controlled). Bielsa-ball player-marking would be high-ball control (individual duels), low(er)-space control. High pressing is in the high-space control, high-ball control quadrant, while a high block that is quick to drop when passes are made through or over it could be an approach taken if you have a squad high on space control but low on defensive ball control.

Clearly you would prefer to control both (unless for aesthetic reasons), and therefore always be in the high-space control, high-ball control quadrant, but that's not always the hand you're dealt.

So, now for the statistics.

If you're lucky enough to remember the 'stats in established media' conversations of, I dunno, 2010-2014 you'll recall the tiresome lines about 'winning the possession', the idea (or, sadly, reality) that some people thought that more passes meant a team was playing better. It's a classic 'correlation doesn't necessarily equal causation' issue. Better players make better teams who have more possession, but maximising the possession figures doesn't mean that the team is good.

In fact, if a team is only moderately good at controlling the ball and controlling space in possession, but they are intent on passing the ball around, then the defending team should often just let them. Their high possession figures aren't coming because they have high ball and space control; they're coming about because their ball and space control is bad enough to render their possession toothless.

'Player talent' isn't unchanging though. You can coach improvement into players in both space and ball control, and perhaps managers who go for 'possession ball' with squads of average-or-worse talent, and who don't fail badly, should be given even more credit than they currently get.

Assessing managers is a slightly higher step up the football food chain than I'm interested in for this newsletter though. I'm more interested in the humble plankton of the industry, hoovered up by the giant bloated whales, the everyday club analyst.

If you, or your boss, has an idea of how you want the team to play, how do you design key performance indicators when you know that they may simply be influenced by the disparity in team strength? Are you measuring execution of a game plan or the inevitabilities of the sport? Do you adjust the metrics if you're playing a team much better or worse than you, or simply adjust expectations for what the post-game numbers will look like?

Reading that back, of course you should change your KPIs, right? A manager isn't going to go out and tell the team to play in the same way against Bayern Munich as they would against Burnley, so, naturally - beyond the essentials like xG and such - you'd change what is most important to measure?

I am eager to be corrected on this, but I think the smart approach would be to develop different stats to look at for different circumstances. Playing the league leaders? Load up spreadsheet template A. Conceded early against a team of a similar strength to you? Spreadsheet template G.

Maybe this whole space control and ball control framework would help elsewhere too, like in finding appropriate metrics to use when scouting players. Maybe it could help people learn to analyse the game better, quicker. (If it doesn't, it's probably not a good framework).

The thing that makes this idea worth approximately 1300 words of your time is not that it's new or industry best-practice. Football is like one big jigsaw puzzle that you haven't been given the picture for. You don't even know what shape it takes, and if you don't know what shape it takes then how can you start piecing things together from the edges?

I think a good framework for understanding football does that: it gives you an idea of the shape of the sport, of its boundaries and dimensions, giving you a starting point to fill in the rest from there.

Where is analytics? What is analytics? An autumn 2022 update.

Thu, 22 Sep 2022 00:00:00 GMT

I hadn't been planning on writing a post-StatsBomb conference newsletter. Blame the train ride home.

Part of the reason for not planning to was not knowing how to write about a conference where you can see most, but not all, of the content. So this isn't going to be a blow-by-blow review, more a combination of thoughts, many of which were percolating anyway, that the day helps to tie together.

The other reason for not planning a post-conference newsletter was a desire not to be seen as being swayed into positive coverage by a free pen and free StatsBomb-branded socks. (A high-quality pen as freebies go; cannot yet report on the quality of the socks).

Despite this second reason, allow me to suggest some taglines for the company's marketing team to use in their post-conference content:

'StatsBomb Conference 2022: A wide array of expertise!'
'StatsBomb Conference 2022: "We don't have to validate our existence anymore"'
'StatsBomb Conference 2022: All analytics is web apps!'
'StatsBomb Conference 2022: All grown up'

These may need some refining.

It shouldn't escape attention that a number of the speakers on the main stage during the day have worked in football virtually, if not genuinely actually, since leaving university. Within the last decade. And these weren't 'how to get your foot on the ladder' type talks, they were 'how to take charge of a department' type talks. StatsBomb itself has 'grown up' from a somewhat sparsely-attended launch as a data company in 2018 to hosting a really-very-large stage within the more attractive bowels of Wembley Stadium. No wonder that the day featured a panel titled 'The End of the Beginning? Where Does Analytics Go From Here?'.

When, during it, Javier Fernández - formerly of Barcelona, currently of Zelus Analytics - said that analytics is in a bit of a teenage moment (sometimes looking very grown-up, other times still a little lacking in maturity), it worked as a cusp-of-adulthood metaphor for much of the room. So you set up a data company/earned a very respectable job/got your hands on tracking data: what next?

Mostly software development, apparently.

The progression of football analytics goes a little like this:

Step 1: Find something useful to say
Step 2: Make sure important people are willing to hear it
Step 3: Enable the important people to hear the useful things without you having to do the same amount of work every time they ask

An increasing number of people in the industry have now arrived at step 3 and require the kind of infrastructure that essentially turns them into mini-tech companies (A new Warholism for you: "In the future, everyone will be a tech company for 15 minutes"). They need database engineering and maintenance. Automated tasks. User-friendly applications.

It's not like this is unique in the sporting world to football. In The MVP Machine - Ben Lindburgh and Travis Sawchik's 2019 book on baseball's drive towards data-driven player development, currently in Get Goalside's 'In Progress' pile - Hall-of-Famer Pedro Martínez gets a mention not for his on-field exploits, but in the form of PEDRO, an in-house analytics application at the Boston Red Sox.

Heck, it's not even hugely new to football. People like Joe Mulberry and Karun Singh were cooking up football analytics apps for the Stats Perform Pro Forum in 2019 and 2020 respectively. But it being the mood music is new. (Also, always keep an eye on what the people who were ahead of the curve several years ago are up to in the present).

If you don't want to call this a turn towards software (as attractive as that may be as a VC pitch), you could sum it up differently: as a drive to consolidate knowledge within football organisations.

In some cases that's some kind of internal app (as mentioned by a number of main stage speakers), with which people could look up stats, visualisations, reports all by themselves. In some places it's developing proper analytics teams, whether by expanding (or creating) through direct hiring or by bringing people already on the payroll together in a more aligned fashion. And in some places it's by going full-centralisation, through developing multi-club models or analytics consultancies where you can get some 'economies of scale' by being able to share knowledge/development-time across numerous clubs.

If you want an attempt at a pithy one-liner, where once analytics was a space of adventurers, now it's a space of entrepreneurs. Insight is one thing, productising insight is another.

But all is not lost for those souls who merely, like Star Trek, want to venture out where no-one has gone before. There will always be more new knowledge to find. There's always an elusive golden apple just out of reach, spaces on the data periodic table still to fill.

Some of this will be filled by the existing data, but the other inevitability is that there's always some new kind of dataset to yearn for. In the olden days - as recently charted by the Post Script podcast* - analytics bloggers considered a proper trove of event data as the thing to covet, the place where new answers were to be found. Then there was tracking data, the gift that would lead us to finally investigate and understand space; the final frontier.

Now there is some movement again, though more subtly.

The new element now is the broad field of computer vision (or as it's probably easiest explained, 'detecting shapes from images'). This is partly because advances in the field mean you don't necessarily need in-stadium multi-camera set-ups to produce good-quality tracking data. There are companies whose whole shebang is making it out of the kind of TV broadcast footage that anyone, in one way or another, can get their hands on.

But computer vision is partly interesting in the same old way that any wave of technology is interesting: if machines and bots can do what humans can do, it'll usually (eventually) be quicker and cheaper to get the machines and bots to do it.

Open-source computer vision packages are getting easier to use and better at what they do, to a level where even a fairly average joe could take a crack at creating their own computer vision-derived football data nowadays. And, y'know, why buy tracking data if you could create your own?

For that matter, why buy event data when you could create your own. We're not there yet (way off) but could a combination of tracking data and body posture detection one day do away with human-collected football event data for good? What's a 'tackle' if not two players in the vicinity of the ball, one of whom is making certain movements with their legs?

For the record, Sarah Rudd - speaking on the previously mentioned 'Where Does Analytics Go From Here' panel - doesn't think you, working inside clubs, should pursue this. And, sure, I take the point of the former StatDNA/Arsenal employee and general analytics pioneer, that trying to replicate the fine-tuning and reliability of entire companies whose sole purpose is this one thing might not be the smartest idea. It's probably a little silly. But then, this newsletter is a little silly.

The general point is that it would seem weird if the increasing accessibility of computer vision programmes didn't end up affecting relationships between data providers and clubs (or other data purchasers) in some way. Even if that way ends up being, as kind of happened with event data, forcing existing companies to adapt and evolve their offering.

That said, that comparison isn't perfect: as Rudd said in her solo talk at the conference, event data is more 'generating' than 'collecting' - "you don't pick it up off the ground" - whereas pure tracking data is much more like collecting, in that players' movements in space are as close to an objective fact as you can get.

The comparison isn't 1:1 in another way too. If you were dissatisfied with event data, as an event data customer, it would be fairly easy to collect new and different things to a good degree of reliability (as the success of Sportscode and similar software can attest) - the problem is in scaling it. With computer vision-produced data it's the reverse: pretty easy to scale, much harder to fine-tune to a reliability where a slightly unusual camera angle won't silently ruin the whole system.

But, like, what if you took the cheapest event data you could find, which was generally ok but a little inaccurate in its event locations or something, and you matched it to video and developed a computer vision programme to improve the timestamps and locations of that basic shot data? Maybe you add in some fancy features that aren't in the cheap event data too; now that the hard part of deciding 'this is a shot taking place' has been done, how hard would it be to add in things like shot height, shot technique... Could you DIY yourself some top-of-the-range shot data from a lower-grade input plus video plus basic computer vision?

(Even if you could though, would all the development time, and opportunity cost of not doing other things, be worth the reduced cost of the event data? That's one for the accountants and sales teams to fight it out over)

This segues neatly into an amendment I need to make to the pithy one-line attempt from earlier. If the balance of football analytics is shifting from explorers to entrepreneurs, then there's still one exploration-themed job that's still very much needed: cartographers. Data cartographers. By which I mean, skilled modellers who can take the cries of "there's something over there!" and work out its exact contours, sea depths, and mineral composition.

Nowhere was this more evident at the StatsBomb conference than the research stage. Investigating how existing model ideas applied to new datasets was a big, big theme, whether that new data be women's football leagues or StatsBomb 360. (The research paper competition winners and links to the papers are listed here).

This kind of work is particularly important, and the dynamic with the rest of analytics going forwards particularly intriguing, because "slow down and run thorough tests across different model types and parameters" gets to coexist with "move fast and break things" (and/or "move fast and market things"). Also, because companies keep producing new types of datasets nowadays.

It may not be glamorous work, but it is essential. Partly because without good cartographers you risk causing a ship wreck or wandering into a desert. And partly because this work being thoroughly done and the findings being written up makes it less likely that the knowledge will vanish from memory (or, at least, that it's more likely it'll be unearthed again if it does). We all need cartographers and we don't give them enough credit.

Anyway, that's a thread of thoughts that over the span of a few months would probably have formed the basis of two or three more well-put-together newsletters. It's a rough stock take of where I think things are at the moment, probably not right at the edge of analytics but sorta two steps back from the edge. The kids are growing up, putting on business suits, and going out into, and shaping, the big wide world.

Speaking of going into the big wide world, for those on the outside of the industry seeking a way in it might seem demoralising that there are fewer low-hanging branches with which you can haul yourself up on. While there are a lot more resources to learn from, it's unlikely you'll be able to sweep to fame and/or a career on the back of a rusty, homemade xG model like you could a number of years ago. It sometimes seems like the skill or knowledge levels required now for meaningful and/or meaningfully-paid positions are very high (which is unsurprising, it's a competitive industry).

But... But but but. Let's flip the skill-requirement idea on its head. If you really want to learn a lot about data science then there's a ton of work based on your favourite sport to sharpen your expertise on now. If you're not maths or coding technical but are interested in communication, journalism, visualisation then you can take some of the nerd-stuff and work on making it accessible and popular. If you're interested in business or product creation, there's now a whole lot of people whose skills and ideas are there but might need a little refining or packaging to take things beyond just 'ideas'.

In the very early days, (public) football analytics was mostly a thinking exercise, an imagination game with not much data to go around. Then there was a code-y mathsy stage that merged, with the increasing openness of tracking data, to a physics-y mathsy stage. At some point, maybe the physics-y people really will 'solve football', or come as close as feels possible, at which point analytics may well become a pure 'product and pounds (£)' business.

But at the moment, it's a little bit of all of those things. It's everything.

Thanks for reading

*Prompted into nostalgia by the Post Script podcast, I've re-opened the least un-worthwhile parts of my old, old blog as a kind of historical document

What if we didn't care about passes?

Wed, 14 Sep 2022 00:00:00 GMT

"If God had wanted us to play football in the clouds, he'd have put grass up there." — Brian Clough

"If God had wanted us to collect passing data, he'd have put abacuses in the press box." — Mark Thompson

A lot of people have spent a lot of time asking why it's taken so long for football to 'have its Moneyball moment'. I have two theories:

Michael Lewis never spent a year embedded in Bolton
Data might have been good quality for a long time, but it wasn't cheap and it wasn't free

The foundational stone in the Moneyball mythology is Bill James, the author of the Bill James Baseball Abstracts, who spent night shift after night shift poring over all the data he could get his hands on. He was using box scores which, although not perfect or extensive, gave quite a lot of information to work with, all of which was publicly printed in newspapers. In football at that time, the data-sceptic's maxim that 'the most important statistic is the scoreline' was almost inarguable, if for no other reason than it was the only one available. Say the phrase 'box score' in 1970s England and people would probably think you meant a successful shot in the penalty area.

That's not to say that there was no, or had never been, any decent football data around. By the time that James was putting together his first Abstract, an Englishman called Charles Reep had been providing data analysis to football teams for two decades (I've written about Reep previously). And in Hungary, newspaper Nemzeti Sport had been publishing data visualisations for decades before even Reep came along. As is so often the case, there's nothing new under the sun.

But their data wasn't easy to collect. Reep noted the number of passes in sequences of play, as well as zones that they started and ended in. Nemzeti Sport published what we'd now call momentum charts, showing what looks like teams' progression up-field throughout the match. Matches which, by nature of the sport, are broken up by pauses only at random intervals of varying length. Baseball - like cricket, American football, even basketball to a degree - are almost designed for amateur analysts. One team does something, the other team perhaps does something in response, and then we all pause while we fill in our stat cards.

There's nothing 'special' about box scores. They're a kind of Darwinian representation of the sport; a distillation of what seems important, what readers find interesting, and what's cost-effective to collect. As ground-breaking as Nemzeti Sport visualisations may have been, as intriguing as Reep's data, as rich as the tracking data of the early 2000s, none of it was that.

What Reep and Nemzeti Sport's work does do, however, is give us a way of thinking about what a mid-20th century data collection might have looked like without being too influenced by what we have in the present. It wouldn't have simply been a streamlined Opta collection process.

Both hit on the idea of zones that a team got the ball to. Both hit on the idea of shot locations. (Again, call 'em possession value and expected goals precursors and you're back to 'nothing new under the sun' territory).

We've got to keep in mind that, to go back in our TARDIS and gift football some mid-century box scores, things need to be simple enough for someone sat at a wooden desk, with milk bottle glasses and a flat cap, to note throughout a match. There are no abacuses in the press box. And that means, as much as Reep and a lot of 21st-century analytics latched onto them, we're jettisoning passes.

I say this partly in the hope that, by writing it, I can manifest into the universe an end to people using the phrase 'tiki-taka' in relation to Pep Guardiola's teams (some people still do it!). But mostly it's because the act of collecting pass data is pretty heavy-duty, and would be even pre-positional play and pre-modern pitch maintenance techniques. Get rid of pass stats. We don't need 'em.

So you're sat in the stands on a cold Tuesday night in Stoke, your warm winter flat cap on your head and a cup of Bovril by your side, and you're the stat collector for the match. You're collecting shots for sure, and some other newspaper has probably claimed a boost in sales after starting to separate shot stats into a couple of basic zones. Outside vs inside the box is an obvious candidate, but it wouldn't be a surprise if they'd come up with some kind of fancy 'danger zone' too, like the width of the six-yard box up to the penalty spot. Three zones is quite enough to be dealing with though.

Shots look quite bare on the page on their own though. You once heard a story that an editor, one of the ones up north, had gone into a rage some years ago because George Best wasn't at the top of any lists. The paper spent half a season paying kids half a penny a game at Old Trafford to keep count of various things he did. After Christmas they were rushing to the evening papers instead, to see if anyone in the league had got more 'forward advances' than him (some kind of basic 'progressive passes plus carries' metric, maybe starting in the centre-circle strip of the pitch and ending roughly in the final quarter of the field).

This all seems manageable - plenty of time to rest and sip your Bovril - so why not take an opportunity to try and crowbar something defensive-based into the mix. If George Best is being referenced in this data collection fanfic then that means Bobby Moore was around too, so it's not inconceivable that there'd have been a real appetite for some defensive statistics. That'd probably end up as some kind of clearance/ball recovery stat. But what about if there'd been some one, or some team, who'd really got a handle on defending Best at source?

In your winter-weather flat cap in Stoke you ready a separate scrap of paper besides your regulation stat sheet. Your grandparents were regulars watching the team; they always talked about the only bad game they saw Stanley Matthews ever play, when he just kept getting turned back where he'd come from. You start keeping a tally of 'turn backs'. Your editor loves it. The punters love it. The crowds start cheering turn backs like goals, betting booths at grounds start offering odds on it, and when you retire some young journalist tracks you down and writes a very pleasant two-and-a-quarter pages about you in their book on the history of something called 'analytics'. It's the best-selling book of 1996.

The following year, across the Atlantic, the new general manager of the Oakland Athletics carries a copy of the book into work on his first day in charge.

There's another starting point that this 'what if' adventure could take, still on the theme of taking the importance of passing stats down a peg or two.

The story of football analytics history always has the start of football analytics blogging as a key moment; for some histories, it's practically 0 AD. Part of this is because it led to, or at least correlated strongly with, an explosion of figures working with data; part of it is just that the way we construct history tends to focus on documentary evidence. There's nothing history loves more than bloggers.

The start of this starting point is around 2009 and 2010, with another peak around 2013 and 2014 before things took off for good. At the time, the dominant teams in men's football were Barcelona (on the club stage) and Spain (on the international stage). Vicente del Bosque and Pep Guardiola possession-ball was the way to play and, by the time of that second surge in analytics bloggers, had been for half a decade. Crucially, it had also been the way to play ever since the beginning of the more detailed Opta datasets that these analytics bloggers could get their hands on.

Is it a surprise, then, that through-balls and cut-backs were identified as optimal strategies, while long balls and crosses were often dismissed?

What if blogging had taken off earlier? What if we cared less about pass data not for reasons of cheap data collection (although, still: StatsBomb, Opta, Wyscout - think about it) but because that wasn't how the good teams won games?

Sam Allardyce's Bolton Wanderers, alluded to at the very start of this post, are the obvious team of interest in this decade 'pre-blogging'. While he later became something of a pantomime villain, more meme than man, long-ball connoisseur Allardyce loved data. As recently outlined in Rory Smith's book Expected Goals:

"Two years before Billy Beane's epiphany in Oakland started to transform baseball and a decade before football clubs started investing heavily in their data departments, Allardyce was writing almost the exact same story in strikingly similar circumstances, distilling what he had found in the data into a set of unorthodox principles that defined how his team played."

Bolton - little old Bolton who had spent just four years in the top-flight between 1964 and their promotion under Allardyce in 2001 - finished between sixth and eighth every season between 2003/04 and 2006/07. Had football blogging hit 2014-levels ten years earlier this is the story that it would have been looking into.

And it's not like Bolton was the end of this story. That period was also the rise of José Mourinho; it had Rafael Benítez - not exactly an exponent of Joga Bonito or Juego de Posición - reaching two Champions League finals in three years, winning one of them.

What if, instead of automatically looking for Lionel Messi at the top of every new attacking metric list, early analytics was looking for Ronaldinho? What if the midfielders to look to as easy examples of 'best in the world' weren't pass-extraordinaires Xavi and Andrés Iniesta, but Frank Lampard and Steven Gerrard?

Early analytics work pointed out that things like long balls and corners were inefficient ways of scoring. This was, by the numbers, true, even if it wasn't the whole story. But at the time that the growing football analytics community was taking its toddler steps, there was little to force them to seriously confront this fact. Even teams like Swansea City were getting (relative) success by copying the Guardiola template.

We, the masses, are getting around to it now though, somewhat belatedly. As the epilogue of Expected Goals points out, the 'analyticsy' teams of the moment - like Liverpool, Brentford - have been embracing set-pieces, once thought a little bit old-fashioned, a little bit, well... Allardyce: "The very inefficiency of set-pieces made them the game's juiciest low-hanging fruit."

Our environment didn't stop shaping us.

It's little surprise - to me, in hindsight, at least - that one of the big unique features of StatsBomb's data provision when it launched in 2018 was pressures and counterpressures. There was a lot of other stuff that they pointed to, but pressures was the big thing. And it made sense, given that ten outfield players are defending at any one time and existing defensive data was very sparse.

But what was the big tactical trend that took place between the possession football era of 2010 and StatsBomb's data launch eight years later? What was 2015-2018's buzzword answer to 'tiki-taka'?

Gegenpressing.

"And language is also, literally, the "containment". The terms we choose - or the terms we are offered - behave as containers for our ideas, necessarily shaping and determining the form of what it is we think, or think we think." — Zadie Smith

Football analytics is on the brink of a new world: 3D

Sat, 27 Aug 2022 00:00:00 GMT

There are a lot of ways to describe the path that new technology takes towards being adopted.

There's the typical 'life cycle' chart, probably a product of business schools, the population (market) divided into (terminology may differ) 'innovators' (nerds), 'early adopters' (wannabe nerds), 'the majority' (normies), and 'laggards' (a stand-up comic's parents).

Then there's Douglas Adams' three rules: 1) Anything that is in the world when you're born is normal and ordinary and is just a natural part of the way the world works 2) Anything that's invented between when you're fifteen and thirty-five is new and exciting and revolutionary and you can probably get a career in it 3) Anything invented after you're thirty-five is against the natural order of things.

And then there's Google Glass.

I looked it up recently and Google Glass, or the big Google Glass hullabaloo, was 2014. Nearly a decade ago. If you said 'Google Glass' to Ansu Fati - aged 19, born late 2002 - he'd probably wonder why you wanted to do an internet search for such a common material and that, besides, he'd sooner look it up on TikTok anyway.

Google Glass may well make a resurgence in time, but it never even got to the early adopter phase of the adoption chart, never got people of Ansu Fati's (current) age thinking that this was something new and exciting and that they could probably get a career in it. It was a little too disturbing, a little too complicated and expensive, a little too 'well what do I do with it?'.

Football analytics has gone from strength to strength since 2014. If Google Glass had had the rise that football analytics had had since then you'd all be reading this a quarter-inch from your left eye while watching TV with the other. Even the laggards know what xG is now.

But in all those years, y'know something that all this data and all these genius minds haven't dealt with much yet? 3D.

For all this time, the rocket ship of statistical soccering has grappled, in large part, with just the X and Y planes of existence. (I say 'just the X and Y' instead of 'just two planes' because this is the kind of newsletter whose readers will write in to point out that football analytics has long dealt with time, itself a dimension — yes, I am familiar with the 'normal chess is 3D chess, 3D chess is 4D chess' argument).

3D data is less important in football than in some other sports (where it's already been worked with) given that most of the action happens at, or in the neighbourhood of, ground-level. It's also a sport of not just low scoring (making it harder to work on repeatable scoring mechanics), but of low 'opportunity to create scoring chances'. Sports get war metaphors a lot, but football isn't a series of skirmishes, it's career planning in the sixth-form guidance counsellor's office. If you lay the right tactical foundations and keep good structure in possession maybe you can land an entry-level job interview at Natwest.

So, unlike basketball, where 'body pose' software was used in 2017 to analyse NBA player three-point shooting motions (Ansu Fati was not yet fifteen), 3D skeletal data in football has been, to Get Goalside's knowledge, scarce.

Is it because, like Google Glass, it's a little complicated and expensive and honestly a little silly-looking? Maybe.

Unlike Google Glass, though, it's getting less and less 'well what do I do with it'. A paper earlier this year used body pose-adjacent data-gathering to analyse Leonardo Spinazzola's Achilles tendon rupture. At this year's MIT Sloan Sports Analytics conference, another paper took goalkeeper body pose data from one-v-one situations to analyse save technique decisions. (I wrote a 'Research in Focus' summary of that goalkeeper paper here).

That second paper is an example of the kind of isolated technique that lends sports like baseball and cricket to such a wealth of quantitative analysis - and, as it happens, use of this kind of data. Driveline Baseball has its own biomechanics lab to use data and luminous skeletons to improve performance. Sky Sports debuted some stick figure analysis in their England-South Africa Test match coverage earlier this week.

Never seen this before in any sport- skeletal tracking (quite literally) in broadcast @SkyCricket pic.twitter.com/jYR0xhwr5e— Simon Farrant (@simonjfarrant) August 25, 2022

In the old days, shortly after Moneyball (the movie) came out, people said that the reason we didn't have an Oakland As or a Billy Beane in football was that baseball was just a more quant-y sport. It was all about numbers, and football wasn't.‎ Some said that data couldn't work in football - or at least not anywhere near as significantly - for that reason. They were wrong. (We now have several Billy Beanes in football, as well as the actual, first-edition Billy Beane).

It would probably be wrong, then, to sit here and say that football - unlike basketball, baseball, and cricket - simply isn't a sport where body pose data would prove to be useful.

Goalkeeping (as readers of the aforementioned paper or attendees of previous StatsBomb conferences will know) is an area ripe for its use. The current trend in set-piece focus also feels like it would meld well with skelly-data analysis.

Other discrete skills are probably a bit more difficult to capture and train in settings where you can obtain high-quality data, but surely not impossible. A step removed from set-piece shooting would be the 'Coutinho' or 'Robben' shots: wide forwards cutting inside to the shoulder of the box, where there is usually some space. These situations seem more replicable than many types of shots, and offer relatively few options on the shot itself. I'd be intrigued at the idea of using this approach to study players' control of bobbling passes too, as a frequent part of the game that seems potentially reproducible in lab-like settings.

To use the Douglas Adams criteria, people like me were born into a world of event data - Player A tackled in (X, Y) location - it feels normal and ordinary and just the way the world works. Body pose data is new and exciting. Potentially revolutionary. Is it in, or approaching, a place where you could make a career out of it?

Do VARs dream of electric football commentators?

Sat, 16 Jul 2022 00:00:00 GMT

It's not often that Martin Tyler and the MIT Sloan Sports Analytics Conference are mentioned in the same breath, so savour this newsletter while it lasts.

At this year's Sloan, two folks from Amazon and one from Fox Sports got together to publish a paper, titled 'Sports narrative enhancement with natural language generation'. Their aim was to turn bits of sports data into sentences that 1) made sense 2) read well.

While it's fairly straightforward to put data into templated sentences, it's much harder to add in variations and keep it all sounding human and un-robotic. At the same time, you don't want to lose the meaning or, worse, accidentally change the meaning of what's being said.

It helped them — Henry Wang, Saman Sarraf, and Arbi Tamrazian — that sport is fairly formulaic. There are set periods of play, set numbers of players, on-field officials to make instantaneous decisions. As much as we like to think of sport as being played on fields of possibilities, there's quite a tight limit on what can actually happen. Nobody's discovering penicillin at the Camp Nou.

This, after all, is why the sporting video games we all know and love are able to have in-game commentary that more or less works for anything you do. Enter Martin Tyler.

"I'm reacting to prompts from the game-makers," he explained to VICE in 2015, talking about recording for the FIFA series. "They'll say, 'It's 3-1, and they've just scored, and there's five minutes to go – what would you say?' So, I go, 'Ooh, Five minutes to go and they've just scored? Maybe they've got a chance now.' And then they'll ask me for another one[...]And then they'll ask for another one.[...]And then they'll ask for another one, and so it goes."

When you're EA Sports, with the money that the FIFA franchise has to burn, you can afford to get someone to spend a few days recording all these variations, which you can then pick and play at random when the time comes in the game. But for anyone else this would quickly become unfeasible and unmanageable, particularly if you're trying to create templates for data to slide into. It's not flashy, but a big part of the Sloan paper's value is the scaleability.

Now, if we're already used to 'automated' human commentary in video games, and we have techniques to create natural-sounding sentences from data, in a sport that has so much of it... how far away are we from this all being combined in real life? How much of a stretch would it be to have properly automated TV football commentary?

Automated reporting (or 'AI journalism' if we want to use the proper buzzwords) is already firmly in use in written media. The scope is still fairly limited, but, for example, the BBC have been using a partly automated system to report on NHS data since 2019.

Sports results were actually one of the first widespread uses of these techniques too, with the Associated Press in the game since 2016. It's for exactly the reasons already mentioned: regular events, ready-made data, largely formulaic outcomes.

The article linked above has an extract of a machine-written AP story, which gives a sense of what the Amazon and Fox Sports engineers were trying to improve on:

STATE COLLEGE, Pa. (AP) — Dylan Tice was hit by a pitch with the bases loaded with one out in the 11th inning, giving the State College Spikes a 9-8 victory over the Brooklyn Cyclones on Wednesday.
Danny Hudzina scored the game-winning run after he reached base on a sacrifice hit, advanced to second on a sacrifice bunt and then went to third on an out.

It's to sportswriting what Huel is to dining experiences: gets the essentials and, granted, technologically impressive, but you'd feel cheap gifting it to your mother.

However, if, thanks to software engineers like the ones who wrote the Sloan paper, this could be made to sound a lot more human, these words could be run through an artificial voice generator (which are themselves getting more lifelike) and put straight into a broadcast.

What this ends up looking like (or, I guess, sounding like) could depend on commentary style too. Some approaches are largely descriptive: player names, general locations on the pitch, basic descriptions of shots or notable passes. With a small enough delay between the action happening on the pitch and the data being processed*, this could easily be automatable.

*(note: this is probably a fairly big assumption, although data companies will know, and boast about, their latency figures better than me; even if it's not possible now, though, if there's the potential for money to be made, someone could always set up near-instant 'commentary data' specifically).

On top of that, you could even put together a separate language generation set-up with shouty, enthusiastic voice synthesiser aimed specifically at big moments. When goals are remarkable it's mostly because they're from long-range, after a long passing move, or came at a dramatic point of the match. All of these traits are easily identifiable in data.

Imagine the scene. You've finished a round of bug fixes on TylerBot v3.1.6 and are setting it loose on a match. "Manchester City on the edge of the box. Attempted pass cut out. Touch from Balotelli, out to Sergio [TylerBot identifies a shot has just been taken] Agüer- [TylerBot identifies that a league title-winning goal has been scored; Big Moment mode is activated] -OOOOO!!! I swear you'll never see anything like this ever again!"

Or something to that effect.

Martin Tyler, the real Martin Tyler, was replaced as the voice of FIFA video games in 2020 (although the replacement (this time around) was another human, Derek Rae). In truth, even if the FIFA games decide to jettison the carbon-based commentators, the real sport of football is random and unexpected and instant enough that humans are going to remain the preferable option. Moments like the Agüero goal, or Eric Cantona's kick into the crowd at Crystal Palace, or, for very different reasons, Christian Eriksen's cardiac arrest need a real person to describe them appropriately.

However, there's an increasing amount of sport being broadcast on an increasing number of channels, and companies will always try and do things on a budget. If a small league wanted to air every single one of its matches somewhere on the cheap, without paying a full slate of commentators every week, could they use automated commentary for some of the games that won't have as large a focus on them?

Another idea (which will feel a little less scary for existing or aspiring freelance commentators — sorry folks) would be to pair a human commentator with an automated co-comm.

If the human on-mic has a more traditional play-by-play style, the AI could be set up as an insights and fun fact-based colour commentator. Or, if the real person is more interested in relaying atmosphere or their own analysis, maybe the AI could be the play-by-play and could be turned off/muted when the human wants to chip in.

This, rather than replacing the Martin Tylers of the world outright, seems like the most likely use of automated commentary, if it's ever going to happen. Perhaps it's something that some of the mid-sized leagues across the world, vying against each other for global attention, will look into.

And it might be closer than we imagine. IBM demo-ed a prototype system back in 2019, although the sporttechie write-up reports that it said the same phrase — "Here comes the cross!" — four times.

With time, the systems will be able to get better. Hey, Siri, the fans are spilling out onto the pitch; is it all over?

How football competitions are their own competition

Tue, 28 Jun 2022 00:00:00 GMT

"Father?" the little child said, looking out past their parent's face and into the distant sky, "do other football leagues proclaim themselves the Best In The World? Or is it just ours?"

In the olden days, football leagues were mere bureaucratic bodies. They organised matches and set rules. At some point the job evolved and they became entertainment businesses, like HBO but with no scripts, dragons, and only occasional nudity (remember that fan's celebration to an Allan Saint-Maximin goal).

The most famous example of this switch is also the closest to home for the Get Goalside newsletter: England's Premier League. In the early 1990s, the big clubs in England thought they'd be able to negotiate more lucrative TV rights if they were doing it by themselves and that, at any rate, the money being shared between 24 teams (at the time) was preferable to being shared amongst the entire 92 of the Football League.* They knew that what they had was an entertainment product, and there was money to be made in providing it to the people as one.

The entertainment business has changed since 1992, though. And so has the business of football competitions.

The big story of the TV biz at the moment is streaming, but another way of framing it would be as a kind of 'internetisation' of television. There are more options available, and more people able to watch the same things (legally or less legally) across the globe. As a result, domestic networks aren't just competing against each other, and not just against Netflix anymore, but against Disney+ and YouTube and Twitch and all the rest of them too.

And so the same in football. In the '90s, the way that football leagues were akin to entertainment businesses was that there was money to be made, via television and merchandise, primarily from domestic audiences. As we all know, the 21st century has seen a rocketing influence of the international audience, to the extent that the Premier League's overseas TV rights will now, for the first time, earn the league more than domestic rights.

So here we reach the main point. Just like in the television industry, football leagues are now competing much more directly with their overseas equivalents. This is why La Liga (not just Real Madrid or Barcelona) are taking it upon themselves to complain about Paris Saint-Germain's and Manchester City's finances. It's also why they have their own analysis and visualisation tool, Mediacoach, which forms part of LaLiga Tech, which launched last September. All a way of trying to make sure that theirs is the best product around.

On a slightly different scale, the relatively recently-formed Canadian Premier League has made a concerted effort to help the entire competition with its own CPL in-house analysts and expertise. For a league to succeed, the clubs need to succeed, so why shouldn't the league help facilitate that. (The CPL is also an example of using the internationalisation of football TV: matches are available to watch in the UK on BT Sport - pretty remarkable for a league whose salary cap in 2021, for players and coaching staff, was $1.2m).

The CPL push is, in part, a way of trying to provide a boost to the Canadian men's national team ahead of the World Cup in 2026, where Canada are a co-host. They're not the only country leveraging the increased competition between the world's domestic leagues for the national team's benefit. Norway's national FA and the associations of their men's and women's leagues are creating a joint project, Europatoppen, which aims to make sure everyone's working on the same page to help move Norwegian football forward.

If you're reading this newsletter you'll probably be very aware of various clashes in global fooball governance. FIFA's desire to expand the World Cup and/or Club World Cup clashes with what confederations like UEFA want to do. In turn, UEFA's desire to expand the Champions League clashes with the space that domestic football has available.

The point of all this has been to show that those Godzilla-vs-King Kong battles aren't the only fights between administrative bodies. Everybody wants to be the winner, and everybody has competitors. Even the competitions themselves.

In some ways, this isn't new. Viewers of English TV are very used to broadcasters touting the Premier League as the 'best league in the world'. The league itself has long tried to make itself marketable and appealing, which, explicitly or not, puts it in competition with other leagues in international markets.

However, as the world gets smaller the market gets tighter. And just like football teams, these leagues are getting increasingly savvy at business organisation, technology, and data in their bid to rise up the ranks.

Postscript

*It's kind of funny that the Premier League is the original example of a group of big football teams banding together to break away from an existing structure in a bid to leverage their size and power for more money, and then it was the Premier League clubs who were the catalyst for the 2021 European breakaway project's early demise because of English club supporters' anger at the idea. If the Premier League weren't so rich - partly coming from them being ahead of the curve on monetisation, which in turn partly comes from their initial 1992 breakaway - then they probably wouldn't have been so secure in saying no to the 2021 breakaway idea or as numerous in the group approached to be part of the scheme (and therefore strongly influential when withdrawing as a bloc due to the nation's cultural feeling).

So... everyone has hybrid data now

Sun, 12 Jun 2022 00:00:00 GMT

Previously on Get Goalside: "With the bombast of a start-up, StatsBomb talked about how their data was going to change the game. And the bombast was deserved. Because it did. Because, nowadays, everyone seems to have pressure data."

It's time to talk about new data again.

Like every time we talk about data collection, we need to introduce 'event' data and 'tracking' data as separate individuals. Event data tells you what player made what action, and where on the pitch. Tracking data tells you where everybody is, but doesn't tell you what they're doing. Event data is collected by people; tracking data is collected by cameras and software.

Or, that was how it always used to be.

If you were smart, you would always have been able to combine the two (if, of course, you had the two types). But not many people had those skills.

The hybrid data era began in 2018 when StatsBomb first launched themselves as a data company (they'd previously been an important analytics blog (which, disclosure, I wrote for occasionally) and a consultancy). Their data was event data — telling you which player did what action just like everyone else — but for shots, they'd offer something unique.

The 'freezeframes' would give the position of every player on camera when the shot was taken, as well as some information on how the goalkeeper was moving at that moment in time. (A StatsBomb blog post on these shot freezeframes is here).

Two incredible saves in normal time and extra time before three penalty saves in the shootout

An all-timer of a performance from Brice Samba in the Nottingham Forest goal last night#NFFC pic.twitter.com/Qd5UdFDa2u— StatsBomb (@StatsBomb) May 18, 2022

That's useful, right? You get to know how many defenders were in the way of a shot, where they were, where the potential pressure on the shooter was, where the goalkeeper was, et cetera. But why stop at shots?‎

Well, three years later, StatsBomb stopped stopping at shots. They started offering freezeframes on all of the events in their event data (which they call StatsBomb 360). I wrote about the StatsBomb 360 launch in this post. As I say in that piece:

"[T]he whole theme of enriching event data makes a lot of sense. It allows you to count more stuff (and more useful stuff at that) while not adding too much extra technical requirement in skills or computational power."

Subscribe to Get Goalside!

This is where I take a rare victory lap: back in the day, at the first Stats Perform Pro Forum I attended in 2018, I remember a conversation where I suggested that maybe in the future sites like WhoScored (which use event data) would also columns for tracking data-based metrics. Tracking data would be snipped up to produce regular statistics, and presented alongside event data. Hybrid.

Of course, by this time StatsBomb will have been well underway setting up their own hybrid dataset, and other people will have been using hybrid systems themselves. But still.

Now, StatsBomb's data might be a kind of hybrid, but it's not 'tracking data' per se. It doesn't need to bother being tracking data, it just takes a snapshot. Earlier this week, Stats Perform (who own Opta) launched their own hybrid dataset: Opta Vision.

Opta Vision is a hybrid system in its 'purest' state: combining Opta event data with STATS tracking data. (Stats Perform's press release on Opta Vision is here). Alongside the event data, insights that are only possible using tracking data can be revealed. Things like passing options and pass difficulty, how a team's shape is changing, and when players are making runs to make themselves available.

Both companies having hybrid data systems that cover the vast majority of a match feels like a turning point in football data. We've crossed the threshold into a new world.

This doesn't mean that non-hybrid data is useless. The choice of data provider that someone uses will always partly depend on price, and hybrid data seems likely to be at the pricier end. It will also always be the case that what you do with the data is more important than the data itself. I would cook an expensive cut of meat far worse than a chef would cook a cheap one.

While the biggest factor in a team's success at using data will still remain 'giving time to good analysts', hybrid data is going to help.

But it'll also make 'giving time to good analysts' even more important. Every time a data offering gets bigger that means there are more things to discover, more metrics you need to decipher the value of. StatsBomb and Stats Perform have themselves only recently adjusted their expected goals models based on data they've had sitting around for a few years.

D'you want to know something that means? It means it's the best time in years to get into this. I said we'd crossed a threshold, but a lot of people are going to take a while to work out what's beyond it.

New types of data mean new types of opportunities.

Addendum

There wasn't a natural place to put this in the main body of this newsletter, but I wanted to touch on StatsBomb's recent American football data announcement too. (A link to StatsBomb's main American football data launch piece is here)

That data is also a hybrid approach, also built on freeze-frames, but time-based rather than event-based. This makes sense, given that pre- and post-snap movement are so important in that sport and you'd miss a ton of vital information if you only took a freezeframe for the snap, a hand-off or pass, and catches, tackles, whatever.

The frequency of events in (association) football probably means that you're capturing about as many freezeframes by taking them per event as you would if you took them per X tenths of a second, as StatsBomb appear to be doing in American football. That said, I'd be really interested to see a similar time-and-event-based approach in soccer.

The other part of this I'm interested in is the balance of computer vision and human collector input and level of accuracy of the data.

Former Milwaukee Bucks director of research, author of The Midrange Theory, and StatsBomb's Director of Basketball (hey, when did that job title change?) Seth Partnow was recently on the Expected Value podcast talking about this new data. One of the things he talked about was that it really matters that you get player identification correct at the line of scrimmage. In soccer, some tracking data systems have trouble at corners. In American football, everything's a corner.

There's one last strand to this American football StatsBomb data that is of interest to me.

In his launch article (linked previously), CEO Ted Knutson says "We’re actually delivering low frequency tracking data in the new data". 'Traditional' soccer tracking data runs at about 25 frames per second, but I've wondered before how many of those frames are actually necessary. How much could you strip back and still have useful 'tracking' data?

I suspect that those at the cutting-edge already, at least, have some idea; but I suspect that everyone else is gonna find out within the next 5 years.

Thanks for reading - if you've enjoyed this please share it around, subscribe if you haven't already, and consider becoming a full supporter of the newsletter if you have. Have a good day.

Upgrade to paid

Ralf Rangnick, and the value of practice

Mon, 30 May 2022 00:00:00 GMT

There are, arguably, only two lines you need to read from Michael Cox's recent column on Ralf Rangnick's time at Manchester United.

"Ultimately, a huge red flag was that, in the decade before Manchester United appointed Rangnick, he had been appointed as a manager twice. The people who appointed him were Ralf Rangnick and Ralf Rangnick."

The pair of sentences reflect how little sense has been in the air around Old Trafford this last seven months. The only role that Rangnick has ended up playing was interim manager, the role he wasn't well-qualified for, and the role he was well-qualified for, a consultant at around director of football-level, is one he now won't do.

This strange set of circumstances has an equally strange, and equally interesting, accompaniment. In November last year, just before taking on his job at United, an interview with Rangnick was published on Red Bull's Red Bulletin website. His observations about club management there contrast sharply with some details of his time in Manchester, as found in Laurie Whitwell's report for The Athletic.

For example, Rangnick's opening line when asked about they key qualities a successful coach needs to have? "Team insight, interpersonal skills and decision-making authority". Compare with the following passage from The Athletic:

"[Rangnick] did not anticipate the dressing room problems he encountered, nor how his interim status would make him feel more hesitant to push through his own vision. Instead, his instinct was to try to accommodate players rather than cause further ructions; something, on reflection, he is understood to regret."

Later in his answer Rangnick says: "It’s not enough to merely delegate tasks to these experts [staff in a coach's wider team]. You are expected to engage in dialogue with them." In contrast, the opening section of Whitwell's article talks about how Rangnick's use of a former colleague still based in Russia created communication and decision-making problems.

And, finally, when he is later asked in the Red Bulletin interview about what a club will need to do in the future to ensure success, his answer includes the sentence "Success hinges on three Cs: concept, competence, and capital." Many assumed that, at the very least, the German credited with huge influence on that nation's tactical outlook would bring that first C, concept. And yet, from Whitwell's article again, "Rangnick began going against his own ideas to find buy-in."

In the former interim manager's defence, the squad he inherited hardly helped. Rangnick's very next sentence in the Red Bulletin interview after his 'three Cs' could almost be aimed directly at the United hierarchy: "Capital is often readily available, but in most cases the project fails due to a lack of competence on the part of the individuals involved and a lack of concept."

But still. Even in a short interview, Rangnick has the answers for how a good manager should manage. Yet it appears he was some way off being able to put that into practice though.

I write all this not to figuratively kick a man while he's down, but to highlight just how difficult these jobs are. There is clearly a lot of respect for Rangnick's body of work within the game. He appears, from the outside, to be a smart man about football. And yet that still doesn't necessarily translate to being a top-level football manager.

One of the striking things about both Pep Guardiola and Jürgen Klopp are the ways they can simply explain tactical concepts, the extent of their collaboration, and the very open emotion they display. Klopp's hugs are renowned. Guardiola's dancing is a "big mood" (especially around 35 seconds).

Are they elite tacticians or are they elite people managers? They're both.

Find out more about becoming a supporter of the Get Goalside newsletter here

The eternal problem with getting good at practising what you preach is the 'practise' part. There may well be a world where Ralf Rangnick is a great head coach, but it's probably a world where he doesn't have to step down, through chronic fatigue syndrome, as Schalke manager in 2011 and then spend the next decade in sporting director-type roles.

This is a problem not just affecting Rangnick, but may well affect the ever-growing number of people looking to get involved in football analytics roles.

While not easy, it's relatively straightforward to get to know FBref, to learn some R or Python, to check out Friends of Tracking YouTube videos. It's slightly harder to get a feel for the types of tasks that would need doing in a professional job; even harder to get a feel for how to act when they inevitably go wrong/get sidelined. (The insights from an array of industry professionals in Get Goalside's 100th edition may help though).

The particular dynamics of this strange little space we find ourselves in also make things awkward. Data roles are new, so there aren't that many people with lots of experience to give; many who do might feel too precarious in their job security to give away inside details. Some insight can be gleaned from conferences (the club-led questions at the Stats Perform Pro Forum are a good hint at the 'real world'), but these are generally special projects, not day-to-day work. And then there's the question about how much attention that day-to-day work actually gets anyway.

Every now and then, someone on Twitter will write a piece of advice for the people who ask them how to get into these types of jobs. One of the central hooks is usually 'do public work', although examples of what kind of public work are rarer (probably for reasons touched on above).

An article about GPS data in sport I saw earlier today had a sub-heading which read: "Research will not yield a good answer until you ask a good question". I suspect that the same applies to pro-analytics-hopefuls too.

For those hopefuls, working out good, realistic scenarios might be more useful development than extra hours spent reading up on theory; for those in jobs, putting out good, realistic scenarios might be the most helpful advice you can give.

Practice makes perfect.

Fear and fatigue in analytics modelling

Thu, 26 May 2022 00:00:00 GMT

Forgive me, for I have been watching Drive to Survive - the Netflix Formula One series - for the first time. And like a new convert, I'll be talking about it.

The amount of footage and array of angles on offer make for an exciting show alone, never mind the personalities and storylines. You really get the sense of how fast the cars are going, how they stick on the track and cannon into corners almost sideways in a way that cars simply shouldn't.

That they do is all because of the downforce that the meticulously-engineered cars generate, quite often only possible because of the high speeds they're moving at. Pushing too hard can cause trouble, but easing off the accelerator doesn't necessarily make things safer.

And so we get to football analytics.

In 2017, a group of analytics researchers released a piece of work on defensive 'ghosts'. It's like creating a FIFA AI from real data, that reacts to developing action just like real players would. (Click here for the full ghosting paper, or see this link for a 'Research in Focus' handy summary).

They also released a short, two-and-a-half minute video. In it, they highlight two examples where the ghosts, which have been trained in a model to be how 'league average' players might play, act more proactively than the real-life defenders actually did.

In the second example, attention is drawn to a team sitting back as a collective unit. This reminded me of one of the earliest ideas I can remember seeing analytics folk advocating: stop sitting back when you go a goal up.

Stop easing off the accelerator.

It's something that is referred to in Christoph Biermann's book Football Hackers: "[Matthew] Benham's stats had convinced [FC Midtjylland manager Glen] Riddersholm that a narrow lead was best defended by going aggressively for another goal; playing not to concede, on the other hand, only decreased the probability of scoring and and increased the opposition's chance to find the net." (chapter 3, Rise of the Outsiders).

It might intuitively feel more comfortable to be less aggressive in some circumstances, but, just like in F1, the best approach might actually be to keep the foot down.

By chance, a few days after I recently revisited the 'ghosting' video a football match I was watching featured a turnaround by one team, partly brought about by a more aggressive approach in their defending after half-time. Their opponents were suddenly getting no time or space, the newly-aggressive team won the ball back quicker, they were able to assert themselves on the match.

It's hardly limited to that match either, it happens quite often. So why is the actually-slightly-counterproductive retreat such a common feature of football? And what can analytics do about it?

I think there are three things at play, two which have particularly interesting and important implications for making analytics models.

The first and most simple is that sometimes it works. Football is a low-enough scoring sport that it can be possible to just ride out a match while defending deeply or passively. (That said, I think the most common version of this is that a team defends this way for about 20-30 minutes and then realises that they have to move up a gear again).

That shouldn't matter too much in modelling though: if it works then the model would recommend it. It's the second and third that are trickier; two F's: fear and fatigue.

The reason why players defend passively is not that they're saying "ah, winning the ball back isn't important", it's that they're saying "we're worried they'll get to the space in front of goal, so we'll give them space further away from goal instead". It's a tactic of fear.

Sometimes fear is sensible. But sometimes it's just a lack of confidence. And in those circumstances, though easier said than done, the best approach is to get confident.

The thing that modelling might struggle to help you with is knowing when the fear is justified. When a defender goes to tackle a forward as they receive the ball on the halfway line, the margins are tight. If the challenge doesn't go to plan, they might need to turn and sprint back quickly. Nobody enjoys that.

Part of the reason no-one enjoys that is fatigue, the second f from earlier. Players get tired, and that affects their speed covering distances and accelerating. Any model that relies on these movements as factors (such as ghosting or pitch control) might also need to consider fatigue. A model that is 'naive' to tiredness might suggest that a player is in the wrong place; the player themselves might say that they were stood where they were stood because they couldn't recover as normal.

Fortunately, although it might be hard to factor this into analytical models, it shouldn't be impossible (even if we don't hear about it in public for a while). [There might already be public research relating to this but I don't have it easily to hand]

For pitch control models - which work on the basis of how long it would take for a player to reach a particular point of the pitch from where they are now - player speed can be built into them. It stands to reason that a fatigue factor could be too. Maybe this would be based on time spent on the pitch, or distance covered at a high-intensity.

For more on pitch control models, follow this link

Although each individual will tire differently, it seems likely that there'd still be some kind of general rule that you could follow. A halfway house between a complete average and an individual approach could be positional group: research shows what we all expect, that different roles have different physical outputs (a recent paper that touches on some of this position-to-physical metrics interaction is here).

If you really trusted the quality of your tracking data, you could even help to manage defenders' fear levels in-game. The next few sentences are a bit of dreamworld, but why not live in dreamworld for a little while?

Let's say you build a profile of an opposition forward with their levels of acceleration and speeds they can hit in matches. Then you monitor their reactions in the early part of a game and see where in the scale their acceleration and speed levels are falling. If they're below their regular levels, then you could tell your centre-backs to have more faith in getting touch-tight.

Going even further, and putting even more faith in the tracking data and modelling ability, maybe you could measure reaction times. Maybe you could run a defender's movements through a model and it could say "they've reacted slower than an average defender would to these interceptable passes and these off-shoulder runs". And maybe you could take that and tell the player that maybe they should either focus a bit more or to ease away from the tight challenges.

I don't think that tracking data is at the level where it could make that possible, but advances are being made all the time.

It seems almost unnatural to be able to coach fear out of people in circumstances where it would be quite reasonable. But sometimes in sport it really helps in winning.

Upgrade to paid

Postscript:

In one episode of Drive to Survive (season 3, episode 8), driver Carlos Sainz is shown graphs, which I presume show different levels of forces the car was experiencing while going around corners. His lines were being compared to his teammate, Lando Norris's.

"So, compared to Lando, this was his fastest through turn one compared to our fastest through turn one," someone who I assume is some kind of engineer says.

"You can see he's challenging the fronts [front tyres?] just as much in this phase," a biro points to the laptop on-screen, "but his style is that he won't extend holding the brake here because he always just programmes himself to come off brakes the same each corner. It's just the subtle difference that he's got the tow[?] here, you've not quite got it."

Imagine being able to break down a video clip of a defender challenging for an interception-duel in a similar fashion. "You've reacted to the opportunity quickly, but the speed of acceleration is slower [than your teammate/than the opponent/than your average]. That was the difference between making the tackle and committing a foul."

We can dream.

When is data 'right'?

Thu, 05 May 2022 00:00:00 GMT

data [noun]: information, especially facts or numbers, collected to be examined and considered and used to help decision-making (Cambridge English Dictionary excerpt)

I

There's nothing better to get rid of the idea that data emerges to us immaculately conceived, like the census.

Each one brings new debates about what questions should be asked, what options should be given, and how things should be phrased. The Wikipedia page 'Classification of ethnicity in the United Kingdom' alone runs to more than 3000 words. Not including footnotes.

Fortunately, football data doesn't have the self-identification sensitivities involved in the census, but most of the same basic problems still apply. The first one being: how do you categorise the data you're collecting?

Categories don't just spring from nowhere, created with some kind of distant objectivity. English-born Opta are, in my sphere of knowledge, the only data provider to classify 'tackles' as a separate and wholly individual event in their data. For others - like Wyscout, StatsBomb, Ortec - what English viewers might call 'tackles' are found under the family of 'duels', a more Continental approach. Neither is incorrect, they're just different ways of seeing the sport.

Can one set of categories be the 'right' one? Or just more appropriate and helpful for what the data will be used for?

II

Categorising the data isn't where the trouble stops. Once you've decided whether to collect tackles or duels you have to define them. How close do players have to be to make a duel? If a heavy touch goes to a teammate, is that a pass? When Sadio Mané slid towards Zack Steffen to score in the FA Cup semi-final, was that a tackle, a block, a shot?

Oh, and if it's not a shot, then that scuppers your nice rule that shots are the only event that can have expected goals associated with them.

Most people outside the data world would probably be confused if two data providers had two different shot counts for a match. The Mané example is far from the only area where definitions might be at odds though. Cross-shots are another obvious one. Contested first-time finishes, off cut-backs or conventional crosses, can be another.

If categories can't be 'right', then definitions can't either, only sensible. The fuzzy unusual cases, like Mané's goal, should also be treated with as much consistency as possible.

Thankfully, there are at least some things that are unarguable. Once you've decided on categories and definitions, assigning the correct player to that action is a matter of fact. Returning to the previous example (sorry Zack Steffen), you could call Mané's action a tackle or a shot, but saying that it was done by Mohamed Salah would be objectively wrong.

There's a kind of 'central source of truth' to things that officials are involved in too: goals, cards, fouls, substitutions, offsides, restarts (e.g. kick-off, goal-kick); although it can sometimes be tricky to tell who the referee has given a foul against. (VAR and playing advantage also throw up some minor philosophical questions — should you collect things which would have been fouls if the ref hadn't played advantage? is it incorrect to assume, therefore, that all fouls stop play?)

Official-related incidents only cover a small portion of a match though. For everything else, it's like the Sadio Mané shot again.

III

If you've read this far then you've already implicitly answered this question*, but you'd be well within your rights to be asking: why does any of this matter?

*Because it's interesting!

It matters if you're setting up your own data collection company, of course. It also matters, to a slightly lesser degree, if you're making a choice of what data source to use. And those decisions have to be made quite frequently.

As this newsletter has outlined, the number of cases where football data can be objectively incorrect is relatively small. Get players and teams right, get goals right, get location coordinates within a forgivable margin of error: sorted.

Most things outside that fall in the realm of 'is this sensible, unexpected but justifiable, or unexpected and faintly ridiculous'. It might be justifiable, for example, to have a slightly wider or narrower definition of a duel than you would use yourself. It might be faintly ridiculous, meanwhile, to [redacted]*.

*Note: no single company has a monopoly on mistakes and mishaps

How can you tell whether you're able to trust a dataset then? (Apologies to any data analysts/journalists reading who may be having unpleasant flashbacks). Diving into the documentation is a simple place to start. Does the way that they've categorised and defined their data make sense to you? Does it align with how you think about the subject, and how you plan to use the data?

Looking at some basic summaries is also good. Goal tallies should match figures from elsewhere, and other stuff, even if the definitions aren't what you'd write, should pass the smell test. Check for inconsistencies or missing data.

If time and data is available, recreating some of your existing processes or reports using a different data source is probably sensible. Fingers crossed that a team doesn't suddenly go from league-leading to mid-table in something. Because if you do then you've gotta work out why, and which version feels closer to 'the truth'.

Ultimately, it's a case of looking for whether the conceptual framework behind the data collection seems sensible and useful. Then a case of whether there are clear errors or inconsistencies. Then? There's a fair chunk of trust involved. You can't check every data point, although you should do your due diligence.

IV

So when is data 'right'?

Data is right when it's correct in matters of straight fact, and sensible, consistent, and useful everywhere else. Many, if not all, football clubs still do some data collection of their own, with video analysts tagging matches post- and/or in-match. That's because existing providers don't match all of their uses, so they need to create some added useful data of their own.

A difficulty is that some potential purchasers may not have the time or know-how to do their own due diligence. There also isn't (at the moment, at least) the kind of public community due diligence like we've seen around pandemic public health data, where it's common to see tweets investigating and explaining quirks in the data (e.g. case reporting delays around holidays, or changes in collection practices).

Really, "when is data 'right'?" is a relatively easy question to answer. "Do you know if this data is 'right'?" — that's the one that takes the work to answer.

The Stats Perform Pro Forum 2022

Thu, 07 Apr 2022 00:00:00 GMT

The Stats Perform Pro Forum is the analytics conference for the dreamers. Started nearly a decade ago as a place for hobbyist stats-watchers to meet and present ideas, it's still fundamentally a party organised by a data provider who gift amateurs with interesting ideas the means to explore them.

Now though, it's an established part of the Forum that presenters have the chance to work with tracking data, with mentorship from an array of impressive names.

There is a page from the hosts, Stats Perform, about the presentations here and an overview from The Athletic here. I'm going to give an overview of my own here at the top, and then spend a bit more time on each presentation afterwards.

After three years of quality newslettering, you can now be a paid supporter of Get Goalside! More information is available here or you can go right ahead and support the newsletter below.

Upgrade to paid

Now for the presentation-by-presentation stuff. I'm limiting this to the on-stage talks and not including the poster presentations, partly for space, but mostly because I only made notes on the talks.

There'll be a summary for each that I'll crib from Stats Perform's intro to the presentation, and then some commentary from me.

Expected Counter

Expected Counter: Probabilistic modelling of the occurrence and danger of counterattacks in soccer before they occur in real-time —Ashwin Phatak, Henrick Biermann, Franz-Georg Wieland

Summary: Aims to predict the success of a possible counter-attack and its potential danger to the defending team, measured by a new metric: Expected Counter.

Commentary: 'Counter' was defined here as a move starting 70 metres away from the opposition goal and ending within 30m it, making the journey in under 30 seconds. Don't get too hung up on that if you disagree with it though, I don't think things would necessarily change much if, for example, any of those boundaries were tweaked.

There were some neat observations about where counters were more likely to spring from, in terms of where the turnovers were. Within the pre-defined 'start of counter' area, the central area closest to the halfway line was the most likely part. That might be unsurprising, but there was a marked drop-off between the centre of that area and the wings at the same height of the pitch.

As well as the locations, the researchers looked at whether the numbers of attackers vs defenders made a difference (it did) and whether the compactness of the team being countered against made a difference (it also did).

This all might appear to be expected, but I think the important thing is the relative differences. With this kind of work, you'd have a better way of gauging the defensive value of keeping a player further back in possession to defend against counters compared to the cost that might have in attack, for example.

This particular type of work could also be done with freezeframe data (a snapshot of tracking data for each on-ball action), rather than the full-fat full-match tracking data that this used. That'd probably save on some computing power (by machine or person).

Breaking a high press

Identifying ways to efficiently break a high press — Bakr Annour, Silvio Matano

Summary: Applying event and tracking data to generate formation and player recommendations based on the pressing styles adopted by an upcoming opponent. The presentation outlined how sequences of interest are identified and how their success is defined, explained how their model was used to 'place them' in a common space, and how the backend of a hypothetical application would interact with the model.

Commentary: A lot of my favourite presentations at Forums in the past have been the ones that edge into being tech talks, and this was one of them.

Identifying typologies and similar situations of high presses with tracking data seems tough. Because as well as just looking at, say, the height of the press, there's all sorts of other factors you need to account for to make the similarity seem right.

Although examples in presentations can always be cherrypicked, their results certainly seemed plausible. There was also an interesting piece of insight that occupying the centre of the pitch close to the halfway line appeared to be an important factor in a team breaking through a high press.

Efficient runs

Measuring KPIs for scoring efficient runs — Caterina De Bacco

Summary: From a defensive standpoint, the model labeled runs based on whether they resulted in attacking players being marked or closed down, or interrupted a potential passing line. In attack, runs were credited based on whether they generated a passing opportunity, attacked vacant space, or triggered movement from an opposition player which in turn created space for a teammate.

Commentary: This presentation combined two things I really like. One was taking an existing fixation of football (distance run) and thinking about it in a smart way. One of the outputs of these models were literally graphs showing a player's distance run throughout a match and their 'effective' distance run (i.e. the amount they ran while doing good things). You could, as De Bacco did, easily create a ratio with these figures and compare players based on what proportion of their running was 'effective'.

The other thing is approaching the problem of football in a smart way and breaking it down into component parts. As with most things in analytics, you could quibble around the edges of the models that De Bacco created, and she pointed out some areas for improvement in them herself. But individually they all made sense, and collectively they covered most things in football.

I think that this presentation was a great example of how thinking about football smartly can really help produce good analytics work.

Reinforcement learning for scout allocation

A reinforcement learning approach to scout allocation and talent discovery in football — Arnav Prasad

Summary: The presentation demonstrated how clubs can build and test various algorithms to inform their own strategic decision-making, in this case with how they allocate their scouts.

Commentary: Reinforcement learning is one of those things that is surprisingly like what it says on the tin. You give a computer a problem and it updates as it goes along; the results reinforce the learning.

Most 'analytics' work focuses on what happens on the pitch, but it's neat seeing things that go beyond that. Crucially, this presentation recognised that scouting operations sometimes have different aims: some want to uncover gems, potentially at risk of certain quality; others don't mind about missing the unexpected upside, preferring the certainty.

If you created values for how much talent you thought might be in each scoutable region, and then used scout reports to update that, perhaps this type of approach could be useful.

Defensive efficiency with event data

Measuring and modelling defensive efficiency with only event data — Abhishek Amol Mishra, Soumyajit Bose

Summary: Evaluated ways of judging the defensive ability of centre-backs with event data, with possession-adjusted existing metrics and new metrics focusing on opposition goal, shot, and threat prevention.

Commentary: Yes, I have a predisposition to like any presentation focusing on not just defending, but centre-backs. I will try not to let that skew me. Unfortunately my notes on this were also not as thorough as they might have been.

There was a broad collection of metrics which (somewhat similarly to De Bacco's presentation) focused on different areas/types of defending.

There were convex hull-based stats (reminiscent of PATCH, although a fresh spin on it). How many passes went into the area a player is usually responsible for defending, and how many attempted passes into it were actually completed.

Other metrics focused on clustering the paths that opponents created chances through. One awkward thing about quantifying defending is that you're often trying to work out some kind of counterfactual: if a defender makes a defensive action, we have no real knowledge of what they stopped from happening.

The clustering technique seems a pretty innovative way of trying to get around this. It's kind of like constructing a plausible 'what if', in a different way to an expected possession value model.

In the results, different defenders tended to do well at different metrics, again highlighting the worth in splitting things up into different components.

Getting out of a counterpress

Pressing Times: Can data tell us when and how to navigate out of a counter press? — Gerald Lim, Ashley See, Zhi Yua Chua

Summary: Quantified counter-pressing strategies through the application of different metrics, the compute the successes of various decisions taken against the counterpress based on key outcomes like ball progression, retention time, and the threat conceded from losing possession.

Commentary: This was another presentation where thinking about the problem smartly, and not being afraid to spend time breaking it down, worked wonders.

The researchers used a pitch control model to quantify the intensity of the counterpress, but they had a few different methods for it. There are different kinds of pressing strategies — ball-oriented, player-oriented, passing lane-oriented — and each of them got their own approach and metric. This alone has potential value in analysis.

They then looked into different methods that a team could use to get around a counterpress: clearances, long passes, short passing, that sort of thing.

There was a point, almost made in passing, that I thought was interesting about 'scouting' for the national team. I know that people have talked before about players from the same club side making for good national teammates, but not every nation is turn-of-the-2010-decade Spain. The point the researchers made here was that you could use these metrics to look for players who played similar counterpressing styles for their club side.

As I said at the start of this, the best way to understand the Pro Forum is as a kind of advanced brainstorming meeting. This was a particularly good one.

If you've enjoyed this newsletter, and the work that Get Goalside has done over the last three years, you can now become a paid supporter. There's more information here, or you can hit the button below.

Upgrade to paid

Get Goalside 100

Mon, 21 Mar 2022 00:00:00 GMT

If the 'posts' figure in my newsletter dashboard is to be believed (and discounting a couple of subscriber survey requests) this is issue 100 of the Get Goalside football analytics newsletter. It started just over three years ago, in early March 2019, and over 1400 subscribers later, here we are.

It's been interesting to think back to that time and how much football analytics has changed since then.

When Get Goalside #1 dropped, we were still two months away from Liverpool's research department getting their shiny feature in the New York Times magazine. The general footballing public were less skeptical of data than they had been, but it wasn't yet exciting to them. A lot of people were putting out possession value models.

There's a lot of things I've learned in these past few years, but I wanted to open this issue out to more than just my perspective. I asked a bunch of people what was something that stood out to them that they'd learned over the last few years, and whether there was something in particular they'd want to learn over the next few.

I'll scatter some brief thoughts throughout, but for now I'll leave you in the expert company of: Javier Fernández, Lydia Vandenbergh Jackson, Jan Van Haaren, Sam Gregory, Arielle Dror, Joris Bekkers, David Sumpter, and Devin Pleuler. My thanks to each of them.

Javier Fernández

Senior Data Scientist at Zelus Analytics, currently focused on the development of cutting-edge performance metrics for football and basketball. Formerly, head of sports analytics of FC Barcelona where he built the club’s top-level analytics department from scratch. Javier holds a MSc and a PhD in Artificial Intelligence, both centered on football analytics.

xG has made it to the clubs, the press, and even the TV. It is not strange anymore to find an opening for a data scientist at successful clubs, many of those not really knowing what they need, but sensing they have to be part of it. However, it is very easy and tempting for the data analyst to settle right now. We are at the risk of creating and immersing the sport in a new and always invisible bubble, the xG bubble.

We are right now in the most exciting era for football analytics. Both the current football analysts and the bright analysts-to-be have the opportunity to make an immense impact in the sport they love. However, for doing that, we have to make a great renovation in three fundamental aspects of analytics. We need a shift in mindset, a shift in focus, and a shift in resources.

Mindset: In this new era, the football data scientist needs to gain confidence. You need to feel that you can sit at the same table as the sporting director, the head coach, the scouts, or the players. And the reason is simple, you have precious things to add to the game. So we have to embrace this and work hard to demonstrate it. It will pay off.

Focus: We can't settle with aggregating on-ball stats or reducing everything to xG shooting metrics. Don't settle with the obvious. Keep curious and focused about what should be the most important: "understanding the game better." At the same time, we need to identify where we can add the most significant value. There is tons of value to add in bringing the best players and getting to the best players (yes, players love any piece of information that can make them 1% better). The head coach is probably not the one you will help the most.

Resources: There are hundreds of smart people trying to make the sport better by sharing their work and ideas in blogs, newsletters, Twitter accounts, public posts. Resources are better than ever but are still scarce. This sport has an incredible opportunity of becoming even more popular and even more enjoyable. Organizations will benefit immensely if they share more data; analysts need to prepare more and better use the data. We will all grow and enjoy more. xG is great. But football is not simple. Don't settle.

Lydia Vandenbergh Jackson

Former professional soccer player and now Analytics Engineer for Zelus Analytics

As a former professional soccer player and collegiate coach, I’ve realized how much my experiences have shaped how I view the game and rate players. An amazing aspect of data analytics is that it isn’t (typically) as biased as I am! We can find amazing insights and make better decisions when using data. Growing up, soccer was always about goals and assists but now with advanced metrics and predictive models, we can evaluate all players on the field more effectively.

Now that we have all of this data, we have to translate it so that key decision-makers understand it. It has to be simple and presented in a way that makes sense to how coaches/players think. Trust is key here. It should be a two-way street with the opportunity to provide feedback. Feelings and intuition are important. Being able to relate to certain moments or scenarios due to my experiences helps me excel in my current role because we have to look at every problem or question from a holistic perspective and take into account all the things we can and can’t measure.

Mark: Something I took from both Javier and Lydia's responses are how they talk about communication. When I was starting out in the analytics space, the focus seemed more about 'these are ways you can win over the coach'. This might have made sense at the time, but feels like it's focus on a narrow goal framed communication as something quite transactional.

The word 'trust' stuck out to me in particular: coaches and players need to trust that you know what you're talking about, but also, considering how new analytics is, I think that they need to trust that they can safely be inexpert about statistics around you too.

Jan Van Haaren

Jan is a Data Scientist at Club Brugge, where he is involved with recruitment analysis, opposition analysis and performance analysis.

Lessons learned

The football analytics community is extremely scattered and disconnected. Fanalysts, football clubs, analytics companies and academics are operating in the same space, but they have different interests. The fanalysts often want to land a job with a club or company, clubs want to outsmart their competitors and win trophies, companies want to earn money, and academics want to publish papers. However, fanalysts, football clubs, analytics companies and academics often do not know about each other's ideas. Fanalysts use social media and blogs, academics publish papers, companies sell commercial products that often lack transparency, and football clubs tend to be very secretive. As a result, people keep reinventing the wheel, which hampers the progress of the field as a whole.

More generally, football analytics is becoming more and more of an engineering discipline. The complexity of the data, models and metrics is ever-increasing. The early datasets (e.g., matchsheet data and basic event data) were easy to work with and the early approaches (e.g., expected goals) were quite straightforward to replicate, but those times are pretty much gone. Working with contextualized event data (e.g., StatsBomb 360 Data) and tracking data, and leveraging more sophisticated approaches (e.g., pitch control, expected possession value) requires more advanced and more diverse skills.

Wishes for the future

I'm hoping for a "better-integrated" football analytics community with tighter relationships and more exchanges of ideas among the fanalysts, football clubs, analytics companies and academics. I understand that each of those groups has to protect their own interests, but I'm confident that more interaction should be possible regardless. The companies and academics would better understand what challenges the clubs are facing, while the clubs would more easily find appropriate solutions to those challenges.

I'm also hoping that the evaluation and validation of analytics methods and products will get more attention. I have the impression that many football organisations (e.g., football clubs, football associations, agencies) are clueless as to what methods or products to use for, for instance, performance analysis, opposition analysis or recruitment analysis. What metric or product is most appropriate or useful to solve a particular task in a given situation? When would you use metric X, product Y or device Z? We could, for example, design a number of benchmark problems that are inspired by real-world use cases inside football organisations. New methods and products would be tested against those benchmark problems to inform the practitioners.

Sam Gregory

Sam Gregory has been working in the football analytics industry for more than five years and is currently the Director of Analytics at Inter Miami and a graduate student at Victoria University studying the intersections of sports science and analytics.

Over the past few years I think the areas I've personally learned the most in have been in some of the less-sexy parts of data science: scalability, data storage, engineering and data pipelines. As data becomes more and more accepted in football the non-technical people in the sport need less hand-holding and want to explore data themselves - this means instead of having to sit down with someone and deliver a custom report every time people want tools, dashboards, webapps and automated reports that they can interrogate themselves. This requires lots of backend work to make sure that users can access the output of your work without constantly asking you for data or visualizations. This is especially true at a club where you will have "users" across various departments in the club (recruitment, performance analysis, sport science) all of whom have different questions. If I were to recommend one area for incoming analytics hires in a club to upskill in it would be database and data pipeline management.

Looking forwards, as I've been saying for the last few years I think collecting data from TV broadcasts (i.e. broadcast tracking) is really the next big thing in sports analytics. There are now multiple companies doing this work so I hope in the next few years I - along with the rest of the analytics community - will have figured out better ways to get the most out of this data and all of the additional challenges it presents (occluded players, predicting off-screen loads, massive data storage etc.).

Mark: I will butt in and take a victory lap here for saying in my start-of-year piece: "2022 will be the year that [an emphasis on data engineering] starts to get really drilled into the public analytics consciousness."

Also, as someone with an analytics newsletter, I too am very much in support of Jan's wish for a better-integrated analytics community.

Arielle Dror

Arielle is a Data Scientist at Zelus Analytics and occasionally contributes NWSL pieces/visualization for American Soccer Analysis

It’s probably a cop out to say that I’ve learned everything in the past couple years since I’m relatively new to this world. Since I started learning about football analytics, though, I think my biggest consistent lesson is that communication skills are perhaps more important than any technical ability you might have. At the end of the day, you need to be able to translate a model or analysis into terms or concepts that folks on the other side of your screen (or across the room) can easily understand.

Sometimes, that might mean opting for a less predictive model or a less intricate visualization — barcharts and scatterplots are great for a reason, you don’t need to reinvent the wheel— but it doesn’t mean that you’re less talented, it just means you’re doing your job well. In a lot of ways that’s not too different than analytics in other fields, but I think that it might be even more important in football analytics since the concepts are so new to a lot of people who are looking at your work.

In the next couple years: From afar, it feels like analytics in the women’s game has recently advanced quite a bit, but lags behind what’s available on the men’s side. I don’t think everything we know from public analytics can necessarily translate directly to women’s football right now. For example, I want to understand the biases that might arise when using models based upon what we know about the men’s game. Do we need to adjust models? And how? Do we need to contextualize the numbers we see differently? I don’t entirely know the answer, but I’m looking forward to learning as the field learns, too.

Joris Bekkers

Joris Bekkers is a Sports Data Analytics Research and Engineering Consultant who has been working with the U.S. Soccer Federation since 2018.

One thing I’ve learned over the past few years is that, especially in smaller teams, you can’t just be a data scientist, a data engineer, or a data analyst. You need to learn to be all these things. You need to learn to be a well-rounded developer that also has knowledge of the football-side, a strong grasp on at least one programming language, know how to maintain databases, create well balanced, visually interesting and information dense data visualizations, keep up with the latest in football analytics research, and learn how to combine all of this to build automated data pipelines.

And for me, the most difficult part of all of this is to learn to take some time out of the week to develop or learn new skills, because it’s easy to get carried away by all the work you have to do, but you need to take a step back sometimes and think about ways to be smarter about the work you do.

In a few years from now I hope to be able to say that I’ve successfully integrated Sport Science and Analytics, such that strategic (on-field) decisions and physical workload are almost inseparable, and tactical tradeoffs are made between energy expenditure and on, and off-ball gains.

Mark: I've grouped these two responses together because of the things that Arielle and Joris mention as wanting to learn in the coming years, because they match very closely to mine.

'Is a women's football xG model different to a men's football xG model?' and 'why are analytics and sports science separate spheres?' are two Get Goalside issues I've wanted to do for a while but haven't been able to do (yet).

(On the former, there's an interesting post from Lotte Bransen and Jesse Davis of KU Leuven here)

Also, Arielle's line about sometimes opting to use less intricate methods to communicate better - "but it doesn’t mean that you’re less talented, it just means you’re doing your job well" - is a great one.

David Sumpter

Professor of applied mathematics in Uppsala; author of Soccermatics and other books; and co-founder of Twelve football

I was most struck by a reminder by Jon Mackenzie that I said five years ago that clubs needed to invest in understanding the basics of football using tracking data. And really that still hasn't happened. There is some progress at Liverpool and Barcelona, and they are on their way at Manchester City and Leipzig. But there still isn't the ground work done that is needed. That doesn't mean there aren't good data scientists at clubs. There definitely are. We talked to many of them on Friends of Tracking. It is just that when, for example, I studied fish or pigeons or even locusts, we were a team of 5 or 6 data scientists and experimentalists (coaches, in footballing terms) working together. I still haven't seen that happen.

In the meantime, in my own research we are making some steps in that direction ... often working together with those data scientists at clubs. But my work is done at a more leisurely pace just now. But if I was in charge of a big footballing club I would say, lets take some of our budget and let's set up a project to truly understand the movements in football using data and then use it as a co-ordinated way to do scouting, tactical analysis, everything... The potential edge for this first team to do this would be substantial.

Devin Pleuler

Director of Analytics – Toronto FC

The most unexpected learning was just how difficult and time-consuming it would be to extract value from raw tracking data. Before full player tracking became ubiquitous across analytics departments, there was a naive optimism that it would serve as some magical panacea for unlocking the intricacies of the game invisible at the event-data scope. But instead, we’re still left waiting for decent defensive metrics.

The reasons for this are not entirely clear. What is for certain is that friction arises between the sophisticated techniques required to work with tracking data and the language found in the sporting theatre. Pitch control and its various flavours come the closest, but it remains non-trivial to implement.

Instead, teams have mostly resorted to utilizing tracking frames to enrich individual event datum while retreating into the safe statistical corner of rates and frequencies. This is a genuine improvement over where the industry stood, but things obviously haven’t gone exactly how we promised.

Mark: Last but certainly not least, two quite focused responses on working with tracking data.

The themes of a lot of other peoples' responses mingle in with this too. The reason why several people touched on data engineering, for example, is because departments are quite small, which David would like to see change.

I also think there's a link between where David says "let's set up a project to truly understand the movements in football using data" and Devin says "there was a naive optimism that [tracking data] would serve as some magical panacea for unlocking the intricacies of the game invisible at the event-data scope. But instead, we’re still left waiting for decent defensive metrics."

Maybe one reason why the impact of tracking data has been limited is because so few clubs have invested in research departments/projects as David describes them. (And, as Jan says, clubs tend to be secretive)

Thanks again to each of the contributors.

For my part, I think the last few years have taught me a lot about the day-to-day of football, and the extent to which good and sensible tech can speed things up (influenced somewhat by the fact this is something that Twenty3, my day job, excels at).

That's not to say that tech tools inherently matter more than analysis or exploration, but it's more about what is delivered. Analysts at clubs might have been using a blunt knife to chop the vegetables before; with a tech tool you can bring them a decent knife from the shop; and that's probably more use to them than the hand-sharpened artisan one that takes four months to be crafted and is probably too expensive.

(There's also probably some recency bias here, I'm sure there are things I've learned and forgotten that I once needed to learn it).

Looking ahead to the next 100 newsletters, the thing I most want to learn is how to take insights from things like pitch control models and make them actionable. What specific things do you draw out? Can you use it to create tactical plans? If so, how much are you reliant on coaching quality to implement them? Could the analysis of the data even help improve the coaches' coaching?

Can we get those artisan knives to the analysts?

Fingers crossed.

Notes

Acknowledgements

Lastly, there are some people who deserve a bit of thanks, without whom Get Goalside wouldn't have got this far, or probably existed at all. They fall broadly into three groups.

The first are those who've encouraged, cajoled, or nudged me at some point of my time in analytics: David Perdomo Meza, Bobby Gardiner, @TiotalFootball, Mladen Sormaz. The second are people whose work has inspired me in various ways: Mohamed Mohamed, Tom Worville, Joe Mulsberry, Vosse de Boode, Karun Singh, Grace Robertson, John Muller. James Yorke and Thom Lawrence fit firmly into both camps. Apologies to others who I've missed.

And the final group is you, dear reader. Particularly if you're an email subscriber. Thank you very much; for reading this, and for reading however much of Get Goalside that you've read in the past. I hope you enjoy what's to come in the future.

When will tracking data have its xG moment?

Thu, 10 Mar 2022 00:00:00 GMT

Five years ago, I wandered into my first Stats Perform Pro Forum.[1] The Pro Forum, if you're not familiar with it, is like Davos for people who read Opta data feed specifications instead of public policy documents. A swarm of analysts, and increasing numbers of 'Head of Analytics'-es, get invited by the data provider to see talks that push the boundaries of what we do with data.

The 2017 Forum was, I think, the start of tracking data being The Thing People Talked About. The buzz around it was like the buzz around NFTs now, only less riddled with scams. The hope was that it would change the game for analytics, especially around defending.

Now, I wouldn't say that tracking data hasn't had an impact in the five years since, but nor has it produced an impact like expected goals. There hasn't been a widespread, noticeable change in the way the game is played that can be attributed to it. So, why is that? Let us count the ways.

1. Secrecy

The most obvious and most frustrating one. Expected goals as we know it was more or less created in public, by bloggers and loudmouth tweeters. The point at which long-range shots started falling away is also, coincidentally, about the point the shouting about xG started (around 2014 and 2015).

Above chart is from a recent piece on The Athletic about the change to shot locations in the Premier League.

Since then, smart people have been snapped up by clubs or assorted NBOs (Non-Balling Organisations) like data providers, agencies, or software companies. Work happens behind closed doors, save for the occasional Sloan conference paper (usually involving people from Barcelona, with their Innovation Hub, or Swedish team Hammarby, with their David Sumpter).

This is obviously particularly annoying for people publicly writing about football analytics, but the thing with xG was that every team could put it into action. You didn't need an analytics research department to read a blog. You just needed to know where the blog was and think it was worth paying attention to.[2]

2. Football is hard

It's really easy to tell whether one type of shot is better than another. But when you get further than, say, 25 yards from goal everything just kind of merges into a shrug. I like referring to Thom Lawrence's phrase when thinking about this, 'the Trough of Meh'.[3]

The value that came from knocking together an expected goals model was simple and relatively big. 'Scoring from long-range shots is way less likely than you think, and gives the ball to the opponent - do it a little less'; 'this player who scored a lot of goals was on a freaky hot streak, think hard about signing them'.

Early days of analytics also highlighted the value of through-balls and cut-backs, things which everyone probably already knew were valuable, but didn't seem to train specifically for as happens now. (Although this is more of a Pep Guardiola influence than an analytics influence).

The public work on tracking data has often been about things like pitch control models (for more on those, read this), which seem to naturally lean towards learning about build-up phases of play. They're simply harder to be punchy about than shooting shots.

3. The technical problems

Long-time analyticsers can probably skip this section, but it's worth repeating. In regular event data each action that happens on the pitch (shots, tackles, passes) are a single row, like in a spreadsheet. In tracking data, a single row is usually around one twenty-fifth of a second of action. A five-pass move taking five seconds will take up five rows of event data; it'd take 100-125 rows of tracking data.

This mass of data raises the barrier to entry, unless someone has created a product to do the hard work for you. You can't do much with it as a beginner coder; your database infrastructure will need to be better; your computer modelling needs to seriously take into account processing speed. You probably need to think about physics.

Data is also less available. There are some free datasets out there, like Metrica Sports' but you can't systematically scrape entire seasons' worth of data in an evening like it would be possible (or so I've heard) to do with event data.[4]

It's all very well having some games of data, but it really limits what you can make out of it. This 2017 article on how players outperform their xG (or don't) by Marek Kwiatkowski had a sample of nearly 200,000 shots and still had big margin-of-error bars.

4. All factors are important

With higher barriers to entry for a public mass of amateur analysts, less data to work with, and therefore the bulk of the work happening behind closed doors, it's no surprise that nothing's gone boom.

But maybe I'm wrong. Maybe there has been a comparable explosion in understanding of the game, that's been applied to tactical plans, and it's the first factor, the secrecy, that's keeping it from public knowledge.

The number of people who've worked with tracking data is growing every year too. And, even amongst the shrouds of secrecy, there are an increasing number of teams of people working with it. There are the stalwarts, at Liverpool and Barcelona's Innovation Hub. Manchester City/City Football Group have expanded their operation (as written about previously). There are more tracking data providers than before. There are big consultancies like the Analyticses, Zelus and Blue Crow (also written about previously).

The other thing is that, even if the technical barrier isn't as high, the pool of people working on this stuff isn't necessarily smaller than when xG was created. Football analytics is far less niche now.

So maybe I've been asking the wrong question. Maybe the question isn't "when will tracking data have its xG moment"; maybe it's when will we hear about it.

Notes

Footnotes

[1] "...my first Stats Perform Pro Forum." || Back then Stats Perform was still Opta, but I've referred to it by its current name throughout for ease of writing.

[2] "You didn't need an analytics research department to read a blog. You just needed to know where the blog was and think it was worth paying attention to." || I think we're beginning to get to something like this with tracking data. This research paper from the aforementioned Barcelona and Hammarby demonstrates that pitch control models can be used to help players with their positioning choices. Projects like this tactics board app from Joris Bekkers, which includes a pitch control model function, could offer coaches the chance to use 'tracking data' without having tracking data.

You could re-create a situation from a game in an app like that, using the pitch control function to demonstrate how the balance can change if positions are changed by just a few yards. I'm not sure if this would lead to something as blatantly clear as the xG effect on long shots, but it'd be a way to spread the influence of tracking data.

[3] "...the Trough of Meh" || The other side of this trough is how risk increases sharply as you get close to your own goal. The midfield is a place of very little immediate risk and very little immediate reward. Even when looking at non-immediate risk and reward there's usually not much going on.

[4] "...you can't systematically scrape entire seasons' worth of data...like it would be possible...to do with event data" || You could try and create your own tracking data. This 2020 Sloan analytics conference paper from Neil Johnson gives a bit of a template (although I can't say I've tried it myself), but I doubt it would be a speedy way of getting your hands on it.

The Theory of Everything (in football)

Fri, 11 Feb 2022 00:00:00 GMT

There was a point in time where people didn't write books on story structure. Nobody knew what an 'act' was, or what pathos was, or that a punchline goes at the end of a list of three. They just told stories, and a lot of those stories will have been pretty terrible. The stuff that has survived to the present day, about mythical figures like Zeus and Hades and Margaret Thatcher, are just a synthesis of the best yarns.

But people started to work out that there were things that were pretty universal to good stories, certain themes and rhythms that audiences just loved. And that helped make the chance of you hearing a decent story much better.

There are times when I think that football and football analytics are still in that campfire phase of storytelling. We've hit on some great stuff but there's still a lot of groping around in hope. You can work out a lot by putting data through a linear regression model, but I want to try and come up with some theory to hang it on.

Part I: Attack vs Defence

Football is a simple sport. You want to score more goals than your opponent.

We can only score goals when we have the ball (barring freak own goals) and we can only concede when our opponents have the ball. This means we tend to think of things as 'attack' and 'defence', but another way of giving these things names could be 'goal scoring' and 'goal preventing'.

On the surface, both of these things are easy to measure: scoring goals means you're doing well at goal scoring; not conceding goals means you're doing well at goal preventing.

But we know that goal scoring is not just done by the players doing the shooting. People like Kevin de Bruyne and Fran Kirby are very important in their teams scoring goals, but they don't often do the scoring themselves. They're called chance creators. And sometimes they're involved in what's usually called build-up.[1] The 'goal scoring' part of football has several fairly distinct concepts.

This is the part where storytelling theory would probably say I should tell a joke to keep you engaged through this exposition-heavy first act.

Why did the chicken cross the road in 68% of simulations? I don't know, you'd have to ask the video analysts.

This is the point at which me spending time telling you things you already know becomes worthwhile. Because these divisions, these ways of splitting up 'goal scoring': there's no such division that people make for the 'goal preventing' side, not in the regular public discourse at least.[2] Isn't that a bit weird?

Defending is just 'defending'. We talk about pressing a bit, but that's a method rather than a theoretical unit, more of an equivalent of through-balls rather than the concept of chance-creating being a distinct part of 'goal scoring'.

So, let's split 'goal preventing' up. There may be better ways of doing it what I'm about to propose, but I'm going to put forward 'block preventing' and 'direct preventing'.

'Block preventing' is a method of goal preventing where teams try to keep their opponents at bay. It's partly done through not allowing good passing options, but also partly through instilling fear and caution. Teams don't pass in U-shapes because they think it's the best way of scoring, or even because they're under direct threat of losing the ball -- they do it because they fear that they might lose the ball if they pass more directly. This is what good 'block preventing' does.

Meanwhile, 'direct preventing' is the kind of 'goal preventing' that is, well, more direct. If everything goes to plan, direct preventing is where you win the ball back; block preventing is where the final whistle does.

Dividing goal prevention in this way subtly changes our idea of how to quantify defenders. Just like how a player can be a primarily a chance-creator instead of a shot-striker, maybe a defender isn't a direct preventer, but could be a good block preventer. Ultimately, both things are part of the same aim -- not conceding goals -- but they are roles tied to distinct aspects of the game.

Public discourse has kind of started to do this already, albeit without a theoretical framework behind it. It's now fairly common to note that a defender's high number of tackles and interceptions means they're 'busy', rather than saying it makes them 'good'. But maybe they are good. Strikers who take lots of shots are 'busy' too, but many of them are good.

The thing that we recognise by differentiating 'busy' from 'good' is that there are different facets to goal preventing. But maybe defining and theorising about these different facets will help lead us in useful directions...

Part II: Hey, it's not my fault

The thing about assessing forwards statistically is that it's really easy. They do, like, one thing. They shoot, and sometimes they score. Expected goals models help work through some of the noise of coming up against good goalkeepers or bad variance for a stretch of time. But ultimately they kick the ball and someone makes a note of that and everything is done and dusted.

Except sometimes, a striker makes a run and their teammate on the ball doesn't see it and doesn't play the pass. Or the cross is slightly overhit. Or some other bizarre event that would be a good punchline to a list of three.

We don't tend to worry about these instances too much because it seems that teammates spotting runs happens frequently enough to give a signal, but the principle is always in the background. Forwards -- or shot-strikers to hark back to the theory in Part I -- can (usually) only strike shots if the chance-creators find them. And chance-creators can often only chance-create if reasonable capacity has been built up around them.

Similarly, a lot of 'direct goal preventing' only happens if defending elsewhere on the pitch has failed. This might either be 'block preventing' failing, where the team's structure has been punctured by the opponent, or it might be that a central midfielder made a mess of a tackle attempt, and now a centre-back has to step out.

This is one thing that makes statistical analysis difficult. Is a defender making a very small amount of tackles because they're a poor direct preventer, or because the rest of their team is so good? Is a striker taking a lot of shots really fantastic, or are they just a slightly-above-average cog in a fantastic attacking machine?

These questions about opportunity aren't unique to defending, and may not even necessarily be more significant for goal preventing than goal scoring. It's just that we haven't found good stats yet, so we think about these problems more.

Part III: I contain multitudes

As I write this, a certain type of Marvel Cinematic Universe fan is annoyed that Spider-Man: No Way Home isn't in Academy Award contention. The merits, or lack thereof, of this Spidey-for-Oscars campaign aside, it's a topical reminder that there are different ways for films, or stories, to be 'good'.

Not only are there different ways for stories to be good, there are a ton of different methods that they go about achieving this. Now, in football what is 'good' is a little less subjective, but there are still some different methods of going about being it.

A player who has the quality of 'good at scoring goals' has a whole set of component skills that they possess, and each of those players has a different set of them. So will each 'good chance creator' player. Technique, perception, anticipation, mentality, fitness, agility -- no two forwards will be exactly the same as each other on all of them. And that's before we break down these skills down into sub-skills, like different techniques for different types of shots or passes.

Similarly, a good 'direct goal preventer' will have different skills. A good 'block goal preventer' will have different skills. They too will need to have good technique, perception, anticipation, mentality, fitness, agility. They too will have to be a good manipulator, or manager, of space.

When we talk about defending statistics, we're far more like to go to the specific skills. Like in Part II, where I talked about the issues of opportunity, the lack of a decent overarching metric has forced people to drill further down. But these questions exist for attackers too.

I say this because there's a temptation, when trying to statistically analyse defenders, to go down a rabbit hole that focuses in on individual skills. Because we (currently) have no overarching stat that flashes a big light saying 'good defender!', the temptation is to get more specific.

What I mean is, a tracking data model of how defenders manage space wouldn't be the equivalent of xG, it would be the equivalent of a tracking data-based model of how attackers exploit space. Either of those tracking data models would be incredibly useful, but also incredibly difficult to produce. Expected goals came quite easily, because of a quirk of both the sport and the data that was already being collected.

Splitting up 'defending' from a theoretical standpoint might help. There are different ways that a player can be good at 'direct preventing', but we probably have stats that fit those jobs (e.g. tackles, interceptions).

'Block preventing' is the harder part, and where we're more likely to try and find the different skills that makes a player a good block preventer. Anticipation, perception, agility, space-management, decision-making. If we had stats that seemed useful for assessing 'block preventing', very few people would be considering these.

Part IV: A New Hope

Where does this all leave us?

I've proposed thinking of football in two modes, 'goal scoring' and 'goal preventing'. This is a bit simplistic, and I'm aware of that, but as a starting point it's easy to understand. I proposed splitting 'goal scoring' into shot-striking, chance-creating, and capacity-building; and 'goal preventing' into block preventing and direct preventing.

I've talked about how shot-striking often involves chance-creating, and direct-preventing often involves a breakdown in block-preventing. Things are less linear than that, but it's worth keeping in mind that these types of dependencies happen on both sides of the ball.

And I've talked about how each of these theoretical units, like chance-creating or direct-preventing, can be achieved through different methods, and each of these methods can be split into individual skills.

I think that what I would do next with this is to think about what block-preventing and direct-preventing entail. I'd also look at existing statistics and consider what extent they are useful for assessing block-preventing and direct-preventing. I would also try to think about which players are doing which type of goal preventing while watching some matches. And, of course, wonder about whether this theoretical framework is correct or useful at all.

I may do this in a future post, but this one has run its course.

A man went to see his doctor.

"Doctor," he said, "I'm having terrible trouble. Life is cruel and confusing. I'm losing all hope. Football, a sport I loved, seems both so primitive to be without meaning and yet so impossibly intricate I'll never understand it. What am I to do?"

"Ah, treatment is simple," the doctor says, "There's this great newsletter I know, Get Goalside, that'll help you figure it out."

"But doctor," the man said, bursting into tears, "I just finished reading one!"

Notes

Footnotes

[1] "[...]what's usually called build up." || I'm not a huge fan of that term, I think it implies that if you do enough building you progress a level to chance creating zone, which I don't think is really true. A more accurate way of describing it is often something like 'we have the ball and we want to score but we also very much don't want to concede the ball here'. Friend of the newsletter Tiotal Football has talked about this part of football as building 'capacity' for an attack, which is a phrase I like too.

[2] "[...]there's no such division that people make for the 'goal preventing' side[...]" || I'm aware that coaches often talk about 'defence' and 'transition defence', as opposites to 'attack' and 'transition attack' -- that's useful for coaching but I don't think it's a hugely useful framework for building a theort about what football is, mainly because it leaves the non-transition aspects too large. This is also a good point to mention why I haven't talked about football as "in possession" vs "out of possession". Again, I think that's useful for coaching but not for football theory - your goal is not to have the ball or not, it's to score or prevent scoring; what you do in and out of possession is just means to those ends.

FIFA's big, big plans (no, not the World Cups)

Thu, 27 Jan 2022 00:00:00 GMT

A lot of people scoff at The Phantom Menace for having a plot revolving around trade policy.[1] But I am pretty sure that you, dear reader, will have strong opinions about the organisation of the international football calendar. If you can find scheduling something to get excited about, you can find galactic tax policy exciting too.

Anyway, this isn't a defence of the prequel trilogy. This is about why FIFA is more of an interesting place than you might think.

Let it first be said that if you want people to trust you, you should avoid massive corruption scandals. But the reason why FIFA keeps pushing their extra World Cups (of the club and national team varieties) is that they don't earn as much as you might think. Their annual report for 2020 predicted that they would reach a figure of $6.4bn revenue for the 2019-2022 cycle. UEFA brought in half that during the 2019/20 season alone.

Most of FIFA's money (83%, according to Sportico) comes from the men's World Cup. This money goes into, yes, the luxurious aspects of some of the administration, but also funding global football development.

It re-jigged the way it doled out this funding in 2016, making a uniform amount of money available to each member association for operational costs and further projects (with a slightly larger amount going to nations that don't have particularly large revenues).

But half of the annually allotted $1m comes with conditions.[2] They include a requirement for organising a women's competition that lasts for at least six months, with at least ten teams, consisting of at least 90 matches. A women's national team has to have played at least four matches that year. Boys' and girls' competitions in at least two age groups need to be organised. And a refereeing programme (that includes women referees) needs to exist.

I say all this not to suddenly pivot to being a pro-Gianni Infantino newsletter. Only today, the day I'm running a final edit on this piece, Infantino has put his foot in his mouth yet again. But I found that previous information out while researching this issue and thought it was interesting. And it's a nice way to remind people of the intended goals of FIFA before going into all of the other things they do. Things that are more immediately relevant to this newsletter...

Subscribe to demystify football in your inbox.

A number of years ago, FIFA set about making sure that the wearable tracking devices that players wear during games were safe to use. This was mainly just a rubber-stamping exercise, but in 2018 they widened their remit, checking the accuracy of these systems and of optical tracking data. Then last year they announced they'd be doing the same for tracking data generated from broadcast footage.

Before the middle of the decade, they'll have a whole range of data quality marks for certified systems. And it's not just for tracking systems; it's goal-line tech, it's offside lines, it's playing surfaces.[3] One of the latest projects is a lower-cost implementation of VAR. Makers of various technologies are able to put on their websites that they have a FIFA Quality mark.

It's not just badges of approval though. The body are working on standard formats for transferring tracking data and standards on collection and protection of player performance data (particularly noteworthy in the context of Project Red Card).

And you probably wouldn't be surprised, by now, if I were to write that it's not just that either, that there is one more thing that FIFA are doing that's worth writing home about. And that it, the latest area they're moving into, might be the most intriguing: data collection.

As last year's Arab Cup started (a FIFA-organised tournament, as of this edition; essentially prep for the 2022 World Cup), the body made a big announcement. Not only would they be collecting event data during the tournament, and supplying it to the national teams taking part, but this dataset would be huge.

“We have a team of 25 analysts working on each game," FIFA group project leader Chris Loxston said in the organisation's article [linked above] on the subject, "so we have one analyst covering one player for the entire period of the game."

For context, the standard set-up elsewhere in the industry is usually one collector per team.

"Typically, a football data set only looks at all the actions around the ball," Loxston continued, "what we are actually able to do here is look at all the actions off the ball and around the ball as well. An average football data set is somewhere around 2,000-2,500 events per game; we collect in excess of 15,000 data points."

This is huge. And very interesting.

Training data collectors is no easy task; companies like Opta and StatsBomb spend months training theirs up. But the advantage they get from this time and expense is a set of collectors who will work for them on match after match for (they hope) years. FIFA are spinning this operation up for individual tournaments at a time. (Although the fact that they have one collector per player may make each person's job much easier, and therefore the training and collection processes less intense).

Assigning an analyst per player also means that FIFA could potentially be giving (some form of) work experience to an awful lot of people at one time. Useful. The global governing body having their own 'public sector' dataset is also a neat way of sharing data amongst all national teams and associations (not just those that can afford 'private sector' event data). Also useful. And might it be a potential media opportunity in future (media of the world, wouldn't you want FIFA's own data for their big tournaments)?

Now, to collect data, you need to have an idea of what you're collecting. Which is why FIFA have developed their own 'football language', which "will be our blueprint for how we [FIFA] analyse football in the future".[4]

The language encompasses a lot more than the on-ball events that most event-datasets do. There's sections on types of movement to receive the ball and offering to receive, even if the player doesn't end up getting passed to. (And I think the site is a gold-standard that all data providers should aim towards in terms of communicating definitions to users, complete with videos).

FIFA's football language even makes a distinction between standing and sliding tackles. Finally.

The reason why I find this particularly interesting isn't just that Arsène Wenger has, single-handedly, addressed an issue I've had with football defensive data for years[5], but for what it could mean on a wider scale.

In England, when (almost) anyone talks about stats they talk through Opta's terminology. Tackles (rather than duels), Big Chances, key passes. These terms may be quite similar to how people talked about football pre-data, but to some people, if you talk about defensive duels or shot assists, they won't really know what you're talking about. Especially if you're talking about data.

From a brief look through FIFA's football language, I don't think it clashes with -- as in, seeks to re-define -- aspects of any existing event data provider's provision that I can think of. But given the broad scope, will future data providers feel pressure to align their definitions and terminology with FIFA's? Will we talk about 'units' of opposition being broken or bypassed, rather than lines? Will the ten types of movement to receive the ball that the global body have identified become standard definitions?

Asking those questions, and coming to the end of this edition of the newsletter, I realise that I may not have done FIFA many favours by opening this issue by mentioning The Phantom Menace. The start of a story where an elderly gentleman asserts his influence in an imperfect but generally well-meaning democratic body, before using it as a platform to become ruler of the whole galaxy? Arsène Wenger doesn't seem like that kinda guy... does he?

"Data collection leads to statistical modelling. Statistical modelling leads to expected goals. And expected goals... lead to the dark side." – Yoda, Attack of the Cloned Github Repositories, deleted scene

[1] "A lot of people scoff at The Phantom Menace for having a plot revolving around trade policy." || In fairness, the opening paragraph of the introductory text crawl, the first cinematic thing that Star Wars fans would have seen for decades, was: "Turmoil has engulfed the Galactic Republic. The taxation of trade routes to outlying star systems is in dispute."

[2] Fifa Forward programme conditions can be found here

[3] "And it's not just tracking data..." || If you want to see the full range of things FIFA do quality checks on, check this page out. It all started with footballs in 1996.

[4] "our blueprint for how we [FIFA] analyse football in the future" || Quote from Arsène Wenger himself, from here

[5] Me crediting Wenger with the standing vs slide tackle distinction is a joke; it being top of my Christmas list to Santa every year is not.

What's in the way of analytics solving football?

Thu, 13 Jan 2022 00:00:00 GMT

Sometimes newsletters are timeless, sometimes they age faster than the popularity of sea shanties. The intro to this one will make it a time capsule.

The nerds are ruining Wordle. Not just the tech sharks trying to monetise a game designed to be free, the coders algorithming up a guaranteed solution. They've solved it. AI solved chess. One day, will analytics do the same for football?

Well, no. I don't think so, at least. But why not?

A lot of people, including this newsletter, spend a lot of time thinking about what football analytics can do. The possibilities it can reach to. But I think that there are limits. Thinking about what those limits are might help work out where the current ceiling can be pushed further.

The problem is the ball

Let's start with why I'm so confident analytics won't solve football.

The first reason is simply that if 'solved it' is the summit of Mount Everest, we're a long way off even getting to base camp. Football is, I think, at least half a decade behind basketball in the use of tracking data, if not moreso. And football has double the number of players on the pitch. And even basketball isn't solved yet. When basketball is as 'solved' as chess is, maybe I'll start worrying a little.

A second reason is that football pitches are just so darned big. Really big. It's what gives rise to the Rafael Benitez's 'short blanket' analogy - you can pull the blanket up high, but then your feet will be cold; you can push your defence up high, but then you'll leave space in behind. Players may get faster, and goalkeepers more prone to sweeping, but it seems unlikely that football will ever get around its 'short blanket' problem completely. And that leaves a variety of strategies on the table.

My third reason, and possibly the most crucial, is (with apologies to Americans) literally in the name of the sport.

1) 'Foot' 2) 'ball'.

The ball bounces around in a hard to control manner. The feet are the appendage permitted to control it; notably hands, the parts of the human body that evolved as expert manipulators of objects and tools, are outlawed.

Even expert players misplace fairly simple passes, or take slightly sloppy touches, in many games. Add to that the fairly permissive officiating, and football is just a sport that, by its nature, makes it very difficult to keep control in key areas.

This lack of control also contributes to the very short amount of time that players have to make decisions. All of these factors make it seem unlikely that there will ever be a definitive set of tactics that should be followed to win football matches. Cycles of evolution will, I think, always occur.

However, analytics will help these evolutions take place. Just like inventions have sped up the social and intellectual evolution of humans, it seems fair to assume the same will happen in sport. I think that the most likely future is one where high-quality data analysis helps teams to find the holes in the game's current 'meta' (most effective available tactic) and identify how to exploit them. This analysis may only take months, or weeks, or even days.

But after that, it's down to coaches and players to implement it.

The problem is people

If we accept that analytics won't solve football, but might speed up tactical evolutions, we now need to think about what limit there'll be on this process.

Let's take a look at how 'playing out from the back' developed. Assuming that more money going into the sport inevitably leads to technical quality rising, football was probably always going to get to a point where defenders looked comfortable on the ball. Barcelona and Spain's possession-heavy dominance of the late 2000s-to-mid 2010s probably gave it a shove. And then the high-pressing trend which immediately followed gave an added urgency.

But footballers don't just become highly technically-adept overnight, even at the top level. We're only just, more than a decade after 'tiki-taka' entered the lexicon, getting to a point where top-division defenders are more likely to be comfortable than uncomfortable on the ball.

This, while an extreme example, helps us frame our thinking. If crossing suddenly became the way to create chances, how soon would it take the wingers and full-backs of European football, starting from a point where barely any of them have been brought up to prioritise the skill, to become experts? Eighteen months, perhaps? Maybe even longer? It doesn't need to be all of them, just enough to make a difference in the majority of teams, but let's not forget that players go entire careers without reliably being able to put a good ball in the box.

And edging slightly away from technical ability, think about centre-forwards. There's a complete dearth in them at the top level at the moment, to the extent that the options last summer for Manchester City, who surely had the pick of the world's talent, appeared to be: 1) Harry Kane 2) an ageing Cristiano Ronaldo 3) no-one. Not able to get number 1, and understandably not wanting number 2, they went for number 3.

If football's market of talent was an efficient one there'd be a spate of players who could fill that focal point striker role - but there isn't, because the market for talent is seeded by the trends of several years ago. Things are liable to go even slower when the gap in the market that needs filling is not only a technical and/or intellectual one, but one requiring a specific physical skillset as well.

I think that there's a chance that coaching approaches could shorten the physical 'time lag' as well as the technical one, but not by much. If a tactical strategy requires an insightful passer, you can train one from the entire pool of players at your disposal: tall or short, freckled or bespectacled. If a tactical strategy requires a tall and weighty striker, you're stuck with what God gave you. (In reality, a way that coaching could shorten the physical 'time lag' is by creating strategies to get around physical 'shortcomings', thereby widening the pool of potential players).

A final metaphor

I'm big on metaphors and analogies so here's one final one.

I don't think football will solve football and provide one tactic that every team will follow. What I do think could be a more realistic (though ultimately false) fear is that analytics makes it more like Formula One.

In F1, the driver is alongside the engine as a major feature. In the olden days of motor racing, they (as well as reliability) were probably the only two key features. But now, aerodynamics is a big thing. Barely a season seems to go by without it being big news that one of the top F1 teams added some random bit of carbon fibre to make air go round the car a little better.

If, in this metaphor, the engine is the players, and the driver is the manager, analytics teams are the... the aerodynamicists(?, fitting given that some of them are former physicists). The analytics team of the future might identify, and possibly even design, new features for a tactical plan that could streamline build-up, help cool down opposition attacks, and generally make things run much smoother.

(If we want to stretch it further, these random pieces of carbon fibre are not always strictly within the rules. In football, analytics types appear to be responsible for the increase (or at least the advocacy of it) of basketball pick-style blocks at set pieces, which are also not necessarily strictly withing the rules. Working out where the rules can be pushed to breaking point and where officiating can be exploited is still, tiresome though it may be to watch its results, analysis)

'Best aerodynamics' is not the be-all and end-all of what wins F1 teams championships though. And football, currently, is probably only just discovering what a wind tunnel is.

Analytics in 2022: what now?

Sat, 01 Jan 2022 00:00:00 GMT

It's a new year in the Gregorian calendar, the year 2022 of the common era. What a wonderful collection of almost-binary digits that is.

2021 can be captured in this excellent blog post of links from Jan Van Haaren (some posts from this newsletter are featured there, so it must be a good post).

But we are now into 2022. And here's what I think is in store.

Analytics settles on possession value

Possession value models are to everything else in a football match what xG is to shots. They quantify things that happen on the pitch. But quantifying anything and everything else is far harder than quantifying the value of shots.

As such, analytics is all over the place. Hell, we don't even have one name for the thing yet.

I think there are two factors that'll make 2022 a big year for possession value models. The first is just that it's another year that people have had to work on them and another year of conference papers. The second is that I think theory is gonna help and that more people are thinking about the theory.

Expected goals were easy. When players take a shot there's only one thing they're trying to do, and there's a very limited number of situations that they can come from. Phases of play only matter inasmuchas it affects pressure on the shooter and a clear sight of goal, things which some data sources are already including.

A year of data engineering

A (small) theme in 2021 was an emphasis on data engineering being important in the practice of analytics. 2022 will be the year that starts to get really drilled into the public analytics consciousness.

One of the reasons for this is that more people are getting more jobs in the professional club space. But another, I think, is that if any areas of the media are looking at The Athletic and considering doing data work, then they're also going to need some engineering to set the data up right.

Heck, it may just be a year of data-as-tech. Tech solutions are nothing new in football, and 'data/analytics companies' have been doing tech offerings for as long as they've been around. But I wonder if the exclusivity of tracking data (both in terms of availability and required expertise) means more people will focus on apps and software rather than the data science.

Analytics people get tired of analytics lip service

There was a time when managers saying 'expected goals' was enough to make analytics twitter very happy. But I think Christian Purslow's video on the Grealish sale at Aston Villa was a watershed moment in 'analytics' being publicly acceptable as a reason for decision-making. And not all decisions are good ones.

Purslow himself wasn't really putting forward a data argument for the signings that Villa made to replace Grealish. But his "we couldn't replace Grealish's talents with one player so we bought three players instead" is oh so close.

At the same time, the structures of football are very much being contested. I don't just mean FIFA's battles for a biennial World Cup. Multi-club structures are increasing in number, mainly involving English clubs as a central hub (as Rory Smith of the New York Times has written about), even as far down the pyramid as Fleetwood Town.

It strikes me that the types of people who would be interested in multi-club ownership structures would also be interested in heavily investing in analytics. And it strikes me that football even vaguely resembling an MLB-type farm system isn't something that most (English, at least) football fans will take to well. It'd just take one owner of a multi-club empire, who'd disregarded traditions just a little too much, to extol the virtues of data for things to get uncomfortable.

And then there's the flip-side of the plucky player using analytics to get a better contract: the bosses using data to stamp down calls for a raise. This seems like something that might happen behind closed doors and not see the light of day, but perhaps it might come out in a briefing war.

Get your "Are we the bad guys" memes ready.

Analytics becomes even mainstreamier

Sure, this is a 'more of the same' prediction. But with The Athletic keeping on going with their data stuff (we got a graph with a p-value in December), Jamie Carragher using xG timelines on Sky Sports, and covid and political polling driving data-use elsewhere, there's no slowing this down.

I won't turn this into a media analysis newsletter, but lemme just draw attention to the fact that The Athletic is more similar to Reach (the ownership group of football.london, Manchester Evening News, and others) than it is to their national paper The Mirror. And in the TV world, Sky wedded to football coverage in exactly the opposite way to BT (who have been rumoured to be selling off BT Sport all year). So it seems unlikely that anyone will rival these two outlets' data use.

But everyone everywhere will keep picking up data stuff. That's just the way the world is now.

Analytics mentoring schemes get set up for working class kids

It's been noted before in football, and other sports, that the 'demographic pool' of players is somewhat different to the demographic pool of the data analysts and scientists. There's no good reason why that should be the case. Given how close clubs are (supposed to be) to their local communities, that could be a useful avenue for helping address this balance.

The analytics sphere is also pretty largely cisgender male, and programmes specifically aimed at encouraging cis women, non-binary people, and trans people into football analytics would be good too.

Credit to Brandon Kent, who has run a couple of Measurables Office Hours programmes to connect professionals in the biz with people from underrepresented groups.

[Please get in touch with any programmes or groups which may be relevant here, and I will add to the online version of this piece]

Oh, and one final one. The Get Goalside analytics newsletter will keep going. Hopefully bigger and better than ever. But really, as long as it has your readership, it'll be doing just fine as it is.

Pick a sport, any sport, take ideas from it

Thu, 23 Dec 2021 00:00:00 GMT

We're into the festive period. Two years ago, I wrote the silliest things you could do with advanced football tech. If you weren't around back then I recommend it for a good mix of fun and insight into what kinds of data is around.

That's exactly what we're going to do again this year.

The subject line of this newsletter tells you what we're gonna do. From a variety of sports, I'm going to draw out what I reckon football analytics could learn. Some may be more light-hearted than others. Each sport's also going to have a 'novice disclaimer warning', where 5/5 means I barely know the sport exists, let alone its analytics landscape.

Anyway, let's go. If you celebrate, have a great Christmas.

NBA/basketball

Novice disclaimer warning: 3/5

Basketball is probably the primary sport that football analytics currently takes inspiration from. They're conceptually similar sports in that a team of players fights over possession of a ball and tries to score it in a goal at the other end of the field. Zonal and individual defence works in similar ways, and the pace seems more akin to football than ice hockey is. Oh, and crucially, through the NBA it's a rich sport with an analytics heritage (unlike field hockey).

Basketball made use of tracking data in big ways before football did, and if you go back to papers from Sloan analytics conference from, at a guess, 2013, you might still find basketball stuff that's ahead of the game in football. How does attacker movement attract defenders? How can repeated patterns be identified? And, lest I forget my friends at [redacted], how can we insert picks and screens into as many facets of the game as possible? (Has anyone tried running one of these on refs to block their view on potential borderline fouls?)

The good stuff to be mined from basketball is probably already being mined to be honest, so I'll try and pick something sideways here. Star players seem really good at getting shots in tight situations, with trademark step-backs and stuff to give them an extra metre's room to shoot. They presumable practice this all the time. Is this something applicable to forwards?

NFL/American football

Novice disclaimer warning: 3/5

There's probably a smart answer to this based on in-depth knowledge of current NFL analytics. This knowledge is not something I currently possess.

However. The dynamic of wide receivers vs cornerbacks or safeties seems an interesting one to look into, from the point of view of centre-backs vs attackers on long-ball counters. Maybe there could be something to learn from the defender's point of view about how to best hedge your bets on where a pass is going.

Tennis

Novice disclaimer warning: 1/5

An individual sport, played with racquets and a net between competitors: tennis could hardly be more different to football. But, once you get settled into a rally, tennis is just a game of angles and rhythm.

There are a lot of difficult-to-appreciate, difficult-to-quantify skills in football, but one of the most-difficult, I think, is tempo control. At least with most other skills you can see their direct effect. Anticipation might result in something like a higher rate of interceptions or shots from cut-backs. Space perception might be able to be quantified with pitch control models. But tempo?

The impact of tempo can be way down the line of a move. A proper grasp of tempo could help set the structure for an attack, or firm up a team to stop them losing control of things. But that, to my knowledge or imagination, isn't going to directly show up in a possession value model.

Tempo, though, is more of a focus in tennis than football. That sport might make more headway. And if they do, football can steal those insights.

(Also: the effects of equal prize money in helping to develop one of the few sports where men and women are held in equal stardom)

Sign up to get the Get Goalside analytics newsletter every other Thursday

Golf

Novice disclaimer warning: 4/5

Pitch maintenance.

Netball

Novice disclaimer warning: 4/5

Netball's a fascinating sport in that movement is inherent to it but the player in possession is (apart from one pivoting step) banned from it. From the little I know the sport can be fluid, but, because of its rules, has a sort of hampered motion baked into it.

If no passing options are immediately open to the ball-holder, I imagine that it can be pretty easy for things to get static. Think of a throw-in in football and how, after some initial movement, the taker will either be visibly searching for options or chuck it long in hope. Netball is like a series of 360-degree throw-ins.

While throw-in coaching is beginning to be adopted in football, there's probably some movement schemes that could be inspired by watching some netball. But I also reckon that open-play might get some inspiration too.

Formula One

Novice disclaimer warning: 3/5

If my 'novice disclaimer warning' was solely about the sport, it'd be more of a 2/5, but I simply know nothing about how Formula One teams use their data. However, I'm going to take a stab in the dark and say they have some incredible data warehousing infrastructure work going on. You can't do good data work if the warehouse is made out of the digital equivalent of cardboard.

I'm also intrigued about how drivers deliver feedback to their team, and the interaction between them and their engineers. Could there be a future where players relay something to their bench in a break of play, an analyst checks some data, and insights can be passed to coaches and back to players?

(This actually goes back to what I said about Julian Nagelsmann's earpiece idea)

That idea isn't necessarily one about a use of analytics, but more about the relationships between parties. Drivers are both the competitors, but also seem to be a kind of qualitative data source, sitting alongside (literally) all of the sensors in the cars. They're presumably educated, during their formative careers, in the ways that they can give useful information back to engineers. Maybe football players can be used in the same way.

Cycling

Novice disclaimer warning: 3/5

[redacted for legal reasons]

Athletics

Novice disclaimer warning: 4/5

[redacted for legal re-], nahhh, just kidding.

The thing with athletics (and cycling to a degree) is that it's about pure body mechanics. Strategy too, yes, but no formations, no space creation as such.

The field of kinematics is so foreign to me that I'm not 100% sure that's the right term, but there's got to be an absolute wealth of information and techniques that could be ported over from athletics to football.

For more on this, read this piece on Danny Ings from 2019 by Carl Anka at The Athletic.

Do we know football well enough to have good defensive stats?

Thu, 02 Dec 2021 00:00:00 GMT

Some problems

There's a fact I've always had trouble fully believing. The one about the ancient Greeks not having a word for the colour blue.

Ancient Greek texts had a curious absence of references to the colour. Sky was even referred to as 'wine-coloured'. When this was first noted, it was assumed that they just couldn't see blue. As it turns out, they could (their eyes are the same as our eyes), they just didn't seem to have called it anything yet.

However, there are also all sorts of weird studies that try and pick apart whether we perceive colours differently depending on the words we have for colour groups (usually blue and green). One involves the Himba tribe of Namibia, whose primary colour groupings group blues and some greens together while delineating more clearly between different types of green. They could, apparently, tell subtle differences in green shades quicker than the difference between blue and green.[1]

How much does our language interact with our experience of what we see? Seth Partnow's new book, The Midrange Theory, had me thinking about this.

Partnow, the former Milwaukee Bucks' Director of Basketball Research, writes in part of the book about how the terminology of basketball stats might be different if they were all created now. For example, shots "can and probably should be divided into two broad categories: attempts at the rim in one group, and jumpers in another."[1]

He continues by making the point that the most accurate 'shooters' (people whose 'shots' most often go in the basket) are those who take shots close to the rim (dunks and lay-ups), even though that's not quite what people picture when they think of basketball 'shooting'. As he puts it: "[...] as measured on the stat sheet, they are "great shooters" even though on the scouting report they are listed as "non-shooters"."

The same principles, I think, apply to football. (Of course I think this, I've written several times before about language and terminology in football (1, 2, 3, [3]))

But what Partnow was saying became all the more relevant to me this week because I was thinking about clearance statistics. Specifically, why the good people of the football stats internet have not risen up in revolt about them yet.

I've long written off clearances as a useful stat. Why?; overwhelmingly they come from heading or booting away opposition crosses. And therefore the most it tells you is how much a team or defender had to defend against balls into the box.

But not all clearances are like this. Sometimes a player could make a different choice. On rare occasions, when they're under no pressure or the cross is weak, this can be when defending a ball into the box. Slightly more often, the alternatives to a clearance are in situations when a player heads away a long ball or feels pressed into a corner near the sideline.

And I'm not just having a go at data providers for the sake of it (for various reasons, it would be shooting myself in the foot to do so), this information could have real value. How many times does a player clear the ball when they didn't have to? Does this mean they're a panicky player? Does this indicate something about team styles? If I had those pieces of info, they'd both be going straight into my pre-match reports.

I don't have full documentation of every data providers' offering to hand[4], but I don't think they have on offer what I'd want.

And while I've always kind of accepted that as the way things are, maybe I shouldn't have been.

Some suggestions

Back in the day, when tracking data still seemed like a dream full of possibilities to the public analytics community, it was assumed that it could solve 'the defensive issue'. It may well do, eventually, but there's still room for better defensive data in the manually-collected event feeds too.

I wrote last week about how pressure data is popping up in more places since StatsBomb started collecting it in 2018. Wyscout and StatsBomb also both appear to have[5] a separate category for tackles which try to stop players who are dribbling at them. This implies that all their other defensive duel/tackle events are initiated by the defender. Useful information! Wyscout also have an 'Anticipation' tag which they attach to some defensive duel events for when a defender nicks the ball off an opponent's toe. This separation from other types of duels and interceptions strikes me as very sensible.[6]

There will, presumably, be some other neat combinations that are possible by diving into providers' full event data and event qualifiers sets. However, that's not where the thinking about the sport takes place. Constructing meaningful basic statistics from the bucket of tags is like Apollo 13 making a DIY air filter. You can do it, you might learn something from doing it, but damn if it wouldn't be simpler and more helpful not to have to.

And while all this tinkering with tags is nice, it's also a little small-fry.

As Partnow says about shooting in basketball, if we were starting from scratch I don't think we'd have the statistical landscape for defending in football that we have now.[7] In fact, as he also teases in a footnote, our lack of conceptual understanding of defending probably holds back how we choose to collect data on it.

If defending is all about space, why are the defensive statistics so much about how a player affects the ball? Maybe we could be counting 'possible passing lanes covered' or something. You'd have some positional normalising to do when analysing those figures (midfielders would be called upon to cover a possible passing lane more than central defenders), but it could be useful, no?

The elite defensive players, the N'Golo Kantés of this world, maybe they'd show up with sparkling numbers because they're able to switch from one possible passing lane to another quicker than other players. It's not that they're quicker or cover more ground, it's that they see where the passage of play is heading and can move sooner.

Maybe you could track something around the speed that a player gets out to press someone and then recover back into position. Or, if it's simpler, when they're too slow to get out or too slow to get back in. The data collecting world is your oyster.

These might not be good suggestions. I don't know if the numbers would be genuinely useful for a start, and the descriptions of them are a bit unwieldy. Speaking of practicalities, the collection process might be unwieldy too.

Annoyingly, if Partnow's suggestion that a lack of language of defensive qualities leads to lack of useful statistics, we might have to wait until football as a whole gets better at judging defenders before we get good stats on it. I hope we don't.

But at the moment, we are like the ancient Greeks, lacking a word for the colour blue. It remains to be seen whether that's because we don't see the world that way or just because we haven't gotten around to naming it yet.[8]

Tell a friend, tell a colleague, subscribe to Get Goalside. Oh, and buy a copy of The Midrange Theory

Notes

Footnotes

[1] Given that this is a single study and I'm vaguely aware of how many 'pop science'-fodder studies get heavily revisited afterwards (marshmallow anyone?), I feel like I should say that I wouldn't be surprised if these findings also got revised at some point. But who knows.

[2] The Midrange Theory, ch. 2.

[3] Those three links not including a lost-on-an-old-blog-but-stored-in-my-files-somewhere piece where I watched lots of examples of Opta 'interceptions' to see whether they all fit the concept I had in mind. They didn't, quite.

[4] Don't have access to full documentation yet. fingers crossed.

[5] I say 'appear to have' because I'd prefer to check my assumption, comparing data to video, to be sure that my idea of what is being collected matches the image of situations I have in my head. Similar sentiment applies to the 'Anticipation' tag Wyscout have that I mention in the next sentence in the body of the newsletter.

[6] I wonder if part of the reason why 'public analytics' sometimes feels stunted is that the data available isn't even everything in the data providers' events feeds. There isn't even good delineation between different types of long passes. StatsBomb's free data is admirable, and is a fantastic tool to help people learn to get to grips with data, but the datasets -- WSL aside, either cup competitions or full seasons for one team/player -- tend not to be great for getting a proper sense of new metrics. I previously tried looking at how you might approach playing the Arsenal Invincibles, and I think that piece stands up well, but I don't know how much I'd be able to apply things in there more broadly. Building models off that data would certainly be awkward.

[7] bloody duels

[8] The circle of this metaphor is only, only just over my threshold of "yeah, that works" to stick with it. But I'm really baffled by whether people see the colour blue or not, so all that about the Greeks has to stay in now.

So... everyone has pressure data now

Thu, 25 Nov 2021 00:00:00 GMT

They say that imitation is the sincerest form of flattery.

Take the phone market: once upon a time you could get brick phones and flip phones and slide-y phones and Blackberrys. And now, almost certainly thanks to the success of the iPhone, every mobile on the market is a five-to-eight-inch slab of metal and glass, with three cameras and zero headphone jacks.

They say that imitation is the sincerest form of flattery.

On 9 May 2018, the former analytics blog StatsBomb went from consultancy to data provider, giving out USB sticks of their data at a launch event in London (which I promptly lost). There were two nifty features of their new data provision that was worth making a fuss about.

The first was 'freeze frame' information for shots. Any time a player took a pop at goal, StatsBomb would provide the location of where every other person on the pitch was (providing they were on camera). No longer would xG models be 'naive' to the position of defenders. Now they would know. They would see the truth.

The same day -- in fact, a little before the StatsBomb event was due to start -- Opta (who'd later merge with STATS to form Stats Perform) dropped a tweet.

[image used in case tweet gets deleted in future, accessed November 2021]

The next day, Opta posted a blog explaining two new qualifiers (tags which they associate with events in their data to provide more information): shot clarity and shot pressure. Two data points, it was noted by observers at the time, which would be possible to get from freeze frames.

Now, these developments of the two datasets by the two companies may have happened purely independently of each other (setting up, or adapting, a data collection service isn't an easy thing to do). And hey, they may even have coincidentally chosen the same date to make their announcements (it was the final week of the Premier League season). But, well, you know what they say...

The second most sincere form of flattery is signing up to newsletters

I thank you for sticking with me through those previous 290 words of analytics history. It's relevant, though, because of the second big feature of StatsBomb's dataset that was launched that day in 2018.

It was 'pressure events', and it would massively increase the number of actions that could be associated with defensive players. At the time, the big player in the football data market was Opta, and their primary defensive actions were tackles and interceptions. The most high-volume players might average six or seven per game. With pressures, the highest-volume players might average four times that. That's useful data.

With the bombast of a start-up, StatsBomb talked about how their data was going to change the game. And the bombast was deserved. Because it did.

Because, nowadays, everyone seems to have pressure data.

Wyscout -- primarily known for their video services but also a data provider in their own right -- have had 'pressing duels' in their data set since 2019.

More recently, the STATS-Opta merger offspring Stats Perform are in on the act. Stats Perform's shop-window content wing, The Analyst, is producing a lot of interesting stuff that uses pressure data.

Newcastle have the highest proportion of pressures in their defensive third & lowest proportion in the attacking third of any team in the Premier League so far in 2021-22.

Will Eddie Howe make them a hard-working side in the opposition half of the pitch? ⬇️ #NUFC— The Analyst (@OptaAnalyst) November 20, 2021

I'm not clear on which of Opta's various data feeds this is on (their main events feed is 'F24', their expected goals in the 'F70'-something range), and it seems probable that it's something that gets added to the mix from the STATS side of this collection marriage. [I wasn't able to get a clarification of the data feed in time for publication but will update the online version of this when I get it]

And that's not all. Sportlogiq -- a Canadian company specialising in tracking data rather than manually-collected event data -- collect information on the amount of pressure players are under, and have done so since (at least) 2018. (I spoke to them, and STATS, and StatsBomb, among others for an article in that year)

It seems likely that, at some point, pressure data will just be standard in any dataset you might be able to purchase. Some form of shot clarity (how clear a view you have of goal) might be too. And if that comes to be the case, what will it be that differentiates the data providers?

My thoughts on this, after the following box where you can sign-up to the newsletter if you haven't done already

Cost, speed (particularly for media), and accuracy are potentially the three big differentiators. But these are all quite boring. It's like saying that a key decision-factor in buying a new phone is build-quality: true but, y'know, at a certain level the phones are all functional enough to serve your purpose just fine.

So then (speaking about phones still, you understand) you get to things that you quite frankly don't use or realistically don't care about. Number of cameras. 4K shooting-quality. Shape of the bezel.

One thing that does set some phones, or phone brands, apart is the software they come with (mainly Apple). But similarly, StatsBomb have their IQ system; Stats Perform have the Trumedia-produced ProVision; Wyscout have data built into their video platform. Perhaps these systems can be unique and valuable enough to lock customers into the provider's ecosystem.

Alternatively, perhaps other 'add-ons' could have the same effect. In recent years, the number of tracking data companies (such as Sportlogiq) has leaped. Most of these (to my knowledge) don't offer a 'complete' set of events (such as passes, tackles, etc), but often do offer the ability to match their data with a full event provider. Maybe your choice between these event data providers will be decided by which tracking data companies they work 'out of the box' with. Or which data provider works with other types of 'add-ons' (how easy it is to link with video? third-party software? VR experiences? database set-up advice? consultancy services?).

The interesting thing about this question is that customers will choose data providers who best serve their needs. But, at the moment at least, everyone has such different types of process that those needs can differ quite greatly, including what type of data they want. Unless a customer is really big or really important though, providers aren't going to change what they collect to suit one client. Otherwise they'd never stop. Their provision will likely settle on something that is generally good for most people, but not perfect for anyone.

And if data providers are all trying to service the same people, will all of their provisions end up looking the same? A row of five-to-eight inch slabs of metal and glass staring back at you, separated by the number of megapixels in the three-camera arrays. And the shape of the bezel.

This is the end of the main newsletter, but subscribers in the data sector might want to continue reading

There's a point in this newsletter where I wrote '(to my knowledge)' in reference to what tracking data companies offer. Most of my awareness of data provider provision comes from several years immersed in and around analytics Twitter, and working for Twenty3, whose software product is built to work with varying different data providers.

However, I'm aware that there are gaps in my knowledge. I would like to fill them; for my benefit, for your benefit, and, to be honest, so that I can be fair to companies who don't have the Twitter clout that some others might have.

I'm aware that this newsletter has a degree of significant readership, and while it's still very much a-thing-I-do-in-my-free-time I want to make as much effort as I can to be thorough.

So, over the coming days I'll be getting in contact with the data providers I'm aware of to ask for access to documentation and be put on any list for product updates. I don't plan to write specifically about updates or what's in the data, but, like with this newsletter, it might be useful to fold the information into a wider piece. The providers are welcome to say no, of course, and I won't hold it against them if they do.

I'd also be open to other companies who I may not be aware of getting in touch with me. If that's you, you can use the email address getgoalside.newsletter@gmail.com

Finally, and only very tangentially related in the spirit of a slight increase in professionalism, while this is still very much a-thing-I-do-in-my-free-time, I do have a ko-fi page where you can (literally) pay some appreciation or encouragement if you wish.

Release time, and expected expected goals

Thu, 11 Nov 2021 00:00:00 GMT

A question for you. Should the player on the ball in the screenshot below A) pass the ball backwards to the supporting player B) turn and play in the teammate on the right C) turn and shoot?

Regardless of 'should', what actually happened was this (below)

Sofiane Feghouli shoots from 22 yards out

Feghouli's shot nestles in the back of the net, the goalkeeper rooted to the spot

It was quite a pretty goal.

If you're familiar with current expected goals models, you'll probably have said that Feghouli should have passed the ball instead of taking his shot. The teammate was closer to goal, and open!

However.

I'm not here to say that Feghouli shouldn't have passed, but by the fact I'm opening the newsletter using it as an example, you might be able to guess that I have a slightly different take on it.

Let us suppose that in the moment below, where Feghouli is facing the goal but hasn't yet started his shooting motion, he decides to pass. I'll mark on the screenshot where I'd say the best place to pass to as a 'first touch' point is for the receiver. The spot would be far enough in front of the player to run onto, allowing them to move centrally with their control of the ball, while hopefully far enough away from opponents to get a clean shot.

However, even if we assume Feghouli executes the pass well, what will the player's touch be like? How much will the goalkeeper and nearest defender be able to re-position themselves in that time? How does the player receiving the ball react to pressure of a defender rushing across in their peripheral vision?

In basketball, 'release time' is something that (I believe) sometimes gets talked about. Assuming they do it all in one motion, how long does it take between a player catching the ball and releasing their shot. This can matter: in such a small court space is important and scarce, while recovering to defend a three-point shot can be relatively easy. Players with quick release times might be able to squeeze out more shots from certain opportunities than players with slower release times. (This isn't an area I know well, so please get in touch if you know more about it).

Football is slightly different: basketball shooters don't have to worry (much) about defenders making contact with them, whereas in football you can shoot first-time. The principle remains though. If you're able to receive, control, and get into your shooting motion quickly, you're going to have a better chance of taking shots and scoring goals.

And that brings us back to Sofiane Feghouli.

If you're familiar with expected goals models, you'll know that there are some things they're (generally) unable to take into account. When you're thinking about alternative options, 'release time' and focus is one of them.

Below is a screenshot of the passage of play we've been looking at which is as close as I can get to (what I think is) the start of Feghouli's shooting motion.

The nearest defender about three yards away, and not moving forwards at any particular pace. Feghouli has had time to spot that the goalkeeper is well off his line. Pressure of playing in front of a crowd aside, you could argue that this is as close to a training exercise as a player is going to get.

Now, if Feghouli passes to his teammate, if the teammate needs to take a touch then they'll have a defender storming at them in their peripheral vision. They'll likely need to set their feet a little for the shot before taking it.

Looking at this scenario in isolation, it's not the case that Feghouli's 'shot from 22 yards with a defender between him and goal' would be replaced by 'shot from 18 yards with no defenders between ball and goal'. It's also 'shot from 22 yards with one defender but easily set in shooting motion' vs 'shot from 18 yards with no defenders but (potentially) not set in shooting motion'.

Even if you could measure this, it might not necessarily make much difference to xG values. My intuition says that it would be significant, but not a linear relationship: players seem to have an average 'clean-shooting' level, a 'scuff area' and then their sweet spot area. And sometimes the scuffs (the small scuffs) help more than a sweet strike would. But still, if strong/weak foot would make a difference, this surely would too.

The bigger question is how to measure it in the first place. The simplistic 'release time' could be done with any type of data where you have timestamps for when a player receives the ball and when they shoot. If you wanted to go more in-depth, you could look at their release times in different circumstances, like receiving from their left vs receiving from their right.

(You could potentially do that investigation on data beyond shots -- the way that a player is able to manipulate the ball to make a pass after receiving from their left or right is probably similar to how it looks when they prepare to shoot).

Of course, this data's not going to be easy to come by. Yes, you could use tracking data (if you have it) but shots have small enough sample sizes, let alone shots where this kind of thing is applicable. First-time shots have no notion of 'release time'. (Although the investigation of how well a player is able to sort their feet out for things coming from their left/right might be somewhat transferable).

Clubs could, in theory, create data from their training sessions, although you'd have to figure out how transferable that is to real matches. Although the data being limited to individual teams would reduce its value (no scouting, for example), it could still be used to help in-game decision-making, identifying training opportunities and tracking development.

I've done the writing cardinal sin with this newsletter and not thought of a satisfactory ending for this piece. So I'll end by saying is, despite what this newsletter's title is, please, please do not make 'expected expected goals' a thing. Nobody needs that.

If you enjoyed this, please subscribe and/or tell others about it. And if you have further thoughts on this, get in touch! You can either reply to this email if you're a subscriber, or find me on Twitter @sportsball_mark or @EveryTeam_Mark

Where to spend your analytics money?

Thu, 21 Oct 2021 00:00:00 GMT

Let's say that you are an analytically-minded fan of a football team. You care about them doing well, and you think smart use of data will be part of that. What should you be on the look-out for: who they hire internally, which data providers they use, or what third-parties they partner with?

Most usually, the answer to that would be who they hire internally. The creation of Liverpool's research department has contributed to that, as has the (related) increasing number of PhDs on football club payrolls. With City Football Group following Liverpool's lead in building out a big department, and Manchester United hiring a director of data science, some clubs are still heading that way.

But that's not the majority of clubs.

On Monday, Sarah Rudd, former-StatDNA (who were bought by Arsenal) vice-president of analytics, posted some job adverts. They were for Blue Crow Analytics, where she is now vice-president, an offshoot of Blue Crow Sports Group set up this summer. [2023 Note: Rudd has since left Blue Crow to co-found a football consultancy, SRC] The 'About Us' section of the job ads reads:

Blue Crow Analytics is a sports technology company focused on helping teams and athletes achieve sporting results through the use of technology, data and analytics.
The company was founded in the summer of 2021 by executives and investors in the technology and sports sector with experience across a variety of sports including football/soccer, baseball, basketball, motor racing and esports.

With the American element (the company is Houston-based) and reference to multi-sport experience, a resemblance to another third-party consultancy, Zelus Analytics, struck me. Whether this is what Blue Crow are actually going for remains to be seen, but it's worth a rundown on the model.

Zelus work with a limited number of teams per division in a number of different sports. They go hard on the R&D and the high-technical hires, like a "we hire the physicists so you don't have to" approach (although I doubt they'll put that tag-line on the business cards). A write-up in The Athletic is here.

This kind of high-end, investor-backed (more on that in a moment) consultancy is an interesting evolution to the industry. But it's not just in the third-party space that this evolution has been happening.

At the same time, over in data provider world, you have an increasing number of companies working with data in an exciting way. Opta and STATS (the artist formerly known as Prozone) merged to create Stats Perform, a company with extensive experience with event and tracking data under one roof; StatsBomb are producing a sort of event/tracking-lite hybrid data feed with freezeframes for each event they collect; the iteration of computer vision has led to numerous companies developing tracking data from single-camera video footage, reducing cost and increasing availability.

So where -- as an analytically-interested, innovation-hungry football fan -- would you look for your football club to be investing its money?

In the interests of transparency, I should say that I work for Twenty3 Sport, who I guess you'd class as a 'third-party' in these conversations. Football clubs should obviously be investing their money there. But besides that...

Internally

To take the clubs' perspective first, the ones at the top are humongous revenue generators, which means they could almost certainly afford to hire a few good people themselves.

An advantage of having people in-house, from the point of people doing the work, is that they (should) know the problems that need dealing with very well. They're working frequently with the people whose lives they need to improve, and that could lead to better work.

Hiring your own people also means that you get to keep all the work for yourself. The flip-side of that is that it might be more likely that your workers and their work are siloed from other ideas.

As well as this, not only is there the oft-mentioned factor about needing to get the internal politics right; but to do the 'innovation', data hires need time away from the hectic schedule of professional football. They don't need to be completely disconnected, but nor do you want them to be doing a ton of work before and after each matchday. If you can't offer them that space, you need to lower the expectations for the innovative, new stuff they can do.

Third-parties

An advantage for outside third-parties, on the other hand, is that they'd be separated from that day-to-day hecticness (assuming you haven't hired them specifically for that purpose).

On top of that, it's possible that with multiple clients consultancies could build up some sort of economies of scale, and that dealing in multiple sports -- or just with multiple clubs -- could help spark more ideas.

But on the other hand, you've gotta chase the money. How much R&D you can do will depend on either how many clients you can get or how much outside investment you attract.

The scale of outside investment presumably varies a lot and matters a great deal. As far as the new Blue Crow on the block goes, nothing about the analytics venture seems to be in the public domain, other than investment or advisement from Marc Reeves (former Nike, NFL, IMG executive; investor and/or advisor to, among other things, Leeds United; involvement known purely through an unrelated press release). Zelus meanwhile are backed by RedBird Capital, according to The Athletic's write-up, a behemoth who have a majority stake in Toulouse FC (via 'Redbird FC') and minority stakes in Fenway Sports Group (who own Liverpool) and reportedly Málaga CF.

There is no particular reason for that previous paragraph in this newsletter other than it's an area I don't know much about but seems pretty interesting and important. Toulouse FC are the only football club that Zelus have publicly (as of this May) named as working with... because they're backed by the company who own a majority stake there. Will they also do work with Liverpool now RedBird have invested in FSG? The stake RedBird have in Málaga is reportedly much smaller, but what about them? Will Marc Reeves' Leeds United connection (a club whose director of football, Victor Orta, was on a panel at the recent StatsBomb conference, so presumably open to this kind of thing) get an in for Blue Crow there? Those last three sentences are all mere speculation, but you can imagine how these connections might be able to be leveraged.

However, if, as a consultancy, you're dealing with multiple different teams in multiple different sports, that could present another problem. How long will you have to spend getting to grips with lots of different data sources, repeating the same processes over and over again because your new client uses X provider instead of Y.

Data providers

That last problem brings us to the data providers themselves. They will, of course, just be focusing on their own data, so have an advantage on that front. In an ideal world, providers would be the experts in their data, and be able to package it up and productise it and experiment with it in interesting ways.

The fact that the data provider space has become increasingly competitive also raises the chance that they'll do this. Cost is a central differentiator, as are things like accuracy and ease of use, but they're not exciting differentiators. But if the provider can give you something new and useful, something football-intuitive, that fits easily and immediately into your workflow? That's attractive.

On the other hand, the fact that data providers need to sell to a breadth of customers might mean that their research is limited. Anything interesting that they want to add to their data needs to be put through a proper product cycle, with all sorts of quality checks and engineering checks. All of that will take time, and if it wouldn't be enough of an attractive proposition to enough customers, that process may not be worth it.

Where would you go?

So, if you were an analytically-minded fan, or a decision-maker at a club, or someone looking to work in the industry, what area would you focus on?

If you're one of the elite, it makes sense to get ahead of the game and get a department set up internally, on the condition that you retain that knowledge. There's no point in the exclusivity benefits of an internal department if you don't make sure you still have it if someone leaves.

However, most clubs are not yet at the point where they're hiring data people and allowing them large amounts of time for research projects. For the majority, it probably makes sense to choose data provider smartly as much as cost allows (and to be honest about whether the shiny things in the data will get used) but then to make strategic use of third-parties.

Those third-parties could be consultancies like Zelus or Blue Crow, but they could also be platform-based like (yes) Twenty3, Analytics FC, SciSports (all three of whom also do bespoke work of some variety as well). And then there's the non-platform, non-Big Investor-backed consultancies of varying shapes and sizes.

Using these would allow you to flex your capacity for 'analytics' as needed, without hiring full-time. As a club, this can also enable you to build up an internal knowledge bank if you make sure that the third-parties work and/or findings get stored somewhere people will remember them.

On top of this, assuming that you're not their first customer, third-parties are also likely to have processes in place that mean you can skip some of the tricky 'training wheels' stage of setting up an analytics department. With the frequency of change that happens at clubs, this might be helpful for new managers or directors of football who need work doing quickly while they and their team get settled in/fully hired.

Sadly for fans, because many of these deals or partnerships don't get made public, you may not be able to keep up with how analytically-innovative your club is being. You're even less likely to know what work's actually being done too.

But from a wannabe-professional's point of view, it might not be a club job that you want to aim for, the fun stuff might be in a data provider or third-party instead. For the time being, at least.

Conference season, autumn 2021 - The What & the How

Sun, 10 Oct 2021 00:00:00 GMT

We know what players do, but we don't know how they do it

-- paraphrasing Vosse de Boode, Ajax's Head of Sports Science, speaking at StatsBomb Conference in 2021

It's the most wonderful time of the year
With the nerds all assembling,
The Voronois tending,
To show up on screen...
It's the most wonderful time of the year

-- paraphrasing Andy Williams, American pop singer, singing "It's the Most Wonderful Time of the Year" in 1963

Yes, it's analytics conference season. The New England Symposium on Statistics in Sports (aka NESSIS) is running online every Friday throughout October, and the StatsBomb conference took place in-person this past Friday as well. There's too much to take each conference talk-by-talk, but keep an eye on the respective YouTube channels (NESSIS, StatsBomb) and programmes for the events are here (NESSIS, StatsBomb).

But if you take both together, there are some interesting trends on view of where football analytics is heading...

As the De Boode quote (/paraphrase, I can't remember her exact line) at the start of this newsletter says, analytics is fairly adept at answering the 'what' questions in football now, but less so at the 'how'. In her talk at the StatsBomb conference this year she gave a good example of what that latter part can look like.

Answering the 'how'

When André Onana arrived at Ajax, he used a different stance to how Ajax goalkeepers were taught to set themselves in when preparing to make a save. Wider. The club could have just trained him out of it, thinking their way was best, but they just so happened to have a movement analysis lab and sports science department to hand.

De Boode and co tested reaction time and dive velocity across a number of their goalkeepers in different stances. They could use motion capture systems to get data on the angle of how the 'keepers stood and how they moved. And then they could analyse that data, and find that Onana was right after all. His wider stance, in certain circumstances at least, was linked to quicker reaction times and faster dive velocities.

Now, this is the most cutting of cutting edges in the field (and perhaps shows how Ajax keep producing such talented players). But although the presenters at these two conferences don't have what Ajax have at their disposal, their talks have also been dominated by questions that go beyond the 'what' and are getting closer to the 'how'. Principally, by investigating player decision-making.

It's all very well to know that players this many shots or make that many passes, but how often do they have those opportunities and choose not to? A goalkeeper may make a number of saves, but how do they manage to pull it off?

Across the 12 talks on the 'research' track of StatsBomb's conference and the first week of NESSIS, at least six of them featured a look at 'decision-making' in some form.

Three StatsBomb research talks mentioned it explicity: 'What drives the goalkeepers' decisions?' from Samer Fatayri, Kirill Serykh, and Egor Gumin; 'The quest for the right pass: Decision making' from Javier M. Buldú and Borja Burriel; and 'Turning with the ball & decision-making under pressure' from Soumyajit Bose and Manas Sarawat. Also of note is the fact all of these focus on a different area of the game.

As well as that, Hadi Sotudeh's talk 'Potential Penetrative Pass' contributed to the theme, developing a framework to identify when a player could have made a penetrative pass but either opted not to or didn't see the option.

And on top of that, two NESSIS talks each brought a slightly different flavour. Craig Fernandes looked at 'Untangling the Relationship between Intention vs Execution' -- but in tennis, rather than football. That said, there's a lot of knowledge and idea transfer that takes place between sports, so maybe someone in the non-racquet sport will take some inspiration from it in the future.

Also at NESSIS was Sam Gregory presenting 'Pace and Power: Removing Unconscious Bias from Soccer Broadcasts'. Here the focus wasn't on what players decided to do, but how viewers decide (unconsciously) to talk about women's football vs men's football and black footballers compared to white footballers.

Although not the application in the talk, you could see a use where scouts could be tested for their biases -- not necessarily even along race grounds, but on anything. The work Gregory presented (which was done alongside a team at SportLogiq and Toronto FC's Devin Pleuler back in 2019) turned video footage into anonymised 3D animations. Maybe scouts are more sympathetic to players they know vs those that don't, or to certain types of athletic build. Perhaps, in time, this type of process could be part of a scouting training course.

Furthering the 'what'

All of that being said, we're not all Ajax, and there's still a lot of 'what' to figure out. Another big theme of the conferences was widening the knowledge there through more specific questions. Defending, goalkeeping, phases of play: all are things that haven't been particularly well-covered by event data analysis so far, and all are things that StatsBomb research talks focused on.

A sizeable part of this was making use of StatsBomb's recently-unveiled '360' dataset [understandably, given it was their conference; for more on 360 data, read this]. Having a snapshot of where every player in camera-view is when every event is made is a bridge between the traditional 'event data' (what did player X do on the ball at location Y) and 'tracking data' (tracing the paths of all 22 players and the ball).

But some of it is just about asking more specific questions. Max Odenheimer and John Harrison scrutinised goalkeeping in an innovative way, while part of Will Morgan's talk looked at analysing defensive strategy while normalising for team strength and home advantage.

At NESSIS, Ethan Baron's talk -- 'Predictive Value of Off-target Shots in Soccer' -- was in a similarly inquisitive vein. Off-target shots are generally given a value of 0 in post-shot expected goals models because the went off-target. However, as Baron reasoned, knowing how close a player misses by might be able to tell you something useful about them. The metric created for the project, taking off-target shots into account, seemed stable year-to-year and a decent predictor of xG overperformance.

What's next?

Something striking about Baron's talk, Odenheimer and Harrison's, as well as the talk delivered by Maaike Van Roy from DTAI Sports Analytics Lab -- which split defensive set-ups into high/low, left-forcing/right-forcing blocks using 360 data -- is how they were splitting data in very football-specific ways.

Football analytics, as it has often been done, usually involves a lot of aggregation (combining all the data together). [I've lightly moaned about this in reference to 'finishing skill' before -- see here]. The advantage of this is that it increases your sample size in a sport where players don't play much and, outside passes, don't do things very frequently. The disadvantage is that you can mush together some situations that lose a lot of their meaning when they're mushed.

Now, you should always be aware of sample sizes. If I get seven heads from ten coin flips, that doesn't necessarily mean I'm an elite heads-coin-flipper (although I maintain that I am). But I think that you can be aware from this when interpreting figures that you get, rather than not making those sport-specific divisions in the first place.

While part of this is because the 'analytics community' as a whole has built up (much better) a base of ideas of what is useful for professional football, I also think it's worth looking in the other direction as well. "I want to find creative players" is something that a football professional could ask. But they could ask "I want to find players who pass the ball into dangerous areas starting from relatively non-threatening areas."

Both things -- data practitioners learning what questions to ask and football practitioners learning how to be specific with the questions they want answered -- have been important to this process. And it'll mean that, as time goes on, both the 'what' and the 'how' questions will be increasingly easy to answer (as long as, y'know, you have the relevant data and technical expertise).

Julian Nagelsmann's earpiece idea might not be so bad after all...

Thu, 30 Sep 2021 00:00:00 GMT

Every now and then, somebody profoundly intelligent in one field of human endeavour says something that makes the world question how wide-ranging their intelligence is.

"American football is much more technologically advanced than soccer," Julian Nagelsmann recently said. "The quarterback has an earpiece to hear his coach; we absolutely need that."

It seems pretty plausible that the German manager was just spitballing and wasn't actually setting out a plan, but let's examine the idea further anyway. The idea didn't get wide coverage, but what it did get was negative. Pretty heavily so.

However, let's take this notion and run with it. Not because I want to give managers the ability to micromanage, but because interrogating an idea like this can be fun. Sometimes you can get something novel out of nonsense.

Forget NFL

The German's quote itself shoots itself in the foot: American football is not like association football. For one, you don't have a quarter-back; for another, you don't have breaks in play. Maybe you'd want to earpiece up the captain, but captains in football aren't always the players who pull the strings of tactics and tempo on-field.

Another sport which might be an interesting comparison is Formula One, which has earpieces but without the NFL-type breaks between plays. Drivers can communicate with their engineering and strategy teams, and that can be useful.

Sometimes a change of weather or an incident somewhere will dramatically change the pit-stop strategy, or the driver needs to be informed of some issue with the car they need to manage. Information can be quickly and succinctly relayed.

The best exchanges though, as a viewer, are the ones where the drivers just tell their engineers to fuck off.

Realistically, I think the way that tactical earpieces would work in football would be for them to only (or at least primarily) used in breaks of play, to relay relatively simple changes in shape or approach. "Switch to a three in midfield", "their left-back's having a bad game, let's send more long balls right at them", "be quicker pressing their defensive midfielder". All the stuff that already is being relayed to players mid-game, but that only happens when they have time to jog over to the sidelines when someone (usually a goalkeeper) is receiving (possibly unnecessary) treatment).

Innovation(?)

Given that managers can already tell their players what to do, the tactical element probably isn't enough to get a push behind earpieces. However, I think there are a number of other ways that earpieces could be used to add something to football.

Innovation Number One: Concussion detection

Now, this is the kind of 'bullshit science that gets sales' option. A consensus statement at an international conference for concussion in sport in 2016 said that "reported mean peak linear and rotational acceleration values in concussed players vary considerably". For various reasons, "The use of helmet-based or other sensor systems to clinically diagnose or assess SRC cannot be supported at this time."

This being said, I believe in science and all that stuff, so I feel like, one day, maybe you could be able to detect 'possible concussion impacts' from earpieces worn by players.[1] We made it to the moon with less processing power than is in an iPhone and all that.

Also, if used in training, they could potentially monitor the allotted number of 'higher force' headers that players in England are supposed to be limited to each week.

Innovation Number Two: Player scanning data

I'm not sure if it's unfair to call 'scanning' a buzzword, but it's certainly a very popular concept to talk about at the moment. There is a whole field of research about it -- about how it makes players effective, and about what makes 'good' scanning.

The data, though, would be difficult to collect. This recent research paper relied upon watching clips of players.[2] Can you imagine?? Watching players?? (These last two sentences should be read in the voice of Vinny Thomas). Imagine if you could sync an earpiece sensor to the vest-worn sensors, knowing where the head is looking in relation to the torso at every moment.

Innovation Number Three: Player orientation data

Thinking about it, earpiece sensors might not help that much with this. The current vest-packs that players wear don't tell you where a player is facing, just where they're moving. But if you could calibrate an earpiece sensor to know where 'north' of the pitch is, you could presumably do that for the vest-packs too. But an extra sensor couldn't hurt, and we need as many spurious benefits to sell these things, so it's going in the brochure.

Innovation Number Four: Mindfulness

We know that there are some players who struggle to perform in front of large crowds or in big moments. This is a human response. Maybe they could press a discreet button on the earpiece and get some breathing exercises piped into them, or some distracting music to get them in the zone.

The highly-paid elephant in the room

Now, one of, if not the, biggest obstacles to earpieces being used in football is the players themselves. It took long enough for the vest-packs to become standard, and they're fairly unintrusive. Imagine being asked to put sensors (somewhat) inside you.

A pretty compelling case would have to be made to them that this was worth their while, and not just another stick that coaches could use to beat them with. I'm also intrigued as to whether the bundle of use-cases I've provided might be less convincing to them than being presented with just Nagelsmann's idea. I'm sure different players would be more and less responsive to different possibilities.

There's also an important question, apart from the use cases, of how comfortable and secure the earpieces would be. Jonathan Wilson wrote off Nagelsmann's idea because he thought it unfeasible that earpieces would stay in, and maybe he's right. I'm no ear scientist.

Do I see the earpieces happening, even after all of this interrogating of the idea? Not really. I suspect that concussion detection would be so patchy, and so infrequently needed, that it doesn't provide much of a bonus; that the bonus of coaches chatting to their players isn't a worthy trade-off for wearing something in your ear for an hour and a half; and that players would be unconvinced by the need for data on their scanning.

But hey, we've all been wrong before.

If you've found this interesting, subscribe to the newsletter and get more thoughts on the future of football sent directly to your inbox

Notes

Footnotes

[1] "[...]maybe you could be able to detect 'possible concussion impacts' from earpieces worn by players." || F1 earpieces already have accelerometers in them, with the manufacturer's website saying that data is used to help make improvements to neck and head restraints. This is slightly different to the one-off forces involved in concussions, but it's at least something down a similar path.

[2] "This recent research paper relied upon watching clips of players." || Some studies cited by that paper used headbands to detect head movement, which is a step towards automated data collection but isn't something that would realistically be used in real-game scenarios.

Creating new data by stealing from basketball

Thu, 16 Sep 2021 00:00:00 GMT

An origin tale

A couple of weeks ago, I read an article on FiveThirtyEight. It was about a bit of basketball tracking kit, the Noahlytics Data Program, which gives you data on the angle of shots and where they go through the rim. (I only partly mention this because the name is a Biblical pun: Noah's arc -- although it doesn't insist players take things two by two).

The nifty thing about the program, according to the article, is that players can get good, concise feedback on why their shots aren't going in, and quickly too. There's an anecdote in the piece, which is by Ben Dowsett, where then-Toronto Raptors player Kyle Lowry was going through a cold shooting patch. The Noahlytics data showed that the arc of his shots was at 41 degrees, whereas when he was shooting well it was around 46 or 47.

From there, I assume basketball coaching takes over, whatever basketball coaching looks like, but I was intrigued by the thought of how this type of thing could be transferred to football.

'Player motion data' -- a term I will coin here -- is not exactly new. The video cameras that capture positional tracking data (think Football Manager dots for every player on the pitch) can also be used to capture players' body orientation. The software needs to be more advanced, and the work to use the data is more difficult, but it's been applied to the NBA for a number of years.

Body pose is damn sophisticated though, and the relative simplicity of Noahlytics struck me as something that might be a more easily applicable concept to transfer. You don't need to look at a computer-generated ragdoll or a bunch of ankle-knee-hip angle readouts; just a few numbers about a repeatable skill.

But, the question is, what is a worthwhile repeatable skill in football to apply this to?

For regular questions like this, subscribe to the newsletter

Oh football, you bringer of despair

The problem with football as a sport (well, one of them) is that shots don't happen very much. On average, any given team in any given match will take about 11-12 shots, give or take a couple. In basketball, the average is almost ten times that. Teams take around 30-40 three-point attempts and 45-60 two-point attempts (let's not count free throws for now).

The flip side of this is that each shot represents a tiny amount of the team's eventual scoring plays, but the fact remains that if you try improving a player's shooting, you'll see results pretty quickly. This is even moreso the case when you consider that shots in basketball generally look pretty similar compared to the mechanical variety in football shooting. For a start, NBA players will never be taking a shot with their head.

According to FBref, teams make around 40-60 passes in the final third per game, a comparable number to three-point attempts, which is nice. These, I think, are completed pass numbers, so the amount of attempts will be a little higher. We have an action here that has comparable frequency to basketball field goals and a not-insignificant impact on the eventual total of a team's scoring. We might even be able to expand this to passes in the opposition half if we wanted a larger sample.

But what would you measure? Unlike in basketball, there's no fixed target or repeatable goal. You can't consider the intended recipient themselves to be a replacement for 'the hoop' in basketball, because what about through-balls? What about when you want to set the teammate up to receive on the turn?

Unlike for shots, where hitting the ball cleanly does seem to be a problem, I can't think of many 'mishit' final third passes. There are also a lot of different techniques involved depending on the type of pass: long balls, chipped passes, crosses, more regular passes too.

An actionable idea?

Let's think about this a little differently. What about if, instead of pass technique we were to look at first touch instead? The reference points could be the player's own body and the nearest defender (within a given radius). Could you identify when a player's first touch puts the ball too close to the centre of their body, not on either foot? Or when it puts the ball practically underneath them, or too far away from them, making it tough to make a next action?

The information about the nearest defender would be necessary to add context of course. A player's first touch might put the ball overly close to them, but that might be because a defender was right on their back. However, maybe in those circumstances you'd be able to say "hey, we know you're receiving under pressure, but if you put your first touch a little further away from you and onto your stronger foot a little more then you'd be able to offload it quicker".

This would also potentially fill a gap in (most of the) existing data provision in football. There's a wealth of data on shots, a just-about-managing of data for passes, but there's almost nothing on how players receive the ball. But what does every player do before they make a shot or pass (barring first-time efforts I suppose)? Take a touch!

Oh football, you bringer of despair (reprise)

This is, as much as anything, a lesson in the difficulty in trying to 'take inspiration' from other sports. Although basketball and football both involve two teams, a ball, the concepts of shots and passes (with no restrictions on where passes can go), they're still different. The way that the games are played means that, while concepts might be similar on the surface, they way they play out in practice can alter things hugely.

Actually, hold on.

The way things 'are' played?

Let's take a step back to where I -- stuck in my sense of the present day -- said that football teams have about ten times fewer shots per game than basketball teams. There are a number of reasons for this: the presence of a goalkeeper; the different dynamics behind getting within shooting range; the fact that basketball hoops are above players, meaning shots have a free path, once they're in the air, to their target.

But another is simply that basketball players use their hands and footballers use their feet (or heads). The difference between humans and other animals is (among other things) our opposable thumbs, not really anything particular about our feet. As far as ball sports go, basketball, rugby, NFL -- they're playing on easy mode. Football -- where players, as Brian Phillips wrote, are forced to 'compete in a state of artificial clumsiness' -- is where hard mode gets turned on.

Considering they are literally paid millions to do this, elite footballers miskick the ball a surprising amount, especially when shooting. Sometimes they plain just fall over in the motion of failing to hit it.

But imagine if, perhaps using something like the Noahlytics program, they were trained in striking the ball cleanly. I'm not expecting perfection, but imagine if footballers, across the board, got tangibly, markedly better at getting a good shooting motion. Would teams still be taking 12 shots a game?

Players getting better at shooting threes in basketball probably isn't the main driver of why three-point attempts continue to rise, but it surely is a driver (at least compared to when they were first introduced into the NBA in 1979). If footballers got 30% better at shots from further than the penalty spot, would they make more of them? I bet they would.

Conclusion

So, maybe you can take an idea from basketball and apply it to football, but needing to change the area you're applying it to. Or maybe you can apply it to the same area, and you might change the sport as a result.

The Who, The What, and The Why of bringing advanced stats to the public

Thu, 02 Sep 2021 00:00:00 GMT

Get Goalside is returning to a regular schedule for the new season. A new newsletter will go out (at least) every other Thursday. early October sees a couple of big analytics events so there may be an extra one or two that month.

As far as false dichotomies go, the separation of arts and sciences is one of the big ones. From Ada Lovelace to Apple to the symbiotic relationship between the sci and fi in sci-fi, it’s not only unhelpful but ludicrous to drive a barrier between the two.

In the same way, the dichotomy between ‘proper football’ and ‘analytics’ is a false one too. In fact, that’s practically old news, an argument that’s long been settled. Football clubs want to win, and they’ll use whatever helps them, regardless of where it comes from (although sometimes with some scepticism and lack of enthusiasm).

But the media doesn’t need to ‘win’. The websites and magazines that we read, the coverage that we watch and listen to, doesn’t need to worry about correlations and robustness studies. The people who do worry about those things sure think that advanced stats are neat and useful though, and like telling other people about them. So how do you mix the two?

The Why

“There’s always a desire to place narrative and analytics in opposition to each other and I just don’t think that that’s true,” Mike Goodman, soccer editor for CBSSports.com, tells me. “My belief about using numbers and stats and analytics in media is, ‘we do want to tell stories but we want to make sure that the stories we’re telling are correct’.

“And what analytics does is it just gives you the tools to make sure that the stories you’re telling are right.”

For purposes of this newsletter, I’ll be using ‘advanced stats’ – rather than ‘stats’ or ‘analytics’ or something else -- as my catch-all term. I’m talking about not just expected goals, but things like possession sequences or progressive passes too. Anything that you could say is numerical and new to most fans in the past couple of years, things that are newer than things like shots, shots on target, possession percentages.

And some of these things certainly are still new to a lot of people. While it’s over half a decade since Arsène Wenger first started namedropping expected goals, the mention of xG can still get an amused “uh-oh” on national radio. Why do we persist in trying to use them?

“I think in most instances it’s because they fill a gap that other metrics can’t fill,” Tom Worville, Football Analytics writer at The Athletic, tells me, before going on to echo Goodman’s point. “I also think it comes from people demanding more from their football content. A talking-head ex-footballer seemingly isn’t enough. They want facts at times, not just opinion. Fancy stats can be a good bridge for that.”

As well as that, there’s a two-way relationship with how fans experience a game. “If a stat, whatever it is, highlights the way a game’s gone, or reveals something, and is easily explained – with that being key – then I think it would be used,” Matt Cheetham, a football statistician at Sky Sports, tells me.

“For us,” Cheetham said, “it’s all about how easily it can be understood, especially if appearing on screen for 5-10 seconds. During a game, you want stats to accentuate a trend in the game that people are seeing – can we highlight that further, with numbers.”

“We probably want to calibrate our metrics to measure the things that fans clap in the ground,” says Worville. “If we as people can recognise a situation and “reward it” (through clapping) then our stats should be able to measure and reward these players and teams too.”

The What

The latest attempt to bring a metric forwards which quantifies something fans already reward is ‘expected threat’. Expected threat, or xT, is a concept that has a long history[1], with the current name coined by Karun Singh in 2018 (you can find his original blog here, and The Athletic’s explainer here). Like all good metrics, the crux is simple – to take a line from The Athletic’s piece: “what’s the chance of scoring in the current possession [passage of play] based on where the ball is now?”.

The article created quite a buzz, leading to a mention on BBC Radio 5Live and an appearance on TalkSport for Worville, getting xT on both major national radio stations for sport in the span of half a week. It is, in a sense, ‘so we’ve got to grips with xG now… what next?’.

There are other ways of creating a metric for this sort of use case though. Between Worville and myself during our conversation for this newsletter, we reference at least four others: VAEP [valuing actions by estimating probabilities, DTAI Sports Analytics Lab], On-Ball Value [StatsBomb], Possession Value [Opta/Stats Perform], Goals Added [American Soccer Analysis]. Why specifically xT then?

Fittingly, given that xT very much sprang from the online analytics community, that community is a large part in it being xT that has made its way to The Athletic. A number of different models from the community were consulted – one provided to Worville previously by Mark (@markrstats on Twitter), a json file of xT values that Singh made available, and the Python package socceraction – to help determine how best to build The Athletic’s own model.

“With [the] socceraction Python package and a good community of people who have been there and done that and can answer questions, it [xT] just felt like the easiest one to build and explain,” Worville told me. He also name-checked a number of people -- Singh, Mark, Abhishek Sharma, and Pieter Robberechts -- in a tweet when The Athletic first unveiled their own version of the metric.

As well as this, xT is probably the simplest and most easily explainable of the possession value family of metrics. You can, as The Athletic have done, divide a pitch into a grid and show people in one image how ‘threatening’ each area of the pitch is.

Image from The Athletic’s explainer piece on expected threat.

Asked about whether some on the more pure-analytics side of things might prefer a different approach, Worville sounds open to future changes. “I think it's fine to probably look back in a year and be like, ‘oh, you know, that was a bad approach or that's tame’ but that's just the nature of science I guess, at times,” he says. “I think there's no reason to have nothing until we have a perfect model.”

A factor to consider is also that The Athletic doesn’t have as wide an array of data, or analyst power, to work with as, for example, the data providers creating their own possession value-type metrics. But this also raises the interesting point that this is a very different introduction of a metric to ‘the mainstream’ than its spiritual predecessor expected goals.

When xG first hit mainstream outlets in 2017, it was because Opta had added it to their feeds, which the likes of the BBC and Sky Sports used. Indeed, one of the reasons why Worville felt now was a good time for expected threat was that “I just thought that's a good opportunity to kind of pounce on, if it's only going to be a short window of time before a supplier brings out something new in a feed, [and] if we kind of get in there and talk about stuff first.”

The Who

Matt Cheetham: “For us [Sky Sports], it [the usage of advanced stats] is generally a case of how much producers, presenters and pundits buy into the stats. A producer [is] driving the show, so their input is key. I can’t talk for pundits, but I assume a lot of them understand how often something like xG is used by clubs, so appreciate its relevance in analysis.”

Tom Worville: “My editor understands stats and analytics, much more now than he did when I started. And I think he has that level of trust in me that this [xT] is going to be something that will interest people.”

Mike Goodman: “When you’re doing what I’m doing now which is driving coverage, I have professional responsibilities for traffic or sign-ups or whatever, but it’s [also] sort of like ok what does coverage look like that our audience wants to see, that differentiates us from competitors.”

When I was speaking to Tom Worville for this newsletter, I wondered aloud whether expected threat was following the same trajectory as expected goals. There had been several years when xG had been something that an online community of enthusiasts used, cobbling together models of their own, before it was productised to serve to the masses. Karun Singh’s xT blog was in 2018, and now in 2021 we have The Athletic introducing it, getting mention on national radio as a result.

Worville didn’t wholly agree. “I’d probably say xT is arguably a bit quicker,” he says. “I think the trajectory is slightly different.”

It certainly will be now. The people involved in earlier waves of advanced stats are making their presence felt. Some who pioneered expected goals and got it used in clubs are helping to persuade pundits that it, and other advanced stats, are worth listening to. Some are helping shape the entire football output for major outlets. Some are bringing a new metric to the attention of audiences who otherwise may never have come across them.

“I think we've probably got a good platform now where a lot of the stuff that we do will hopefully seep into broadcast at some point and other outlets will be doing it,” Worville says. “And that's only going to be positive, I guess, as an analytics community trying to get more people interested in using this stuff.”

Worville also points out that, while an increasing number of members from the online analytics community go quiet because they get hired (by clubs, data providers, or consultancies), a lot of people are new to this world, new enough that they might have missed Singh’s original blog in 2018. The community is still going strong, like with the individuals Worville mentioned in his xT tweet (and Omar (@placeholder1966 on Twitter) who he also mentioned during our conversation), and the new waves of people who might be being introduced to advanced stats with xT.

The What (reprise)

Each of the three people I spoke to for this newsletter would probably be displeased if I left you all with the impression that using advanced stats is all about using the latest xMetric. Heck, you don’t even have to mention them out loud at all.

“I think at some point what happens is you want to be knowledgeable of them [advanced stats] and to run in the background of what you’re saying,” Goodman says. “I’ll talk to people who call games for a living and they’ll say this, even with expected goals, they’ll say ‘I don’t want to mention expected goals on the air but I do want to be able to say with confidence that that shot that that guy just took relatively rarely turns into a goal’.

“I think if it’s informing your analysis and it’s used correctly to inform your analysis I don’t really care if you’re explicitly saying the number or explicitly writing the number.”

But where advanced stats are explicitly mentioned, Goodman is also clear that they shouldn’t be just about an xNumber, as coverage can sometimes (I say as someone guilty of this myself) become: “My desire for expected goals would be for it to be less of the sort of sun that everything else revolves around and more of a tool in an arsenal,” he says. Expected threat may help that, of course, as Worville suggests: “xG is good but I guess it gets to a problem where you wanna learn a bit more, and it goes beyond xG, it goes beyond progressive carries and passes, which I think is something that we use a lot [at The Athletic].”

“Let's get away from xG a little bit and just kind of compare [sides like Brighton, who underperformed against expected goals last season] to how other teams can build up and how other teams build and stage attacks,” he goes on to say, about xT.

But expected threat certainly isn’t going to be the last new metric that will enter mainstream coverage. “A lot of the more advanced metrics feel on the fringes of regular coverage at the moment,” Cheetham says. “The more they underline a theme, or feature in pre/post-match analysis, I think the more often they’ll appear.”

Notes

Footnote

[1] The example that probably hits the middle of the Venn diagram of ‘earliest’ and ‘most significant’ may be then-StatDNA Sarah Rudd’s 2011 presentation at the New England Symposium on Statistics in Sports (NESSIS) [slides available here]. ‘Non-shot expected goals’, which had a peak in interest around, at a guess, 2016 is also something I’d consider a familial relative of xT.

Not all duels are created equal, nor are they easy to adjust

Thu, 26 Aug 2021 00:00:00 GMT

It is 1694. Edward Wilson dies in a duel he asked for over the favour of a woman. It is 1804. Aaron Burr's shot tears through Alexander Hamilton's liver and into 21st century pop culture. It is an unknown year, close to the present. Someone decides to collect tackle data in a 'duels' framework.

It is 2021. I am writing this newsletter. All of these decisions are arguably regrettable.

I do not like defensive actions being organised as ‘duels’. I think that it’s a categorisation of defensive actions that doesn’t leave much room for a lot of defender-attacker interactions. What is a duel? It’s a contest between two parties, with a suggestion that each party is engaging equally. But when an attacker receives a ball and the defender, behind them, pokes it away from between their legs, is that a ‘duel’?

The point of a duel was also, in its early modern European form, for each side to be on a broadly equal footing. Rules were agreed upon and weapons were matched. But duels on a football pitch aren’t like this.

In the 2020/21 Premier League season, Christian Benteke was the forward who had the best aerial duel success rate (out of players who'd contested 50 or more, per FBref). He got just under 60% of his headers. However, a list of those with a better rate than him includes Andy Robertson, Thiago Alcántara, and Victor Lindelöf, none of whom are particularly known for their heading ability. There was a cigarette paper between Andy Carroll's success rate and Aaron Wan-Bissaka's. Sergio Reguilón was 'better in the air' by this metric than Aleksandar Mitrović.

Duel win percentage is one of the last remaining defensive statistics that I see commonly used as an indicator of quality. For things like tackle and interception numbers, a lot of people (wisely) use them to gauge style or role rather than quality, but duel success rate seems just so temptingly simple.

But simple it is not.

As far as aerial duels go, it's the long balls forward from goal kicks, free-kicks, or clearances that help defensive players out. With more defensive players than attacking players high up the pitch, defenders can delegate while attackers need to move, forcing them to take their eye off the ball and approach things from an awkward angle. They’re also usually aiming for flick-ons, where a jump that’s just too high is a negative; defenders just want anything on the ball, and can leap as high as they like.

Quite simply, its the same principle as for expected goals: not all shots are equal, and not all duels are equal either.

Dribbles and tackles (two sides of the same coin in many datasets) aren’t exempt from this principle either. Prospective tacklers are less likely to be dribbled past if they’re on the wing, and they have a higher likelihood of being dribbled past in the centre of the pitch.

This is shown in the image below, which I’m repurposing from an old post on a now-nonexistent blog. It uses data from StatsBomb's open dataset[1], and is comparing their Tackle and Dribbled Past events[2], split into a grid of roughly 9 yards by 9 yards. The size of each circle indicates the number of events in that zone, and the deeper the shade, the higher the tackle success rate. The range is from around 50-75%.

There’s an obvious reason why it’s harder to dribble past someone on the wings, and it’s the sidelines. As a defender, you can predict a winger’s movements more easily because they have fewer directions to go in. However, this means that comparing players’ tackle and dribble success rates will depend on what kind of tackles/dribbles they’re attempting.

Even for players in the same ‘position’ it might not be easy. One centre-back may have to cover for their full-back and make tackles by the sideline; another might be tasked with stepping forwards into midfield and try to make ‘trickier’ tackles in the centre of the pitch.

Unfortunately, the difficulty in digesting data doesn’t stop there. Here’s another repurposed image of mine using StatsBomb free data on where interceptions are made. A heavy concentration of them are made in very particular areas.

It therefore seems likely that the areas a player plays may affect how many interceptions they make. This doesn’t necessarily mean that it’s defenders making these interceptions. If an attack is high up the pitch, the defenders will often have retreated to deeper/narrower than these peaks of concentration. This might also mean that this visualisation is telling us something about the situations of interceptions too (or, potentially, just the way the data’s being collected!).

Subscribe to the newsletter for where football is heading, on and off the pitch

Similarly, pressures, like tackles (and for similar reasons), are more likely to be made by the sideline.

So, how to deal with this.

You could choose to make ‘expected goals’-type models that account for factors like position on the pitch, but you might also think that is overkill. Specifically on duels, there are other ways of modelling them. You can try and place values on the individual duellers' ability, which will likely take part of the above positional effects into account, though probably not all, but might give a better picture than modelling for position on the pitch alone.[3]

As far as the ‘counting stats’ (number of tackles or dribbles made) go, ‘possession adjustment’ is a well-worn option too. I wrote a few blog posts on the same now-nonexistent website as I took the above data visualisations from, so allow me to repurpose some of it.

There are smatterings of work out there on various forms of possession adjusting. A piece from Mike Goodman in 2013, this on American Soccer Analysis in 2014, this from Ted Knutson later that year. They're surprisingly varied, and if you're interested in this topic then they offer a range of ideas to draw from.

One of these ideas was that, by adjusting for a team’s share of possession, you could get a more reflective number of a player’s defensive actions. The rationale is that players on teams with low possession have more opportunity to defend, so they might have defensive stats that are inflated by this (and vice versa). Adjusting for possession share, in theory, levels the playing field.

I have long had my doubts though, and wrote more about my quibbles with the theory at more length in another newsletter essay. But if we're possession adjusting defensive stats because we think players on teams with low possession naturally make more actions, we should at least check.

When I did this analysis, I only used the two seasons’ of WSL data that StatsBomb had in their open dataset. I figured that the Barcelona data in the Messi career dataset would, with the Catalans’ possession-heavy style, skew the results. For the method, I took the share of passes that a player's team had when they were on the pitch, and got a basic correlation between that and the number of defensive actions per 90 minutes that the players made. Goalkeepers were filtered out because they’re a particular exception.

What you would expect to see if you wanted evidence in support of possession adjusting would be a general trend where the action per 90 minutes goes down as pass share goes up. This does happen, though quite lightly.

Tackles per 90 vs pass share, a slight negative correlation

Interceptions per 90 vs pass share, a slight negative correlation

Recoveries per 90 vs pass share, a slight positive correlation

Pressures per 90 vs pass share, a slight negative correlation

Now, back in the day (circa late 2014), I experimented with a kind of possession adjusting of my own. I was working exclusively on central defender statistics at the time (as I was copying numbers into Excel by hand), and just using possession didn’t make sense to me. Players like Chelsea’s John Terry and Gary Cahill played on a team that had quite a lot of the ball, but they still didn’t do much. So I decided to factor in the number of shots on target and shots blocked that centre-backs were on the pitch for too, as a proxy for ‘work they might actually be involved with’.

I don’t think it was a good method. But I do think it was good thinking. As I would later find out when doing the pass share correlations, the picture looked different for different positional groups.

To start small and make things easier to digest, I split players into broad ‘defender’, ‘midfield’, and ‘forward’ categories. As far as tackles went, pass share seemed to make no difference (in this limited sample size) for midfielders, but it had a slight effect for defenders and forwards.

Tackles per 90 vs pass share, split by position, a slight negative correlation for each

For interceptions, again not much for midfielders but seemingly some signal for defenders.

Interceptions per 90 vs pass share, split by position, a slight negative correlation for each though stronger for defenders

The previous graph for recoveries showed the opposite effect to possession adjustment theory’s expectation: a rise with pass share. This is because the way recoveries are defined in StatsBomb’s data (and some other data providers’ data) necessitate the team keeping the ball for a given number of passes for it to count. However (again though, small sample size), this seems more true for defenders and midfielders than forwards.

Recoveries per 90 vs pass share, split by position, a slight positive correlation for defenders and midfielders while slight negative for forwards

And finally, pressures, where each positional group had a similar trend. This is the least surprising. You could, hypothetically, make an infinite number of pressures without winning the ball back, meaning you can keep adding pressures for as long as you don’t have the ball. For other types of defensive actions, you usually interrupt the opposition’s possession by making a tackle, so the theory about ‘increased opportunity with lower possession share’ may not quite work.

Pressures per 90 vs pass share, split by position, a slight negative correlation for each

This method of possession adjusting isn’t the only one though. As well as pass-based possession, there are ‘possessions’, plural – spells of possession where one team or another had a ball. In this methodology, a string of two passes and a string of ten passes both count as a single ‘possession’.

Got this far? You'll probably want to subscribe to the newsletter

If we go back to the theory of possession adjusting briefly, that defenders who have more opportunity to defend will ‘unjustly’ have higher defensive stats, adjusting for the number of possessions may make sense. A bitty game, where neither team holds onto the ball for long, would surely see increases in things like tackles or interceptions etc.

I did some correlations for that too.[4] The trend we’d be looking for here, to support adjusting per-possession, would be tackle numbers going up alongside number of possessions per 90 in matches a player played in.

For tackles, it was a more distinct trend for midfielders than other positions, but a very slight upwards relationship for all.

Tackles per 90 vs possessions per 90, split by position, a slight positive correlation for each

For interceptions, it seemed to make no difference for forwards (although again: sample size), but seemed to be some kind of light signal for midfielders and defenders.

Interceptions per 90 vs possessions per 90, split by position, a slight positive correlation for each, weaker for forwards

Recoveries is the weird odd one out again

Recoveries per 90 vs possessions per 90, split by position, a slight positive correlation for each, stronger for forwards

And this time the trends for pressures looks more similar to the trends for recoveries than anything else.

Pressures per 90 vs possessions per 90, split by position, a slight positive correlation for each, stronger for forwards

I present these not as definitive takeaways to apply to your own work, but to show whether possession adjusting makes sense depends on both the stat and the position of the player.

My current opinion on possession adjusting is that this kind of complication may not make it worth it, but each to their own. I echo Ashwin Raman’s tweet about how an extra layer of abstraction might make it difficult to work out whether the adjustment is working or not:

So yeah, basically, tl;dr: be wary of more complex forms of adjustment, because they can often be less useful than simpler adjustments because of the harder-to-contextualise skewed numbers.— Ashwin Raman (@AshwinRaman_) August 25, 2021

That’s my opinion though, and for those working in the professional industry, it might be worth trying out some adjustments. But it’s important to know:

That possession adjusting worthiness can change by position and stat
That duels can be easier in some areas and situations than others
That I prefer the duel songs in Hamilton more than using duels and tackles as synonyms, and the duel songs in Hamilton are some of my least favourite.

Read this information; absorb this information; go forth and stat.

If you've enjoyed this, please consider passing it along to someone you think will find it interesting

[1] “It uses data from StatsBomb's open dataset“ || WSL and Messi data only. I can’t remember why I only used that data, but it is what it is.

[2] “…comparing their Tackle and Dribbled Past events…” || There will also be tackle attempts under Foul events, but I couldn’t work out how to separate out just the ones that would be applicable here.

[3] “There are other ways of modelling duels by trying to get each dueller’s ability – this will likely take part of the above positional effects into account, probably not all, but might give a better picture than only modelling for position on the pitch.” || This actually seems like a fun, if pretty technical, research project - which is the bigger influence on duel success: circumstance or dueller ability?

[4] “I did some correlations for that [possessions adjusting] too.” || I don’t seem to have kept a note of how I defined my possessions, but I don’t think that methodology should matter too much.

A history of 'pitch control'

Thu, 27 May 2021 00:00:00 GMT

When I was – at a guess – seven years old, there was a kid's TV programme that tried to make science cool. One episode they stretched out, to full length, some fake intestines. In another, they demonstrated the extreme knowledge of physics that David Beckham must have to be able to bend the ball up and down and swerve over a wall at a free-kick like he did. Rubbish, I thought. It's practice not physics. Stop trying to crowbar science into my football.

Over a decade later, here I am writing about science in football. And not just any science, damned physics. If the mid-2010s of football analytics were about mathematical modelling, the mid-2020s are going to be much more about physics-based models. It all started in 2016, at the OptaPro analytics forum, with a model called 'pitch control'[1]. It would soon become an underpinning feature of a lot of other exciting work, and this newsletter is going to tell you all about it.

The 'pitch control field' was introduced by William Spearman, then of Hudl, and it does what it sounds like – measures who has control over areas of the pitch[2]. The result looks like abstract art, with each team 'painting' areas of the pitch where they have the best access to.

A slide from Spearman's 2016 OptaPro forum presentation, showing a pitch control model in action

This is all physics. Using tracking data (which uses cameras to see where every player is multiple times per second) you account for player position, current direction and speed. Factor in some general acceleration and top speeds that players can reach, chuck it all in a model, then you get your results.

With this being the fast-paced world of football analytics, things would soon change.

Two years later, in 2018, the MIT Sloan Sports Analytics Conference featured not just one but two research papers that included pitch control modelling. One came from FC Barcelona's Javier Fernández alongside then-Sacramento Kings VP of Strategy and Analytics Luke Bornn[2][3]. The other was by Spearman himself[4].

Fernández and Bornn's pitch control method took into account player location and velocity, conceptually quite similar to the model that Spearman had presented in 2016. There are some differences on the computational power and data requirements of the two 2018 models though, with Fernández and Bornn's aiming to be a little lighter.

Things diverge even more from there.

Both 2018 papers realised that the 'basic' pitch control model that had debuted two years earlier could be improved on. For example, that original pitch control field had been based on the hypothetical idea of a football placed at every location on the pitch and who would get there first. This method also gave huge amounts of 'control' to goalkeepers, which might be factually accurate but not exactly useful. Each paper tackled this in different ways.

Spearman adapted Hudl's pre-existing pitch control model into a 'potential pitch control' model. The probability of who controlled what space was now affected by things like how long it would take the ball to get from its current location to all of those other locations on the field[5]. A lot of this built on work that he and the team at Hudl had done a year earlier, on a physics-based approach to the likelihood of passes being intercepted[6].

Fernández and Bornn's approach, meanwhile, was a two-layered one, and is the point at which we diverge from pitch control as a model in and of itself.

They first started with the assumption that the defensive team's positioning is indicative of the most valuable spaces on the field at that particular moment in time[7].

Two example visualisations of 'pitch value' from Fernández and Bornn's paper using the above method (p.8)

However, it's also clear that that isn't quite enough. In the first visualisation of the above image, with a ball placed inside the left-hand penalty area, the position of maximum value is within the left-hand half of the pitch, quite far from the goal that is being hypothetically defended on the right-hand side.

The second way that Fernández and Bornn overcame this 'relevancy' problem of the basic pitch control calculation was by adding a layer to take into account the distance to goal.

Visualisations from Fernández and Bornn's paper showing the distance to goal pitch value and it's effect on the example when the ball is in the left-hand penalty area

In the rest of the 2018 paper, Fernández and Bornn use this 'pitch value' to look at who occupies these valuable spaces that their model can identify. There's a good article by Bobby Gardiner on FiveThirtyEight about some of the applications and findings of the paper here.

Spearman also added a couple of extra layers to his (potential) pitch control model to look at the value of areas on the pitch. The approach he took was pretty different though. While one of his additional layers focused on scoring probability from areas of the pitch (and so looking similar to Fernández and Bornn's 'distance from goal' layer), the other was about where the ball was likely to go next[8].

While Fernández and Bornn looked at a 'purer' kind of pitch value, Spearman's model was about – as he dubbed it – off-ball scoring opportunity. A subtle but significant difference.

From this point onwards, the work has been more about the 'value' side than the control. Fernández and Bornn teamed up again the following year (along with Dan Cervone, then of MLB's Los Angeles Dodgers) to produce a framework for 'expected possession value'[9]. Things didn't stop there by any means, but it's just an example of how things keep moving.

In 2020, meanwhile, a pitch control model was opened up to the world through Friends of Tracking. You can watch the video here, with a link to their Github page for the code here.

The full extent of 'what came next' is a subject for another newsletter, but before this one ends we should take note of some of what came before Spearman's 2016 OptaPro forum presentation.

From 1996 to 2000, there were a series of influential papers by reseachers Tsuyoshi Taki and Jun-ichi Hasegawa[10]. The papers (cited in several of the works I've already mentioned) are astonishing for their breadth, encompassing the development of the tracking technology needed to collect the data as well as the methods to analyse it. All this around a decade and a half before the OptaPro analytics forums even started (in 2012).

They too were building on existing knowledge about applying 'Voronoi tessellation' (by drawing lines marking the boundary exactly between data points) to get a sense of 'control' of an area. This, as Taki and Hasegawa pointed out, assumes each player would take the same time to get to this 'boundary line', when in reality players are moving at different speeds starting in different directions. Just like expected goals, pitch control has a much longer history than you might think.

Notes

Footnotes

[1] || "In 2016, the OptaPro analytics forum featured a presentation that that would mark a new mini-era in football analytics." || Slides of the presentation are available here, but the video of the presentation appears to have been taken offline by Opta.

[2] || "...it weighs up the probability of each team controlling the ball if it was in that location" || As Spearman's 2016 OptaPro forum presentation only has slides available currently, there isn't much detail on how this formulation is done. However, for a detailed comparison of different methods, see Ulf Brefeld, Jan Lasek, Sebastian Mair, 'Probabilistic Movement Models and Zones of Control', copy of the paper available here

[3] || "...alongside then-Sacramento Kings VP of Strategy and Analytics Luke Bornn" || Bornn was also an assistant statistics professor at Simon Fraser University at the time, and had been Head of Analytics at Roma for a season prior to switching to the NBA and joining the Kings.

[4] || "One [paper] came from[...]Fernández [...and...] Bornn..." || Javier Fernández and Luke Bornn, 'Wide Open Spaces: A statistical technique for measuring space creation in professional soccer", MIT Sloan Sports Analytics Conference 2018; copy of the paper available here -- for pitch control method, pp. 3-6.

[5] || "Two years later, Spearman's paper at the MIT Sloan Analytics Forum..." || William Spearman, 'Beyond Expected Goals', MIT Sloan Sports Analytics Conference 2018; copy of the paper available here -- for pitch control method, pp. 3-7

There's also a video Spearman did for Friends of Tracking in 2020 in which he talks about the pitch control model and the things this paper introduces [link should go to the correct time; if not, the relevant section starts around 8:35]

[6] || "...on a physics-based approach to the likelihood of passes being intercepted" || William Spearman, Austin Basye, Greg Dick, Ryan Hotovy, Paul Pop, 'Physics-Based Modelling of Pass Probabilities in Soccer', MIT Sloan Sports Analytics Conference 2017; copy of the paper available here.

[7] || "They [Fernández and Bornn] started with the assumption the position of the defensive team is indicative of the most valuable spaces on the field at that particular moment in time." || 'Wide Open Spaces', pp. 7-8.

[8] || "While one of his additional layers focused on scoring probability from areas of the pitch [...] the other was about where the ball was likely to go next." || 'Beyond Expected Goals', pp. 7-10.

[9] || "...to produce a framework for 'expected possession value'."" || Javier Fernández, Luke Bornn, Dan Cervone, 'Decomposing the Immeasurable Sport: A deep learning expected possession value framework for soccer', MIT Sloan Sports Analytics Conference 2019; copy of the paper available here.

[10] || "...a series of influential papers by reseachers Tsuyoshi Taki and Jun-ichi Hasegawa." || There are numerous papers that are relatively similar and build incrementally on each other, and I can't find links for all of them, but I think any and all are worth looking at if you're able. They are:

(1996) 'Development of Motion Analysis System for Quantitative Evaluation of Teamwork in Soccer Games', Taki, Hasegawa, and Teruo Fukumura, Proceedings of 3rd IEEE International Conference on Image Processing [link to IEEE page for the paper here]
(1998) 'Dominant region: a basic feature for group motion analysis and its application to teamwork evaluation in soccer games', Taki and Hasegawa, Proceedings of the Society of Photo-Optical Instrumentation Engineers [link to the SPIE page for the paper here]

(2000) 'Visualization of dominant region in team games and its application to teamwork analysis', Taki and Hasegawa, Proceedings Computer Graphics International [link to IEEE page for the paper here]

Finding Messi-esque dribblers and making choices while clustering

Thu, 20 May 2021 00:00:00 GMT

Lionel Messi is to dribbling a football what Vincent van Gogh is to layering colour with oil paints: not simply a conventional expert, but with caresses of genius that mark the work as something different. Messi's ball-carrying and close control isn't simply effective, it's beauty -- the fact that it's both is what makes him so unique.

But just how unique is that?

Knowing full well that it wouldn't be a perfect system, I wanted have a go at answering the question, to see if I could find the players that carried the ball the most-similarly to the Barcelona maestro.

One imperfection is insurmountable: the type of breadth of data. I'd be using StatsBomb's free dataset, so I'd only be seeing ball-carrying for or against Barça in matches Messi featured in. The data also only notes events, so the grace of players will mostly be missed.

Still, the dataset gives a lot to work with, and this is where the second problem arises: how do you choose what to use?

Although data and algorithms have the air of nicely in-human objectivity, that's not quite true. The first computers got their names because they just computed maths calculations that humans were already doing by hand. They were calculators more than anything.

Using algorithms isn't too distant from that conceptually. The 'k-means clustering' technique that I'd be using takes a look at the data and sorts it into a given number of groups. You could, I'm sure, take the equations that underpin that and (with a big blackboard) do it by hand, but you'd be going for a very long time.

The human element is the inputs. I knew some things that I didn't want this clustering analysis to result in as I started.

I didn't want it to turn into 'who dribbles in the same areas as Messi' -- I wanted to leave open the option that someone might be particularly Messi-like while spending their time close to the centre-circle
I didn't want it to have too many different clusters -- I wanted the results to be fairly easy to digest

In the formulation that I ended up with -- which I'll detail in a moment -- it turned out that the most similar to Messi was Ronaldinho, which was a comforting result.

After a lot of deliberation, I ended up using three variables to cluster on:

The number of take-ons in a sequence of ball-carrying[1]
The 'directness' of a carry (the progress towards goal as a percentage of the carry's length, so if you moved 5 yards towards goal in a carry of 10 yards, that'd be a directness value of 0.5)
Whether the carry was 'long' or 'short', decided somewhat arbitrarily as a carry over 20 yards or not[2]

The reasons why I chose these related to both Messi and the data in general. When I think of the Argentine's dribbling, I think of him getting past players in tight spaces, and generally being pretty effective (direct). The long/short binary was much more of a trial and error though.

It seemed clear to me that any analysis of ball-carrying had to include some sort of indication of distance carried. It might not be a major thing I associate with Messi, but it is clearly a significant differentiator between different dribbles.

The problem I found with this is that it became too big an influence on the clustering, relative to how important I wanted it to be, pretty much swamping anything else. That might be a more accurate reflection of the distribution within the data, but it didn't feel like an accurate reflection of the problem I was trying to find answers for.

So: number of take-ons (which more or less coalesced into 'presence of take-ons or not'); directness; long/short.[3] I chose seven clusters, ran the clustering with kmeans in R, and then applied those clusters to every carry sequence in the StatsBomb Barcelona dataset.

From there, it's easy to find what clusters make up what percentage of each player's ball-carrying, and therefore to find a hypothetical 'average' player. This average player isn't important for anything other than comparing others to. For example, here's a graphic for Messi's ball-carrying clusters ordered by how they differ from the average.

A graphic showing the seven cluster types and how Messi differs from the average -- his short dribbles involving take-ons make up a particularly larger share of his dribbles than average

The visualisation above shows a pitch for each cluster type of dribbles, with some examples of Messi in each -- dots represent the start of the carry and blue dots represent take-ons. There are 100 dribbles on the graphic as a whole, with the number split between the seven clusters according to how much of a percentage of Messi's dribbles those clusters represent. (For example, the 'backwards, take-ons' cluster has a small number of dribbles shown because it makes up a small percentage of Messi's dribbles, but that percentage share is larger than it is for the average player).

With that information, we can find players who are similarly different-from-the-average. We can see that Ronaldinho shares Messi's relative propensity for short carries with take-on events, even though these seem slightly different (less central) than the Argentine's.

A graphic showing Ronaldinho's seven cluster types, ordered by difference in percentage share of his dribbles to the average. The cluster where he has the lowest share compared to the average are short take-ons that are towards goal but not direct.

There are, however, some downsides to these results -- at least, compared to my starting idea. It turns out that the location of the pitch and the type of dribble are pretty strongly connected (although in hindsight this seems obvious to say). Longer dribbles, by necessity, are more likely to start from deep; take-on carries are likely to happen high up the field or close to the wings. It's therefore unlikely that I'll find a weird central midfielder who just so happens to dribble 'like Messi' in the centre-circle.

However, it's still fun playing around with the results. The next person on the list[4] is one who wasn't a teammate, Sergio García (not the golfer). He only made 123 carry sequences when playing against Messi, but those 123 had a similar balance of types.

Sergio García's cluster graphic; his carries may come in different areas, but the distribution of the clusters as a share of García's total is similar

Despite the downsides, these clusters do achieve much of what I wanted them to. They seem to cover the different types of dribble enough, although I'm sure they could be improved upon. They're also pretty simple to understand, such that I was able to give them short names to put in the graphics' headers.

Perhaps if doing it again from scratch I might take out take-ons as a feature of the clustering. Although it's something I associate with Messi, I'm not sure how I'd characterise the difference between take-ons of a similar length with and without them. (The number of pressures encountered on the route might be a more interesting/worthwhile factor here).

This is, in the end, a reminder of the decisions that are made when dealing with data. That just because something has involved an algorithm, doesn't make it an objective truth. There are always touches of the analyst in there. In some cases, even (not mine), touches of the artist.

[1] || "...The number of take-ons in a sequence of ball-carrying" || StatsBomb's data has 'Carry' events which start in one location and end in another as well as 'Dribble' events, which are the moment of a take-on. For this analysis I strung these together where applicable, i.e. if a player carried the ball, completed a dribble, then continued onwards this all became one 'carry'

[2] || "...Whether the carry was 'long' or 'short', decided somewhat arbitrarily as a carry over 20 yards or not" || To be honest, 'yards' is a bit of a fudge here. I intended to convert StatsBomb's 120x80 coordinate system into the dimensions of the average/FIFA standard pitch (105x68 (which is metres anyway not yards)), but for various reasons I didn't get to it. The principle of above/below 20 'units' of measurement applies.

[3] || "So: number of take-ons[...]; directness; long/short." || I should also add that there were a couple of other filters I used for my clustering. Carry sequences had to be five 'units' or longer (see [2]) OR have at least one attempted dribble event in them. They also needed to end in the opposition half. I realised quite early on that including carries in the defensive half, where teams are usually under much less pressure, massively skewed things in a way that was very far away from what I was interested in.

[4] || "The next person on the list..." || I limited the results to players who had 50 or more carry sequences in the dataset. Bear in mind that these are matches played with or against Messi in La Liga, so non-Barça players will probably have needed to play several matches against Barcelona to qualify.

Analytics is older than you think: (re)introducing Charles Reep

Thu, 13 May 2021 00:00:00 GMT

A blog by Sam Green from 2012 is often credited as kicking off the ‘expected goals era’ of football[1], but the first xG calculations are surprisingly older. As in, ‘almost as old as the Premier League’ older.

The adage that there’s nothing new under the sun probably holds true for this earlier work as well[2], but check out this diagram from a 1997 paper. Look familiar?

Screenshot from 1997 paper, of probability-of-scoring rings around the 18-yard box

This ‘weighted shots’ work began with a mere observation that shots taken close to goal and central went in more often than shots from range, but the calculation grew from there. In the paper, ‘Measuring the Effectiveness of Player Strategies at Soccer’, a number of factors were considered in the further analysis: distance from goal; angle to goal; first-time shot or not; less than a yard to the nearest defender or not; shot from open play or a set play. As the caption above indicates, kicked shots were also separated from headers.

My friends, this is, in some ways, more sophisticated than some of the earliest xG models from the ‘expected goals era’ that would come a decade and a half later. The authors of the paper: Richard Pollard, and one Charles Reep.

Charles Reep has a history even muddier than expected goals. He was indisputably a pioneer of data collection, starting in 1950 with a pen and a notepad (and sometimes, at night games, a miner’s helmet). He worked with Brentford and then Stan Cullis’ Wolves in the 50s, a decade in which the Midlands side won three league titles.

But Reep’s loudspoken articles in the 1960s, principally in World Sports magazine, looked like he was misinterpreting his own data. They highlighted how many goals were scored from short sequences of possession, but implied that this made short sequences more effective without noting that short sequences were primarily just more common.

Since then, his name has become associated with what happens when data analysis can go badly.

The work of Reep’s career in totality is fascinating though, and his position within the English game says something not just about it but about us. We’ll get to that 1997 paper again soon, but let’s take a brief step back to where it all began.

Reep is often referred to as an RAF veteran, usually with his rank of Wing Commander, but he was an accountant by trade before joining the military. That does make all of this data-recording make a bit more sense, the accountant airman.

He was inspired by a talk given by then-Arsenal captain Charlie Jones in 1933 about Herbert Chapman’s tactics, but it was 1950 that he first started creating data. This being the mid-twentieth century, Reep had to physically be at the games to do this, although was later able to record data from major matches via the magic of television.

This physical necessity, and the ‘completeness’ of the collecting, are unlike modern collection practices of course. A 1968 paper[3] which uses some of the data gives an insight into its limited breadth — 12 Wolves games from the 1953/54 season; 15 miscellaneous games in 55/56; two full seasons of Sheffield Wednesday (while working for them); 18 miscellaneous games… It goes on like this through the whole 1953-1967 period.

It’s slightly difficult to know exactly who, in the professional game, Reep worked with during his life. He definitely did work for Brentford, Wolves, and was full-time with Sheffield Wednesday for three seasons after retiring from the RAF.[4] He met Graham Taylor several times, rather than a more formal collaboration, but has been erroneously attributed as working with Cambridge United, Sheffield United, and Wimbledon.[5]

So while Reep wasn’t a string-pulling svengali of Ye Olde English long-ball football, his public writings perhaps made it easy to think he was (several people who were inspired by him did work more extensively in football too, but more on that in a moment).

While in the 21st century the ‘analytics community’ was more aligned with possession-based football, Reep was anything but. Thanks to a 2018 Duncan Alexander tweet, we can read one of Reep’s World Sport articles in full (Alexander bought the magazine as part of writing his 2017 book Outside the Box).

“Reflection upon the significance of all this,” Reep wrote in 1962, referring to his data, which was reproduced in a table in the magazine, “must cause one to question many tactics now widely praised as ‘good football’.” The article ends in a series of questions, culminating in this:

Have British observers been deceived for years by too readily accepting the assurances that “Continental style” football is superior to the English direct style (as it was before 1953*) and therefore has to be imitated?

Statistics prove the answer to all these questions is definitely YES.

* [I think by ‘as it was before 1953’ Reep is referring to the English style prior to that date]

Over 40 years later, Reep’s analysis was critiqued by two foundational books for the modern analytics ‘movement’, David Sally and Chris Anderson’s The Numbers Game, and Jonathan Wilson’s Inverting the Pyramid. The latter book gives the most straightforward and succinct summary of the problems with what Reep’s analysis appeared to be:

If, as [Reep’s] figures suggest, roughly 80 percent of goals result from moves of three receved passes or fewer, but 91.5 percent of moves consist of three received passes or fewer, then it surely follows — even within the unsubtle parameters Reep sets out — that moves of three passes or fewer are less effective than those of four or more.[6]

However, recent years have seen Reep’s influence on English football massively overstated. Christoph Biermann’s 2019 book Football Hackers, says “[i]n England, Charles Reep’s long-ball dogma influenced the game for a number of years even though it was based on wrongful interpretations of the data.”[7] One 2016 FiveThirtyEight article ran the title ‘How One Man’s Bad Math Helped Ruin Decades Of English Soccer’. There’s a danger that we slip into thinking that Reep is responsible for England’s long-ball history and tactical insularity.

His analysis was faulty, but any institutional ruination of English football doesn’t lie with the accountant airman. There’s an argument that we could blame Charles Hughes, who was inspired by/ripped off Reep’s work (depending whether you asked Hughes or Reep) and was the FA’s director of education and coaching between 1983 and 1994.[8] It seems likely he had an influence, but (particularly given that I’m affording a nuanced take of Reep here) it seems a stretch to lump all the blame on Hughes.

This is partly because, importantly, it’s clear from Reep’s own experiences that he wasn’t alone in his beliefs on the game.

The previous quote from his World Sport article, about ‘Continental possession football’, could have been written at any point from then, in 1962, to now, in 2021. Back in the 1950s, Wolves manager Stan Cullis already believed in the long-ball approach that Reep waved a spreadsheet at to back up. Reep surely approached Graham Taylor in 1980 for the same reason. Presumably he recognised a kindred spirit (who might listen to a man trying to introduce statistics into football) and figured that might be a rare opening to work in the game that so clearly fascinated him.

England has always been welcome to the type of football that this pioneer of analytics was proposing. However, we should note that it’s not like English football ever totally succumbed to it. In 1983, an article in The Times which was Graham Taylor’s Watford, as well as the Charleses Hughes and Reep, notes:

There were not a few insults inside the Cottage [Craven Cottage], directed from Fulham’s supporters among a 22,000 crowd at Watford’s tactics, which are persistently breaking with currently accepted thinking, weathering the criticism from such as Keith Burkinshaw [Tottenham manager] and Malcolm Macdonald [Fulham manager] and doing very nicely, thank you.

The article later introduces Charles Hughes with the sub-clause “whose theory turns what is supposed to be the fundamental principle of Liverpool’s success, possession football, on its head.” At the time that the journalist, David Miller, was writing, Liverpool were on course to win their sixth league title in eight years, a period of time that had seen them win three European Cups already. A fourth would come in 1983/84.

English football, therefore, knew the success that a more possession-oriented style could bring. It’s not as if any of the ‘successful’ long-ball teams, such as Wimbledon (who beat Liverpool in the 1988 FA Cup final) who popular either. England didn’t embrace them as a proud embodiment of the nation’s footballing philosophy.

So no, Charles Reep didn’t ruin English football. And so we enter the second half of his career — the co-author of academic papers — and approach the groundbreaking work we saw earlier.

If it’s hard to know who Reep worked with in the professional game, it’s not exactly easy to know what work he did on the papers either. The first appeared in 1968, then another in 1971, and some more in the intervening years until the one this post opened with in 1997.

Those two early papers are tables of data more than anything. From a modern viewpoint they look quaint, but — given that both are now half a century old — they’re absolutely foundational.

The first (Reep partnering with Bernard Benjamin[9]) is primarily a presentation of Reep’s data around length of passing moves and which passing moves led to shots. Pass moves got less and less frequent with each added pass; goals from moves starting in the final quarter of the field accounted for 50+% of all goals; things like that.

The 1971 paper (with Benjamin and Richard Pollard) was more academically mathematical in output, noting that the distribution of passing move lengths (1 pass, 2 passes, 3 passes, etc) seemed to have a ‘negative binomial’ pattern. The pattern isn’t really important in itself, but it’s statistical knowledge about football being committed to the record, and that is important.[10]

The culmination, in 1997 (when Reep was 93), wasn’t merely a precursor to expected goals. From his early days analysing the game, our accountant airman had noted that obtaining possession of the ball in different areas of the pitch had a big impact on chances of scoring. The 1997 paper expanded on this, introducing a (very) rudimentary kind of expected possession value system, based on where a sequences of play began. Pollard and Reep called it the ‘yield’ of possessions.

Here’s an excerpt from the paper: “In terms of probability, the yield of a team possession is the estimated probability of scoring a goal minus the estimated probability of conceding a goal, based on the outcome of a possession.” A yield of 0.025, they write, would mean that 1000 possessions starting in that particular zone would lead to 25 more goals scored than conceded.

The paper even included some analysis of different tactical approaches for different types of possession:

[Image seems to have been lost when transferring website providers]

The sample sizes are, perhaps, quite small but consider that this is 1997. This was over a decade before something as simple as possession percentage became a statistic of discussion in the public realm. A number of the people reading this newsletter will not have even been born when this was written.

Why didn’t it have more of an immediate impact? Possibly availability of the paper; possibly availability of large datasets to the people who were likely to have read it. One of the main takeaways I have from reading this work — and other early analytics research — is that a lot of the ideas existed before, but the data and technology to allow the exploration of that idea might not have. History usually takes its course on the technology side, and professional data collecting operations sprung up too.

However, there’s another takeaway I get from the story of Charles Reep that can have a more immediate impact regardless of where history is: the significance of collaboration.

I’m currently reading The Innovators by Walter Isaacson; the book is subtitled ‘How a group of hackers, geniuses and geeks created the digital revolution’ but Isaacson has a deeper motive than just history-telling. In the book’s introduction he writes:

This is the story of these pioneers, hackers, inventors, and entrepreneurs — who they were, how their minds worked, and what made them so creative. It’s also a narrative of how they collaborated and why their ability to work as teams made them even more creative.

The tale of their teamwork is important because we don’t often focus on how central that skill is to innovation. There are thousands of books celebrating people we biographers portray, or mythologise, as lone inventors[…] But we have far fewer tales of collaborative creativity, which is actually more important in understanding how today’s technology revolution was fashioned.

I think the same may be true of Charles Reep.

While his sometime co-author Richard Pollard fiercely defends Reep in a 2019 paper[11], the former accountant/Wing Commander didn’t do himself any favours in his early ‘60s articles. The papers co-authored with Benjamin and then Pollard contain much more focus, and much more considered insight. Reep may well have brought some of this himself[12] but it seems reasonable to say that his co-authors also brought something to the table (Bernard Benjamin became president of the Royal Statistical Society in 1971; Richard Pollard completed a PhD in statistics applied to football analysis in the 80s).

From the time I entered into the football analytics sphere around 2013, little groups of people have always been around advances. The StatsBomb blog brought a lot of people together either to write or at least to read under one roof. Companies like Opta and Prozone/STATS (now jointly Stats Perform), Hudl, 21st Club (now Twenty First Group), Decision Technology (and others I’ve probably missed) had groups of people working on problems. American Soccer Analysis is probably the best current, public example of collaboration feeding creativity and analysis.

Individuals may produce good work on their own, but it tends to be when they gravitate to discuss their ideas with others that these get refined and improved, made applicable to the game or pushed even further.

Charles Reep was a pioneer of performance analysis and a driven, methodical individual. But while driven and methodical individuals can do good work on their own, it’s usually when they have someone to bounce their ideas off that it becomes great work. A 1997 paper containing rudimentary expected goals and expected possession value calculations…?

That’s great work.

Notes

Footnotes

[1] This wasn’t the first work done on xG in football (related: footnote [2]), but I would argue that Green’s blog does hold this status. Part of this is probably because of its place on Opta’s blog pages — which have a larger prominence and reach than other blogs and papers — and of who’s principally been telling the history of it (i.e., them).

Take this from Opta’s head of editorial Duncan Alexander in a 2017 FourFourTwo article as an example: “Opta first came up with the concept of expected goals when one of our data scientists – Sam Green[…]– devised an analytical model based on similar things being done in American sport”.

I don’t begrudge Opta for this, although the people doing work around 2009-2012 might feel otherwise, but it’s seeped into the story of the stat. In the 2019 book, Football Hackers: The Science and Art of a Data Revolution, this is how xG is introduced: “‘Expected Goals’ was invented by the Englishman Sam Green, who first described the idea in 2012.” The book The Expected Goals Philosophy, a book specifically about expected goals and emerging as the central (only?) pop-science book on the metric, says similar: “In April 2012, an analyst called Sam Green posted a blog article on the OptaPro forum[sic*] […] introducing the idea that shot quality might be just as important as shot quantity.”

*This may be getting confused with the ‘OptaPro Analytics Forum’, an annual conference (since 2021 known as the Stats Perform Pro Forum); I don’t believe that OptaPro’s blog was ever a ‘forum’ in the sense of being set out similar to how e.g. Reddit is. [link to Wayback Machine recording of the original blog post, albeit from 2015]

[2] The Wikipedia page for ‘Expected goals’ (link here) notes a 1993 paper that investigated the effect of artificial pitch surfaces. This appears to have been a use of the phrase ‘expected goals’ more than an ‘expected goals’ model, however. There are a number of works between 1997 and the 2012 Sam Green blog, including two works from 2004: one by Jake Ensum, Richard Pollard (a co-author of the 1997 paper), and Samuel Taylor; and one in ice hockey by Alan Ryder.

[3] C. Reep and B. Benjamin, ‘Skill and Chance in Association Football’, Journal of the Royal Statistical Society, 131, no. 4 (1968), pp. 581-585.

[4] See Keith Lyons’ 1997 paper on Reep, ‘The Long and Direct Road: Charles Reep’s analysis of association football’, a reproduction is on his blog here.

[5] Jonathan Wilson, Inverting the Pyramid; Richard Pollard, ‘Charles Reep (1904-2002): pioneer of notational and performance analysis in football’, Journal of Sports Sciences, 20 (2002), pp. 853-855.

[6] ‘Chapter Eight: The English Pragmatism (1)’

[7] Chapter ‘Revolution’

[8] Inverting the Pyramid, ‘Chapter Fifteen: The English Pragmatism (2)’

[9] Benjamin would become the president of the Royal Statistical Society in 1971

[10] It wasn’t just football where negative binomial distributions appeared, the 1971 paper found these patterns in cricket (runs scored), ice hockey (goals), baseball (runs per inning), and to a lesser extent in tennis (length of rallies) too.

[11] Richard Pollard, ‘Invalid Interpretation Of Passing Sequence Data To Assess Team Performance In Football: Repairing the tarnished legacy of Charles Reep’, The Open Sports Sciences Journal, 12 (2019), pp. 17-21.

[12] The people I’ve read who personally knew Reep all speak quite glowingly of him, although it seems feasible that a degree of this may be defensiveness of a friend whose intellectual reputation has taken a battering.

Bibliography:

Directly referenced, main body

Sam Green, ‘Assessing the performance of Premier League goalscorers’, OptaPro blog (April 2012) [link to version on Stats Perform’s current blog here] [Wayback Machine link to OptaPro’s old site, albeit a 2015 version, here]

Richard Pollard and Charles Reep, ‘Measuring the Effectiveness of Playing Strategies at Soccer’, The Statistician, 46, no. 4 (1997), pp. 541-550

C. Reep and B. Benjamin, ‘Skill and Chance in Association Football’, Journal of the Royal Statistical Society, 131, no. 4 (1968), pp. 581-585

Duncan Alexander tweet [with photos of a 1962 Charles Reep World Sports article] (2018), link here

Duncan Alexander, Outside the Box: A statistical journey through the history of football (2017)

David Sally and Chris Anderson, The Numbers Game: Why everything you know about football is wrong (2014)

Jonathan Wilson, Inverting the Pyramid: The history of football tactics (2008)

Christoph Biermann, Football Hackers: The Science and Art of a Data Revolution (2019)

Joe Sykes and Neil Payne, ‘How One Man’s Bad Math Helped Ruin Decades Of English Soccer’, FiveThirtyEight website (October 2016 – NB: writers don’t write headlines) [accessed May 2021, link here]

David Miller, ‘The possession game turned on its head’, The Times, 3 February 1983, p. 20 [link to archived edition here]

Walter Isaacson, The Innovators: How a group of hackers, geniuses and geeks created the digital revolution (2014)

Richard Pollard, ‘Invalid Interpretation Of Passing Sequence Data To Assess Team Performance In Football: Repairing the tarnished legacy of Charles Reep’, The Open Sports Sciences Journal, 12(2019), pp. 17-21.

Directly referenced, footnotes (if not previously referenced)

Wikipedia, ‘Expected goals’, link here

James Maw, ‘No, seriously: what the heck is expected goals (xG)?’, FourFourTwo (November 2017 issue) [link to online version here]

James Tippett, The Expected Goals Philosophy (2019)

Richard Pollard, ‘Charles Reep (1904-2002): pioneer of notational and performance analysis in football’, Journal of Sports Sciences, 20 (2002), pp. 853-855.

Keith Lyons, ‘The Long and Direct Road: Charles Reep’s analysis of association football’, reproduced on his blog here – the original paper was written in 1997, but this reproduction and caveats in 2011/2012.

Not directly referenced

Alan Campbell, ‘Don’t Shoot the Messenger: The first football analyst was a pioneer 50 years ahead of his time’, Nutmeg Magazine, issue8 (2018) [accessed online 2021, link here] [Note: Campbell was founder and editor of The Punter magazine which published articles from Reep in the 1980s and 90s]

Charles Reep, ‘The great Magyar myth exploded’, The Times, 29 May 1982, p. 18 [link to archived edition here]

Barney Ronay, ‘Grim Reep’, When Saturday Comes, issue 196 (2003) [accessed online 2021, link here]

Øyvind Larsen, ‘Charles Reep: A Major Influence on British and Norwegian Football’, Soccer & Society, 2, vol. 3(2001), pp. 58-78

Keith Lyons blogs, ‘Bernard Benjamin’ (2015) [accessed 2021, link here] and ‘The Charles Reep and Bernard Benjamin Paper 50 Years On (1)’ (2018) [accessed 2021, link here]

Soccer at Sloan 2021: A summary

Mon, 03 May 2021 00:00:00 GMT

Greetings readers, it’s time for some analytics.

Last month was the MIT Sloan Sports Analytics Conference. In pre-pandemic times, this was one of the big ones of the sports analytics conference circuit. If you’re into tennis, think of the Slams — it was one of those. (Also, if you’re into tennis, which of the Slams was it most like? Twee, traditional Wimbledon? Fun, ‘Happy Slam’ Aussie Open?).

Although it’s an American conference with US sports taking prominence, soccer has long been represented among the research papers submitted. This year, soccer-related papers were judged first- and second-best in the research paper competition too. I’m going to be summarising one of them, and two other soccer analytics papers, here. First will come a brief summary for those just interested in the headlines from researchers at the forefront of the field, and then I’ll tackle each in a bit more detail for those who want it.

The papers I’m looking at are:

‘Routine Inspection: A playbook for corner kicks’ by Laurie Shaw (Harvard University) and Sudarshan Gopaladesikan (Benfica) [link here]
‘Making Offensive Play Predictable: Using a Graph Convolutional Network to understand defensive performance in soccer’ by Michael Stöckl, Thomas Seidl, Daniel Marley, and Paul Power (Stats Perform) [link here]
‘Leaving Goals On The Pitch: Evaluating decision-making in soccer’ by Maaike Van Roy, Pieter Robberechts, Wen-Chi Yang, Luc De Raedt, and Jesse Davis (KU Leuven) [link here]

I’ll note here that there may be a chance I’ve misunderstood something in the papers. In that case, check the online version of this for updates. (If I make any really bad errors I may send a clarification email out)

The lowdown

I can’t think of many unifying features of these three papers so this lowdown will be slightly fractured. Each paper uses looks at different things, using different techniques, and seeking to do different things as well.

Making Offensive Play Predictable from the folks at Stats Perform is, as a research paper, kind of a ‘foundational’ work. Using tracking data, it creates three models — two fairly familiar to the analytics field (expected pass completion, expected threat), one quite original (expected receiver) — and tantilises us with their combined uses in analysis. It also introduces the detection of ‘active runs’, where a player makes a burst to notably increase their likelihood of being the player who’ll receive the next pass. In terms of defending, you can see when defences are forcing attackers to make active runs away from goal (i.e. dropping deep) to receive; when players are forced to change their mind about who to pass to; and whether a defending team is forcing their opponents to pass to different areas or make less safe passes than they usually do.

Routine Inspection is also somewhat foundational, but in a much more specific way. Instead of creating models or metrics that are used in open play, it creates a kind of dictionary of run types used at attacking corner kicks. On the defensive end of things, the paper also identifies whether players are marking zonally or player-to-player.

Meanwhile, Leaving Goals On The Pitch is much more focused on application, specifically on whether teams should actually shoot more from distance. Since the rise of expected goals it’s been analytics orthodoxy that teams should shoot less from range, but this paper sought to see which situations it was actually better to take the shot. Spoilers: if you’re a bad team, it’s probably not worth hanging onto the ball if you’re already in range; and, in general, the benefit of shooting more is quite small, but there.

Together, these three papers really do give a good thematic indication of what ‘analytics’ can do. It can poke an assumption about the game and seek to confirm or refute it; it can run tasks to free up time for analysts to do other work; and it can create models and metrics to help further analyse the game.

And now for a research paper minute-by-minute liveblog

Well, not really, but a more thorough summary. I’ll tackle them one by one, and hopefully this won’t run longer than newsletters are permitted to go.

A playbook for corner kicks

‘Routine Inspection: A playbook for corner kicks’ by Laurie Shaw (Harvard University) and Sudarshan Gopaladesikan (Benfica)

[link here]

Although I skimmed over this paper in the summary above, it was the deserved winner of the research paper competition. What it does doesn’t sound like a sparkly headline, but it seems like really meaningful work.

To start with, Shaw and Gopaladesikan identified the ‘target locations’ of players’ attacking runs at corners. ‘Target location’ was defined as where the player was either one second after the first on-ball action of the corner, or two seconds after the corner was taken, whichever came first. The researchers split these locations into seven clusters of ‘active’ players, with other locations outside the box not considered as part of the analysis.

Once they’d identified the players who were actually involved in making runs at the corner, they used them to identify starting location clusters too (they found 6). Both of these used a Gaussian Mixture Model.

There were two things I found interesting about this. The first is that they were computationally finding these zones, rather than using a coach-led system. One isn’t necessarily better than the other, but I think it’s always worth noting when these decisions are made. The second is that, despite using tracking data, Shaw and Gopaladesikan weren’t using the paths of player runs, as some NBA research has done. I imagine that doing it in this method makes the computation easier, but corners have always seemed to me like the part of the game that would most likely cause problems in the tracking data (and I imagine it could depend a lot on your provider).

Those six starting and seven target zones gave Shaw and Gopaladesikan 42 possible runs (hello to Douglas Adams). However, runs usually occur together in regular patterns. The researchers used non-negative matrix factorisation to help create 30 run combinations, or ‘features’. Each of the 1723 corners in their sample could be constructed using combinations of these 30 features.

The next part of their paper does use domain experts alongside pure computerising. Shaw and Gopaladesikan decided that, for coding defensive corners, they wouldn’t focus on deciding whether the system as a whole was ‘zonal’ or not — after all, as they point out, systems are rarely wholly zonal or wholly player-to-player. Instead, they sought to work on the individual player-level.

Analysts at Benfica worked with the researchers, doing two things:

come up with metrics that could be used to predict the role of a defender at a corner, to be used as parameters for a model
watch 500 corners and tag whether the defenders were marking zonally or player-to-player (resulting in 3907 defenders) to form a training set for the model

Shaw and Gopaladesikan used XGBoost (the coincidentally named implementation of gradient boosted decision trees) and found they could determine whether defenders were marking zonally or player-to-player with a classification accuracy of 83.4% (±2.1%).

This kind of methodology allowed them to show an example in the paper, comparing four-zonal player systems and two-zonal player systems and the shots they conceded. The four-zonal player systems conceded more shots, but of worse quality.

As the two authors of the paper note, the next step would be to determine which types of attacking corner are more effective against which types of defensive corner. I also wonder whether something as simple as outswinger vs inswinger could have an impact on success of defensive systems.

Making Offensive Play Predictable

‘Making Offensive Play Predictable: Using a Graph Convolutional Network to understand defensive performance in soccer’ by Michael Stöckl, Thomas Seidl, Daniel Marley, and Paul Power (Stats Perform)

[link here]

It’s worth noting at the start here the team also made a presentation at the Stats Perform Pro Forum on this work (I wrote about that here). Their presentation at the Forum covered some different ground to the paper; a video of that presentation is here and there’s a tweet thread from Paul Power about the paper and the wider work here.

Although this paper, and the associated presentations, introduce a bunch of models and applications for them, the centrepiece of the paper itself appears to be the methodology. As the title says, the researchers use a Graph Convolutional Neural Network to create their models, and there’s a section of the paper that explains why.

Explaining what a Graph Convolutional Neural Network is, and the reasons for using one, is slightly beyond my level of understanding, but this is what I gather:

Tracking data is ‘unstructured’ — you can’t stick it in a table, which means you can’t use machine learning techniques that are based on tabular datasets
Graphs* are a way of dealing with this unstructured data
As well as this, some previous work has used tracking data frames as images. Using graphs reduces the amount of computational power required

*For those reading this who, like me, only know of one type of ‘graph’ (things like line charts) this type of ‘graph’ is simply connecting nodes with lines. In the case of this paper’s methodology, the nodes for defensive players were masked out for some of the models.

Being Stats Perform, their sample was large, 1200 matches of tracking data. That’d be a hell of a lot of frames of data, but they don’t need all of them. For the xThreat model they just used “the frames relating to the moment of passing events”, while the xPass and xReceiver models additionally included frames taken a half-second and one second prior to passes.

The reason for this is because, at the very start of a passing action, “players’ movements already indicate where the ball will be played to some degree”. Taking the half- and full-second prior to the action helps to prevent ‘overfitting’ the model.

There’s more detail on the modelling process in the paper, but they show a table with accuracy and logloss figures in comparison to similar metrics created through different methods. “The loss and accuracy of all three GNN models were better than or the same as the metrics of the respective baseline model,” they write.

After creating the models, the paper then discusses a ‘disruption map’ — essentially a heat map of the models’ results for a team that, for example, can be compared between single games and a full-season sample. Through that, you can get a sense of how well a defence performed against their opponent.

There’s a section that expands on that idea and gives examples of applications, but I’ll skip over that for the sake of space. The authors of the paper then wanted to look at whether defensive players were player-oriented or ball-oriented in their defending. Instead of trying to train a model, they got domain experts to give definitions which they could work from. This is also where they got the definition for ‘active off-ball runs’ (“an attacker moves at high speed to increase their probability of being a receiver”), which the paper then illustrates some uses of.

In terms of the methods used to deal with tracking data, this paper’s a really interesting one to look at. It offers a different approach to ones that it appears others have used in the past, while discussing some of the previous methods as part of the paper. This discussion, and the references section, make for a tremendous starting point for anyone looking to get up to speed with the field.

Leaving Goals On The Pitch

‘Leaving Goals On The Pitch: Evaluating decision-making in soccer’ by Maaike Van Roy, Pieter Robberechts, Wen-Chi Yang, Luc De Raedt, and Jesse Davis (KU Leuven)

[link here]

The third and final paper on this newsletter’s list is the one most geared towards actual findings, with some pretty interesting headlines. It features a fascinating stat early on, that high-volume long-range shooters like Christian Eriksen, Paul Pogba, Harry Kane, Kevin de Bruyne, Heung-Min Son, Eden Hazard, and Gylfi Sigurdsson combined for a long-distance conversion rate of 6.5% across the 2017/18 and 2018/19 seasons. Meanwhile, the possessions where these players made a touch in that range but didn’t shoot only resulted in a goal 2.1% of the time.

This leaves open the question about whether teams should be making more shots from distance. Doing so would potentially forego better-quality shots later in the move, but who knows if those chances would actually arrive? That’s the problem with tackling this question, you’re dealing with counterfactuals.

To address this, the researchers model how a team plays generally by taking two seasons’ worth of data and training a Markov Decision Process model to see what their tendencies in possession were.

By my understanding, this approach looks at the likelihood of a player/team either shooting from a location or moving the ball to another location, or losing possession. These ‘locations’ are, in this case, a 22x17 grid in the attacking half of the field (given that this model is interesting in scoring patterns, the defensive half is assigned as one singular zone). The paper designates ‘long-distance shooting locations’ to be more or less the width of the 18-yard box and from 18-to-30 yards from the byline, although this’ll be most important later on when we’re thinking about the counterfactual shots.

So, the researchers ran this model on 17 teams’ data (the ones who were present in the Premier League in both 2017/18 and 2018/19) so they had a 76-match sample of how each team moves the ball around the field. The first obstacle in the problem, tackled.

(Interesingly, they slip in a method for determining the intended end location of incomplete passes, using Gradient Boosted Trees Ensembles and “the characteristics of the actions and what has happened prior to the actions”).

But there’s another obstacle that the paper attempts to overcome too. If a team decides that it wants to shoot more often from long-range, there’s no guaranteeing that they’ll keep up their current quality of long-distance shots. They might just start swinging whenever they get the chance, and that might mean worse-quality attempts. What should you take as the goal probability of shots that never happened?

For each zone, the researchers looked at the distribution of xG values for shots in that zone and used it as the basis of assigning xG to ‘new’ shots. For example, if a team is increasing their number of long-range attempts, the researchers reason that the shots will likely be of lower quality and so the xG assigned is at the lower end of the distribution. If a team is decreasing the number of long-distance shots, they’ll take away ones from the lower end of the xG distribution.

And so with the model of teams’ play tendencies, and the method for adding or taking away counterfactual shots, all you need to do now is run the code (with which the researchers used PRISM, a probabilistic model checker).

The results are interesting. Yes, teams do seem to be leaving goals on the table, but not many. A uniform increase in shots across the entire ‘long-distance’ area of 20% would result in just an extra 0.5-1.0 goals per season, and only for the top-half of teams in that table. This gets better if teams are focusing on just the long-distance areas they do comparatively well in, but only increasing to 1.6 at the very top.

Interesting as well is the comparison of different teams. The paper displays a heat map of the long-distance zone for four different teams (Chelsea, Everton, Huddersfield, Man Utd) with the probability that they’ll create a better shooting chance later in a sequence of play. For Chelsea and United, there are large patches where they could have something like a 20% chance of getting a better shot in the same sequence. For Huddersfield, it looks like they barely crack 15% anywhere.

As the paper points out, an extra goal or two in a season is not nothing, even if it isn’t a lot. But, I think, this is a pretty thorough approach to a very interesting question, and one which challenges analytics orthodoxy as well, which is worth doing every once in a while.

There are questions I’d have, like what happens if a team changed tactics partway through that two season sample used in the Markov model and how that affects the tendencies. But I also feel like one has to cut some slack to people looking to investigate this kind of analytical counterfactual.

If you’re reading this text, thanks very much for (presumably) taking the time to read all of this. I truly hope that it has been useful and/or informative to some of you (although my greater hope is simply that it’s an accurate reflection of the work).

London bus analytics conferences

Thu, 25 Mar 2021 00:00:00 GMT

Honestly. You wait all year for a football analytics conference/event and then two come along in the space of a week. On 17 March we had ‘StatsBomb Evolve’ and on 24 March we had the Stats Perform Pro Forum.[1] In all the weeks in all the months in the world, they just happened to walk into almost the same one.

Both brought announcements, both brought fine analytics minds, and during both I suffered internet issues (on my end) while watching. So I won’t do a play-by-play for either. However, the post-conference dusk makes for a good time to think about what they indicate for where analytics ‘is’ and where it’s heading.

This newsletter will have a brief summary of each, linking out to the company’s material on the event. Then I’ll talk about what it all seems to mean.

The events

In chronological order

StatsBomb

StatsBomb Evolve was split into four speakers and five basic parts (after a brief ‘Previously, on StatsBomb’ intro):

StatsBomb 360 (the new product, ‘where lots of players are at the moment of each event’)
StatsBomb data collection
StatsBomb data quality control [imo these two sections were far more interesting than they sound; this is the sole bit of editorialising I’ll do in this section of the newsletter]
StatsBomb data science (mainly discussing their OBV, or ‘on-ball value’, model)
Q&A

There’s a StatsBomb summary of each section, along with select slides, here.

Stats Perform

The Stats Perform (née Opta) Pro Forum followed the same structure as previous years: presentations from members of the analytics community (two of which followed a research question proposed by a pro analyst working in the game), then a guest speaker. The presentations were:

Ola Lidmark Eriksson – Volatility and calculation of risk-adjusted return in football scouting
Aditya Kothari – A physics based measurement of defensive contributions
Caterina De Bacco – Identifying and evaluating the efficiency of each player during the pressing phase against an opponent’s controlled build-up play
Stats Perform’s very own Paul Power, Thomas Seidl, and Michael Stöckl doing lots of things with tracking data (presentation titled ‘Making Offensive Play Predictable’)
Debangan Dey, Rahul Ghosal and Atanu Mitra – Enriching event data: A semi-supervised augmentation approach using location information
Vignesh Jayanth – Identifying and evaluating strategies for successfully penetrating a high opposition press from short goal kicks, played inside the box, to move the ball into the opposition half

Laurynas Raudonius won the inaugural Dr. Garry Gelade award, named in honour of the late analytics thoughtleader and Forum stalwart, and presented his poster ‘Recognizing and evaluating opportunities in counterattacks using tracking data’. Delivering the guest talk was Mo Bobat, Performance Director for the England Cricket Board.

A more detailed description of each of the main presentations by public analysts is here.

The analytical themes

On the analytics front, I think there were two main themes from these two events.

One of them was “data that is somewhere between the two traditional camps of ‘event’ and ‘tracking’ data”. To explain what those two types are briefly: if event data is someone smooshing a paintbrush against a canvas to mark every pass someone makes, tracking data is twenty-two sets of hands[2] keeping their pens to the paper for the whole match.

Having all of that tracking data is, in theory, great. You know everything! But just like having the internet’s boundless knowledge on hand 24/7 doesn’t make you a genius, tracking data doesn’t necessarily solve all your problems either.

Approaching the problem from one end we have something like Dey, Ghosal, and Mitra’s Pro Forum presentation, ‘Enriching event data’. They had a multi-step process to take insights from tracking data and find them in event data. The potential uses for this could be for scouting purposes if you have tracking data for your own matches/league but not others: you use the tracking data that you do have to work out how to use the event data in fancy and helpful ways.[3]

Approaching it from the other direction is StatsBomb and their ‘360’ data. Instead of using tracking data to enrich event data, they’re new thing is using their event data collection to identify moments to snapshot tracking data. For every event in their dataset, they’ll essentially capture a single frame of tracking data.[4]

Instead of you having to work with lots of frames to work out, say, how close the nearest defender was when someone receives a pass, you have that single snapshot connected to each event in the data.

Both approaches, the whole theme of enriching event data, makes a lot of sense. It allows you to count more stuff (and more useful stuff at that) while not adding too much extra technical requirement in skills or computational power.

The other main theme was defending.[5] Two of the Pro Forum’s presentations were, by the title, about that side of the game, and a further two were at least pretty directly applicable.

StatsBomb also made a point of saying how their ‘360’ freezeframe data will allow for more defensive analysis. At first this’ll be on a team level (maybe holes in a team’s defensive structure, where they’re more likely to appear), and then more player-level stuff will be possible when StatsBomb assign player IDs to their 360 freezeframes.[6]

The reason for this second theme is, I think, a subliminal bonus theme: the maturity of the industry.

I feel that some of the focus on defending is coming now not necessarily because of new ideas, but because of technical or computational improvements that allow people to investigate much more easily. People have long wanted to investigate defending from the point of view of ‘space management’, it’s just been quite difficult to do.

However, nowadays best practices of storing and dealing with tracking data are being worked out and passed around; Friends of Tracking’s YouTube and Github repo have given everyone the opportunity to toy around with a pitch control model and tracking data; Stats Perform had a very fancy looking animation of their own data and models which has clearly been worked on a lot; StatsBomb have been developing the data collection processes and computer vision tech to enable them to collect the 360 data.

If this isn’t too grand a comparison, it’s a little like Europe in the early modern period: without the printing press, a lot of stuff would’ve struggled to get off the ground.[7] We now seem to have the printing press.

The business themes

But yes, these events weren’t merely for the advancement of science. I’m beginning to suspect that it wasn’t even a coincidence they were just a week apart.[8]

It shouldn’t take me to tell you what the business rationale behind StatsBomb offering more data is. I won’t do their sales pitch; I’m neither charismatic enough nor on their payroll enough. What I will say is that it’s an interesting shift in the market.

Assuming the pricing, execution, and delivery[9] are right, it’ll mean that StatsBomb are straddling the event and tracking data spaces in a fairly unique way. Of course, they’re not the only company to straddle this divide — tracking data companies generally seem to produce ‘counting’ type metrics that are essentially event data, and the other company this newsletter rests on, Stats Perform, is a merger of a tracking data and an event data company after all. But StatsBomb’s approach is far more of a ‘pure hybrid’ than others’.

Another worthwhile note from the ‘Evolve’ event is that StatsBomb are going to be making their OBV (on-ball value) model a free add-on to the rest of their data. One might wish to say that that sounds like a way of undercutting competitors who are currently developing their own model and/or planning to charge for it. One might.

However, here we should make a brief digression back into analytics to talk about what OBV actually is and therefore why we should care about it being free.

The OBV model is part of a growing family of ‘possession value-type’ models. I’m talking non-shot expected goals, expected threat, on-ball value, expected possession value, goals added, possesion value added. They all do similar things. Hell, I’ll just crib StatsBomb’s one-line explanation from their blog of the event (link again here) because it’s good and works for pretty much all of them:

…valuing every event that happens on the pitch based on how it changes a team’s likelihood of scoring or conceding.

Each model tackles things in slightly different ways and not all of them value events based on likelihood of scoring or conceding like OBV does (some more basic/specific ones just do the scoring half).[10]

Where OBV is concerned, an interesting point was made by Dinesh Vatvani, StatsBomb’s Head of Data Science, in his part of ‘Evolve’ about team strength effects. Some possession value-type models include information about ‘the possession so far’ as a feature (e.g. what happened in the previous three passes or how long the string of possession has already lasted). This is done to act as a proxy for opposition defensive structure, however Vatvani demonstrated that by including this information you may end up polluting the model with the stylistic tendencies of the strongest teams. (For clarity’s sake, this is my summation, not his words). OBV doesn’t include this ‘historic possession’ information then; some others do.

StatsBomb also briefly talked about other models they have, such as role classification, pass clustering, and an expected passing model. OBV was the main attraction though.

Back to business.

[Edit: this paragraph originally said 'the OBV model is still a bit of a WIP' - this has been amended below]

It’s worth noting here that the OBV model that will be made available for free is something of a stepping stone. Ted Knutson is clear that OBV is a worthwhile standalone model, even without the information from the StatsBomb 360 freezeframes, which it doesn't incorproate. There will be an OBV model that incorporates that data eventually, and will 'live alongside 360 data subscriptions'. That information — where a bunch of other players are on the pitch, remember — will surely be very handy to determining value of events, and gets rid of the need to proxy the opposition’s defensive structure.

So, free OBV won't include 360 data; the version that will won't be free. (It seems to me that StatsBomb’s pressure data, collected since their launch a few years ago, may already act as a bit of a proxy of opposition defensive structure anyway).

StatsBomb’s 360 data is new, even to them, and they were open during the event about the fact that they would be finding new things to do with it as time went on and as they gathered more of that type of data.[11]

On Stats Perform’s front, the presentation during the Forum by their own employees showcased a number of things:

their tracking data, which is captured in-stadium (and therefore always has all players on the pitch)
An ‘xReceiver’ model, based on how likely a player was to be the intended recipient of a pass at any given moment
An ‘xPass’ model based on how likely a pass from one player in the tracking data to another is to be completed
An ‘xThreat’ model, one in that family of possession value-type models I mentioned previously
When players are pressuring the ball, or passing options
‘Active runs’, which are when a player makes a run that increases the likelihood of them becoming an xReceiver above a certain threshold

And also, something slightly different, they’re launching a website![12] By the time you read this the site might be online, but for now I can only link to a Twitter page. Like StatsBomb, I’m sure they will produce good content with all of their various product- I mean, toys - on show. If it’s good enough for Twenty3 Sport (the people I work for) it can be good enough for them.[13]

The community themes

But while there is pure analytics and pure business involved in these events, the heart of them — and I genuinely mean this — is the community of people in the public and private analytics world.

The Pro Forum started life as a way for Opta (as it was then) to give some recognition (and some data) to the hobbyists, some of whom went on to write, and then work for, StatsBomb. During StatsBomb’s event, I thought it was a nice touch to give so much (deserved!) space to the folks at Arqam, the company StatsBomb worked with and then acquired to run their data collection operation.

On the community theme, it’s clear that people will give you time and help, and that barriers to being involved in analytics are falling. Partly aided by the virtual nature of the event, the presenters at the Pro Forum were from a varied field of backgrounds, and on more than one occasion they took the time to thank a string of people who’d helped them with their presentation.

Outside of that, there are collections of resources like Friends of Tracking, Devin Pleuler’s Soccer Analytics Handbook, and this amazing bibliography of expected goals literature. For group support, Lydia Vandenbergh Jackson and Arielle Dror have started a Slack for people from underrepresented genders interested in women’s soccer (and other sports) and data analysis. Original tweet on that group and how to access is here.

I started a slack channel with @arielle_dror for women and other underrepresented genders who have an interest in women's soccer and data analysis. You don't have to know how to code to join! If you're interested, DM me 🙂— Lydia Vandenbergh Jackson (@lvanden) June 24, 2020

Another lesson that I think can be taken away from these two ‘conferences’ is that event data ain’t going anywhere. Companies — like StatsBomb and Stats Perform — will probably increasingly add event-type stats to their data feeds which are derived from tracking data, so that you don’t have to. That means that, while it might be interesting and advantageous to know how to deal with tracking data, you don’t need to to work in the analytics space.

Finally, if you want some inspiration of work you could do, there’s an interesting specificity to some of this year’s Pro Forum presentations. Ola Lidmark Eriksson’s was about determining how good and consistent a player’s metrics are (very useful); the club-led proposals, presented by Caterina De Bacco and Vignesh Jayanth, were about very specific in-game situations. (Here’s the link to the list of presentation titles and descriptions again). You don’t need to solve football to be useful; often, just improving a small part of it by a little can help a great deal.

On where to get data: the Friends of Tracking repository has some free Metrica data; StatsBomb have made some data available; there are some free samples of Wyscout data (albeit in .txt form) here; Football Reference is a tremendous starting point as it lets you download tables easily as .csv files. [NB: I’m likely to update this paragraph with other data sources I might have missed]

I said in my review of 2020/predictions for 2021 newsletter that:

In general, then, there’ll just be more. More use of data, more people working with data, more understanding of data, more little innovations.

I didn’t even know how right I was.

[1] The name of the latter event doesn’t need quotation marks because I don’t feel embarrassed writing it.

[2] This isn’t always 22. Some companies collect the tracking data in the stadiums, but some do it based on video footage they gather from other sources (often TV broadcasts from somewhere or other). In these situations they may not — in fact, for TV footage, are likely to not — have all 22 players all the time.

[3] (or, alternatively, if someone, somewhere has tracking data and shares the features of the model around you might not need any tracking data at all)?

[4] It’s worth noting here that StatsBomb’s messaging has said that these 360 frames will have ‘every player on the pitch’, a statement which seems more like an aim than a reality. The examples in the presentation were clearly taken from broadcast footage, but on more than one occasion it was mentioned that the company are working to get more wide angle footage.

[5] Yay!

[6] At the moment, the freezeframe data they’ve had around shots (which they’ve had since launch in 2018) has player ID for all players in frame, but 360 is launching without it. As Ted Knutson noted in the event, there are a hell of a lot more non-shot events in football matches than shots. I don’t believe they gave a timescale for when this information would be added.

[7] Other historical analogies will no doubt also apply

[8] Who can forget the original launch of StatsBomb data when Opta also dropped news of some data upgrades on the same day?

[9] StatsBomb will be offering the 360 data in a feed, as well as deriving their own metrics and visualisations to put into their StatsBomb IQ software product, so that’s two different avenues for ‘delivering’ this.

[10] I want to note two things here. One is that I find this variety in pretty similar models very interesting and am very curious about how it’ll play out — I believe that things will evolve and eventually there’ll probably just be a couple of different approaches called different things under a ‘possession value’ mini umbrella. The second is that I spoke to Vatvani on Twitter about the OBV model as I wanted to clarify some things I hadn’t fully followed during the presentation in case I wrote about more details of the model here. He was very helpful and I feel bad now that I haven’t ended up writing about the model apart from its current lack of 360 data. Another part of his presentation that I found interesting was the use of OBV for players shooting, working out the difference between OBV and post-shot xG to gauge how much a shooter had made of their opportunity.

[11] I find this exciting. Like the Pro Forum, one of my favourite things about analytics events or announcements is the feeling of opportunity and exploration of how things can be used and what stories or insights they can bring.

[12] I’ve just realised an amusing symmetry. StatsBomb are moving more firmly into the event/tracking data hybrid space that Stats Perform have been operating in; Stats Perform are moving more firmly into the content marketing space that StatsBomb was kinda born out of given its origins as a blog.

[13] It’s nothing to do with us at Twenty3, but I do enjoy the coincidence that since we started producing public content to showcase our product: StatsBomb have hired a content marketer guy; Analytics FC have started content marketing across various platforms; Stats Perform are opening a website EDIT This kind of comes full circle. Content marketing isn't new, and this joining of random dots will miss a lot, but OptaJoe is long-established content marketing. Content marketing is very old, but it seems to be ramping up recently.

A fab early-February for football analytics aficionados

Thu, 04 Feb 2021 00:00:00 GMT

February 3rd & 4th 2021 were big days in the football analytics-sphere.

It started off with a new piece on The Athletic about running stats. We don’t usually get running data in the public domain, but Sportlogiq — via the charts and texts of Tom Worville — had enough to keep us going for many a month.

The other big media piece came the following day, with Ashwin Raman’s spot in the BBC

BBC front page alongside a global superstar ffs, Taylor has truly made it pic.twitter.com/5eCbSXEqbz— Alex. (@alexfrco) February 4, 2021

Then outside the media sphere there were two BIG announcements from what could probably be fairly described as the two biggest event data providers in the space: StatsBomb and Opta (giving their names in chonrological order of public messaging).

Both were announcing a live, virtual event in the middle of March (what a coincidence). For StatsBomb it’s a product launch; for Opta (proper name, Stats Perform) it’s the annual analytics Forum. Here is a link to the StatsBomb announcement; and here is a link to info on the 2021 Stats Perform (née Opta) Pro Forum.

Both are pretty big; big enough to be worth a newsletter. Here’s why…

Again, I stress, the ordering here is purely based on the chronological order of their public announcements.

StatsBomb

In December, StatsBomb teased ‘StatsBomb 360’, and this week’s post was revealing what it was. It’s a new data offering, and if you filter out the mentions of the words ‘ground-breaking’ and ‘industry-changing’ what you get is actually as conceptually simple as it is exciting.

The way that event data is usually collected is that some human people sit in front of a screen, watching football, and click their mouse and press buttons when things happen. StatsBomb’s original ‘revolution’, upon their launch as a data company, was their shot freezeframes: when a shot was taken, you wouldn’t just get the information about the shot, you’d get information of where every player in view was. See the below example from a blog from the DTAI Sports Analytics Lab

Screenshot taken from DTAI blog post ‘Enhancing xG models with freeze frame data’, Sep. 2020

In practice, this might’ve proved most useful from StatsBomb’s point of view as a way of improving an xG model, rather than its application within clubs, but hey ho. It was genuinely quite exciting, and probably quite important.

This info isn’t collected by the human people sitting in front of the screen though. It’s gathered through computer vision software, a phrase which sounds unfamiliar but is surprisingly Ronseal. You ‘show’ an image to a computer and it sees things — in this case blobs of pixels that are football players and pitch markings.

StatsBomb is now expanding this computer vision-based freezeframe tech to “every event we collect - approximately 3300 events per match”. Readers of this newsletter may already have joined me in guessing this, given that the following paragraph appeared in the “Where will analytics go in 2021” post from New Year’s Eve:

StatsBomb have dropped their own teases, including the below image from an internal hackathon, which looks a little like some kind of passing option snapshot with cover shadow (or something). Or, as Ted Knutson says in the article, an illegal soccer rave.

I was off the mark in a sense: they weren’t just taking freeze frames of pass events, but all of them. Tackles, clearances. All of it.

360 is obviously very exciting for data-minded people. For a long time, data-ites have looked towards tracking data — info of where every player is at every millisecond of the match — as the panacea to all problems. Then some of them got their hands on tracking data and realised that it wasn’t.

What StatsBomb seem to be offering with 360 is the middle ground between event data and tracking data. (To give others their due, a middle ground that other companies, like Sportlogiq, have trodden, and that Stats Perform are somewhat on their way to too). The StatsBomb announcement post lists a bunch of stats that they can get out of this data, like line-breaking passes and my favourite acronym of the year: Defensive Island Events (DIEs).

Some clubs might find this useful. I’m sure they’ll love line-breaking passes and ball receipts in space. The devil will be in the (pricing and product-packaging) detail.

StatsBomb are pushing this, understandably, but it’s overshadowing something else that’s re-mentioned in the 360 announcement. It’s something that I think might be the more visible influence on the data landscape to those of us outside of the professional team environment. Live data.

If you read this newsletter, you know who Opta are. Part of that is because they’re a longstanding data provider with a reliable and sensible data offering. But a lot of it is because they offer live event data. Why is that important? Because data coming in the following day is little use for broadcast and online media.

Sky and BT Sports (the main UK broadcasters) put up stats in-game all the time. I don’t think there’s any real chance that they’d use a provider whose data wasn’t updated live if a provider with live data was available. And while Football Reference has gained a ton of traction in its short, StatsBomb-powered lifetime, an advantage that the Opta-powered WhoScored has is that you can see the stats of games that are in-play or just finished.

StatsBomb won’t suddenly be plastered all over Sky just because they offer live data — league coverage and accuracy/reliability will also be important — but they can operate in that most-visible of spaces now.

Stats Perform

This brings us to the lineup of the 2021 Stats Perform (née Opta) Pro Forum. I gotta say — as someone who’s been to several of these and helped present last year — this slate of presentations sounds incredibly good.

Also.

It seems like one, coincidentally, harks to StatsBomb’s 360 announcement.

“Enriching event data: A semi-supervised augmentation approach using location information” will be presented by Debangan Dey, Rahul Ghosal and Atanu Mitra, and here’s the blurb:

This presentation introduces a method for utilising tracking data to extract more information from an event-only match dataset. Using a merged dataset where both tracking and event data is available, this project will take a semi-supervised approach to creating predictive models that capture hidden patterns from within this dataset with the objective of drawing inferences for tracking data in an event-only dataset.

It’s using tracking data rather than computer vision, but ‘enriching event data’ could be StatsBomb 360’s tagline and it’s a lovely coincidence to see the two announcements share this theme. (Naturally, it also says a lot about where analytics is headed, and maybe I’ll write about why analytics is headed this way some other time).

It’ll be really interesting to see what this group draw out in the presentation, particularly to see whether enriching event data with tracking data can bring out valuable insights which computer vision freezeframes can’t.

Next in the announcement we have Caterina De Bacco with one of the club-led proposals: “Identifying and evaluating the efficiency of each player during the pressing phase against an opponent’s controlled build-up play”. (I won’t give the blurbs for each of these — though recommend reading them through the link)

Stats Perform had two ideas for presentations that came from people working within clubs, and the specificity of the phases of play is an indication of that. Again, this’ll be fascinating to watch, because breaking down analytical insights into the different phases of play seems like one of the big ‘next steps’ for using data in football.

Another big ‘next step’? Defending! And that’s what Aditya Kothari is doing with the presentation “A physics based measurement of defensive contributions”. This one is near to my heart for the subject matter, so check out the blurb:

Focusing on pass and carry prevention and shot prevention, Aditya will build on existing pitch control modelling work with the aim of identifying how well defending teams and individual defenders perform in particular situations during a game, identifying weaknesses and lapses in the defensive system and picking up on other unusual occurrences.

Can’t. Wait.

Of course, collecting metrics is good, but the next presentation on the list, from Ola Lidmark Eriksson, seeks to look at how volatile key performance indicators are. Titled “Volatility and calculation of risk-adjusted return in football scouting”, it’ll take inspiration from the financial sector* to see which players are more consistent than others.

*Mention of the application of financial practises to football is a good chance to plug friend-of-the-newsletter Tiotal Football’s newsletter, Absolute Unit

Rounding out the list of five presentations is Vignesh Jayanth, with “Identifying and evaluating strategies for successfully penetrating a high opposition press from short goal kicks, played inside the box, to move the ball into the opposition half”.

This is the second of the club-led proposals. Not only is this Vignesh’s second year presenting in a row, it’s the second successive year presenting one of the club-led proposals. He’s one of those rare smart people who is as much into knowing about football as data science.

Finally, there’s the Dr. Garry Gelade Award winner, named after the late early leader in the field and fixture of Forums, given to recognise an outstanding submission from a university undergraduate. Laurynas Raudonius’ findings “Recognizing and evaluating opportunities in counterattacks using tracking data” will be shown as a virtual poster during the event.

The predominant theme of this year’s Forum is tracking data, but another is specificity into phase of play. Add StatsBomb’s announcement into the mix and the themes stay pretty similar: going beyond event data, and specificity into phase of play. One of the metrics that StatsBomb’s 360 post mentioned is “Defensive shape around every event”. It’s not clear what shape that will be in, but it sounds like ‘this pass was against a settled defence, that pass was in transition’.

Hell, add in Tom Worville’s piece and the themes are exactly the same:

Going beyond event data (in-stadium(?) tracking data for Stats Perform, freeze frames for StatsBomb, tracking data from broadcast footage for Worville and Sportlogiq)
Specificity into phase (for Worville’s piece, the types of running information he’s looking at give a nice indication into how different players might be used in different types of phases)

And then maybe the article on Ashwin Raman adds in a third, which might come into sharper focus amidst this pandemic: it doesn’t matter who or where you are to work in this field.

Raman is a teenager in Bangalore; some of the Stats Perform Forum presenters hail from India, the US, and mainland Europe; Sportlogiq are a Canadian company, and Worville himself started off, much like Raman, as a guy on Twitter; StatsBomb was a blog before it was a data provider.

You do need the skills required — skim through the list of Forum presenters to see how many of them are working full-time in data science somewhere — and I won’t pretend that the acquisition of these skills is a perfectly fair meritocracy.

However, there was a time when professional football was too club-like to admit even the predominantly white male analytics nerds. Professional football is now, increasingly, not so club-like anymore, and the analytics sphere isn’t quite so restricted in who’s involved. It’s not perfect, but the realisation of that third theme felt quite inspirational to me and I hope — for those just recently approaching it — football analytics is a welcoming space.

Where will 'analytics' go in 2021?

Thu, 31 Dec 2020 00:00:00 GMT

2020 has been the moment in time that best embodies the use of capitalisation to signify heavy intonation. It’s been, very much, A Year.

It’s been a bit of A Year in the football analytics space, although for different reasons. Expected goals has succeeded in setting up shop in the mainstream, creating room for things like PPDA and Mikel Arteta’s win percentages. The first lockdown of the pandemic sparked Friends of Tracking, which — between its YouTube channel and Github code repository — is probably the single best one-stop shop for analytics learning out there.

Jan Van Haaren has put together an incredible list reviewing 2020 in football analytics (all the more incredible for this newsletter being included), so I won’t cover the same ground.

The amount of soccer analytics content has spiked in 2020. My latest blog post lists the research papers, blog posts, news articles, events, invited talks, webinars, podcasts, Python libraries and newsletters that I liked the most!https://t.co/gHdP2znJL8— Jan Van Haaren (@JanVanHaaren) December 30, 2020

However, there are a few things I want to touch on, before going on to think about what 2021 might have in store.

For the past several years, it’s been a semi-recurrent grouch of ‘my wave’ (c. 2014) of football analytics people that ‘there isn’t much public work being done anymore’. 2020 has blasted that complaint out of the water.

Whether through the increasing number of free datasets and learning resources, inspiration being taken from high-profile successes, people gravitating together after being online in the space for a bit, or just sheer numbers, there’s a lot of people doing neat stuff. It doesn’t need to be groundbreaking, and a lot of the useful stuff doesn’t need to be.

Something as simple as Tom Worville’s point about ‘true tackle success rate’ in Opta data (see here for a reference) is a great example. Opta have specific info about which fouls are the result of failed attempted tackles. Making sure you’re using the complete set of data available to you is such a simple thing, but isn’t necessarily obvious.

Speaking of Worville, him joining The Athletic towards the start of the year was a pretty big sign of where things are going. Reach PLC (think the Mirror, Liverpool Echo, Football[dot]London etc) had hired a couple of analytics writers previously, but this felt slightly different.

Part of why it felt different was just because it was The Second, i.e. not the first. This was no longer an outlier event, it was the makings of a trend. Also significant was that The Athletic, with their Hollywood names poached from the establishment of the UK football media, suddenly became Worville-central. Outta the way Ornstein, heave-ho Honigstein and Horncastle, we want Worville. We want his stats. (Disclosure: I’ve known Tom for several years and am personally happy about his success; that aside, those last two sentences remain unimpeachable media analysis)

While the media isn’t the professional game, it’s probably a good bellwether, particularly as the two sides have a degree of overlap. At data companies, the people helping out with enquiries for articles or TV shows like Monday Night Football will also likely work on projects for pro clubs. MNF, and Sky Sports more generally (among other outlets), are increasingly using more and more complex data to help discover and illustrate points.

And then, to take a sharp turn, there’s this other thing: the big potential lawsuit. ‘Project Red Card’, announced in the summer, brought together hundreds of former and current players to seemingly test the waters on what kind of consent needs to be sought from footballers to gather, use, or sell data about them. It seems like a test case that’s in discussion and negotiation stages at the moment, and there’s a decent video explainer here about it.

I haven’t heard much about it since the initial reporting though. This August article in Wired says that PRC was yet to name any specific data companies. Unrelated to the ‘Project Red Card’ itself, in November, Russell Slade, figurehead of the group, tweeted in reference to Zlatan Ibrahimović’s complaints about his image being used in FIFA games.

High profile case recognising data being used without consent or permission. More & more players understanding data ownership and the value of what it means for them historically and moving into the future. https://t.co/OAo5nWDzc1— Russell Slade (@Russell__Slade) November 23, 2020

As it turns out, Ibrahimović’s (and others’) comments were probably more to do with a battle between their agents and FIFPro than sincere concerns about data rights (although these players may well still have those). It’s a thorny area, and this seems like a good time to turn to 2021…

2021

If the mini fuss caused by Ibrahimović’s tweet is anything to go by, then this issue of data rights/privacy could get very messy if (when?) certain interests decide to get involved. What would happen if Mino Raiola pointed all of his clients in Slade and Project Red Card’s direction? What happens if Ibra starts tweeting about how gambling companies are making money off him winning a corner, or how broadcasters might be showing his distance or speed stats in coverage?

In the same kind of area, I responded to this article in the New York Times about data’s role in the future of football with one of my own. I questioned — in a daily newsletter I was doing elsewhere at the time — whether football clubs had the right processes in place to be collecting and storing data on things like player well-being. To quote myself:

[I]f clubs are collecting increasing amounts of personal information, how sure can players be that their data practices are secure? Are football clubs, which relatively regularly make clearly bad manager appointments or transfers, definitely going to be storing this data properly?

[…] If the thing coming over the horizon is an intensified monitoring of individuals by their employers, we should really think about what that means before it arrives.

However, I’m not outright predicting that this particular player-data issue will be a thing for 2021. Like Project Red Card’s question about consent over data collection and sales, this kind of thing will be an issue bubbling away out of sight until someone eventually decides to open the door to it. That door might open in 2021, but I imagine it would probably take a specific incident to prompt it.

Meanwhile, data companies are just, uh, collecting even more data….

Over at Stats Perform (née Opta), they are/will be adding player ‘controls’ to the data, meaning (I believe) that we will finally know all of the times that they touch the ball.

Diving into our player possessions data this evening :) pic.twitter.com/PFVXOeqg7D— Peter McKeever (@petermckeever) November 27, 2020

StatsBomb have dropped their own teases, including the below image from an internal hackathon, which looks a little like some kind of passing option snapshot with cover shadow (or something). Or, as Ted Knutson says in the article, an illegal soccer rave.

Screenshot from the StatsBomb teaser article

Wyscout, I believe, are also doing their own improvements to their data, and other data providers are both existent and available, I am just less aware of their plans for 2021.

Outside of the provision of data, there’s the application. Data providers tend to also offer their own software or services around them. The most interesting space on this front is probably Second Spectrum, a tracking data company with official deals with the NBA, MLS, and the Premier League. They’ve done bits and pieces of augmented reality stuff with the NBA for a while now, but in November had their first ‘enhanced broadcast’ in football/soccer during the MLS play-offs. Note, in the attached tweet, the FIFA/PES-esque player map and hovering names. (Similar has also been offered for certain customers on BT Sport for Premier League matches too).

We are live with our first @SecondSpectrum enhanced broadcast. Watch along on https://t.co/KfNRVHaiK8 and the Fox Sports App pic.twitter.com/S4PD4fuEDY— Chris Schlosser (@ChrisSchlosser) November 21, 2020

I haven’t had the chance to experience this for myself yet, and apparently there are kinks to work out, but I can see how both features pictured here can be immensely useful. Player names would be particularly useful for more casual fans who don’t know everyone (I’m particularly thinking of the possibility of getting people more easily acquainted with players in the women’s game), and it would also offer commentators the option of different styles.

I would certainly like to see a ton more of this in 2021, but it’s kind of dependent on the broadcasters. Anyone who’s watched some La Liga TV coverage will know that their match broadcasts are pretty experimental compared to a lot of others, with things like heatmaps being overlaid on the pitch. The Bundesliga’s broadcast feed has used tracking data for average positions, but not been so avant-garde (to my knowledge) as to splash the information across the entire screen mid-match.

I’d love for this kind of innovation to be more widespread, but it probably won’t be front and centre. So far, the Second Spectrum stuff has mainly been on certain settings within apps rather than the regular TV coverage that the majority watches.

There’s also the company I work for, Twenty3 Sport. To avoid sounding totally like an ad, I’ll use us as an example of wider trends in the industry. While I’ll try and avoid saying totally like an ad, I will say that our Toolbox software product has many tools but, more pertinent to a discussion of the industry, works off multiple data providers.

That the demand for football data is large enough to feed a supply of an increasing number of data providers shows how healthy the sector has become. Organisations are no longer seeing data as something that they should probably have if they can afford it; I think they’re much more likely to be active in seeking data and applications that suit them. The larger number of providers offers that, with more choices to choose from based on various needs of speed, breadth, depth, and price.

Where Twenty3’s Toolbox can also be used as an opportunistic jumping off point for future-gazing, the appreciation and demand for better-designed and on-brand data visualisations will only grow. For good topics (Tom Worville at The Athletic) and bad topics (all coronavirus, but particularly John Burn-Murdoch’s work at the Financial Times), 2020 was a big data vis year. These visualisations being well-designed means they’re more effective at communicating their message and being shared, and being on-brand means that people know who’s doing the communicating. The ‘club side’ may not care as much about this (although clubs are increasingly media outlets of their own), but they definitely care about effective design.

This is all about data communication, something that we and The Athletic share but perhaps Mikel Arteta could use some pointers on. Worville confirmed in The Athletic that the win percentages that the Arsenal manager mentioned in press conferences were based on expected goals modelling. Ideally, this [waves hands generally at Arsenal] would all be better communicated.

There are signs that clubs are improving in this regard, even if their managers confuse things a bit. Job adverts — or, at least, the ones I’ve seen floating across my social media feeds — have improved markedly. Descriptions of data-related roles seem consistently focused and thought-through now in a way that they haven’t always been. (On this subject, I wrote about two City Football Group job ads back in October).

Having trouble viewing this tweet? Click here to open original tweet.

With not just more roles, but more roles that are paid attention to, there’ll also be more coverage of the individuals. And, through this, there may also be more instances of questionable descriptions of what these individuals, or departments, do. It’s probably worth mentioning that for 2021.

In general, then, there’ll just be more. More use of data, more people working with data, more understanding of data, more little innovations.

2020 was the year we firmly got past needing to convince people that data or analytics was useful. That stage, I think, had already long-passed, but there was still a residual feeling that there was something to prove.

In 2021, people will be more open to analytics which will mostly be a blessing, but only mostly. For the charlatans (of which I’m sure there are precious few), it could mean they need to switch up their act. For employees at clubs or player agencies, it may mean frustration at being asked to provide statistics to back up a crummy argument. At the same time, might even more be expected from data analysts because of the success of the research team for Liverpool men’s team?

Mostly, though, it should bring more opportunities; hopefully more collaboration (probably not very publicly, but in group chats or employment); more refinements and more ideas.

Thanks very much for reading, both this piece and any others you read during 2020.

That was a look at 2021 in analytics, but while writing it, a tweet floated across my timeline, with a line that struck a chord with me personally.

"Are sports worth it? Sports are only worth it if we use them for good."

My last column of 2020 for @TSN_Sports https://t.co/UJRjJHisjP— Shireen Ahmed (@_shireenahmed_) December 30, 2020

2020 has thrown up different things for different people, and, at the risk of getting more personal than I’d like to, this has been one of mine. While professional sport itself can be said to serve a community purpose (local or just purely social), analytics sometimes leaves me cold. The relentless optimisation that analytics is based around feels, in essence, to be about winning in a way that feels tangibly different to whatever essence is left in top-level sport itself.

With the Premier League playing every game on TV in different timeslots, empty stadiums are a thoroughly familiar sight, and while some have commented that it’s a symbol of the league’s potential to be a fan-less vacuum, I sometimes feel like it’s similar symbol of analytics. It would be extremely tenuous to argue that developing a good EPV model benefits the community of a football club, except maybe if being without an EPV would send the club into extinction. Which seems unlikely.

The analytics community is also generally middle-class, very male*, and, in the UK, at a guess disproportionately white. This isn’t said to chastise, it’s just…… I’ll circle around to the point.

While the pandemic was in its first peak, there was a phrase that went around in relation to postponements, I think, that google tells me was a line from Washington Nationals’ pitcher Sean Doolittle: ‘Sports are like the reward for a functioning society’.

In the same way, they’re often a reflection of our dysfunctional society. The pandemic, of course, but also the racial injustices that Shireen Ahmed writes about in the piece linked to above. The gender and gender identity inequalities within sport (related, how many of the analytics hires at clubs work across both men’s and women’s teams?). Disability access both at football grounds and within playing sport more generally.

Arrigo Sacchi said that football is the most important of the least important things in life, and sometimes I wonder where analytics fits among that. Regardless, there’s an increasing number of people getting involved in it, and football has a powerful place in a lot of cultures around the world. To paraphrase Ahmed, maybe analytics is only worth it if we use it for something good.

*On this subject,

I started a slack channel with @arielle_dror for women and other underrepresented genders who have an interest in women's soccer and data analysis. You don't have to know how to code to join! If you're interested, DM me 🙂— Lydia Vandenbergh Jackson (@lvanden) June 24, 2020

Why diving is bad

Wed, 18 Nov 2020 00:00:00 GMT

I must be old.

The kids today, or the ones on my twitter feed, keep going on about this thing they love. It’s ‘diving’. I’d only just learnt why people like TikTok.

I suspect that there’s some kickback against older generations in their appreciation of it. The moans about ‘simulation’ do get a little wearying, and so it’s somewhat unsurprising that there’s some teenage-esque rebellion going on. ‘What even is a foul anyway’, the kids will ask in postmodernist disparagement. And ‘isn’t your disdain for diving simply built on societal expectations of masculinity, which dictates that the ‘proper’ way to be a man is to ignore all pain? Is this the hill upon which you wish to die?’.

TikTok sounds more fun than this. They’re creating a whole Ratatouille musical over there.

A couple of years ago, I wrote about another much-scrutinised area of on-pitch footballing morals: why defenders should be (at least partially) excused for clattering forwards after they take a shot. I say that to establish, or acknowledge, my credentials as a defender of traditional British attitudes to things in football. With that in mind, allow me to give my case for why diving is Bad.

First thing’s first: there’s more than one kind of dive. We all know this to be true, and it’s unhelpful to talk about diving without acknowledging this.

Because of the current preferred nature of football, referees will tend to allow play to go on as long as players are on their feet. Even if an advantage isn’t signalled, this is essentially what’s happening when a player is tripped near the box but manage to stay on their feet and keep the ball. In these situations, I think that officials regard the possibility of play ongoing as ‘advantage’. Crucially, although slightly tangentially, I think that the word ‘advantage’ is possibly a bit misleading: play can reasonably continue, so therefore it shall.

Now, if referees will always insist that play goes on even if an individual is being somewhat impeded, going to ground forces the official’s hand. With the impeded player on the floor, play can’t reasonably continue (unless the referee decides that actually they weren’t sufficiently impeded for it to be a foul). This is Dive Type 1.

Another type of dive is much easier to describe, partly because it’s one that’s already been discussed for several years. If a heavy challenge is coming towards a player, they will sometimes dive in order to get out of the way and avoid the potential for injury. Dive Type 2.

Then there’s a third type of dive, which is, broadly speaking, just deception. Blatant dives are in here, but so too, along the spectrum, is ‘winning a foul’* and overly-dramatic reactions to things that may already have been called as fouls. *(e.g. leaving a leg in front of a challenge, stopping suddenly to force a defender to clatter into you). Dive Type 3.

There are grey areas and spectrums between all of these. A player might feel they’re being impeded and go down, but might do so in a slightly unnatural and theatric way because they have to force the referee to realise they’re being impeded: is this an example of Dive Type 1 or Dive Type 3?

Dive Type 2 — the injury preventor — gets even more tangled in hypotheticals. It’s become quite common for players to dribble the ball almost into the path of a tackle, only to poke the ball away at the last moment. Often, this poke will take the ball far away from where they possibly could have collected it. Here we sometimes get a Diver’s Paradox.

The challenge coming in is impeding them, but if they skipped over it and landed on their feet it wouldn’t be called as a foul (see Dive Type 1) and they also wouldn’t regain the ball. But the impediment of the challenge that they’ve skipped over is only an impediment because they’ve poked the ball so far away in an uncontrolled manner — had they tried to keep the ball under their control, they may well have been tackled. The challenge is only a foul because the forward has poked the ball irretrievably far away; but the forward has a severely reduced chance of getting the ball because of the challenge.

In fact, put like that, some of these ‘injury preventing’ dives are a case of ‘winning a foul’. They entice an opponent in for a challenge that the attacker themselves makes dangerous by moving the ball away and moving their own body into the path of the challenge.

All of this is to say that diving is an area filled with different motives, not to mention different levels of impediment (a lightly tugged shirt vs a full-bodied whack on the ankle). It’s difficult. But diving does exist.

The difficulty diving creates is the most uncontentious reason why diving is bad: it makes life almost impossible for referees. The way that football is played is already so steeped in grey area that the shadows that diving casts can make things impossibly murky.

Things usually work themselves out, partly because decisions in most areas of the pitch don’t matter much. However, if the baseline expectation is that anyone will go to ground once they reach the box, it’s anyone’s guess what’s going on. Once people start diving, referees have to contend with everything in the previous section, which is a lot more philosophy than I want an official to be considering on the pitch.

There may be some arguments that other types of fouling make things difficult for referees too. Dissent is the clearest one that springs to mind, but I think the dynamics are meaningfully different. With dissent, a referee’s authority to control a match is challenged quite simply; with diving, the diver is forcing the referee to call them a liar. I would suggest that this is far harder to deal with.

That brings me onto the main reason why I think Diving Is Bad: dishonesty. Now, the kids may roll their eyes at me and say ‘Old Man, what a square dweeb you are, thinking that honesty is still something to value’ but, call me a boomer, I think it is.

Diving is inherently dishonest. As discussed, there’s a lot of grey area, and players are only making semi-conscious decisions, but it is what it is. Most other types of fouling are not dishonest: as much as I dislike tactical fouls for being lazy and breaking the flow of the game, they’re not exactly hiding what they are.

Sometimes when people talk about diving they talk about other types of fouling where people try and ‘get away with it’ outside the referee’s sight (e.g. pulling shirts at set-pieces). I don’t think that this is honesty per se, or at least not in the same way: you’re not trying to deceive anybody, you’re just trying to foul when the authorities aren’t looking.

However, there is some other dishonesty around in the game. Players appealing for a throw-in when the ball goes out off their own body is an oft-cited example. This is true, although I think that there are a few reasons why this doesn’t provoke the same anger as diving:

They’re throw-ins. Outside of Thomas Gronnemark, nobody really cares about them
For that very reason, you’re very unlikely to actively want to deceive a referee into winning a throw compared to, say, winning a penalty
Also, any deception invariably comes after the decision has already been made, rather than causing the decision in the first place (this is also the case about players protesting about not having fouled someone when they clearly have). In other words, a lie about not pulling a shirt is even more meaningless than a lie to try and win a throw-in. A lie to win a penalty is immensely more meaningful.
I also think that there’s often a hope in these appeals, not simply that the decision goes their way, but that reality is actually in their favour. For handball appeals, for example, I expect many players appeal because their teammate is or after having seen the incident from a bad angle. This is getting a little deep, even for me, but in these moments I think sometimes players are appealing more to the universe than the referee, which isn’t exactly deception

Having said all of this, I do think it’s worth considering the other reasons why diving, in English football culture, is disliked, reasons which I don’t think are valid.

There’s the masculinity one, which I alluded to earlier. I think that this is part of some peoples’ response, although I wonder how much of it could be made much less problematic if the phrase “man up” was replaced by “grow up”.

Play-acting, which diving sometimes is, could quite reasonably be considered childish, and I would assume that people who dislike diving for this reason would have a similar reaction to female footballers diving. If they do, does that mean they hold up women’s football as a showcase/embodiment of masculine values too? Maybe disliking diving because “it’s not how men act” is (sometimes) more complicated than simple gender expectations.

I think that there’s also the ‘foreign’ aspect of diving that contributes to the dislike. By this, I don’t mean that it goes against some kind of ingrained British/English values, but just that it’s ‘not what we do around here’. A good example of what I mean by this separation is waving imaginary cards.

We seem to have a particular dislike for players or managers waving for a booking in England, but it strikes me that this is just another way of dissenting. Perhaps it seems slightly more explicitly presumptive than ‘English dissent’ — i.e. shouting expletives — but in its essence I think it’s the same. It’s pushing the referee to do their job differently (often genuinely as a frustration release valve, sometimes merely under the guise of it). Dissent is not ‘un-British/un-English’, but waving a card is ‘not what we do around here’.

Similarly, while I have some fundamental objections to diving, its unfamiliarity in the British/English game is undoubtedly a factor in some peoples’ dislike for it. It’s likely a factor in mine too, although I do think that I’ve illustrated justifiable reasons for disliking it.

Finally, I wonder if there’s something about diving that feels like a cheat code. Dislike for diving might, in part, stem from a fear that it’ll overtake the sport and football will become about who can get near the box and fall over the best. I don’t know how well-founded this is, and feel like I don’t have the knowledge of football history to gauge whether the game is going that way, but this may be a factor in diving’s unpopularity.

This topic is, you may be able to tell, something I’ve thought quite a bit about. Perhaps too much. While I imagine many people may disagree with parts of my opinion or reasoning, I think some things here are undeniable and should form the base for all future discussions about diving:

There are different types of dive, of varying forms of legitimacy
Diving makes refereeing more difficult, for obvious ‘muddying the waters’ reasons
Diving is a deception, although I expect a lot of disagreement over how much deception there is elsewhere in the game and how much this matters

Different people will have different feelings and values that they put on certain things within the game. Even though this is my view on diving and why it rubs me up the wrong way, I don’t think that it is necessarily wrong for someone else to see it differently. It just isn’t how I view the sport — diving is bad, to me.

This has partly been a rebuttal to my Twitter timeline, which I expect is far more receptive to diving than the norm. However, I think this is a worthwhile exercise regardless. It helped me organise some thoughts, will hopefully help others, and at the very least means they’re all in one place when the next incident and debate inevitably comes along.

Now I can just link this piece every time, rather than trying to fit it into a tweet to aim at the kids on my timeline…

On language, stats, and football

Sat, 14 Nov 2020 00:00:00 GMT

Americans call aubergines ‘eggplants’. For those familiar with the deep purple skin of the vegetable(?), this seems peculiar. To British ears, it’s one of those “oh, America!” things to roll ones eyes at and quietly mock, like ‘pants’ and ‘soccer’ and the missing ‘u’ in ‘colour’.

And then you learn that there’s a variety of aubergine that is white, and small, and generally looks quite like an egg.

Image from Wikipedia

It doesn’t matter that Americans call the purple version an eggplant even though it doesn’t look like an egg. It’s a little confusing to people unfamiliar with the term, but once you pair the word with the object it’s, well, just like any other name. Carrots don’t look much like cars.

The name ‘expected goals’ has been debated for almost as long as the statistic has existed, from back when it was calculated on stone tablets in the far distant past of the Allardyce Age. Despite the name’s detractors, the metric — which is a computer-modelled judge of chance quality — remains to be named that way on the screens, airwaves, and pages of an ever-increasing number of outlets.

Y’know what name from that era, when ‘expected goals’ was first coined, hasn’t stuck? xG2.

If you’re into stats, you might know xG2 by the name ‘post-shot expected goals’ or ‘expected goals on target’. It’s different from the other/normal expected goals stat because that one is only built to judge the quality of an opportunity at the moment a player hits a shot. If a striker makes a hash of it, we still want to know that it was a good chance. Therefore, the model for xG doesn’t factor in where a shot ended up.

But xG2 does. Or, did, when it was called that. Anything off-target gets a post-shot xG value of 0, because the trajectory it was on means it would never turn into a goal. But if two shots were exactly alike at the point a striker hit them, post-shot xG will give a higher value to the one that goes towards the top corner than the one that goes down the middle of the goal.

xG2 died as a name, replaced by the more descriptive post-shot xG or xG on target. (Neither of those two options has definitively won yet, but my preference, as well as the frontrunner, is the former).

Seth Partnow, in a presentation at the 2019 StatsBomb conference, said that a good name for a statistic doesn’t ensure its take-up, but a bad one ensures it won’t be. I think I agree (with the caveat that absolutes make for good quotes but bad models of reality). All you want to do when naming something is to pick a moniker that won’t get it laughed out of town or put on a government watchlist. Who is ‘Aston Martin’ and what do they have to do with cars; how should I be expected to know that an ‘Apple Macintosh’ is a computer; why do Americans still call aubergines ‘eggplants’? It doesn’t matter.

The resistance to the term ‘expected goals’ is real, but I would guess that it’s primarily an opposition to the stat itself rather than the name. Remember when people made fun of ‘Twitter’ and, uproariously, quipped that even if they did join the platform they wouldn’t know what to twit and twat about? The problem wasn’t peoples’ resistance to ‘Twitter’, the word, it was peoples’ resistance to Twitter, the concept.

Language evolves in a strange way. Definitions of words guide how we use them, but these definitions are also shaped by usage. You also have very similar words meaning subtly different things in sometimes quite similar contexts.

Take football, and the phrases ‘a good finish’ and ‘a good finisher’. A ‘finish’ can be almost any shot that leaves the viewer scrambling for a synonym for ‘shot’. Usually it will be a placed effort, with the instep, but that could mean it’s struck from eight or 18 yards. It could also mean a header, or a volley.

A ‘good finisher’ is a much more narrowly defined description. Tap-ins and one-on-ones are the things judged here, usually. A player could dribble down the wing, cut inside, and curl a shot into the top corner from the edge of the box and that would be a great finish, but that wouldn’t be what comes to mind if they were described as ‘a good finisher’, I believe.

And then there’s ‘finishing skill’. In statistical circles, this would be ‘how does a player’s goalscoring compare to their expected goals figures’. These would include every type of shot: tap-ins, one-on-ones, curled strikes, volleys, placed headers, power headers, directing a pull-back from 12 yards, driving a strike from 20.

Disclosure: I strongly dislike ‘finishing skill’ being used in this context.

It doesn’t take a single skill to score goals from all those types of chances I just listed. It takes a whole range, and there are countless players who are very good at some but not so good at others.

It is confusing that ‘a good finish’ and ‘a good finisher’ have slightly different uses. It (to me) is confusing that a player’s ability to score headers and mazy dribbles and tap-ins might be lumped together as one single skill. It (to me) is doubly confusing to add the latter to the former linguistically and confuse things even further.

But what if ‘finishing skill’ takes off, as a way of saying ‘over/underperformance of expected goals’? Will I have been wrong to have disliked it?

Statistics aren’t the only area where this clash of languages is happening in football. By coincidence — or by Twitter — statistical and tactical concepts butted heads with the mainstream at about the same time. Talk of ‘pressing’ and ‘halfspaces’ riled the kinds of people who were also riled by ‘expected goals’. Just as I would argue that people took against expected goals rather than ‘expected goals’, I think people took against the presence of new terminology rather than the new terminology itself.

‘Pressing’ and even ‘counterpressing’ are firmly in the English lexicon now though (although some still use the German gegenpressing when referring to the latter). ‘Halfspace’ has even crept into the mainstream. It might become fully normalised, it might not.

One of the reasons it might not, in Engand at least, is because English football already had the term ‘channels’. ‘Running the channels’ is a phrase that seems as old as time to me, and which, while it doesn’t explicitly mean the same thing as halfspace, has a fair amount of overlap.

(I *think* that ‘running the channels’ means running into the physical channel left between centre-back and full-back in a back four, which tends to be a diagonal run from the centre of halfspace and ending in the halfspace or on the wing. It’s a slight problem that ‘channel’ could refer to a space between players or a more firmly defined area of the pitch, but it wouldn’t surprise me if English football adopted ‘left channel’ and ‘right channel’ to mean ‘left halfspace’ and ‘right halfspace’)

Halfspace or channel, neither matters. What matters is that you can communicate effectively with the people you need to communicate effectively with. Even within the professional game, clubs all have their different things they do and talk about that other clubs don’t. Phases of play, apparently, is a big one. ‘Build-up’, like beauty, is in the eye of the beholder.

As long as they have the term defined enough that everyone understands it, that’s fine for them. For media audiences, things are a little different because the net is cast a little wider. People are always, unseen, coming and going too, meaning that it’s hard to keep everyone up to speed. Yet media entities find a way of helping their audience through it (if they’re good).

The latest attempt by a niche to change the way people speak about football is StatsPerform (aka, Opta’s parent company).

Having trouble viewing this tweet? Click here to open original tweet.

As Mackriell was referencing, James Gheerbrant of The Times recently wrote a piece referencing two models: DAVIES from Sam Goldberg and Michael Imburgio, and StatsPerform’s own Role Discovery model.

As much as I believe these models may be useful, I don’t think they’re going to be become the ‘new descriptive norm’, for two reasons.

One is that I don’t think they’re changing much; the other is that the change they make might not be a good one.

I have a few disagreements with Gheerbrant’s piece, although I’ll flag very clearly that I think The Times’ sports pages are a far better place for having hired Gheerbrant, and the problem is mainly created from trying to introduce these models to the public, which is never easy.

The first disagreement is the set-up to the piece, which is really a way of talking about statistical modelling but disguising it as talking about the English men’s national football team:

The argument against picking five right-backs [as Gareth Southgate has done] is an excess of homogeneity — in other words, having five guys who all do the same job is a misuse of the finite places in an England squad. And if you accept the principle that players who occupy the same position in the team have the same function, this argument holds true.

First point of departure: I don’t think that ‘accepting the principle that players who occupy the same position in the team have the same function’ is at all widespread. Maybe people who don’t spend much time thinking about football think that Phil Bardsley and Dani Alves fulfil the same function, but I think most people accept that different right-backs have different skills.

If this was just Gheerbrant segueing into talking about modelling I’d let it slide, but it’s also the way that I’ve seen some people talk about this type of work. And that bugs me. A person who talks about midfielders as if they do or can all play the same role — who hasn’t already taken notice of the various ways we describe midfielders who do different jobs — isn’t going to change their vocabulary because StatsPerform ran a cluster analysis.

My second point of contention is with this:

But the rigid language that we use to label players (and this is partly a linguistic issue — other football cultures are permeated by much more descriptive terms such as regista in Italian and enganche in Spanish) still shapes how we think about them. Alexander-Arnold, a primarily creative player, is saddled with the tag of “defender” and the expectations that go with it.

I don’t necessarily think Trent Alexander-Arnold is saddled with expectations of ‘defender’, and if he is I don’t necessarily think that’s a language problem. There are other full-backs that we know don’t do much defending because they’re really attacking. We call them wing-backs.

Maybe Alexander-Arnold doesn’t fit ‘wing-back’ or ‘full-back’ perfectly, but some simple variation would probably work. ‘Attacking full-back’ or ‘high full-back’ or something.

According to DAVIES, he’s an ‘offensive wide progressor’. Per Role Discovery, a ‘wide active playmaker’. I dunno. If somebody doesn’t acknowledge that different right-backs play in different ways, I don’t think they’re gonna start calling them wide active playmakers in a Damascene moment.

I think that these models do have uses. If you’re recruiting and looking for players that can fit into your system, the option of searching for ‘wide active playmaker’ rather than simply ‘right-back’ could be really useful. Similarly, with regards to Mackriell’s tweet again and his reference to teams playing in ‘shapes’ rather than ‘formations’, that could have a real impact in opposition analysis. Think ‘find me all the clips when team X is in a disorganised transitional moment after failing to counterpress in an asymmetric 3-1-3-3’ rather than ‘when they’ve just lost the ball in a 4-5-1’. Or something like that (I confess, I haven’t seen much of their work on this aside from their ability to use tracking data).

But… None of this matters. I recently wrote about formations, and why we call something a 4-3-3 and not a 2-3-5. I tweeted the link out earlier today and someone sent me a tweet of theirs in reply.

Having trouble viewing this tweet? Click here to open original tweet.

I agree with it. We all know that formations and players play in different ways, even if we don’t talk about it in sophisticated language. Nobody has ever expected a team’s shape to look like table football. Gheerbrant says it himself in his piece; “we all know, for example, that Aymeric Laporte and James Tarkowski have vastly different roles in their respective teams.”

We know that players have different roles. It’s only been a decade since Michael Cox coined ‘inverted wingers’ and it stuck. Maybe we could use more Cox-es to make labels for player roles, but I flat out refuse to believe that people aren’t talking about roles and shape already.

I’m on the final stretch now.

I think that expected goals is a fair name, you’ve just got to be able to describe it well enough to push past peoples’ initial unfamiliarity. That’s the same with anything, though, xG or Twitter or TikTok.

But if you’re using a term that can be confused with existing vocabulary, you’re on uncertain ground. ‘Finishing skill’ is so wrapped up in the linguistic fuzziness of ‘a finish (shot)’ and ‘a finisher (player)’ that the concept would need to be really solid to break through. And I’m just not convinced it is. I don’t believe that the concept of a player over/underperforming their chance quality on all varieties of shots they take is something people talk about.

‘Halfspace’ is easy to describe, but there are existing words that sum it up that people can grasp at if they don’t like new things. ‘Left/right channel’ or ‘left/right of centre’ can be used. I think that ‘halfspace’ is an example that the new language you’re using needs to prove its usefulness, and perhaps it doesn’t hold a ton of use to mass media football talk.

Most people consume football via television nowdays, and even on radio detailed descriptions will only usually come to describe goals. A new term for an area of the pitch isn’t a high priority. But it’s a word that describes something; we’ll just have to see how the evolution of language takes its course on that one.

And with role descriptions, it’s something we’re already talking about. If people mistakenly think that Jack Grealish can be interchanged with Raheem Sterling or Jordan Henderson, that’s because they don’t know what Jack Grealish does. They’re relying too heavily on his position as a heuristic of his role, rather than not believing in roles at all.

There are questions to be asked, too, about whether a statistical model that measures on-ball output should be used to describe roles of players who do more than just on-ball stuff.

This has been quite a lot of me talking about what I don’t like, so I think it’s only fair that I offer up something that (while I don’t like it) I may have already proven to be wrong about. It’s the statistic ‘touches’.

‘Touches’ in football-data parlance doesn’t actually mean how many times a player’s body connected with the ball; it means how many on-ball events they did. Passes, take-ons, shots, that kind of thing.

I don’t like it. It’s not that it sounds bad, it sounds false. A player who’s only touched the ball three times may not have actually touched the ball three times; they could have touched it twenty times running the length of the pitch but might only be ‘credited’ with a couple of ‘touches’ for the entire motion.

And yet, in my more contemplative hours, I’ve tried to think of a better name. ‘On-ball events’ sounds too unfamiliar. ‘Actions’ also sounds awkward, like a slight mistranslation from German or Spanish. It’d probably stick if it had to, but it’s not strong enough to supplant ‘touches’. And maybe that’s fine.

Seth Partnow’s talk at the 2019 StatsBomb conference — which you can watch here — was titled “Analytics as Vocabulary: Giving Stats the Power of Language”. I like the title (well, I like the talk too) because that’s how I like to think of stats. Language is communication, and you need to be able to communicate the numbers.

Partnow’s bit about naming things — that a good name won’t ensure take-up but a bad one will stop it — reminds me of touches, and reminds me of giraffes.

I don’t like the way people talk about evolution. Even though it’s scientific theory, every interesting fact about how an animal has evolved is spoken about as if it’s a Grand Design, as if it’s clever. Things like “giraffes have long necks so that they can reach the leaves on branches that are high off the ground”. No; giraffes have long necks because the ones who had short necks died.

Americans still call aubergines ‘eggplants’. They haven’t died out because of that.

xG in FM? What does this mean?

Tue, 03 Nov 2020 00:00:00 GMT

Hello reader, welcome to Get Goalside!. If you haven’t already seen it, here’s a video of an AI camera mistaking a bald linesman’s head for the ball.

*Electronic Arts voice*

Ex. pected. Goals. It’s in the game.

Ok, so this isn’t about EA Sports, but xG is in FM. A nice big announcement from Football Manager said they’d have the metric in the 2021 edition of the game. There’s a proper data company edge to it too, thanks to a partnership with Dutch company SciSports.

If you want to read a selection of ‘the stats community’’s reactions, you can check out the responses to my tweet when the news first broke:

Having trouble viewing this tweet? Click here to open original tweet.

In terms of gossip, there was also a reasonable amount of raised eyebrow when the very first screenshots came out (a few days before the announcement of the SciSports partnership) and the xG map colour scheme looked a lot like StatsBomb’s.

Ted Knutson explained away the similarity:

Having trouble viewing this tweet? Click here to open original tweet.

And the next video had a different colour scheme (as well as a nifty SciSports logo):

It isn’t too surprising that the original mock-ups might have had a coincidental resemblance to StatsBomb’s work. Firstly, nerds love FM and nerds love StatsBomb. Secondly, named data companies have been in the game before, with STATS (now StatPerform after merging with Opta) being the ‘in-game data provider’ in previous iterations.

And now it’s SciSports’ turn (it seems that if you want the attention of Sports Interactive, who make FM, you need to be a company beginning with ‘S’. My employers Twenty3 must be kicking themselves at being alphabetically so close.)

It’s quite a sweet story. SciSports founder Giels Brouwer says in the press release,

I have always been a passionate Football Manager gamer and it is a great source of inspiration for our entire team. When we founded SciSports, we always said that our ultimate goal was to build Football Manager with real match data for pro football clubs.

But although the partnership has only just been announced, this hasn’t been a rushed relationship. I spoke to the company’s Chief Analytics Officer, Jan Van Haaren, about the partnership and how it had come about. “SciSports and Sports Interactive have a long-standing relationship that goes back many years,” he said, “Delegations of the two companies have visited each other's offices on multiple occasions throughout the past few years to explore potential collaborations.”

SciSports aren’t quite building the xG that will appear in-game though, the data company acting more in an advisory capacity to Sports Interactive. “Sports Interactive developed their expected goals model in-house using data that was produced by the Football Manager match engine,” Van Haaren explained.

“However, the SciSports and Sports Interactive data science and development teams met regularly to discuss the progress and challenges. SciSports' prior expertise with building advanced metrics such as expected goals helped the Sports Interactive team to identify information that is predictive of the outcome of a shot, and to improve the accuracy and robustness of the expected goals model.”

It sounds like an interesting challenge though. The apparent beauty of working on this kind of model in a game like Football Manager is how rich the data is, “such as access to the spatial locations and body orientations of all the players at all times,” Van Haaren said. However, “the computation of the expected goals values needs to be fast as many matches are simulated in the game while the computational resources might be restricted on some devices”.

The section of stats twitter that were wondering about Football Manager’s match engine in response to my tweet will also be interested in this next bit.

“Since the Football Manager match engine is not deterministic [it doesn’t just flip a coin and say ‘goal’ or ‘not goal’] and the outcomes of shots are influenced by many different factors,” Van Haaren explains, “the expected goals values do provide insights into the performances of teams like they do in real life.

“In addition, the expected goals values also help to further improve the realism of the Football Manager match engine. That is, over a large number of matches, the distribution of the expected goals values for real-world shots should be similar to the distribution of the expected goals values for the simulated shots.”

Neat.

If you’re an avid FM player you might have already seen the latest news on the match engine AI improvements that’ll come in for the 2021 edition of the game. I spoke to Van Haaren before that news dropped and didn’t know it was coming, and I’ll leave any speculation as to how linked the two are up to you, dear reader.

It all raises an interesting point about a new frontier of video game realism. I confess, one of the reasons why I spent more time on the AI improvements is that I don’t really play FM. A minor part of that, and something that might tempt me back if it was properly addressed, is the lack of women’s teams, leaving devoted fans to put databases together,

New database now uploaded to Steam. Likely to be the final update of FM20.https://t.co/59muEQm81V

Link to a new face pack here to go along with the update.https://t.co/QgoMpkXq1y pic.twitter.com/U8ye88quGc— Football Manager - Women's Database (@FMWomensDB) October 3, 2020

But part of it is just that I don’t enjoy it very much. Mostly because I’m not very good; a little bit because there’s a lot more to do than when I played it when I was younger (c. 2005-2012).

FIFA and Pro Evo have similar, although slightly different, problems. How much should they aim to reflect ‘real football’ and how much should they consciously diverge to make a fun game?

I’m not saying that xG will take the fun away at all, and actually think it would improve my enjoyment if I was able to work out whether my tactics were actually working or not. Let’s just hope that FM 2022 doesn’t involve the capability to do this:

the models were built in-game by the digital versions of the analysts— Joe Gallagher (@joedgallagher) October 16, 2020

Pass it on if you’ve enjoyed this newsletter, and subscribe if you think you’d like it again in future

Where are all the English football players?

Tue, 27 Oct 2020 00:00:00 GMT

Hello reader, welcome to Get Goalside!.

According to reports, the FA doesn’t make any sense.

OK, let me try again: according to reports about what FA chairman Greg Clarke wrote to Premier League clubs, his letter makes no sense.

With Great Britain going through the turbulence of Brexit, English football’s rules around foreign signings need to be re-written.

“Government requirements are that players coming to play here should be internationally established at the highest level,” wrote Clarke, “making a significant contribution to football; and not taking the place of already settled talent.”

Further on (in Sky Sports’ report, which I’m assuming is quoting his letter linearly), he writes:

The new system gives a 500% uplift on guaranteed access to non-European talent, whilst also allowing the European talent in. However, the system doesn't allow access to lower quality foreign players who block the development pathway for home-grown players - and who rarely get to play.

Now I am confused on two levels. The first is that Clarke is talking about reducing access to lower-quality foreign players while also boasting a 500% uplift in access to non-European talent. The second is that he cites government requirements that foreign players don’t take the place of ‘already settled talent’ and then noting that the system won’t block the development pathway of home-grown players. Oh, while also saying: “Please be assured you still have access to talented young players; they have not been excluded.”

Clarke apparently wants to reduce foreign imports while boosting them, so as not to stand in the way of settled or developing domestic players (the ageing ones though, presumably, can get stuffed), although the development of home-grown players shouldn’t stand in the way of access to talented young foreign-born footballers. Simple.

Subscribe now

One thing that the FA are accurate in noticing that English players make up a limited amount of minutes in the Premier League. Out of the big 5 European leagues, the EPL has the smallest share of home nationality minutes.

Percentage of minutes played by home nationality in top tier, 2019/20 season (source: FBref)

Spain/La Liga: 59%
France/Ligue 1: 49%
Germany/Bundesliga: 41%
Italy/Serie A: 38%
England/Premier League: 35%

But while Britain is an island, there are still ways off it. Footballers who want or need to are free to move abroad to play in the best competitions that they’re able to. Yes, compared to other European leagues (even beyond the big 5) the English don’t make up much of their home top tier, but English players don’t play abroad much either. They made up a miniscule share of the minutes played in the other ‘Big 5’ leagues (i.e. excluding the PL) last season, at around 0.5%.

NB: ‘Foreign Big 5 min. share’ applies to all leagues — for example, for French players it’s their percentage of minutes in the Bundesliga, Premier League, La Liga, and Serie A.

In the bottom right, you’ll see the Russian Premier League. Interestingly, Gabriele Marcotti pointed out that that league has tried a strictly nationalist rule and it didn’t quite work.

Russia introduced a rule a while back (minimum 5 Russian players per team on pitch at all times) in an effort to help the national team. It didn't work. (And they're changing it). https://t.co/Q3zMveGEJc— Gabriele Marcotti (@Marcotti) October 24, 2020

My take is that if English players were good, they’d play elsewhere if they were crowded out of the Premier League.

Fortunately, one of those places is still at home: the Championship. By clubelo’s ratings, England’s second tier is a better-quality league than the Eredivisie (marginally).

English players made up 53% of the Championship’s minutes last season, a much larger share than in the Premier League. The FA/British government may wish to point out that this is still quite a small share, and that Spain and Germany’s second tiers (who are also strong by clubelo ratings) had a ‘home nationality share’ of 76% and 67% respectively last season.

Get ‘Get Goalside!’ in your inbox every Tuesday and subscribe

Subscribe now

I would party grant them that point. It might be noted that the differences between home nationality share in those second tiers are roughly similar to the differences in home nationality share in the top tiers. England, low; Germany, higher; Spain, highest.

The trouble with all of this — and I say this after spending a few hours in circles — is that it’s quite difficult to logic it out and show one way or another whether English players are having their development blocked.

Maybe England just isn’t producing players up to the required standard. Maybe there’s no actual problem at all: England reached the 2018 World Cup semi-final after all; one of the most exciting generations of young talent is coming through; seven of the top 12 goalscorers in the Premier League (at time of writing) are English.

And things are far better in terms of playing abroad nowadays too. In 2019/20, there were a good smattering of English players featuring across the Bundesliga, Eredivisie, 2. Bundesliga, and Primeira Liga. Five years previously, it was pretty much just Ashley Cole and Micah Richards in Serie A. (Bizarrely, Ashley Young and Chris Smalling bring that cycle full circle).

But most tellingly of all, in 2014/15 English players actually made up a smaller share of minutes in the Premier League and Championship than 2019/20.

I’m unsure of 2019/20 was a blip or a sign of a trend. At time of writing, 2020/21’s figure is 36%, higher than last season, but we’ll have to see whether that holds up.

Subscribe now

One final thing to note.

It always felt, from my vantage point, like it was assumed that the Premier League would be able to essentially pay their way out of Brexit impact. That the heft that the competition has would mean that its part of the football industry would get favourable terms in any new rules.

But, despite Greg Clarke’s numerous references to the British Home Office in his letter, it may not be the government that the Premier League is tussling with:

FA hope Boris Johnson desire to host + win 2030 World Cup will earn crucial government support in battle with Premier League over post-Brexit rules on non-homegrown players. FA want high entry bar, #EPL lower. Decision due end of Oct @TheAthleticUK #BREXIT https://t.co/jktXzvx4wI— David Ornstein (@David_Ornstein) October 19, 2020

All politics, ain’t it.

Liked this? Send it to a mate who’d enjoy it too.

And subscribe to the newsletter, if you haven’t already, below

Subscribe now

Shout-out corner

A brief shout-out corner this week as I’ve left finishing this newsletter off much later than usual. I’m a sucker for a data visualisations-breakdown thread, so here’s one from The Athletic’s Tom Worville.

The latest on the other newsletter, Mark’s Notebook, was about why Chelsea might end up being fine this season despite a questionable manager.

Oh, and also big shout-out Marcus Rashford, and everyone else who’s pitched in - in the absence of the state - to make sure that kids don’t go hungry over the school holidays.

You can donate to FareShare, the charity that Rashford works with, here. A tweet with the latest statement from the Child Food Poverty Task Force is below

Statement from the Child Food Poverty Task Force.#ENDCHILDFOODPOVERTY pic.twitter.com/xNJHAFCWOU— Marcus Rashford (@MarcusRashford) October 25, 2020

Be kind.

Is this the death of formations (as we know them)?

Tue, 20 Oct 2020 00:00:00 GMT

Our dark crime thriller opens…

The shadow of Pep Guardiola looms large over the bleeding body of Formations.

He covers the body with the cardigan he discarded earlier, now sodden, storing his bloodied knife in his cargo trousers. It was a methodical kill, but he doesn’t need to disguise it — everyone will know that it was he that cast Formations to the land of the dead.

As Guardiola straightens, the setting sun’s glow frames his head. A halo, or devil’s hellfire?

Yep — formations aren’t what they used to be.

The resounding reaction to Manchester City’s system against Arsenal in their latest meeting was “…what?”. After a few minutes watching, I had a go and described it like this:

Having trouble viewing this tweet? Click here to open original tweet.

After the game, I asked people to write in with what they thought it was. The answers were many and varied, with around nine different formations being suggested. Here’s the tally:

Assorted jokes (7)
3-3-4 (4, although including formations that could be variations of a 3-3-4 it was 7)
Assorted points about no such thing as formations really (2)
(All 1) 2-1-3-4, 3-1-4-2, 4-2-3-1, 4-4-2, 3-3-2-2, 3-1-3-3, WM, 3-diamond-3

I think that consensus was about right. Out of possession (although City didn’t spend much time out of possession), it settled into a 4-4-2, but when they were on the ball it seems that ‘3-3-4’ would be the best way of describing Guardiola’s set-up.

And, for formality’s sake, this is the out of possession ‘4-4-2’:

This system — a pretty clear example of ‘in possession’ and ‘out of possession’ formations in use — raises interesting questions about formations. One is about which of these two ‘formations’ would we use to describe City; another is about what a ‘formation’ even is.

I should stress, these are not new questions. In fact, I was thinking of writing about this subject anyway, days before Guardiola shoved it into the spotlight. The day before the City-Arsenal match, I’d asked on Twitter whether people thought that the formation notation we use is closer to describing a team’s shape when defending or when attacking.

Having trouble viewing this tweet? Click here to open original tweet.

The results (as they don’t seem to embed in the newsletter) were:

Just show the results (voting in a poll shows the results before it closes): 9.6%
Defensive shape: 77.3%
Attacking shape: 13.1%

When people responded with theories about why the answers were leaning this way, the trend was that defensive shapes are more rigid whereas attacking ones are more fluid. That makes sense.

However, when asking people for City’s formation they settled on the 3-3-4, the attacking one*. How does that fit in?

*Some people gave multiple formations and I’ll be coming to that in a second…

Well, you could refine ‘people use the defensive shape for formations as that’s more rigid’ and refine it slightly. A tweaked version could be that people name the formation based on what pattern they see the most.

For example, if a game is fairly close, then people will notice the defensive pattern more (because defensive shapes are more rigid and regular). But with City, in games they dominate large parts of, it’s the attacking patterns they latch onto more.

An advantage in theoretical terms, but a disadvantage in practical ones, is that this means cultural baggage can have an influence. A lot of people mentioned to me that 4-4-2s are often 4-4-1-1s in practice. It feels like the reason why people still call them 4-4-2s would be cultural.

This theory that it’s about ‘what patterns do people see most’ would also explain the divide between calling a team a 3-5-2 or a 5-3-2. Teams expected to do a lot of defending will get the 5-3-2/5-4-1 treatment; teams who are higher up the league get the honour of being 3-5-2/3-4-3.

All of this being said, the Man City game makes me feel dissatisfied with the ‘formations are the pattern people see most’ explanation.

I think a more accurate description would be that the formation notation — the 4-3-3 or the 3-3-4 etc — is about giving us a starting point to know what roles everyone will be playing. This requires a little bit of football knowledge, but it’s stuff we all pick up. We might use 4-4-2 instead of 4-4-1-1 not because we’re unaware that one forward drops while the other stays high, but because we know that; using 4-4-2 signals that either forward could be dropping, whereas 4-4-1-1 would indicate a solid ‘attacking midfield’ role for one of the players.

Of course, none of this really matters. Reducing things down to a single formation is a little nonsensical, as Manchester City showed. Within actual clubs it would make more sense to talk more specifically — about in possession vs out of possession shape; about individual roles within a system. It’s kind of just for us normal fans that the ‘number-number-number’ formation notation makes sense.

For articles like this in your inbox, subscribe:

Thanks to tracking data — where the position of every player (and the ball) is collected multiple times per second — the world of stats is catching onto this.

At its most basic, we have what the Bundesliga offered during Project Restart. A few minutes into the match, they’d show a little graphic: initially it’d be the team laid out in their 3-4-3/4-4-2 ‘shape’; then the players would be moved to their average position in the match so far.

Here’s a video here showing it in action:

Outside of TV broadcasts, researchers and data companies have also been hard at work. There’s a presentation here of Laurie Shaw and Mark Glickman’s work and Ben Mackriell at StatsPerform vaguely alluded to something here. I won’t break down the research, but there’s a really good article summarising Shaw and Glickman’s work that here.

The upshot of it, though, is that instead of relying on someone to say “I think this team is playing in this formation”, you can get the actual shape a team is playing in. Shaw and Glickman’s work with tracking data also differentiates between attacking and defensive shapes, which (to my knowledge) data providers don’t currently do. [Of course, at clubs you could ‘collect’ this data yourself for your own games, and could get analysts to do it for the rest of the league]

Edit: It’s been pointed out to me that this work has been done since tracking data was first collected in the 90s. Another paper on the subject, from 2015, is here. I’ll continue to update further things noted to me here.

Separating attacking from defensive formations would open a small realm of analysis. You can — as the researchers do in their poster for the Barcelona Analytics Conference — look at which offensive formations happen alongside which defensive formations. Perhaps a team regularly look like a 4-2-3-1 in offence, but switch their defensive shape around based on personnel, for example.

With the groundwork for ‘formations as told through tracking data’ already now done, we can speculate on what could come next. Perhaps researchers will be able to create algorithms to detect the main rotations that teams use. That could be immensely useful for defences, like being able to know who they should pick up if a striker drops.

And because of the way football’s going, that information is only going to become more important. It’s not just Guardiola either — it’s Marcelo Bielsa at Leeds, Chris Wilder at Sheffield United, and that’s just covering Yorkshire.

In five, ten years time we might talk about formations in a totally different way. We might talk about in- and out-of-possession formations as standard; we might wait for the tracking data to give its verdict for us. It’s all starting now though.

Are Man City muscling in on Liverpool's analytics crown?

Tue, 13 Oct 2020 00:00:00 GMT

In Captain America: Civil War (stick with me), the MacGuffin that drives the plot towards Cap and Iron Man throwing punches is a proposal: superheroes should register with a government authority. There follows, naturally, disagreement.

The AI-made-flesh character Vision pipes up, pointing out that since Iron Man arrived on the scene, the number of ‘enhanced’ people and the number of aliens wanting to smush Earth to dust had risen exponentially. Maybe the two were linked. “Our very strength,” he says, “invites challenge; challenge incites conflict; conflict,” a thespian melodramatic pause, “breeds catastrophe.”

It wouldn’t be a surprise if, somewhere on the internet, there’s an image of Liverpool FC’s very own Avengers: Michael Edwards and Ian Graham at the head of an unconventional team of ‘enhanced’ minds, their faces superimposed onto Chris Evans and Robert Downey Jr’s bodies. And, like the Marvel team, their very strength has invited challenge.

Manchester City are hiring for their very own best and brightest (see, I told you to stick with me).

To be more precise, City Football Group are hiring. Two job adverts recently went up for a computer scientist and an AI scientist, two roles not dissimilar to ones you’d find in Liverpool’s famed department. You can learn a lot about a club just from adverts like these — linked here and here, for the duration they’re online — so let’s dive in to what we can glean.

The starting point of who’s doing the hiring is important to note. These will be CFG hires specifically. Yes, they’ll be assisting Man City, but they’ll also be helping out the wider group of clubs.

(At present, that group stands at ten: ownership of Manchester City, New York City FC, and Melbourne City, along with investments in Yokohama F. Marinos (Japan), Montevideo City Torque (Uruguay), Girona FC (Spain), Sichuan Jiuniu (China), Mumbai City FC (India), Lommel SK (Belgium), and Troyes (France))

The nature of CFG means that it’s kind of a surprise they haven’t invested more heavily in this kind of brainpower sooner. The job descriptions for the computer and AI scientists state that they’ll be working on things like identifying talent, informing development pathways, helping manage injury, and helping evolve tactical principles. While clubs at different levels may have different needs and situations, a research and analytics team like the one CFG are now hiring for could produce insights that can be used across the board.

And, of course, all paths in the City Football Group lead to The Big Club back in England. If CFG can improve development patterns at Montevideo and Mumbai, that will be in service of the City brand globally and in presenting pathways, possibly, to Manchester itself.

The AI scientist advert is the more interesting of the two. It, and not the computer scientist one, has the point in the job description that reads: “research & develop AI models that will evolve the tactical principles utilised by our teams across CFG”.

I speculatively wondered aloud, only being half-serious, nearly a year ago whether Liverpool’s analytics department could be behind the team’s tactical switch to using full-backs as crucial attacking outlets. Trent Alexander-Arnold being damn good at kicking a football obviously helps, but there were puzzle pieces to put together if you were after a conspiracy theory to latch onto.

How this would work in practice is probably that the analytics/research team would do a project and then findings would be presented to and filter up the hierarchy. At Liverpool, this would be first to head of research Ian Graham, and then maybe sporting director Michael Edwards. We know that Edwards (who reportedly nudged Liverpool to hire Graham) is best buds with Jürgen Klopp — thanks to The Athletic, we know that they play padel together so often that they arranged for a court to be built at the new training complex (yes, the one that the women’s team might or might not be allowed to use). There would seem to be a pretty direct research team -> Graham -> Edwards -> Klopp line of communication at Liverpool men’s team then.

Sadly, we do not have reporting on whether City’s Director of Data Insights & Decision Technology, Brian Prestidge, plays, I don’t know, golf with Guardiola or checkers with Txiki Begiristain. I would imagine, though, that the path from research department to manager is a lot less smooth at Manchester City (or, quite frankly, any club) as it is with Liverpool’s men.

Both the AI and the computer scientist roles have this phrase in the job description: “Play a key role in developing our Narrow AI platform to identify talent, inform development pathways and understand & predict the football transfer market”. ‘Narrow AI’ is a specific term we don’t need to worry about*, but what we can all take from this is that CFG have a platform of sorts, that incorporates AI, to do transfer stuff.

*it just means that it’s focused rather than a broad use of AI

To pre-empt any rash headlines though, ‘AI’ is a very easy term to BS about. I’m pretty sure a spell-checker is ‘AI’. CFG are probably using something more sophisticated than Clippy, but it’s not necesarilly HAL 9000 either.

The mention of ‘predict the football transfer market’ is particularly interesting. I think that everyone reading this newsletter will be aware of how analytics could help in identifying talent, and developing player pathways is also something that has been talked about publicly (Memphis Depay used Dutch consultancy SciSports to help him choose where to go after Manchester United).

But ‘predicting the market’ is something else. Is this about finding where the value in the market is going to be in, say, a year’s time? Picking the best moments to buy or to sell? Maybe, particularly in the present landscape, it could be predicting where market values are likely to settle and therefore work out what a ‘good price’ is likely to be.

Another interesting point from the AI scientist description is this: “Drive further development of our machine learning models that enable our Human Performance practitioners to better manage fatigue, injury & illness.” The word of interest here is ‘further’, implying that this is something that the CFG are already doing.

However, the thing that catches my eye the most from these pair of ads is in the one for the computer scientist: “Play a collaborative role within the DIDT team in advancing our computer vision, simulation & reinforcement learning environments.”

‘Computer vision’ is really what it says on the tin: feeding images into a computer for it to ‘see’ things. I may be simplifying things slightly here, but camera-based tracking data works by using computer vision algorithms to ‘see’ where the players and ball are on the pitch.

Traditional data companies like Opta and StatsBomb are also using it to help supplement their data gathering. The latter company have written about their method and its use in helping the accuracy of their location data — and ‘freezeframes of surrounding players — for shots.

Are Manchester City developing similar systems to produce their own data? This wouldn’t necessarily be in creating an in-house Opta, it might simply be to collect metrics that other providers don’t collect, linked to existing data sources. For example, we know that Guardiola is big on counterpressing: you could link a feed of event data to the video to pinpoint moments when a City player loses the ball, and then use a computer vision system to look at how many teammates are in a close enough vicinity to counterpress.

There’s a bunch of stuff in the computer scientist job ad that’s functional but important. Lines in the job description about ‘placing the power of our AI models in the hands of stakeholders’, ‘conducting research through applied experimentation and modelling’, ‘productionize our AI-based solutions’. This all seems like ‘you’ll do research, then help us make sure people within CFG can understand it’. Fair.

Meanwhile, the AI scientist role is more of a senior job, requiring more years of experience in a related field than the computer sciencist job and it being desirable to have “previous experience in playing a leadership role within a team.”

The two jobs represent a new stage in a rapid expansion of the team. Over the summer, Ravi Mistry was hired as Football Intelligence Officer and John-Mark Sisman as Performance Physicist (short bios on the Training Ground Guru website here). Sisman presented at the 2018 OptaPro Analytics Forum on pressing styles. I don’t know where I made notes, and didn’t tweet about it, so here’s someone else’s:

John-Mark Sisman’s presentation on analytical approaches to pressing styles was top drawer. Passing lane disruption, with video analysis of Sergio Aguero doing it perfectly, fascinating stuff #optaproforum pic.twitter.com/29HsH8UrMr— Nick Goff (@nickgoff79) February 7, 2018

With these two new hires added in, CFG’s DIDT department will be pretty bulky, and will cover: physics, computer vision, machine learning, productising the insights, and data communication. Four members will have joined the team (who below to a ‘department group’ called ‘Football Know-how’, that I find too amusing not to mention) since around July.

Screenshot of the CFG job listings, with the Department Group given as 'Football Know-how'

The challenge to Liverpool’s analytic-minded strength is coming. Yes, it’s coming from one of the richest clubs in the world that also own a stake in nine other clubs across five continents. Yes, that might be a disturbing sign about football’s ecosystem. But yes, this does also mean that there will be a new contender in the Premier League for the biggest collection of brainboxes stuck in a room somewhere who may or may not be listened to by the famous people that everybody else cares about.

It was… inevitable.

So, how did the pandemic transfer window end up?

Tue, 06 Oct 2020 00:00:00 GMT

Welcome to Get Goalside! I’m glad to see you.

The transfer window. It is closed. Although in England it’s still kind of open for a little while longer for domestic moves (and as far as I know that could be the same elsewhere), so it’s sort of drafty. Which is probably good for air circulation.

I kid, but we should take a moment before we get into the millions spent on football players, to acknowledge how terrible the pandemic has been for so many. The 2020 summer window will surely be come to be known as a turning point for clubs, who either got the best out of a strange market or who kneecapped by it, but even within football it feels crass to speak of this as a moment of change of significance.

Lower down the football pyramid, clubs are struggling for their existence or have already been wound up. Many individuals have lost their health, jobs, or lives. It’s uncomfortable to talk about this, but it would be worse to talk about the pandemic’s impact on the 1% of football without acknowledging it.

Figures correct as of Tuesday morning UK time.

At the very, very top of the footballing tree, this transfer window has, indeed, been a different one. Anecdotally, it feels like there’s been a lot more talk of loans between big clubs, a lot more discussion about deals being structured as ‘buy now, pay later’. And yet, the expected ‘pandemic pinch’ hasn’t seemed to be felt everywhere.

All four of the non-Premier League ‘Big Five’ leagues have a league-wide net spend of close to zero, but the Premier League appears barely touched (Ligue 1, I presume, is splashing some of the cash they’ve built up from selling a lot of good talent in recent years).

Part of this lack of drop in Premier League net spend is probably the way teams have chosen to structure deals. For example, Liverpool reportedly only paid around ten percent of the fee up-front for Diogo Jota, but that still appears as a ~£40m outlay on transfermarkt.

However, if the lack of drop-off in Premier League net spend is largely a factor of deal structuring masking reality, it seems interesting that other leagues aren’t doing the same. Perhaps some in the Premier League are just more sure of their financial future.

Total spending in the Premier League has dropped (by transfermarkt figures),

It’s just that nobody seems to want to buy — or the clubs are reluctant to sell — Premier League players. Apart from the 2018 summer window, this is the lowest amount that Premier League clubs will receive from transfers since 2013.

If you like this nifty chart, share this post far and wide and subscribe to the newsletter

It seems remarkable that a league that spends so much on transfers has been receiving so little, but I gave that chart the title I did for a reason. Arsenal’s struggle to shift — or really do anything — with Mesut Özil seems emblematic of a brief window of millstone-sized contracts. These players (and Özil certainly wasn’t the only one) were hard to sell previously, and are presumably much more so in a pandemic economy.

The next few years will see the impact of this play out, and it’s interesting to see how cycles change. On a club level, Tottenham Hotpsur have had to enter a high net-spend era, after close to a decade of keeping that figure low thanks to selling off big assets and mostly keeping expenditure at a similar level to transfer income.

This visualisation type was originally seen in Get Goalside! just over a year ago when I wrote about 'the dangers of being frugal'.

But how does the Premier League end up so far ahead of the other ‘big’ leagues in Europe in the first place? Is this all driven from the top — the clubs fuelled by oil, consumerism, and questionable moral values — or is it something happening across the board?

Jason Burt of The Telegraph highlighted that around a third of the EPL’s total spend this summer is coming from Chelsea and Manchester City. It’s a far cry from a choice quote he picked out, of Saint-Etienne manager Claude Puel talking about their sale of Wesley Fofana: “Between saving the club and its employees, and keeping Wesley, there was no debate.”

A quick visualisation shows that the Premier League as a whole has a different pattern to other leagues. Over the past couple of years, it seems that ‘Big Five’ net spend in the summer window has been around +€50m to -€20m, with two or three big-hitters stretching up above that. But not the Premier League…

England’s top flight has a ‘middle class*’ of teams and their spending — much like the other major teams. But whereas the Continental middle class seemed both smaller and closer to the pack, the Premier League’s seems to spread between the +€50m and +€100m net spend for the summer window. On top of that, unlike in these mainland European leagues, the Premier League ‘middle class spend’ hasn’t dipped this year.

*I should note, this ‘middle class spend’ isn’t necessarily Top 6 clubs. Here I’m specifically referring to spending capacity rather than performance.

It’s astonishing that the Premier League has appeared so untouched by the pandemic in terms of transfers, particularly when some clubs have been laying off staff.

Love to demand to be treated as a business when it benefits me but also as a custodian of a public trust when it benefits me. https://t.co/C31JcnoLij— Mike L. Goodman (@TheM_L_G) October 4, 2020

Will we see the impact of this pushed down the road into the winter window (which is always a bit weird anyway) or to next year? And/or is this just confirmation that the Premier League is, for good or for ill, in a league of its own in Europe?

The nerd's case for exciting football

Tue, 29 Sep 2020 00:00:00 GMT

Welcome to Get Goalside!. It’s good to see you.

Times continue to be hard - consider giving to your local food bank if you are able. If you need to, you can find a foodbank in the UK here.

The people who plug numbers into computers and point to spreadsheets like E.T. points to home are much maligned. Short corners, an overreliance on retaining possession, buying Roberto Firmino: the list of things the nerds can be blamed for is ever-added to.

In particular, there is (or, judging from other sports, soon will be) fears that analytics homogifies styles of play into one ‘Right Way to Play’. There are many features of the game that get given approval or disapproval by numerical experimentation: cut-backs - good; crossing - bad (revisionist takes are available); through-balls - good; long shots - bad.

However, football is a fluid enough sport that I think you could make an argument for playing an exciting style of football which is backed up by ‘analytics’.

First of all, take the Red Bull clubs (other brands of energy drink are available, and encouraged). Although their high-pressing, quick-transition approach to the game didn’t stem directly from number-crunching, it’s one that analytics bods like because it’s so clearly effective.

You don’t need me to tell you that losing the ball close to your own goal is bad, but the statistical work has, indeed, been done on it. If we look at these numbers from American Soccer Analysis, winning the ball just inside the final third could be four times as valuable as winning it just inside your own half. [Full piece here. Team attacking the goal at the top in the left-hand pitch, and defending the goal at the bottom in the right-hand pitch. Methods of similar metrics vary, image used as illustration of the idea]

And as for attacking quickly, as the sugar-water-sellers endorse, well expected goals models have long used ‘speed of attack’ as a proxy for an absence of defenders around a shooter. A shot from 18 yards generally has several opponents in the way, except when it comes at the end of a counter when it might be close to a 1-on-1. Taking advantage of your opponents being out of their shape is undeniably good.

Not only could you find some stats to back up this style of play, but analytics could help keep the fun going for longer. Players only have limited amounts of energy, and knowing the most effective ways to press means that this precious resource can be used when and where it counts.

The first football club to incorporate body pose software — where you give the computer the video and it can tell you what way all the players are facing — could have a major advantage here. Suddenly you can have the direction that every player faces when they receive every pass. Who are the players who receive on the turn, and who are the ones who are closed off?

Not only can that latter group be pressed, but if there are players in the first group who receive on the turn in predictable ways, then they could be vulnerable to having their pockets picked. I pity the central midfielders who are playing the game when this first comes in. There’ll be no escape.

Enjoying this? Share it around:

Now, we need to talk about passing. This is an area where analytics — the use of data — seems to have been mixed with the analytical mindsets of certain schools of coaches.

There’s definitely a case to be made for retaining possession as a method of ‘defending’ (in fact, several moons ago I wrote a piece about ‘sterile’ possession being Pep Guardiola’s version of parking the bus). Having possession of the ball is also, generally speaking, a requirement of creating chances.

However, football’s a sport where possession switches hands (or, feet) around 100 times a match, and only three goals are scored, on average, per match. Is it really so bad to lose the ball? Elsewhere, Kees van Hemmen has argued that the good teams should look to create open games to give them a better probabalistic chance of winning.

A part of that could be through another undeniably fun facet of the game: dribbling. Coincidentally, the rise of what I shall call the ‘Wyscout generation’ of stats-curious online people has also seemed to lead to a rise in interest for dribbling. Dribbling is fun, and checking which players dribble the most is fun.

But dribbling is also a perfectly good way of getting the ball up-field, which is valuable. Not only that, but the type of ball-carrying that central defenders can do when they’re given time and space can disrupt the opposition’s defensive shape.

This latter thing is another of the things that ‘analytics’ (or: the use of data) can quantify, with tracking data and the expertise to use it becoming increasingly available. In 2019, Mladen Sormaz and Dan Nichol gave a presentation at the Opta Pro conference where they looked at the damage to team shape that off-ball runs made; presumably the same thing could be applied to on-ball carries causing an opponent’s midfield structure to collapse.

Feasibly, you could construct an analytics-powered game model of high-press, high-transition, high-risk, high-fun football.

Although, that said…

And Dribbling for Math may be even worse than not dribbling at all.— Jared Wade (@Jared_Wade) September 27, 2020

Subscribe now

Shout-out corner

Forgive me for shouting out my own thing. I’ve launched a brand new newsletter, going out each Friday, where I write specifically about the football on-the-pitch.

Get Goalside! will continue into what it’s kind of morphed into: writing about the data/tech/theory of football and where the future of football might be heading.

Mark’s Notebook will be just football. The first edition was about how Manchester City’s counterpress worked in their 4-2-3-1 against Wolves (this was, you’ll note, two days before it became such a big deal in their defeat to Leicester).

If that sounds like your thing, feel free to subscribe and/or share the news. There’s a link to the newsletter’s site here. There’ll occasionally be short pieces that go up on that site, like this about Wolves being Wolves-ed by West Ham, but the newsletter itself will be one full article every Friday.

Is football a 'weak link' sport?

Tue, 22 Sep 2020 00:00:00 GMT

Once upon a time, there was a book called The Numbers Game, by David Sally and Chris Anderson. In it, among other things, was a chapter on football and O-ring theory, teaching a young generation of nerdy football watchers two things.

One: the story of the Challenger disaster, which tragically exploded, killing seven crew members, due to a tiny component failing in cold weather, the O-ring. Two: that football is a sport where the weak links make the difference.

The theory — which started off in that most sporting of realms, economics — is that if you have a bunch of workers who each have tasks to do in a project, one person who’s worst has the most effect of the bunch. If you have someone who’s half as good as the rest, they’re going to hold everyone else up; but if you have a worker who’s twice as good, their work will still need to rely on the normies pulling their weight.

The book’s application of the idea to football took hold, appearing in FourFourTwo magazine and an episode of the Revisionist History podcast. In a sense, it’s echoed by José Mourinho and other coaches who are said to believe that the team who makes fewest mistakes wins. In this school of thought, football is a ‘weak link’ sport, and if there’s a trade-off to be made between maximising a star player and covering a frailty then it’s the frailty that must be protected and patched.

But there’s a problem.

In 2016, Stefan Szymanski and Guy Wilkinson published a paper which argued exactly the opposite: it’s not the worst players on the team who count most, it’s the best. (Wilkinson’s doctoral dissertation is available online, the first section of which is essentially the 2016 paper).

Their work isn’t quite a case of ‘superstar theory’ — which would be that the best player on the team matters most — as it argues that the top two players are equally important, with the third-best close behind. But these top few are clearly the ones that matter in their work. Their ‘optimal strategy’ says that these three should account for around 70% of the wages on the team.

Hopefully that last sentence made you go ‘…wait, what?’.

Because now we exit the ‘these are the technical papers’ phase of this post and into the ‘what do the technical papers actually mean’ phase.

If you’ve got this far and are finding it interesting, why not subscribe?

One reason why Szymanski and Wilkinson’s optimal strategy may have had so much of the wage bill focused on so few players is that they just used starting XIs. They note that, in reality, teams may choose to spread their spending a bit more evenly in case their top players get injured. Or that it might be good management of a workforce not to spend 70% of the wages on less than 30% of the starting line-up.

Neither study used any hugely sophisticated methods for gauging ‘player value’ because, to be fair to them, they couldn’t. Football still doesn’t have an all-in-one performance metric that can be used for work like this. Possession value models[1] are promising but are still very much in the ‘iteration’ phase, and I’m not sure how many deal well with the Valley of Meh issue. Also, running the model for the number of seasons and competitions that would be needed for this work may be beyond the scope of many researchers.

This doesn’t mean that Sally and Anderson or Szymanski and Wilkinson’s work is fundamentally flawed, just that they would appear more as works of theory than works of experimentation. That said, when you have two studies that say completely contradictory things, the method does kinda matter.

It’s also kind of a problem that both theories seem plausible. You could quite easily argue in either direction.

For O-ring: “Teams regularly build game-plans, in part, around protecting their worst players, and quite often their better players have to limit themselves to accommodate worse players. Besides, Messi at Burnley wouldn’t make Burnley into world-beaters.”
For ‘get a good main trio, forget the rest’: “Better players don’t just add their own skill to a team, they help unlock talent in others. And, y’know, good players are good! Besides, Chelsea had the worst goalkeeper in the league in 2019/20 and still finished fourth; if your top players are good enough you can mask a weak link.”

To me, football is too much of an inter-connected sport for things to be as simple as coming down to the strongest link or weakest link.

If a team plays an intricate possession-style, it would probably do them good to avoid a weak link in midfield. But a team playing a zonal defensive system might find it better to have one or two stronger links to help organise the players around them than improve the weakest player. A poor finisher in a strong attacking team might be better than a very strong poacher in a weak attacking team. A weak link at full-back might be less serious than a weak link at centre-back.

Maybe instead of the theory argument being strong link vs weak link it should be strong selection vs weak selection, or maybe strong unit vs weak unit. Anyway, it seems not to be as simple as O-rings after all. But at least I still know what the Challenger disaster is, I guess.

Kindness is what you do. Be kind to people.

[1] || "Possession value models..." ||They do what they say on the tin, model the value of actions in a spell of possession. Examples include American Soccer Analysis’ Goals Added, Opta’s Possession Value added, Karun Singh’s Expected Threat, KU Leuven and SciSports’ Valuing Actions by Estimating Probabilities

Why we could be close to a return of hoofball

Tue, 15 Sep 2020 00:00:00 GMT

As the ex-pros from the ‘80s made us very well aware a few years ago, nothing is new under the sun in football. ‘Pressing’? They did that. And pretty soon footballers of the ‘90s will be lining up to tell us how they used to specialise in the trendy tactics of today: deep blocks, deep crosses, and counters.

Look around the Premier League and what do you see. Liverpool turned their midfield into a doldrum of competency to shift the creation to their full-backs; Manchester City trademarked obscenely curved crosses from the shoulder of the box; Sheffield United and Wolves both sprung from the Championship to Europa League contention with deep blocks, heavy use of wing-backs, and, in Wolves’ case particularly, dangerous counters.

The symbol of this turn in trends, from a midfield possession focus to 'something else', could be Mikel Arteta’s Arsenal. Having grown up under Arsène Wenger and Pep Guardiola, the natural assumption would be that his teams would be clocking up 60% possession shares each game.

But no. By FBref’s numbers, his team and Scott Parker's Fulham had a comparable share of their touches in the final third and penalty area: Arsenal on 23.7%, Fulham on 22.1%. This isn’t because the two teams were of comparable quality; it was because Arteta’s team often dropped into a deep, compact block. Fulham were able to move up the pitch relatively easily, getting touches in the fringes of the final third, but it was all part of the away side’s plan.

When Arsenal get the ball, they show part of the reason why this is a wise move. Although Arteta’s side are an extreme example of this, teams are getting better and better at playing out from the back under pressure. Not all of them go ‘full-Gunners’, passing across their own six-yard box as if they were doing a training drill, but quick exchanges around the team’s own penalty area are seen across most teams in the league.

If teams are better playing out from the back, the benefits of pressing high necessarily drop. And, because high pressing was always a high-risk, high-reward strategy, the lowering of benefit also necessarily means an increase in danger.

And, therefore, Mikel Arteta’s Arsenal are content to sit deep, even against Fulham. The siren call of the deep blocks may well tempt more sailors as time goes on.

But crossing: surely Liverpool and City latching onto that is, in large part, a symptom of excellent crossers in Kevin de Bruyne and Trent Alexander-Arnold?

Well, maybe not, because there’s a veteran Premier League practitioner of the crossing game who’s something of an unlikely analytics darling. Tony Pulis.

In 2017, Neil Charles gave a talk at the Opta Pro analytics forum where he said (quoting from this Sky Sports piece):

"Everything that West Brom play into the box, virtually, comes off the shoulder of the penalty area on the diagonal. Everything. People have pointed out it is so that he cannot get counter-attacked."

Charles's model then looked at the location of West Brom's shots too and an interesting connection between the two emerged. "Their shots come from an area just outside the six-yard box and central," he added. "West Brom don't shoot from anywhere else, pretty much.

"What does the model think you should do from that (crossing) position? It thinks you should hit a point just outside the six-yard box. It's almost spot on as a strategy. It may not be the prettiest strategy but it's an effective one and we can quantify that."

No doubt that further work has been done in the meantime, but if football tactics are a world of evolution that it shouldn’t be surprising that there are things of value to be mined from the past. That includes deep blocks, and it includes Pulis-esque crosses.

One of the reasons for Pulis’ choice of cross — the defence against counter-attacks — has become even more vital in the slightly-more-modern age of 2017. There are an increasing number of ‘top teams’ whose main tactical feature is their ability on the break. An attacking strategy that minimises your vulnerability to this makes sense.

However, evolution doesn’t return to trends of the past for sake of being retro. It takes what it needs, jettisons what it doesn’t, and splices in things from outside that prove to be useful. The big addition that we might see in this ‘English football 2.0’ is… basketball.

Basketball is a sport that interests a lot of people in and around football for a number of reasons. First, it’s fun. Second, and more importantly, the control and manipulation of space is similarly important in it as it is in football[1].

Still, the principle is neatly summarised in this video: offences space themselves to stretch the defences as much as possible. Doing it opens lanes to drive towards the basket, either to get a high-quality shot (see this for values of shots from specific areas) or to draw defenders and kick the ball back out for someone for an open three.

It’s taken basketball a long time to get where it is today (see this for the evolution of offensive spacing, and much thanks to friend of the newsletter Mohamed Mohamed for thoughts and links). It would make sense for football to take a while too. But football working out how to maximise the space around the penalty area, taking inspiration from the NBA, seems very possible.

The start of it’s already here, in those very crosses that Kevin de Bruyne specialises in (although he isn’t the only one, and City aren’t the only team). Those ‘shoulder of the box’ crosses are often created by teams pulling defenders away from the arae where those balls will be delivered from, the creator-in-question strolling into the space just as the pass is played backwards to them.

The medium-term future of football tactics could be deep blocks, deep crossing, and basketball. Not quite hoofball, but the former hoofballers will be sure to tell us that they were doing it all first.

[1] || "Second, and more importantly, the control and manipulation of space is similarly important in it [basketball] as it is in football" || That said, three-point shots make things different for two reasons. The first is that long-range shots are worth more points, but the difference in probability of a three-pointer being made vs a shot from inside the paint is much different to that of an 8-yard shot vs a 20-yard shot in football.

The case against buying Lionel Messi for €100m

Sat, 29 Aug 2020 00:00:00 GMT

Hello, and welcome to Get Goalside!. Glad to see you, it’s been too long.

This week’s charity is Macmillan Cancer Support. Please consider helping them if you can.

After writing this late on Saturday 29th, I decided that due to the probably-fast-paced nature of this situation I’d send this out straight away, rather than waiting for the usual Tuesday release.

This is not a ‘look at me’ hot take (although I am glad you’re reading it).

I’m starting this post assuming you’ve seen the rumour that Manchester City may offer €100m plus various players for Lionel Messi (pending whatever legal shenanigans are involved in his contract dispute with Barcelona). As the title of this newsletter says, I want to offer the case against buying Lionel Messi for €100m. That’s regardless of any other makeweight players. That €100m on its own is too much.

But I also want to be clear on the specifics that this statement, just like the rumour, applies to Manchester City.

Look, I understand the pull of the “if Lionel Messi is available, you buy Lionel Messi” point of view. As I’ve noted on my other substack, Messi’s an incredibly large part of Barcelona’s attack. His goal contribution rate has been better than all of his teammates put together in no fewer than six seasons during his career. Even approaching 34, he’s really good.

But as I’ve noted on the ol’ work blog at Twenty3, he doesn’t defend.

[Image taken from blog post linked above; data from Wyscout — stats for this and all future visualisations are from the 2019/20 league season]

“Yeah, sure,” you’re probably saying, “but he’s Messi. He doesn’t need to defend.”

You’re right.

But allowing Messi not to defend means that everybody else has to.

In the blog that the above image comes from, I was arguing that Barça needed to get rid of Luis Suárez for that exact reason. Neither he nor Messi do much defensive work — whether through will or, more probably (in Suárez’s case at least), physical capability — and I think it limits the team quite a lot. You can’t press high if you have two forwards who don’t put in a shift, for example. And we know how much teams love to press high these days.

One problem at City is they already have a forward who doesn’t do much defensive work:

I’ll grant you, this problem isn’t insurmountable. Particularly because City already have a centre-forward who very regularly puts in a shift when out of possession…

[Image taken from the Twenty3 blog post — here’s a link again]

However — and this is the real point — City are already really good. Although Liverpool very well may be doing something that expected goals models aren’t catching*, pretty much every model going had Pep Guardiola’s team ‘ahead’ of Jürgen Klopp’s last season.

[*A very reasonable theory that could account for a lot of this is that Liverpool manage games after going ahead whereas City are more likely to continue racking up chances. Stat-padding, but for teams. Nonetheless, City are really good.]

At the very, very least, City are clear title contenders in the league, and one of the best teams in Europe. All while not needing Messi to be a part of the finely calibrated system. I’d also bet money that, prior to a few weeks ago, Lionel Messi wasn’t in any of their plans for how the squad would be built and how the team would be constructed.

So you’re going to pay €100m for:

A 33-year-old
Who doesn’t defend
Whose lack of defending may require you to bench your main forward
Whose arrival you haven’t planned for
And which might (probably will) throw off your short- and medium-term squad-building plans
When you’re already one of the best teams in Europe

If you haven’t been completely put off by this fiery contrarian take yet, please consider sharing and/or subscribing — it’s much appreciated

Subscribe now

There are a couple more boring tactical points about Messi’s arrival at City that aren’t interesting enough to spend much time on.

The first is that Messi’s tendency to drift inside means someone needs to be providing width high on the right wing the whole time. Can 30-year-old Kyle Walker do that?

Alternatively, Messi could be given a free role behind a striker in a 4-2-3-1, which would at least have a full-time right wing player. But then they have the central midfield duo problem that they’ve struggled with this past season.

I should stick up for Rodri and İlkay Gündoğan here because I think they’ve partly suffered from circumstance. By my eye, it was City’s counterpressing that enabled them to fit in four attacking midfielders to their steamrollering 4-1-4-1, with Fernandinho as the anchor, and trouble comes when that pressing doesn’t work. It’s perfectly possible that Rodri and Gündoğan are quite good, it’s just the press in front of them is letting them down. But Messi’s not going to help that much…

‘It’ll pay for itself in shirt sales…’

The whole ‘shirt sales will pay back the transfer fee’ line has roundly been debunked over the past few years, but we might be able to replace ‘shirt sales’ with ‘VIP boxes’. I’m also not being wholly serious, because it clearly wouldn’t pay for the transfer fee plus wages, but the City move for Messi has carried with it the rumour of a move to NYCFC after a year or two.

Imagine… Lionel Messi, in a twilight of a career that’ll still surely be sparkling, in New York City — how much would the richest of the rich in NYC pay to see him play?

Pep Guardiola can probably work out a way to fit Messi into his team, back in Manchester. Messi will probably carry on working hard for his one-time Barça boss. Even if it isn’t smooth sailing, City will probably still have more than enough to compete for the titles they care about.

And, when all is said and done and the nerds have finished blogging, “if Lionel Messi is available, you buy Lionel Messi”.

Subscribe now

Shout-out corner

Shout-out to Millie Bright’s banger in the Community Shield, particularly for my own nerdy context around it in the game.

Manchester City had just gone down to ten players and had gone to a 4-1-4-0 out of possession. Amateur tactico here was thinking ‘this’ll be interesting, they’re ceding a lot of space to the centre-backs and forcing them to make decisions on the ball’. And then Chelsea and England centre-back Bright steps up and does this.

If you don’t have ESPN+, take the minute to watch this Millie Bright goal for Chelsea because goddamn. pic.twitter.com/wH9gV7gaAn— Meg Linehan (@itsmeglinehan) August 29, 2020

And, not sport, but considering events in the US my thoughts go to this piece from June on the New Statesman by Rebecca Omonira-Oyekanmi: In Britain, we have our George Floyds too.

This week’s charity is Macmillan Cancer Support. Please consider helping them if you can.

Donate to Macmillan Cancer Support

James Milner, Hugh Grant, and versatility

Tue, 11 Aug 2020 00:00:00 GMT

Welcome to Get Goalside! The first rule of Get Goalside! is that you’re always welcome to a second biscuit.

This week’s charity is Refugee Action, who support refugees and people seeking asylum in the UK.

At the weekend, I asked a question on Twitter. Who — outside of James Milner, who would surely be the obvious choice of my largely-Premier League-focused followers — is the most positionally versatile player out there at the moment?

Then my notifications exploded

Say no more https://t.co/3aYbTv9Znl pic.twitter.com/mCdLy96uo0— Aymeric Laporte (@Laporte) August 9, 2020

The player who really inspired the question, though, (via Kim McCauley on the Shirtless Plantain Show podcast) was Crystal Dunn. For her country, the American plays as a starting left-back; for her clubs, she’s an indispensable central midfielder or forward.

It’s a slightly different type of versatility to Milner. A few of the (many) responses to my tweet reckoned that Milner was jack of all trades, master of none. Whether that’s a fair opinion is for you to decide, but it has an air of truth about it (a kinder way of putting it might be that he’s second or third choice everywhere on the pitch). What’s for sure is that the same certainly can’t be said for Dunn.

As the responses rolled in, other kinds of versatile players were mentioned. There were ones like Joshua Kimmich and David Alaba, who’ve settled into playing in pretty consistent positions but who it’s widely agreed could play anywhere.

There were others, more in the Milner mould…

Sergi Roberto, the Catalan James Milner.— Grace Robertson 🏳️‍⚧️ (@GraceOnFootball) August 9, 2020

I think of this category as ‘X plus a lot of Ys’. They’re people who predominantly play in one position (for Milner, central midfield; for Sergi Roberto, right-back), but are surprisingly adequate at filling in elsewhere.

The line between these and another category — the John O’Sheas of the world — is pretty thin. They get appearances because they’re versatile which they wouldn’t if they weren’t, but if O’Shea had spent his career lower down the league than Manchester United then he’d have probably secured a consistent position for himself. For some, maybe ‘versatility’ is a sign of (useful) substitute status.

Enjoying this so far? Share the love

Young players are tantalising for this reason. Both Bukayo Saka and Ainsley Maitland-Niles were mentioned very frequently — two players who’ve played a variety of positions for Arsenal in their short careers, mostly on the flanks but with the variety of roles that means it’s not as simple as ‘a winger playing further back’.

But the players who interested me most were the ones like Crystal Dunn: Cardiff City’s Callum Paterson, Wigan’s Kal Naismith, and West Ham’s Michail Antonio. All three are players who, like Dunn, have been regular starters at full-back and then, with a different manager or different team, switched to playing a completely different position (centre-forward for Paterson and Antonio, attacking midfield for Naismith).

With most ‘versatile’ players, there’s some kind of pattern to their versatility. Although James Milner, for example, can play in a lot of positions, there’s a certain inescapable Milnerness to his performances. They’ll be positionally disciplined and fairly dependable, with strong hints of defensive solidity. Others players, despite moving around the pitch, are very much ‘wingers who defend’ or ‘centre-mids who can play wide’.

None are quite as fun as ‘full-back who also plays striker’, and it also leaves me wondering to what extent it makes sense to summarise the concept of versatility using positions at all.

Subscribe now

Kyle Walker (hello Aymeric) has played as a ‘centre-back’ almost as much as ‘right-back’ for England in the past several years. There’s no doubting the football intelligence it takes to do this, but his centre-back role is quite full-back-y. One could argue that his developing versatility between inverted right-back and bombing-down-the-wing right-back is at least as impressive, if not moreso.

In a similar vein, I remember someone responding to my tweet that Kevin de Bruyne could probably play anywhere. I’m firmly in the ‘After Messi, KDB Is The Best In The World’ camp partly because of this: not only could the Belgian play in a bunch of different positions, but I think that he’d be able to fill very different roles within those positions too.

De Bruyne has the playmaking ability of a traditional number 10 (whose skillsets, as David Silva’s career progression has shown, can be moved around the midfield area); he has the delivery of a sweeping ball-progressor, whether that be in the style of Trent Alexander-Arnold or Paul Pogba; he has the pressing intelligence and energy (although this has dropped a little from its peak) of a more defensive-minded player.

Maybe this is it: maybe the mark of true versatility isn’t how many positions you could play, but how many different types of player you can emulate. It’s like acting: Hugh Grant is more or less Hugh Grant in any role he’s in. And then you have, like, [googles ‘actors with range’] Charlize Theron.

James Milner is like a slightly more versatile Hugh Grant. He can play left-back, or right-wing, but he’s always still James Milner. The Kal Naismiths, Callum Pattersons, Michail Antonios are more like Charlize Theron or [more googling] Ralph Fiennes: they can be in one role one day and a completely different one the next.

And the Kevin de Bruynes and Crystal Dunns are like Meryl Streep: incredible range, Oscar-worthy winning performances, and the first name you’d put on the poster whatever role they’re in.

Subscribe now

Shout-out corner

Shout-out to Parth Athale for just casually putting a webapp together to identify statistically similar players.

So I've resorted to Colab, here's the link: https://t.co/Qw7n47SSfk

You just need a Google account for this, and run the three cells to run the app, it's pretty easy - you can do it on your phone as well.

Really wanted non-coding people to access it easily, hope this does it :) https://t.co/nAAekJQnr3— Parth (@pathaleee) August 6, 2020

Also, to Will Thomson, for just up and making a sophisticated algorithm, like, a ton faster than some pros have it:

This appears to be a genuinely huge contribution to the soccer analytics community. My fastest implementation is ~80x slower. https://t.co/PbKqLQp0Xl— Devin Pleuler (@devinpleuler) August 5, 2020

On a different subject, this past week The Athletic wrote an article about Robinho that was… flawed. [CW mention of sexual assault below]

In 2017, the Brazilian was sentenced in Italy to nine years in jail for rape; he’s been out of the country and appealing the verdict since, so hasn’t served any time so far. Still, the article from TA was written in a way completely absent of the sensitivity one would expect when writing about a person convicted of such a crime. No surprise as (one of) the writer(s) admitted to not having remembered the incident:

At least he owned upp to not doing his research. pic.twitter.com/ZFVe4oWYWs— Wiebe (@WieWout) August 10, 2020

I wrote a thread of elements in the article that, in my view, should’ve been edited better, but the crux is:

Having trouble viewing this tweet? Click here to open original tweet.

Anyway, the actual shout-out here comes because I would have a far less well-developed sense and sensitivity of the problems in the article were it not for things like the Burn It All Down podcast. It’s currently on a summer break, but I highly recommend subscribing for when it comes back, and checking out past episodes.

This week’s charity is Refugee Action, please consider helping them.

You can donate to Refugee Action here

The return of the 'carry the team' centre-forward

Tue, 04 Aug 2020 00:00:00 GMT

Hey, welcome to Get Goalside!. Sit down, I’ll fetch a drink.

This week’s charity is the Trussell Trust, a UK food bank charity. If you can spare it, please give the amount you’d spend on a meal out, or just a Big Mac.

All stats in this post from FBref

Most Premier League seasons see at least one team’s highest scorer account for at least a third of their goals.

Not 2008/09.

2009/10 only managed it thanks to Darren Bent’s remarkable haul of 24, an even half of Sunderland’s goals that year.

2010/11 only just met it, with Carlos Tevez scoring exactly a third of Manchester City’s goals. These were doldrum years for the Premier League in terms of strikers who could be relied on to carry their teams on their backs.

Although Didier Drogba scored 29 in 2009/10 (a season when Chelsea scored an incredible 103), the golden boots either side of him were won by Nicolas Anelka with just 19 in 2008/09 and Tevez & Dimitar Berbatov with 20 in 2010/11.

But now the ‘carry the team’ striker is back. In 2019/20, we had:

Danny Ings, 43% of Southampton’s goals (22 of 51)
Teemu Pukki, 42% of Norwich City’s goals (11 of 26)
Pierre-Emerick Aubameyang, 39% of Arsenal’s goals (22 of 56)
Jamie Vardy, 34% of Leicester’s goals (23 of 67)
Raúl Jiménez, 33.3% of Wolves’ goals (17 of 51)

That’s five teams whose high-scorers have made up a third or more of their side’s goals. And in five of the past six seasons, there have been four or more such teams/players (last season being the exception, with just two — Glenn Murray and Jamie Vardy (again)).

It’s the most sustained period of ‘main man’ strikers that the league has seen since the mid-90s and the likes of Alan Shearer, Ian Wright, Robbie Fowler, Les Ferdinand, Homes Under The Hammer’s own Dion Dublin.

You might have noticed that there’s a real mix in the clubs that are represented. While there’ve been some different trends over time, there’s very little skew as to what kind of team is likely to see this kind of season.

Some of the aforementioned trends: the mid-90s spell accounted for a large chunk of the ‘top four’ places; 2014/15 was an incredible year where three of the bottom four had one of these ‘main man’ strikers: Charlie Austin at QPR, Danny Ings at Burnley, Christian Benteke at Aston Villa.

Subscribe now

But the ‘main man’ tag that I’ve infrequently been using is inaccurate for a couple of reasons.

The first is that fitness plays a part. Given that these are raw goal totals, the player has to be fit for the vast majority of the season, and players like Sergio Agüero may have managed more than the one of these seasons that he has to his name (2015/16). Harry Kane would surely have a few more if he’d stayed fully fit too.

The second is Sam Kerr.

For three years in a row, immediately before joining Chelsea at the start of 2020, Sam Kerr put up huge ‘carrying a team’ seasons in the NWSL regular season:

2017 [Sky Blue FC]: 40% (17 of 42)
2018 [Chicago Red Stars]: 42% (16 of 38)
2019 [Chicago Red Stars]: 44% (18 of 41)

[As the NWSL season only runs to 24 games before the play-offs these might seem low to regular PL watchers — in a 38-game season these would be like 25-29 goal seasons]

Since the league’s inception in 2013, there’s been one of these seasons from a forward per year (apart from 2014). The Washington Spirit high-scorers of 2013 and 2015 did that job, with Diana Matheson and Crystal Dunn respectively; in 2016 it was Shea Groom for (now defunct) FC Kansas City.

But while Sam Kerr may struggle to score 40% of Chelsea’s goals in a very talent-rich squad, the NWSL doesn’t hold a candle to the FA Women’s Super League in this respect.

The NWSL’s seven seasons have seen an average of just over two ‘high-scorer carrying their team’ per year, but the WSL’s three years since switching back to a winter schedule in 2017/18 have averaged just over four.

Part of this could come from the (relative) parity that the NWSL has baked into its structure. With talent spread more widely, good strikers are more likely to have a good team around them who’ll contribute more goals. Bristol City are a case in point of the opposite being the case in England.

There have been just two occasions in the WSL’s recent winter schedule (and ‘on FBref’) era where a player has scored more than half of their teams goals. Both have been Bristol City forwards.

2017/18 [Bristol City], Lauren Hemp: 54% (7 of 13)
2019/20 [Bristol City], Ebony Salmon: 56% (5 of 9)

As with Sam Kerr’s numbers, a shorter season is a factor in these single digits, but City have also been a team who aren’t very good but are just good enough to stay up. Think Sunderland in their Darren Bent/Jermain Defoe eras. (Across a 38-game season, Hemp’s seven would be 14 or 15 and Salmon’s five would be 13 or 14).

Salmon was far from alone in this covid-hit 2019/20 season for carrying her side though. The full list reads:

Ebony Salmon, 56% of Bristol City’s [10th of 12] goals (5 of 9)
Rachel Furness, 50% of Liverpool’s [12th] goals (4 of 8)
Aileen Whelan, 45% of Brighton’s [9th] goals (5 of 11)
Chloe Kelly, 43% of Everton’s [6th] goals (9 of 21)
Vivianne Miedema, 40% of Arsenal’s [3rd] goals (16 of 40)
Abbi Grant, 40% of Birmingham’s [11th] goals (2 of 5)

Subscribe now

This is a really interesting, really weird bunch. Four of these teams simply struggled to score goals, and Liverpool’s troubles get put in an even worse light when I add in the information that Furness joined mid-season.

Abbi Grant and Aileen Whelan are — with no offence intended and mainly to skip over them quickly — just players who scored a few on bad teams. Ebony Salmon and Chloe Kelly are both more exciting: quality youngsters playing from a left-wing position and putting up decent underlying numbers.

Kelly’s already joined Manchester City. Salmon could well join her. Remember Lauren Hemp, the other Bristol City striker to score more than half her team’s goals for the season? After that 2017/18 campaign, Hemp moved to Manchester to join the Sky Blues.

I think that’s the reason I enjoy this very simple stat so much: the strikers who carry their teams to such a large extent are always interesting players.

In the Premier League, I’ve mentioned Danny Ings and Christian Benteke (well, and Austin), as well as alluding to Jermain Defoe and Darren Bent. Matt Le Tissier got 51% of Southampton’s goals as they finished 18th in the 22-team 1993/94 season. Andrew Johnson scored 51% of Crystal Palace’s goals in 2004/05, although ten (ten!) of these were penalties.

And in the 2017/18 WSL there was another forward, alongside Lauren Hemp, who had a big season, contributing half of her team’s league goals. Ellen White scored 15 of Birmingham City’s 30 as they finished fifth.

Whatever happened to her after that, huh?

GOAL! England 1-1 #USA

She's the first England player to score in five World Cup games in a row - Ellen White sweeping in at the far post to slot home. She's now the tournament's top scorer.

Game on!

Follow live ➡ https://t.co/yJU6xyiO0b #ENGUSA #Lionesses pic.twitter.com/5JbaDRUuGC— BBC Sport (@BBCSport) July 2, 2019

fin.

Subscribe now

Shout-out corner

The Football Manager 2020 database that @WomenOnFM has put together. It’s an incredible effort that’s so valuable given that Sports Interactive (unless I’ve missed something) keep resisting adding women’s leagues to the game. Maybe tweet them to ask them when it’s gonna happen to show that it matters to you. Personally, the next time I spend money on FM will be when it includes women’s teams.

A return to last week

Last week I wrote about whether Tim Sherwood could have, had things gone differently, pulled off a successful ‘Moneyballing’ at Aston Villa in 2015/16.

I mentioned that when he left in the October of that season, the goal difference wasn’t that bad even if the points were. It was a pretty basic metric, but it suggested that maybe either he or the club flounced off a little too soon.

With some expected goals numbers acquired, things didn’t seem that bad either. Taking out a heavy xG-thumping of Sunderland (which actually turned out a 2-2 draw), they were maybe fourth or fifth-worst side in the league, rather than dead last that the table had them as.

Villa were robbed. We were all robbed of this marvellous stats-savvy Tim Sherwood era we could be living in.

This week’s charity is the Trussell Trust. Please consider helping them.

Donate to the Trussell Trust

Could Tim Sherwood have got 'Moneyball' to work at Aston Villa?

Tue, 28 Jul 2020 00:00:00 GMT

Welcome to Get Goalside! I was hoping you’d be able to make it.

This week’s charity is, well. Usually I’ve come across charities that I then point to here. Wiley’s antisemitic tweets have made me realise I don’t actually know any anti anti-semitism charities.

The Antisemitism Policy Trust have a couple of guides on myths and misconceptions about Jews and antisemitic imagery and cartoons that seem informative. If you wish, please let me know of other resources or charities. By the time this email goes out, I’ll be 27 hours into a 48-hour ‘walkout’ on Twitter that some are doing.

Please, let us all do better.

Last week I tackled what the ‘true’ takeaways from the book Moneyball were; this week I want to stay on that theme.

In the summer of 2015, Aston Villa spent (according to Transfermarkt) nearly £60m on:

Jordan Ayew
Jordan Amavi
Adama Traoré (yes, that one)
Jordan Veretout (lotta Jordans)
Idrissa Gueye
Rudy Gestede
Scott Sinclair
Joleon Lescott

The last three of this list were bought from English clubs; the first five were bought from France or, in Traoré’s case, the Barcelona academy. The average age of the last three was 28; the average age of the first five was 22. It seems fair to say that there was something different driving these two sets of signings.

This was the brief ‘Moneyball’ era of Aston Villa, under CEO Tom Fox*, Head of Recruitment Paddy Riley**, and Henrik Almstadt***. It ended when the team were relegated with just 17 points in 2015/16, an exact half of what the team above them managed.

*arrived in 2014 after five years as chief commercial officer at Arsenal

**arrived in 2014, having previously been an analyst at the club, from a short stint at Liverpool

***arrived in 2015, having previously worked with Fox at Arsenal, and is now working at AC Milan

Were these bad signings? Were Villa unlucky? Is there a chance that a data-driven approach to football could have taken off four years before Liverpool once-and-for-all legitimised it with their Premier League triumph?

A few months after the summer 2015 transfer window closed, Brendan Rodgers was sacked from Liverpool. The Daily Mail’s Neil Ashton wrote a story, which I won’t link to, with the following title: “Liverpool's head of technical performance Michael Edwards is the laptop guru who did a number on Brendan Rodgers”.

The article has not aged well. But this was the atmosphere of the time.

(In a quirk of fate, the two teams were connected in that summer window, as the Merseyside club bought Christian Benteke from Villa for a reported £32.5m; it seems unlikely that Benteke was one of Edwards’ signings)

Villa were clearly embarking on something similar to Liverpool in selling a key asset (in Liverpool’s case, Raheem Sterling) and trying to rebuild with analytics-driven purchases. As a quick overview, here’s where those five buys are now.

Jordan Ayew - Crystal Palace
Jordan Amavi - Marseille
Adama Traoré - Wolves
Jordan Veretout - Roma (on loan from Fiorentina)
Idrissa Gueye - Paris Saint-Germain

As individuals, their careers have gone in slightly different directions, but all have taken steps up from Aston Villa. The group as a whole have done well enough that Ayew, whose career has been with meh-to-middling Swansea and Palace sides, looks like the failure of the bunch.

That Tim Sherwood was the manager at the time means that it’s fairly well-established that he was (publicly) not happy with the transfer business in the summer of 2015. As well as the eight players previously mentioned, Villa also brought in José Crespo from Cordoba for a miniscule fee, Tiago Ilori on a loan from Liverpool, and Mark Bunn and Micah Richards on frees.

Of the five that look like fairly clear ‘analytics buys’, three of them were pretty much regular starters. Idrissa Gueye started 35 league games that season, Jordan Ayew 27, and Jordan Amavi started nine of 12 before tearing his ACL in November 2015.

Veretout started 21 games, a hefty share, and it’s only Adama Traoré — still a teenager when he joined the club — who didn’t start at all. There was only one match, a 3-1 defeat at Tottenham Hotspur, where none of the group were in the starting line-up (although Jordan Ayew came on to get Villa’s only goal).

Subscribe now

Granted, Sherwood had left just before this, and under him, only Gueye and Amavi were regular starters; Ayew and Veretout were more bit-part. Still, they started four and three times respectively in Sherwood’s ten league games to start that 2015/16 season.

They may not have been exactly the players he wanted, but that hardly seems the sign of a manager throwing his toys out of the pram for being given dud parts.

Perhaps the problem wasn’t the analytics influence, but another, quite fundamental, part of squad-building…

As has been mentioned, Aston Villa brought in a lot of players in that summer of 2015. Quite probably too many.

Nearly 44% of Villa’s minutes in the 2015/16 Premier League season were played by people who hadn’t played at all the previous campaign. This was ahead of the second-highest figure by seven percentage points.

Teams ordered by percentage of minutes played by ‘new’ players; the 17 non-promoted teams only. To save me time going through transfermarkt manually, I’ve used ‘didn’t play in the previous season’ rather than ‘new signings’ per se, so this could include academy graduates

It seems notable that the three teams at the top of this list finished in the bottom four that season. On the other hand, Southampton and West Ham had two incredible years, although neither of those teams had more than a third of their league minutes being played by newbies.

Subscribe now

With Sunderland and Newcastle, their high figures (37% and 35% respectively) are probably signs of squads in need of work. Villa will have fit into this category too, in the transition between 2014/15 and 2015/16 seasons, but nearly half of your minutes being played by new players seems like a sign you’ve overdone it.

This is an aspect of Moneyball that is probably overlooked. In a team sport like football, it seems likely that this degree of turnover is counter-productive. It’s often noted how international football seems a lower standard than club football because compatriots may be compatriots, but they don’t play together enough to know each others’ games. Signing as many players as Villa did in summer 2015 probably creates a similar effect.

I’m sure that the minds at present-day Liverpool, Manchester City, Arsenal etc have done work on ‘good’ levels of squad turnover. Managers like Sir Alex Ferguson may have understood it intuitively, or through experience, but by now these clubs (and others) will certainly have some kind of data on it.

I don’t know what the interpretations of this data analysis are though, or particularly how they’d apply to clubs like Villa, Newcastle, or Sunderland in those years. If your current squad is, to put it bluntly, bad, then you might not care about disrupting connections between established players by chucking half of them out. Getting new, better players in might be the most important thing.

Yet on the other hand, chopping and changing the squad doesn’t just affect the players being shifted in and out of the side. The week-in-week-out regulars will have got to know the movements and tendencies of all of their (granted, subpar) teammates. One or two new players might be one thing, but if they’re scattered all over the pitch, that’s a huge amount of the 11x11 connections that are affected.

Subscribe now

I was going to leave it there. The theory made sense to me. A lot of turnover equals a self-inflicted shot in the foot that negates some of the benefit of potentially good signings.

But then I did the sensible thing and watched some of the games, and went to check the table for Sherwood and Rémi Garde’s reigns.

When I watched bits from a couple of Sherwood’s games, I saw a team who were a little porous in midfield through a lack of structure, and a team who just struggled to piece anything together in the final third. The midfield could be decent on the ball though, and if they tightened up a bit out of possession then there could be something about them.

The most granular I can go with data for this season (without spending more time than it’s worth on it) is just goal difference on WhoScored. And that tells an interesting story.

In Sherwood’s first ten games, despite only picking up four points, the team’s goal difference (-8) was a similar level to the bunch of bottom five teams that stretched up to Norwich on nine points. I dunno what more advanced stats might suggest, but maybe Sherwood got a little unlucky. Six teams had conceded as many or more than them, although 15 had scored more.

It’s the Garde era where it all went wrong. If we take the 24 matches that encompassed the rest of Villa’s season where there was something to play four (they were relegated mathematically in mid-April), their goal difference (-34) was the worst by far. The next-worst was Norwich, who’d played the same amount of games, on -18.

Did luck put a stop to Villa’s ‘Moneyball’ era? Could Tactics Tim, with a bit more time and a few more lucky bounces, have overseen a melding of old- and new-schools…?

We’ll never know. But it’s a hell of a ‘what if?’ to debate about.

Subscribe now

Shout-out corner

Let’s do quick-fire bumper one after I forgot to do one last week:

The NWSL Challenge Cup final was on Sunday. I asked before the tournament who to support, Kim McCauley suggested Houston Dash, Houston Dash won. Shout-out Kim, here’s her Patreon; shout-out this thread from The Athletic’s Meg Linehan on the women’s soccer media to be consuming.
This review/preview of PSG’s season is excellent. It’s in French, but stick it in Google translate and it’s well worth your time.
The Measurables podcast has put together an incredible list of analytics pros who are making themselves available for virtual office hours for people of color, members of the LGBTQ+ community, women, and others underrepresented in sports analytics.
A lot of the open football-related coding help is in the coding language R, but PySport are doing neat stuff for Python. See below:
We are very proud to release version 1.0.0 of kloppy!
Introducing
- Metadata models
- Project website with quickstart notebooks

Thanks @brunodagnino @fishnets88 @pratik_thanki @mr_le_fox

Checkout: https://t.co/XUh2cw2ub5 #sportsanalytics #socceranalytics #pysport pic.twitter.com/FBoVKdAIXE— PySport (@PySportOrg) July 26, 2020

A reminder of the Antisemitism Policy Trust’s guides on myths and misconceptions about Jews and antisemitic imagery and cartoons that seem informative. Please, let us all do better.

The real meaning of Moneyball

Tue, 21 Jul 2020 00:00:00 GMT

“The first thing they did was run you.”

That’s how Michael Lewis opens his seminal book Moneyball, which charts the story of the Oakland A’s 2002 season and the wider movement of data in baseball. It’s a good first line; it grabs you.

But it has nothing to do with what ‘Moneyball’ really is…

Everyone’s at least heard of the phrase, the M-word, by now. Applied to football, it conjures images of transfer committees, spreadsheets, and clubs you’ve heard of signing players you haven’t. Maybe even Alex Stewart’s ‘Football Manager meets Moneyball’ series which, in actual fact, used tips from Soccernomics (Simon Kuper and Stefan Szymanski) and The Numbers Game (Chris Anderson and David Sally) rather than Moneyball itself, although there was conceptual overlap.

Still, Moneyball is one of those books that everyone will say they’ve read even if they haven’t. It — the title — is a wonderful word. But what are the actual messages of the book?

“The first thing they did was run you.”

I said that the opening line has nothing to do with Moneyball but I was perhaps exaggerating. It’s related, but the way that Michael Lewis uses the line sets the book up on a narrative path that diverts away from the real meaning of what the A’s were doing.

Michael Lewis spends the opening pages of the book crafting the character of Billy Beane, painting a vignette to the reader to try and explain why he is how he is. Billy Beane was a future star, this section tells us. Not only was Billy Beane a future star, Billy Beane looked like a future star. Tall, muscular, handsome; to use a phrase from later in the book, he could sell jeans. But he never made it as a successful pro-baseball player.

The way the book tells it, Beane struggled to find a dispassionate enough sporting mentality that would allow him to fail and improve, and ultimately succeed. He had a fiery anger that he turned inward on himself whenever he couldn’t hit the ball at the plate. And when his career ultimately flamed out, Beane turned that fire on the old-school system that had sold him a dream — the scouts’ dreams — and set him up to fail.

‘Billy Beane versus baseball’ is the story the book is telling when it opens with “The first thing they did was run you”. ‘Billy Beane versus baseball’ is not a message of whatever philosophy the word ‘Moneyball’ is meant to represent.

The chapter of the book on Bill James and the history of sabermetrics has a similar tone: a man, engaging in person but angry at the world, rages against the establishment. Reading the Afterword, this isn’t a surprise: “As far as they [the people at the Oakland A’s] knew I [Michael Lewis] wasn’t even writing a book about the Oakland A’s. I was writing a book about the collision of reason and baseball.” It was only after talking to people within the sport that he realised that Oakland was (one of the) only places that collision was happening.

The reason why I say that this ‘Billy Beane versus baseball’ message isn’t one that should be internalised is encapsulated in this passage from chapter 5:

Billy Beane was a human arsenal built, inadvertently, by professional baseball to attack its customs and rituals. He thought himself to be fighting a war aginst subjective judgements, but he was doing something else, too.

At one point [A’s scout] Chris Pittaro said that the thing that struck him about Billy — what set him apart from most baseball insiders — was his desire to find players unlike himself. Billy Beane had gone looking for, and found, his antitheses.

Billy Beane searching for ‘his antitheses’ is a bit of dramatising, but it’s an idea Lewis returns to again and again. That not only was Beane inoculated against some of baseball’s biases by being an example of their failures himself, but that he was actively choosing to go against them.

In reality, Beane was overlooking the superficial aspects of a player (good looks, mainly, it seems) to look for the skills that matter. But the way the book portrays it, this project is about finding weirdos specifically because they’re weird; and that seems as potentially a damaging bias as the one Beane was supposedly fighting against.

Interestingly, there’s one aspect of the old-school scout that the book (very quietly) shows to be useful: character assessments. Going purely by the storytelling of Michael Lewis — because that’s all the knowledge of Beane’s playing career I have — it should have been abundantly clear that this prospect was struggling.

The moment Billy failed, he went looking for something to break. One time after Billy struck out, he whacked his aluminum bat against a wall with such violence that he bent it at a right angle.

The next time he came to the plate he was still so furious with himself that he insisted on hitting with the crooked bat.

That sounds like something that the old guard would’ve, could’ve, and should’ve caught on to. The way it’s written, anyone turning up to watch Beane play more than a couple of times was likely to see this side of him, which would have sparked a concern that they wouldn’t have needed to consult the stats to double-check.

But baseball scouts, Lewis tells us, looked for the five ‘tools’ in young players. Could they run, throw, field, hit, and hit with power.

There were two problems with this.

The first is encapsulated by kid-Beane, and by one of the prospects that the A’s scouts suggest to adult-Beane before the 2002 amateur draft. Kid-Beane could run, throw, field, hit with power, and he could hit, but erratically. In the drafting scene, a name gets thrown about who “[is] an athlete”… but can’t hit. In both cases, scouts were blinded by how the player looked, and what they imagined these strapping young men could become, and couldn’t properly see their actual skills.

But the second problem is that the skills weren’t actually the ones needed. Little point in kid-Beane being able to hit with power if he’s turning up at the plate with an anger management problem-induced crooked bat. Moneyball makes the point several times that pitchers shouldn’t be solely judged on their velocity, that variety and deception are key to getting hitters out. For the hitters themselves, knowing when to leave balls outside the plate, or that they can’t do anything with, is a key skill too.

The concept of a ‘Five Tool Guy’ — a phrase used in the book — could probably still be applicable, it’s just that the tools have got to be useful. That’s one of the keys of Moneyball: what are the useful skills?

You can find out what the useful skills are with statistics, which is what most people take away from Moneyball, but that needn’t be the only way. (To an outsider like me), it should have been obvious that a pitcher’s variations, in addition to their pure speed, is what makes them dangerous. It was obvious enough to the players:

“You know how many times [Jamie] Moyer jams guys with an eighty-mile-an-hour fastball?” says [Scott] Hatteberg. “All the time. It’s because he sets it up with a sixty-nine-mile-an-hour change-up.” He fast-forwards to a slow curve, and an even slower change-up. “See,” he says, “All this other shit is what makes his fastball look like ninety-four.” [Chapter 8]

But the value of knowing which skills are important would have been far less useful to the A’s were it not for one of the underrated skills the book shows. Billy Beane’s soft skills.

It helps that Michael Lewis had a history in finance and was drawn to the trading aspect of Beane’s job. Who knows how much writerly polish there might be on these scenes, but when Beane is wheeling and dealing it’s compelling and, by all accounts, it’s effective too.

He also comes across as a great manager of people, partly thanks to many of the things — the charm, the intelligence, the knowledge — that made him such a good player trader. But also:

Most GMs hadn’t played the game and tended to be physically intimated in the presence of big league players. Billy had not only played, he might as well wear a sign around his neck that said: I’ve been here, so don’t go trying any of that big league bullshit on me.

Hell of a soft skill.

My favourite example of Beane’s people management, though, is sort of an aside in chapter 11. Lewis tells us that the GM had stopped by a player called Ramon Hernandez’s locker and made a bet with him where one would pay the other each time Hernandez dealt with a certain type of pitch in a certain type of way.

The point of the exercise, Billy now says, is “it gives me an excuse to henpick Ramon. It’s a subversive way for me to keep nagging the shit out of him without him knowing it.”

Honestly, signing up for a communications class is as much of a takeaway from Moneyball as signing up for a computer class.

And now, finally, the stats. It is amusing, laughably so, how simply baseball seems to be solved. On the hitting side, it seems like it all comes down to on-base and slugging percentages. And Paul DePodesta, Billy Beane’s quant guy, works out the heresy that one extra point of on-base percentage is worth three times an extra point of slugging percentage. The dude ran one formula and came to a(n apparently) correct conclusion that even the other stats nerds hadn’t found out.

From my limited understanding of baseball, it’s a sport where the scoring system builds in a more linear way than football. Four players in a row getting singles can be a run batted in. There’s a lot of tangible value in getting on-base.

Football analytics has shown just how little tangible value there is in the vast majority of things done on the football pitch. It’s a sport of ~150-200 possessions per match and the two teams will combine to score three goals. It’s a sport that’s designed to make you wonder about the point of your own existence. It’s hard to find what’s undervalued when the value of almost everything seems to be next-to-zero.

That happy thought brings me to Beane’s five rules for player trading, number 3 particularly [for the record, I’m paraphrasing the book]:

Change is good; not having money means only short-term solutions exist; always be upgrading
Number two is actually two: i) when you have to do something, you’re screwed, because you’ll make a bad deal ii) you can recover from a player you don’t sign but you might never recover from a player you sign for the wrong price
Know the value of every player. You can put a dollar value on it
Know what you want and get it
Ignore the media

There’s another rule I want to add in, that comes from Lewis’ Afterword to the version of the book I have (written, presumably, after the original publication of the first edition in 2003). It’s part of an answer to a question posed by a hypothetical objector to the book, about why the A’s don’t score more runs.

Still, the A’s on-base percentage retains one important trait: it’s good for the money. And the point is not to have the highest on-base percentage, but to win games as cheaply as possible. And the way to win games cheaply is to buy the qualities in a baseball player that the market undervalues, and sell the ones that the market overvalues.

That’s Moneyball.

But I’ll bring this part-review, part-blog about Moneyball to a close with the person both the book and the movie finish on: Jeremy Brown. I’ll put the scene from the movie here to watch if you want, because it gets me every time.

If you don’t want to watch it, I’ll summarise: Jeremy Brown was one of the prospects signed by the A’s in the 2002 draft. He was drafted by the A’s because of two things:

He had great on-base percentage
He was fat

People weren’t interested in Brown because of his size, and people made fun of him because of it. But the kid could bat, as he proved in the minor leagues. He wasn’t an athlete, and so he shied away from running past first base when he got the ball in play. But this one time, this one time, he connected with such quality, he thought he’d do it. He’d round first and go to second. But he tripped. He fell, and he scrambled back to the safety of first.

The reason why this part gets me is because, in the book, Lewis fills the pages with characters who don’t believe in themselves because of some funky quirk that makes the baseball establishment overlook them. Brown’s is that he was fat and that he couldn’t run. And when he finally gets the idea into his head that he’ll run to second, he’s quickly reminded by the universe that running is not the thing he was put on this earth to do. You can hear the inner monologue from here.

But Brown had to keep on running anyway. Because, unbeknownst to him as he scrambled back to first in the dirt, he’d hit a home run.

Moneyball the movie uses the scene to remind Beane of the romanticism of baseball at a time when he’s not feeling all that positive about it.

But the thing with Brown is that, regardless of whether he’d hit a homer or not, he was already doing what he’d been drafted to do. His stats — the ones that the A’s cared about — were great in the minor leagues. Lewis uses Brown hitting a home run as an inspirational story, that the fat kid did good. But, according to ‘the Moneyball philosophy’ he was already doing good anyway.

And then he stopped doing good.

Brown’s first and last MLB appearances were less than a month apart, debuting in early September and making his last major league bow on the 1 October. He retired in 2008.

Michael Lewis’ use of Brown as an epilogue encapsulates some of the ‘problem’ with the book Moneyball. The thing that Lewis — a fantastic writer and storyteller, don’t get me wrong — was most interested in wasn’t the thing that the A’s were most interested in. And Jeremy Brown didn’t turn out a major league success.

There may be reasons for that, and there may be reasons for other failures of that 2002 A’s draft. The course that baseball, and other sports, have gone on since then shows that Beane’s general approach was the right one, even if it may have been overblown by the book.

But Michael Lewis opened the book with “the first thing they did was run you”, not “a book that sold 75 copies [Bill James’ first Baseball Abstract] changed baseball”. He could have feasibly done the latter, and, because he’s Michael Lewis, written a pretty compelling story. But Moneyball, as a Michael Lewis book, isn’t about sabermetrics and rationality; it’s about Billy Beane taking on the establishment.

Even so, there are things to take from it.

Question. Question the received wisdoms. Do this with thought experiments, do this with data, but do it. It might even turn out that the received wisdom is correct, but at least you’ll know that it is.
To use a capitalistic, finance-y phrase, exploit the market. “The point is not to have the highest on-base percentage, but to win games as cheaply as possible”
Be organised. Beane’s third rule for player trading wasn’t “know the value of players”, it was “know the value of every player”. Being organised also guards against Billy Beane Rule Of Trading number 2.i, that when you have to do something that’s when bad things happen. Being organised helps you know the value of things, and it helps avoid digging yourself into holes.
Soft skills matter. Whether it’s Beane’s trading and player-management, or Ron Washington’s coaching of Scott Hatteburg, Moneyball shows the value of beating hearts as much as it shows the value of cold computers.
The data has to be good data. There’s an admission, hidden in the later chapters of Moneyball, that part of the reason why old-school baseball may have been suspicious of data was that they’d been dealing with a lot of stuff that didn’t matter. Bill James railed against the uselessness of fielding data. Despite a chapter devoted to the 2002 amateur draft, the book never touches on how reliably college data translates to the professional game.
Numbers need to take on the power of language.

On this last one, I’ll end this post with a quote from Bill James in the book, the length of which and number of sub-clauses giving a sense of just how much of a frustrated-writer he was. His point is about how some statistics are just numbers; they exist but they don’t tell us anything.

But

[w]hen the numbers acquire the significance of language, they acquire the power to do all of the things which language can do: to become fiction and drama and poetry…And it is not just baseball that these numbers, through a fractured mirror, describe. It is character. It is psychology, it is history, it is power, it is grace, glory, consistency, sacrifice, courage, it is success and failure, it is frustration and bad luck, it is ambition, it is overreaching, it is discipline. And it is victory and defeat, which is all that the indiot sub-conscious really understands.

How would you play against the Invincibles?

Tue, 07 Jul 2020 00:00:00 GMT

As a person who plays with data, it’s always a treat when someone releases a free dataset. StatsBomb did that recently with their open data project: Arsenal’s 2003/04 Invincibles season.

Due to the difficulty of getting match footage, it only includes 32 of the 38 league matches that season, but it’s still the majority of the campaign. We could use it for a lot of things but surely, given the magnitude of what Arsenal achieved that season, there’s only one thing we want to know: how would you beat them?

If you’re going to play this 2003/04 team, you probably already know who the threats are and where they operate in. A heatmap of where this Arsenal side created chances from is simple, but the novelty of having data from the early 00s makes it very worthwhile.

And that aforementioned novelty of early 00s data is also what makes this particularly fascinating. That’s such a central concentration, with some pretty notable minor peaks in cut-back/low cross areas. I’d venture a guess that this isn’t too dissimilar to what Manchester City’s might look like nowadays.

A closer look at who’s setting up these chances shows just how key Thierry Henry and Dennis Bergkamp were, leading the way for chances created in open play. Henry operated all around what we might now call ‘the ten space’; Bergkamp a little deeper and to the right.

Sidenote: Interesting to see a bunch of ‘chances created’ passes ending so far from goal, likely meaning that the recipient dribbled a fair distance before shooting.

When we check the locations that Henry and Bergkamp received the ball, a similar pattern appears as is evident in the shots they were setting up: the Frenchman further up though still covering a large area; Bergkamp a little deeper and wide-ranging.

But while we’re on Henry, I wanted to check something else. Firstly, here’s his general shot map (with his feet - I’m not fussed about his headers; shots shaded to give indication of expected goals value).

Interestingly, given that Arsenal’s chance creation map looked so modern, Henry took very few shots in the Invincibles season around the penalty spot. Today’s forwards are much more likely to have shots there.

But the thing I really cared about was the ‘Henry finish’. A one-on-one, to the left-of-centre of the box; did Henry really open his body up and curl it into the opposite corner as much as we remember?

It may seem so. He had eight of these chances in the 32 games covered, missed the target with just one, and scored five, all of the goals going in that bottom right corner.

Edit: However. It turns out that half of these shots were taken with Henry’s left. Maybe we just remember this very specific type of finish from the Frenchman because it was so dang pretty.

But enough of Arsenal’s attack. We’ve seen how it revolves around Henry and Bergkamp to a large extent, in terms of pure chance creation at least. What about how we might attack them?

Arsenal’s opponents during the 2003/04 season also had a leaning towards the left-hand channel in where they were creating open-play shots from. Maybe this is just something common to all football teams; maybe Arsenal were just weak in that area.

I was also intrigued by the secondary peak on the right-hand flank, from the perspective of the Gunners’ opposition.

Let’s plot this heatmap as individual points, and highlight the ones in this area — is it a case of players getting in behind Ashley Cole?

That might be unfair on the left-back. Let’s instead look at chances that were definitely painful to be giving up. Any threshold of expected goals value we pick is going to be somewhat arbitrary, so let’s go with 1 in 6 chances (or 0.16666667 expected goals) purely for a ‘roll of the dice’ vis title. (also, Arsenal only conceded 7 open-play shots worth more than 0.25 expected goals in the 32 matches in this dataset)

Huh, maybe Cole and/or Robert Pires are a little vulnerable, as well as direct passes just to the right of Arsenal’s centre in midfield. They were pretty strong in their right-back and left-of-centre positions though.

This has just been a skimming of the surface using some data, but there are some conclusions:

give Henry a reducer early on — he’s such a key part of Arsenal’s attack that you’re probably best off kicking him off the pitch
Bergkamp, Pires, and Freddy Ljungberg will all drift around quite a lot. Annoying.
If you want a spot to focus your attack, catching the Gunners in transition either down your right flank or through your left channel seem to be the best options.

If only we’d had data back in 2003. Maybe the Invincibles wouldn’t have been invincible.

3 fun looks at home advantage

Tue, 30 Jun 2020 00:00:00 GMT

Welcome to Get Goalside!, it’s good to see you. Nice face covering.

This week’s charity is FND Action. FND stands for Functional Neurological Disorder, which FND Action raises awareness for.

The first few matches post-lockdown created an excitement about the lack of home victories. It had all started with the Bundesliga, whose low opening home win percentage led to #takes and #claims.

There have been only three home wins in 22 Bundesliga games played behind closed doors, which suggests home advantage is wiped out without fans.

We have worked out just what that potentially means for Premier League clubs 👇— Sky Sports (@SkySports) May 27, 2020

And though it started off looking the same in the Premier League — amusingly with the only two home wins in the first eight games both coming against Arsenal — things have rebalanced.

At time of writing, there have been 25 matches post-restart, and ten home wins. If we take games in bunches of 25, that’s not far off the average home win percentage.

Even when the only home wins were coming against the Gunners (a period with a home win rate of 25%) that wouldn’t be the lowest rate in English men’s top tier history. There was a run of 25 matches in December 1994 where there were only three home victories (12%), and a run as recent as 2015 where there were only four (16%):

So, while the absence of crowds might affect home advantage (I’m not ruling it out!) I wouldn’t be using these matches as evidence of it.

But while I’m looking at home advantage data for the English First Division/Premier League… Which team has the longest streak of consecutive home wins? Current Liverpool! (since February 2019)

But prior to them?

Also Liverpool! (I promise that we’ll get away from Liverpool next week, although hello to all the Reds fans who subscribed after last week’s newsletter)

Read last week’s post on putting this title-winning 2019/20 team into some (impressive) historical context. Or if you’ve already read it, share this one!

The Reds had 21 straight home wins from January to December 1972. Funnily enough, considering jinxes of recent years, both the match that started the run and the final win in it were against Crystal Palace. They finished 3rd in the 1971/72 section of this streak, just one point behind winners Derby County in an incredible four-horse race; but in 1972/73, as like now, they finished on top.

Both Manchester clubs came close to matching that 70s record from in the past decade, but Manchester City fell one short (20 wins, March 2011 to March 2012) and United fell two short (19 wins, October 2010 to October 2011).

Here are the dates of the longest league winning streaks at home for the other teams who’ve managed one of 10 or more [and the league positions in the seasons the streaks covered]:

Everton (14): November 1967 - August 1968 [5th | 3rd]
Newcastle (14): May 1995 - February 1996 [6th | 2nd]
Tottenham (14): January 1987 - October 1987 [3rd | 13th]
Wolves (14): March 1953 - November 1953 [3rd | 1st]
Arsenal (13): February 2005 - November 2005 [2nd | 4th]
Aston Villa (13): February 1983 - October 1983 [6th | 10th]
Blackburn (13): October 1993 - April 1994 [2nd]
Chelsea (13): April 2005 - December 2005 [1st | 1st]
Leeds (13): November 1968 - August 1969 [1st | 2nd]
Derby (12): October 1971 - April 1972 [1st]
Nottingham Forest (12): February 1980 - September 1980 [5th | 7th]
Blackpool (11): November 1951 - April 1952 [9th]
Ipswich Town (11): October 1961 - March 1962 [1st]
Portsmouth (10): December 1948 - April 1949 [1st]
Preston North End (10): November 1952 - March 1953 [2nd]
West Ham (10): October 1985 - April 1986 [3rd]

If your team has been in the top-flight, and/or you’d like to know the longest home win streak in another English division, leave a comment and I’ll respond with the answer.

Shout-out corner

I listen to a lot of podcasts, so I feel I have some minor authority to say that the guys on The Beautiful Game podcast and Joe Devine on the Tifo football podcast are some of my favourite interviewers at the moment. Their guests always seem incredibly at ease and in long, conversational interviews they don’t over-insert themselves.

Also, @VenkyReddevil is doing some daily visualisations which I’m enjoying seeing. I’ve been learning things from them, in the choices taken and the way they aid the reader or control the pace of reading the chart.

Premier league defenders who are the most successful in aerial duels.

Data comes from @StatsBomb via @fbref

Link: https://t.co/lFFgS8aWwk #dataviz #rstats pic.twitter.com/feQiDnAPM2— Venkatanarayanan (@VenkyReddevil) June 25, 2020

Finally, to share another new newsletter, Kim McCauley has a newsletter on women’s football called the Gegen Press for All for XI. Also follow Kim on twitter.

Once again: this week’s charity is FND Action. There’s a page on what FND is here.

Donate to FND Action

Liverpool 19/20: A historically great side?

Tue, 23 Jun 2020 00:00:00 GMT

What’s all this about the European top divisions being boring, we’re about to have a first-time Premier League winner, baby!

*ahem*

There are a variety of ways Liverpool can/will win the league within the next week or so. There’s not much point me breaking down how they’ve done it, because that’s what everyone will be doing. So instead, I want to put them in some historical context.

For a start, the fact that this could be the Reds’ first title in 30 years might be distracting a little bit from just how dominant they’ve been. They’ve won 27 out of 30 games, which is incredible, and they look set to top a four-year run of truly tremendous title-winners.

Perhaps more impressively is that this is their second season performing this well. Not only are the Reds the best point-getters in Premier League history, they’re (narrowly) the best over a two-year period too.

This gets all the more impressive when put into the Premier League’s recent context, which has got a lot more competitive at the top in recent years.

While it’s often said that it’s difficult to defend a Premier League title, one could extend that adage to simply managing consecutive title challenges. In the first two decades of the Premier League, Manchester United, Arsenal, and Chelsea all ensured that the top two stayed pretty familiar year-to-year. Other challengers popped up every now and then — and the latter two of those clubs obviously had quite definable eras — but it was all quite regular.

However, this all changed around 2013. Since then, finishing first or second has become far less of a predictor of where you’d finish the following season.

It’s no wonder that fans of big clubs have started coming to terms with spending a season outside of European competition to allow them to mount a title charge.

It could be that, in this current ‘Big Six’ era, there’s either too much competition to solidify a top two spot or it takes too much energy to obtain one. That Liverpool and Manchester City look set to be the top two for two seasons in a row is actually a Premier League oddity.

None of this is really a comment on the current team at Anfield though.

Some have pointed to Liverpool and Manchester City’s respective expected goals figures (+0.88 and +1.47 per game respectively per FBref) to suggest that this has been a bit of a lucky victory parade. As a statsy person I note that, but think there’s more to it than that.

Firstly, somebody on stats twitter whose tweets/article I can’t currently find pointed out a while ago that City were racking up goals and expected goals against smaller teams [case in point: I write this as City score their 5th against Burnley]. Liverpool, meanwhile, have taken things easy after going ahead, and have spent minuscule amounts of time trailing. That all makes a difference to the top-line xG stats.

Secondly, Ryan O’Hanlon’s recent No Grass in the Clouds newsletter pointed out that Liverpool’s ‘post-shot expected goals’ difference is actually better than City’s. Post-shot xG adds ‘where in the goal the shot was headed’ to the calculation, as well as things like shot location and build-up that the regular expected goals models have. Post-shot models, because they feature fewer shots (anything off-target or blocked gets a 0) can be more volatile but over a large-ish sample they can be interesting.

To nick the figures from Ryan’s newsletter, as of last Friday the company Stats Perform had City on +1.46 expected goals per game and Liverpool on +1.0 (xG models vary from place to place a little).

However, post-shot expected goals changed this dramatically, putting Liverpool on +1.45 per game and City on +1.28. A narrow gap, but still a gap that has Jürgen Klopp’s team ahead.

I’m also, just generally, of the opinion that ‘well actually’-ing any title winner with mentions of luck is a bit pointless anyway.

Have Liverpool got lucky getting a 2.77 points per game pace this season? Very probably. But getting lucky bounces here and there (and/or your rivals not getting them) is also very probably part of every league triumph.

Upgrade to paid

So, the league is all but done. The next challenge is sustaining the success even further. Below is a chart of all the teams who’ve finished the season with 2.0 or more points per game in Premier League history. Liverpool are set for their second-straight season, but they’ve a long way to go to match the major dynasties:

The nine consecutive seasons that Manchester United managed between 2004/05 and 2012/13 might be too big a goal to aim for right away. Maybe four-in-a-row should be the goal, which is the longest chain that Manchester City and Arsenal have managed. Even just a third-straight year would put Liverpool above Blackburn (1993/94 and 1994/95) and Spurs (2016/17 and 2017/18) on this particular arbitrary metric.

Liverpool’s squad is in a great place to be dominant now, and many of these players will be good for another couple of years too. But they don’t have much coming up behind them. The vast majority of the team’s regulars are in what’s known as the ‘peak age’ zone, and several of them are towards the latter end of it.

(streaks indicate how long player has been at the club, and kudos to Robin Wilhelmus for putting together the web app for this)

Maybe we’ll (finally) see more from Keïta. Maybe Minamino will become a more regular rotation option next season. It’s two and three years down the line that Liverpool need to be planning for and how they approach the next couple of transfer windows will be really interesting.

But one thing’s for sure: they’ll always have this season. And, barring an extreme nosedive, they’ll go down in the record books as one of the best teams that English men’s football has ever seen.

Can you teach tactics in a lockdown?

Tue, 16 Jun 2020 00:00:00 GMT

Hello, and welcome back to Get Goalside!.

Now that GG is back, we’ll be encouraging readers to donate to a different charity each week. This week’s is Mermaids, who support trans young people under 20 and their families, and provide informational resources.

If an international pandemic wasn’t so awful, it’d be fun to joke about how most teams would love 100 days to plan how to face Manchester City, as Arsenal have now had. That’s 2400 hours, or 1600 matches back-to-back-to-back -- even more if you fast-forward over the stoppages!

Fitness levels have, of course, dropped. Much of the coverage has focused on the physical side of the game, but with so much free time will players have been able to turn to study the sport instead?

Mentions of tactical training in interviews seem few and far between. Gabriel Jesus hinted at something in an interview with the City club website, while being coyer than a koi fish:

“The concept [of our training] is always the same. There are one or two things that have changed, but I will not be revealing them here!

“Pep is creative and intense, so he always has ways to make sure that the players’ fusion always works.”

But that’s pretty much it.

Granted, fitness is an understandable focus, given how much energy is expended being a professional footballer. But given the length of time between the last matches that were played and the ones set to kick-off this week, could this be the most tactically well-prepared we ever see Premier League footballers?

Maybe.

Although there’s been a lot of time to be watching video -- and at least some squads in the English football pyramid have been getting players to review and report on previous games -- not being on the training pitch could affect things. Viewing football on a screen is quite different to experiencing the positions you should be in on the field itself, so there’s no guaranteeing that sending a player clips on their positioning in build-up, say, will translate to them absorbing it in practice.

Making it as easy as possible for them to imagine themselves in those scenarios could be very important then.

“I think the effect can be close to normal training if you can visualise the actions for each player from their perspective as they would have it in a game,” Judah Davies, writer at Spielverlagerung and a coach, told me.

Subscribe now

“I think a lot of tactical and decision making things are about familiarity,” Davies said, “particularly given that a lot of them are performed intuitively due to time constraints. So to give the most familiarity, the player needs to be put in the situation in question, and see how it should appear.”

The use of virtual reality systems created specifically for footballers is still in its infancy, but perhaps if a whole team had taken Michael Antonio’s lead in getting a set-up installed in their own home they could’ve had a real advantage. They might not be able to run around, but being able to stand on a pitch (of sorts) and look around as play evolves around them would surely be a bonus.

Along the lines of cutting-edge things in football, read my piece from Christmastime last year, about the silliest things you could do with the most advanced data and tech in the sport.

This is assuming, of course, that the motivation is there. Shadow play -- moving through positions without an opponent -- may not be players’ favourite type training, and with the added emotional and mental complications of lockdown it’d be intriguing to know how many of them would take this up if it were available to them.

Some tactical aspects are presumably easier than others to study and internalise, which could have influenced what players have been up to behind closed doors when the Instagram cameras are off. I imagine that positioning in more static phases of the game is one of these, while counterpressing -- where every situation is different and quickly evolving -- could be harder.

But there’s the elephant in the room: fitness matters. We’ve all seen the early-season (or in summer international tournaments, late-season) games that become increasingly raggedy in the last 20 minutes. Fitness undeniably helps keep fatigue at bay.

“Perhaps if [the break and opportunity for study] was just a couple of weeks, it might be a slight advantage,” Davies said. “But the decrements in fitness will just be too big to balance I think.”

And, importantly, fitness is a cyclical issue. Not only (to put it in simple terms) does the state of being fit mean players can run more, but it allows them to build on their progress more easily.

“The fitter you are the better and/or more you can train with no drop in quality,” Patrick Eibenberger, fitness coach at Borussia Mönchengladbach told me. “This helps to accumulate more quality actions in training. Having said this the second benefit is that usually you recover quicker in between sessions. Therefore being fit benefits you twofold: in a vertical way (within a session) and horizontally (in successive sessions).”

Subscribe now

Perhaps, as well as the crowd noise, this interconnectivity is why some of the matches in the newly-returned leagues have seemed a little off. If the benefits of fitness accumulate like this, then the penalties of lack of fitness will surely follow a similar pattern. It might be easy to think of fitness as something linear, like the way they’re shown as status bars on video games, but the effects can be more complicated than that.

It might be convincing, then, to say that the fitness impact of the lockdown period will be far more important than the opportunities to knuckle down and study that lockdown also could have yielded. It wouldn’t be surprising (especially considering some attitudes towards tactical focuses in more traditional parts of the media) if some went further, to say that this is definitive proof that football is more about being fitter and faster than it is about Xs and Os.

But in the realm of physical conditioning as well as the tactics board, it pays to be strategic. Instead of asking about how fitness will affect the gameplan, “you could ask the other way around, like how does my strategy affect fitness training,” Eibenberger said.

“Do I want to follow a holistic periodization model or will I break decision-making strategies into small blocks and overload a certain physical quality more often? It is a little bit like a chicken-or-egg question. The perspective from where you’re trying to observe is paramount.”

…

I was, perhaps, a little optimistic when I thought that this could be a revolutionary three months in player tactical education. For a start, it’s the coaches who’ll have had the most extra free time. They may have had more time to study, but as far as the results on the pitch go they’ll still be relying on the players to be able to internalise the instructions.

On match-specific matters, there’s also no guarantee that the league’s teams will play the same way when the season begins again. Whether because of fitness concerns on one hand or the opportunity to tweak things themselves, there’s a chance that studying for upcoming opponents based on matches played three months ago is just a slight waste of time.

But it will be fascinating to take a look at these teams in their first games back and play a (very lengthy) game of spot the difference.

Subscribe now

Shout-out corner

I really enjoyed this vis below from @CrumpledJumper, given that I’m someone who almost exclusively uses Twitter on mobile.

On Monday @NylonCalculus I wrote about how LA and OKC are taking different approaches to this CRASH-or-GET BACK conundrum:

Every player who attacks the offensive glass is one less player who can get back on defense to stifle fast-break opportunities.

Here's a summary thread: pic.twitter.com/iXLtONKkHa— Todd Whitehead (@CrumpledJumper) June 12, 2020

Landscape charts are so often impossible to read on mobile (usually because people make them sitting close to a larger, landscape-orientated screen) but this one is incredible readable.

See something you like?* Email (reply to this one) or tweet (@EveryTeam_Mark) and let me know!

*(that relates to the Get Goalside! newsletter - chocolate cake recipes may be appreciated but are unlikely to make it into Shout-out corner)

Subscribe now

A reminder of our charity of the week, Mermaids.

Donate to Mermaids

What MUFC's future DoF should copy from Liverpool

Tue, 21 Jan 2020 00:00:00 GMT

I started this blog post as a thought experiment about what you’d do if you were hired as the Manchester United Director of Football (assuming you placed job over whatever club loyalty you might have and decided not to burn the whole place to the ground).

But then I got interested by what Liverpool did. So let’s look at that.

I assumed, when you became the MUFC DoF, that the business side of things would be running fine (thanks Ed — genuinely!), and that it’s just the football side you need to worry about. Also, because that’s the way football is set up, that you’d focus on the men’s side and leave the women’s team as is (maybe you’d increase their funding radically by, like, docking two week’s wages from any random member of the first team).

United have already tried to get their team back on track by buying big stars, of course (Pogba, Di Maria, Alexis Sanchez-ish). Not quite full Galactico, but moreso than other teams. The fanbase are unlikely to give you time to go the Chelsea youth team route, buying all the kids you legally can to find one Tammy Abraham. So we’re at the Liverpool method.

What is the Liverpool method? Let’s start by looking at Sadio Mane: at the time, it was reported that he cost Liverpool ‘around £30m’. £30m in 2016 feels like, what, £40m-£50m today? That’s kind of still what Liverpool are buying their first-team options for. Alisson and Naby Keita were both just over £50m, and Fabinho was just over £40m. Virgil van Dijk is an exception, partly because of the whole tapping-up scandal, and partly because that delay gave time for Van Dijk’s star to rise.

If they’d managed to sign him when planned, the rumoured fee was around £60m — still a record for a centre-back, sure, but noticeably less than the around-£75m he eventually went for.

The price isn’t the only thing, though. All these major signings for Liverpool in recent years have been around 23-27 years of age when they’ve moved. Old enough that you’re not buying at-present-not-yet-fulfilled potential; young enough that you’ll get pretty much the entirety of their peak years.

Hmmm.

Lemme break into my workplace and use Twenty3’s Content Toolbox *eye emojis* to do the most basic search for the 23-27 year-olds performing well on expected goals and expected goals assisted in Europe’s top leagues*.

*sidenote: this is another thing with Liverpool’s signings — it’s not as if they’ve had to look far. Hoffenheim (Firmino), Southampton (everyone), Roma (Salah and Alisson), Leipzig (Keita), Arsenal (Oxlade-Chamberlain), Monaco (Fabinho), Newcastle (Wijnaldum): barely any of Liverpool’s players have been gems unearthed in some far-flung league.

Let’s see:

Mauro Icardi (PSG, loan from Inter; possible, but baggage); Serge Gnabry (at Bayern); Timo Werner (Leipzig; interesting…, we’ll come back to him later*); Neymar (PSG; nope, for many reasons); Morata (Atleti; interesting). Along with Raheem Sterling and Mohamed Salah himself, those are the big hitters. Let’s throw in Coutinho too, even if his rep and/or stats isn’t quite as big as the other players mentioned here, as he’s at the upper end of the age range and at a major club already.

*I originally forgot to come back to Werner. I’d aimed to point out that Liverpool had been rumoured to be interested in him but potentially put off by the price. LFC know their market.

So then you have a string of recognisable-ish names: Arkadiusz Milik (Napoli), Paco Alcacer (Dortmund), Kevin Volland (Leverkusen), Alassane Plea (‘Gladbach), Angel Correa (Atleti), Andre Silva (AC Milan).

And some names you might not know(/that I didn’t): Karl Toko Ekambi (Villarreal), Sargis Adamyan (Hoffenheim), Lucas Alario (Leverkusen), Adrien Hunou (Rennes), Goncalo Paciencia (Frankfurt).

Those lists are just from this season, and you’d want to check whether they had a decent track record of course, but it’s these two latter lists that you’d be working from to follow ‘the Liverpool method’, I think. People might be getting excited about your Christopher Nkunku or Marcus Thuram, but if they leave their respective clubs (both German ones, interestingly, and both recently bought) then the transfers be materially different types of risk.

With this in mind, let’s look back at Manchester United’s transfer history since the departure of Sir Alex Ferguson (using rough fees and ages as I’m taking from transfermarkt without checking actual at-the-time figures. this is a newsletter after all).

2013/14: Marouane Fellaini (26, £30m); Juan Mata (25, £40m). In age and fee, for 2013, Fellaini probably fits the ‘Liverpool signing’ mould. Mata… similar, but more expensive.

Here is a good time to mention something else that’s very particular about Liverpool, though: they always seem to have a very good idea of where they want their signings to play, typified by Wijnaldum and Oxlade-Chamberlain. If United had bought them for those fees, I feel sure they’d have been underwhelming, played in roles that didn’t suit them. Current-day Liverpool wouldn’t sign a 26-year-old Fellaini of course, because he wouldn’t fit well in the system (never mind the Everton connection), and that’s kind of the point.

2014/15: Angel di Maria (26, £70m); Luke Shaw (19, £30m); Ander Herrera (25, £30m); Marcos Rojo and Daley Blind (both around 24 and £15m). Herrera fits the mid-range mould, and I think United fans would all agree he proved to be a valuable signing. Di Maria’s the big bet, although I’m sure his stats were good at the time — but Liverpool’s only signing that big has been Van Dijk, a defender (who age better than forwards) who cost so much for unusual circumstances. Rojo and Blind: squad players. Shaw cost far more than Liverpool have bet on a youngster in a long time, interestingly.

2015/16: Anthony Martial (20, *googles, for once* around £35m that had an incredible amount of add-on clauses that were reported at the time as ‘up to £58m’). Morgan Schneiderlin (26, £30m); Memphis (21, £30m); Matteo Darmian (26, £15m); Bastian Schweinsteiger (yeah, we all know how this went, he was old and cheap).

Martial and Memphis — way more expensive than Liverpool have paid for a youngster in a while. United lost money on Memphis (caveat: Louis van Gaal) and it may only be thanks to inflationary fees that they’d make money on Martial above the roughly £50m they’ve probably paid for him in add-ons by now. Schneiderlin’s a ‘get what you see’ buy; Darmian and Schweini squad player profiles.

2016/17: Paul Pogba (23, Mino Raiola); Henrikh Mkhitaryan (27, £30m); Eric Bailly (22, £30m). Hmm. Pogba transfer has been discussed to death and is, for various reasons, enough of an outlier not to discuss again. Mkhitaryan is at the top of the age range, but feels like one of those ‘if he’d been in the right role he could have worked’. Bailly seemed a decent buy, but has been very unlucky with injuries.

2017/18: Romelu Lukaku (24, £75m); Nemanja Matic (29, £40m); Victor Lindelof (23, £30m); Alexis Sanchez (29, Henrikh Mkhitaryan). Their age and fee profiles, in order: bit pricey; wtf; decent; I guess he was better suited to the role United had open than Mkhitaryan so if we count it as a free it’s ok…?

2018/19: Fred (25, £50m); Diogo Dalot (19, £15m). Interestingly, neither from Top 5 leagues. Dalot is probably pricier than Liverpool would go on a teenager; Fred is at the pricier end of the profile, but came from Ukraine, which is a far harder comparison to the English Premier League than the major European leagues, where you have much more cross-country info to go on.

And now 2019/20: Harry Maguire (26, £80m); Aaron Wan-Bissaka (21, £50m); Daniel James (21, £15m). In order: United’s Van Dijk (in transfer profile, I mean…); faaaar more than Liverpool would spend on a young full-back; maybe a reasonable punt on a youngster, albeit from the Championship.

A few things strike me:

Luck/’luck’. Although some of the profiles aren’t exactly what a ‘Liverpool method’ might go for, Di Maria, Mata, and Mkhitaryan are all players I feel could have worked out if things had gone slightly differently (on and off the pitch, to varying degrees). Maybe you can add Schneiderlin and Fellaini to this: with a flip of a coin and/or a slightly different role, could they have been another Herrera? (well, Fellaini kind of was, in how often he was used, just not appreciated anywhere near as much by fans). Think about the reverse too: what would we think of Liverpool if Salah and Wijnaldum, to pick a semi-random pair, had had as slow starts at Anfield as Keita has.
Big-ticket youngsters. Having looked at the two recent transfer histories side-by-side, it’s striking the difference here. There are other factors at play (United’s continued financial muscle; Liverpool starting their process at depths United still haven’t quite reached), but even so. There’s always a higher chance that youngsters won’t pan out as planned than with older players; and United have consistently paid relatively big money for young and/or relatively unestablished players.
Plan? Not only did United spend big on youngsters, but they did so at the same time, which feels like a lot of your annual budget to spend on risk. Also, the amount of full-backs points to a real problem. Shaw was terribly unlucky with his leg break, but that aside a teenager is still a gamble (and one United have repeated), Darmian shrug, Dalot is just an indication of what had gone wrong previously.

All of this isn’t even taking into account quality of the players when they moved, or the shape of the squad David Moyes inherited, or the ageing of the squad as the years have gone by. But it’s interesting how so little of United’s transfers in the past 7 years have been that mid-range, early-peak age profile that Liverpool seem to have perfected.

I want to stress that Liverpool’s method shouldn’t be copied religiously, and that there are still interesting questions to answer about what they’ll do in two or three years when those stars start to leave their peak age range. But when all is said and done, the £45m-25-year-old-from-a-Top-5-league is a profile that United should perhaps look more closely at.

Postscript: I have no idea how this profile scales across leagues. Does a similar principle apply lower down the Premier League, for clubs like Southampton and Newcastle, that the likes of Liverpool then buy off? Is it the £20m-25-year-old they want? Or does the financial landscape mean they need a different age range as well as price point? Ditto for leagues lower down the pyramid across Europe. Some clubs definitely seem to have worked it out though.

How good is 'good' and how 'good' is Messi?

Tue, 31 Dec 2019 00:00:00 GMT

It’s an unspoken necessity whenever you’re debating who the ‘greats’ of football are that you first need to define ‘great’. Not so much because the definition itself matters, but just so that everyone’s talking from the same starting point.

I’ve been interested in that kind of thing for a while, and this study from YouGov from 2018 only added to it [image below taken from the article].

Last week, I tried doing the same for football, tweeting out a Google Form where I asked people to rate from 0-100 (inclusive) a bunch of different players: Peak Messi, the average Premier League player, Jeff Hendrick, Virgil van Dijk, Lucas Digne…

Here are the results.

While I’m aware that 260 respondents to a Google Form isn’t the most scientific way of doing things, I deliberately tried to nudge people into structuring their thinking.

The preamble to the survey specified that 0 should be the worst player to ever play in one of the top 5 European leagues, while 100 should be the best player ever. I then asked for ratings of these ‘players’ to force people into setting their own benchmarks (and to remind them of their benchmark part-way through the survey).

The ratings 0-100 are along the bottom, and the height of the shape is the distribution of people who chose each rating:

Headlines: respondents choose Peak Messi over Peak Cristiano Ronaldo, and there’s still a huge range of ratings. People are likely to think of yer average UCL or EPL winner as somewhere between 75-85. Maybe part of the wide spread is because an ‘average’ player within a given category is quite difficult.

Let’s look at some of the top current players that I added in the survey. These were players that most people are going to be pretty familiar with — or at least have fairly firm opinions of — and as a result the range of ratings was relatively narrow.

This was interesting to see. Kevin de Bruyne and Virgil van Dijk have a pretty similar distribution, with most rating them somewhere between 85-95. That’s pretty cool, and, to me, is an intuitive place to place them: somewhere between the average Champions League winner and peak Messi/Ronaldo.

Everyone else has a bit more of a spread, to varying extents, which is fine. People have different opinions on players, as well as their differences in definitions. That was the main reason why I included Pogba and Henderson, because they’re two players I knew might have a large range of opinions about them. Henderson has most of his distribution between 75-85, about what we saw for an average Premier League or Champions League winner, but he has a significant tail stretching towards 60, which looks more like the general opinion of the average rating of the whole squad who wins a Premier League title.

And finally there was ‘the field’, a bunch of players from up and down the Premier League who people may not know particularly well. I wanted to have some players in the survey who people might think of as a ‘bog standard average’ player, as well as players who may be likely to be on relegation candidate teams, and this was just a random selection I could think of:

One thing that interests me here is that most of these players still have a peak of their distribution around 70. It may just be that peoples’ default is to think that Premier League players are around 60-70/100, regardless of who they are.

A couple of people on Twitter said that FIFA ratings had been in their minds while filling this in. It tended to be mentioned in a way to imply (or outright state) that they’d been trying to avoid thinking about those ratings while filling this survey in, but it was still on their minds.

I wonder to what extent FIFA (or Football Manager, who rate players in matches with 6/10 as the starting point that players are expected to improve on if they play well) influences this. The respondents, coming to the survey from Twitter, are likely to be people who’ve also played one of those video games.

While many parts of those games might be infuriating, if their general benchmarks for player category ability seeps into popular consciousness it might not be a bad thing. That was, really, the whole reason why I wanted to do this survey in the first place, to work out how it would be best to talk about players and prospects.

Would people understand what you mean if you say ‘an average Premier League player’? Apparently not. But they (or my 260 respondents at least) have a much more similar understanding of what an average Premier League starting XI player is. Just don’t ask them about Isaac Hayden.

Postscript:

This whole exercise has a use within actual football clubs, in that it shows the importance of setting benchmarks for scouts.

I’ve heard of some clubs that use their own players as benchmarks for various skills in players they’re scouting, and judging from my (albeit small and unscientific) survey that looks like it makes sense. The consensus among the respondents was much narrower for players that they were likely to know well, and arguably more narrow than abstract ‘average’ concepts.

Stat types who write publicly often bash the football establishment, but I think these particular clubs should probably get some credit for this system.

The silliest things you can do with advanced football tech

Tue, 24 Dec 2019 00:00:00 GMT

It’s Christmastime for those celebrating it [semantically, is it also Christmastime if you don’t?], and so I needed a newsletter topic that I could write well in advance. A recent conversation led to me having such a silly idea that I knew, just knew, that I needed to run with it.

So here it is: the most utterly silly things that you could do (in theory) with the most advanced tech and data in football.

Merry Christmas, happy holidays, or just have a nice December day. And if you enjoy this, please give it a festive share.

Which player runs the weirdest?

(using body pose technology)

This is the idea I had to kick it off, although I’m not entirely sure how I managed to get to it. ‘Body pose’ technology takes in video of sport (it started off in basketball, but is starting to be applied to football too) and essentially produces a stick figure of each player. And now you have shedloads more data to mess around with!

The main uses of body pose that are being talked about are knowing which direction players are facing at any given time, because that’s not in many (or any) datasets at the moment. You could also use it for other genuinely useful things, like scientifically studying a player’s technique of a given skill.

But you could also, presumably, get very very big computers to analyse how every player runs, and then run that through another very very big computer to see who the statistical outliers are. Is Raheem Sterling’s Velma-esque running really so unique? Is Jordan Henderson’s gait, as discussed by Sir Alex Ferguson, really so different to that of other modern footballers?

Now we’d be able to tell.

Who’s the best ballkid?

(using optical tracking data)

This is a combination of two of the most on-trend things in (different circles of) football right now (as of writing, and may have already dated).

In one sphere, José Mourinho and Duncan Ferguson have led to a massive increase in the number of Premier League managers hugging ballboys (granted, from a non-existent starting point). In another sphere, there are a rash of companies promising the ability to get ‘tracking data’ — knowing where players are on the pitch at any millisecond in the match — from cameras placed around the stadium. Some are even getting it from broadcast footage, requiring just one camera angle.

It all works by training software to identify which blobs of humanoid-shaped colour are football players, and so presumably it could do it to identify ballkids sitting at the side of the pitch as well.

And if it can pick up data on where the ballkids are and what their movements are, then it can give results on how long they take to get the ball back in play, which could be pretty cool. Useless, but cool.

This is where what stats nerds call ‘score effects’ come in though. If a home team’s losing, then they’ll want the ball back quickly; if they’re ahead, they’d prefer to dawdle. This would need to be factored in to any ballkid rankings.

(NB: I wouldn’t be surprised at all if there have been clubs in the past who’ve done basic analysis on this kind of thing, but this would take it to the next level).

Who are the best tactical foulers?

(using pitch control models)

‘Pitch control’ models are, like, sooo 2018, but they could be a really important building block for quantifying a football match. In short, you calculate — based on things like ball position, player positions, player trajectories — how much control each team has of each area of the pitch. (Image taken from this blog post from the ever-intriguing Barcelona Innovation Hub)

An image of a pitch control model in action

With these models, one is able to calculate who has the balance of control in any part of the pitch.

Now, teams who are being counter-attacked will often commit tactical fouls somewhere in midfield before the counter has time to develop. My idea is that tactical fouls made when the defending team’s control in their own half is weakest are the highest-value tactical fouls. If the defending team has high control in their own half when they make a tactical foul, it’s probably a waste.

(Let me know if I’m wrong, but this seems like the most tangible benefit of a pitch control model based on the (low) current levels of public knowledge about them)

Which goalkeepers are the most needlessly showy?

(using tracking data and body pose technology)

Simple. We all know that some goalkeepers, sometimes, pull out a save for the cameras.

With a mixture of tracking data and body pose technology, you could work out how far goalkeepers had to move to make a save, the time they had to make up that distance, and therefore which saves feature unnecessary flourish.

Who are the best at wasting time when they get substituted?

(using GPS/tracking data)

This is the least high-tech, given that players have been wearing GPS units in their shirts for a number of years now. They generally keep the data in-house for various reasons, apart from number of sprints and kilometres covered being released.

But why stop there!? Why not tell us the average speed of timewasting substitutions? And who are the worst, who’s letting the side down when they need to be timewasting? And which referees are the best at hurrying up players in skooting them off the pitch?

The public demands an answer.

Who gives the best hugs?

(using ‘wearables’)

Wearable technology is what it sounds like: it’s all the technology that players can wear. At the moment it’s mainly those sports bra GPS sensors, but boot trackers are coming in too.

With so many sensors all over the place, and with a sensor unit already strapped to the small of players’ backs, it surely can’t be too difficult to add another chip in there and measure data on how players and coaches hug.

Of course, for this you’d need to collect data on how the objectively best huggers hug, so presumably Jurgen Klopp would have a busy couple of days. But I’m sure nobody would mind too much.

Possession and anti-possession can be friends

Tue, 17 Dec 2019 00:00:00 GMT

I must be finding my subscription to The Athletic worth the (almost permanently 40% reduced) subscription fee because I’m using it yet again as the starting point for a newsletter. This time it’s Michael Cox’s piece on Newcastle vs Burnley, a game which everyone apart from me seems to instinctively recoil against.

The hook of the article is the question ‘what does a match look like when it’s between two sides who don’t want the ball?’. The bit that interested me is the following two paragraphs:

I don’t know whether Cox chose these quotes just to summarise the two managers’ wider philosophical outlook on possession or not, but taking them purely at face value it intrigued me that both statements can be true.

Sean Dyche is very correct when he says that possession doesn’t win games, and that it’s a myth that has long-been debunked. But Steve Bruce may well be right when he says that his team needs to learn how to keep the ball better as part of their general improvement. Where things get fuzzy is his use of the possession statistic to back this up or, maybe, the way it was phrased.

Lemme just present that quote again:

Last season Newcastle were second bottom of the Premier League in respect of keeping possession. If we’re to improve, it’s something we have to get better at — we have to keep the ball better.

Granted, Bruce doesn’t actually specify the metric for ‘keeping possession’, but let’s assume that it’s their average share of possession in matches as that’s the possession statistic that’s most frequently used, and it seems like a fair assumption to make.

Based on the way that their teams are currently playing, I don’t imagine that there’s too much of an ideological difference regarding possession between Dyche and Bruce. I just think that they’re taking different things from the possession percentage statistic.

Dyche, I imagine, has been irked by Burnley’s low share of possession being thrown at him. Quite rightly he points out that possession, in and of itself, means nothing.

Bruce, meanwhile, has identified a problem in his team. He’s reached for a statistic to back this up. It seems like he’s implying that a low share of possession shows their need to improve at keeping the ball, and it may well do, but it may also mean that Newcastle were playing a style which deliberately let their opponents have the ball for long spells of time.

In terms of a manager speaking at a press conference, I think this is fine. Bruce knows what he means, and I think the general message comes across. However, you wouldn’t want to use it in serious analysis within the club. Cox’s article also highlights David Moyes:

And there are various examples of managers focusing too much on the passing figures, as Rio Ferdinand memorably recalled of the David Moyes era at Manchester United: “Moyes wanted lots of passing. He’d say, ‘Today I want us to have 600 passes in the game. Last week it was only 400.’ Who cares? I’d rather score five goals from 10 passes.”

I’m not really sure what Moyes was wanting from his team in that second-hand anecdote, but for Bruce, I imagine that what would be more useful would be stats which focus on where and how his team are giving the ball away. They could look into possession sequences where the team fails to a) retain the ball b) play an unpressured long ball* for X seconds after gaining possession of the ball.

*playing a long ball may be a strategy, and so it shouldn’t be counted as ‘losing possession’, but a long ball played under pressure may be an indication that the team have played themselves into trouble.

This rather specific example probably wouldn’t make for a good answer at a press conference, and it probably wouldn’t make for a good message to players either (just tell them what they need to do rather than the metrics they’ll be judged on), but I just found it interesting that these two seemingly opposite quotes from Dyche and Bruce could, in theory, happily coexist.

Making predictions in an unpredictable age

Tue, 10 Dec 2019 00:00:00 GMT

It’s the UK General Election week and any comparison between the unpredictable nature of the British political landscape and the unpredictable nature of the Premier League table is probably best left unmade. And yet…

At eight points, after 15 games, this is the joint shortest gap there's ever been in the Premier League between 17th and 5th.

2003-04 was also eight
1998-99 was nine

They're only ones in single figures

average is 13 points.

Last season was biggest ever at 22.— Miguel Delaney (@MiguelDelaney) December 7, 2019

Things haven’t changed much after another game in the bank — Manchester United are fifth with 24 points; Southampton are 18th on 15. And amidst all of that tightness in the league, there are the stories.

Manchester United fans have gone from wondering whether Mauricio Pochettino — or even the new-look José Mourinho — would be a better choice in the dug-out than Solskjær to jubilating that he is, once again, at the wheel, all in the space of a week. Mourinho himself has been transformed from grumpy git to ballboy bestie; Everton have reversed their slump now that Duncan Ferguson, himself a friend-of-the-ballchild, is in charge; and many a pundit and tweeter throughout the land (myself included, I imagine) are now eating their words about Steve Bruce’s managerial quality after Newcastle have got four wins in six.

But of course, with so much uncertainty in the league, we could get to the end of the month and all of these figures could be the subject of #[Name]Out tweets again.

How do you make predictions and analyses of things when it’s all so topsy-turvy, and when being ‘wrong’ doesn’t necessarily mean that you were wrong.

You get this every transfer window. Ángel Di María at Manchester United was a great example. Did he end up as a good signing for the club? Not really. Could he have been? Absolutely. He didn’t seem to take to Louis van Gaal (considering how LVG’s United team ended up playing, who can blame him) or Manchester (considering that his house was broken into, his wife was reportedly too fearful to return to it, and it was up for sale within a month of the incident, it’s hard to blame him much on that count either — not to mention the weather).

Point is, it would have been hard to predict either of those things before his move to Manchester. Even the old trope about foreign players not liking England is hit and miss, particularly in terms of how much it affects their performance. It’s far from unknown for South American players to thrive in the north of England, and a happy work life can make imperfections elsewhere a little more bearable.

The problem is that there’s only one Di María to Manchester United transfer to test your theory on. Even if you thought that he wouldn’t adapt to Manchester, it’s tough to know what the chances really were that the Argentinian wouldn’t take to the many varieties of grey that the city has to offer.

Things are a little easier when you make comments about a team, because they play multiple matches, every now and then against the same opposition. But it’s still difficult. (I realise that this sounds a lot like a SAMPLE SIZES!!! post).

The obvious answer, of course, is to watch as much as possible of a team/player/manager’s teams, check the statistics and read around the subject to inform your opinion, and proceed, cautiously, from there. But even professional pundits don’t have time for that. Goodness help the rest of us with pesky things like day jobs that don’t revolve around watching football.

I say all this not because it’s revolutionary, but because it feels like there are going to be a lot of narratives flying around about various teams which’ll come falling down within a few weeks of first airing.

Has Steve Bruce really fixed Newcastle? If so, how? Has Ole Gunnar Solskjær got Manchester United back on track? If so, how? Has José Mourinho etc etc.

The ‘how’ is the important part. If it’s unexplainable, it may not be properly fixed. The ‘how’ should also come before the answer to the main question (the one that people actually care about).

Has Mourinho fixed Spurs?

How: it looks like Tottenham are benefitting from a genuine new manager bump where previously weary-of-boss players have a new (though limited) lease of life. But some of the same problems of progressing the ball and aspects of defensive solidity still seem present. Playing a bit more of a direct style seems to have bypassed this first problem, and the main strength of the Tottenham squad is, in my opinion, in their attacking midfield, but it’s not completely consistent.

Answer: He’s probably got them performing back around the level of the sum of their parts, but not above it, which is what Pochettino had originally done.

Has Solskjær got United back on track?

How: Scott McTominay and Fred have been in great form of late, just looking completely ‘on it’. Timing challenges, good positional decisions, reading the game to make interceptions and supporting runs when the team’s in possession. This has helped on the defensive and offensive side of things. Rashford has also looked on it, and the quality of his shooting against Tottenham was fantastic. That I’ve only mentioned players being on top form gives me concern about how sustainable this is — players can’t be at the top of their game all the time, that’s why it’s ‘the top’.

Answer: I… don’t know? I don’t think it’s entirely by his managerial genius, but I certainly don’t think that the results of the past week have just been luck.

Has Steve Bruce really fixed Newcastle?

I’m not going to lie about the amount I watch Newcastle. I really don’t know what’s going on there.

It’s a wonder why columnists and weekly pundits, at least, don’t revisit past predictions of theirs and examine why they thought what they thought and what parts they got right and wrong, if only because it would fill column inches and air time.

It wouldn’t necessarily need to betray any gaps in knowledge either, as the general public are sensible enough to realise that one individual may not quite have an intimately detailed expertise in all 20 clubs in the Premier League, as well as the odd couple in the Championship or Champions League.

There’s probably a market for a pundit to have a pair of weekly columns: one on a Friday, say, giving a preview for the weekend’s action; and one on a Monday or Tuesday that reviews and takes stock of what they predicted, what surprised them, what they foresaw, and what it all means for the future.

Hmmm. That actually might be too good an idea to be giving away…

Trust me. Just trust me.

Tue, 03 Dec 2019 00:00:00 GMT

I forgot to plug this in the past couple of weeks but my colleagues at Twenty3, Daniel Girela and David Perdomo Meza, were at the Barcelona Analytics Conference with a kickass poster featuring some of the cool work we/they are doing there.

I’m taking the ‘standing on the shoulders of [Arsenal-associated] giants’ approach this week, with the post heavily built around two extracts from David Ornstein articles on The Athletic from the past few days.

The first is one of the less-exciting paragraphs from his and James McNicholas’ piece on the demise of Unai Emery at Arsenal (emphasis added).

That month, it emerged that Arsenal’s head of recruitment Sven Mislintat would be leaving the club. The German had been assured he would graduate to the role of technical director after Wenger’s departure, but with Raul Sanllehi staging something of an executive coup, Mislintat found himself sidelined.

The second comes from his Monday column, rounding up some of the juiciest bits of goss from around the grounds in England. The below comes from a section titled ‘Rogue companies turning search for marginal gains into a minefield’.

For example, one Premier League club recently wrote off an annual six-figure sum they were paying to a firm for performance data that was eventually found to be riddled with mistakes.

The partnership had been running for several years before the club decided to recruit specialists to thoroughly scrutinise the information being provided. The feedback was damning.

They alerted the company to the problems and demanded answers, but all that came back was a “thank you” for raising the concerns, a pledge to implement improvements and a reference to the small print of the contract, which essentially stated there is a margin for error and no recourse.

And, alongside these two extracts, Unai Emery’s hiring and firing itself can be chucked into the mix as we ask: how can we trust people in football?

These three instances — executive coup, cowboy company, and unsuccessful manager — aren’t exactly the same thing, of course, but they do share commonalities. The coup and the company are clearly in the game for their own benefit at the, seeming dispassionate, expense of others. The company and Emery are both outsiders who win their way into a club by promising something that they fail, albeit for different reasons, to deliver — if the clubs in question had had more knowledge about these figures in the first place, maybe they’d have never been hired.

How does a club, or an individual within a club, know who to trust when backstabbing, underperformance, and downright grifting are all around?

I suppose that this is why so many people hire people they know: you might be able to find someone better outside of your network of knowledge, but you might also find someone worse who just interviews well.

This isn’t anything unique to football of course, but what is a more unique problem is not having that network, not knowing who to ask, or who to ask who to ask, about who to hire.

(Presumably, this is why the performance data company were able to get away with dodgy data with a Premier League club. Data of all types is still a relatively new field within the professional game, and so how are you supposed to know who to trust?)

Leicester City’s Head of Performance Innovation, Paul Balsom, spoke at the recent Training Ground Guru conference about how the club came to hire their Head of Analytics Mladen Sormaz. Balsom was advised on what to look for in the Head of Analytics role and not to rush the eventual candidate when they first joined the club, as the first few months would have to be devoted to sorting out the data engineering and infrastructure.

That’s sound advice. I wonder how many Premier League clubs are seeking advice like this, not just about who to hire but about the entire concept of using data within football. I also wonder how many have the kind of networks where they can get advice from someone knowledgeable and trustworthy in the first place. Having been in and around the football stats sphere for about six years now, I’ve seen a number of job descriptions for different types of data-related roles at clubs and they have differered significantly, so it’s clear that not everyone is doing what Balsom did.

Now. I’d written the above on Sunday evening, with the aim of tying it to a neat close on Monday. But then on Monday morning, StatsBomb CEO Ted Knutson decided to give stats twitter a good ol’ shake to wake them up for the week. [As I know tweets don’t always show up right when embedded online, for the avoidance of any doubt Knutson is quote-tweeting a tweet asking people for ‘a thing that everyone in your field knows and nobody talks about because it would lead to general chaos’].

Having trouble viewing this tweet? Click here to open original tweet.

It should be noted that, as the CEO of a data provider himself, Knutson isn’t without a dog in this fight. That said, I more or less knew the first tweet to be true and I trust that he’s telling the truth on the second.

These are similar problems to that in the extract from David Ornstein’s article: data companies providing data with flaws. The scale of these flaws differ hugely both in severity and how widespread within the data they are, but flaws are flaws.

(Although it should be noted that the company that made potential errors of judgement several years ago is more likely to have learnt from it and have consistent data now than the company who seemingly under-collected shots last season. Companies will also respond pretty promptly to any problems that you do find that have slipped through the net. I should also say, in this paragraph that’s rapidly becoming a pre-emptive fire extinguisher for any bridges I’m close to burning, that issues in data is a bit like refereeing errors — sure, the big ones get a hell of a lot of attention, and deserve to, but the vast, vast majority of the data is absolutely fine).

You don’t need me to elaborate on why ‘can we actually trust the data’ is a worrying sentence for football clubs. I also don’t want to piss off the data companies more than I already have by dwelling on it, BUT Knutson logged back on later in the day to do it for me [the whole tweet, here, wasn’t relevant so I’ve quoted the important bit]:

The more the professional football world moves toward possession and ball progression value models, the more they care about tidy possession sequences and who has possession of the ball where and when.

I’ve written previously on the topic of ball progression-type models here. To nick part of that post as a quick intro to the concept:

[These types of models] have generally sought to measure how much value players further back in the goal-scoring process (ie, midfielders or defenders) add to the chances of goal-scoring. […]

Across the board, these models have shown that passes in midfield have little (if any, really) direct value in scoring goals. I have some quibbles about the fact that midfield is very dynamic and questions about whether the models capture this […] however in terms of the descriptive value of the model I think that it’s valuable, and true-to-life, to say something like ‘by and large, taking an average of events in this area, actions in midfield have little direct impact on scoring goals’.

The mess of midfield and fine margins in the impact that actions appear to have is why the people doing the modelling need the very best data.

So. On one side of the equation we have data companies who have accuracy and/or consistency issues; on the other, a cadre of data scientists needing accurate data that they can rely on for their models to be trustworthy. Not only do they need it to be accurate for the models to be right, but the problem with modelling is that it can get pretty tough to troubleshoot and work out why something’s gone wrong, or even if it’s gone wrong in the first place.

And it’s not just the big data companies who might give people trust issues.

[Note: QA = Quality Assurance]

Having trouble viewing this tweet? Click here to open original tweet.

Knutson’s right, clubs collect a bunch of their own data with an army (of varying size) of young analysts. They can be noting things that data companies don’t collect, or more subjective information that’s tough to construct from the data feeds they’re buying.

As a football club, though, I imagine that you don’t really want to spend much time on reading up on, and enacting, quality assurance practices, which should be part of the benefit of buying data in from a provider: it’s their job to collect accurate, reliable, consistent data so they should be the ones who have quality control down to a T.

Yet, seemingly, they don’t. And if you’re a potential customer, it seems unlikely that the provider is going to tell you about their issues, particularly given the response that the performance data company from The Athletic article gave when confronted with their own dodgy data.

Who do you talk to to find out about which data company is most accurate? *shrugs*.

And don’t even get me started on the rash of companies popping up offering to turn video footage into tracking data. That, dear reader, is a conversation for another time.

This post wasn’t meant to be a write-up of the problems that data companies can have, though, it was about trust and how clubs are meant to establish it. It’s clearly tricky terrain with few, if any, maps to guide you. To discover who to trust, it seems you first need to know who to trust.

On the Twitter-sphere, and particularly the stats twitter-sphere, we can be quick to snark at the old men (always men, of course) who’ve been in the game for several decades and whose large contact books keep them in work. But, on the Wild West of the data frontier, it’s the same sort of people, who know everyone’s secrets and everyone’s strengths, who could be some of the most valuable.

A defence of early-analytics' mistakes

Tue, 19 Nov 2019 00:00:00 GMT

The analytics ‘movement’ gets a lot of stick sometimes (although generally from people who are never going to give it much credit anyway it must be said). The analytics ‘movement’ also over-thinks things too much. I’m going to indulge both by writing something of a review and an apology of the things that ‘analytics’ has gotten wrong over the past five years or so.

Disclaimer: This will all be based on my recollections, which may be faulty, but probably little more so than anyone else’s. I’ll also likely slip into referring to ‘analytics’ as a singular entity with singular opinions, which it isn’t, but meaning is broadly clear.

‘Don’t talk about G, talk about xG’

There was a moment in time on analytics twitter when it seems you could scarcely move for discussions about whether ‘expected goals’ was a good name, why it was a bad one (the ‘why’ was always why it was bad), and what a better name might be (there weren’t many good options put forward).

In retrospect, I think that there was something accurate in these discussions — a struggle to communicate well the fruits of analytics — but a missing of the mark as to what the actual problem was.

In his talk at the recent StatsBomb Innovation in Football conference, Seth Partnow (former Director of Basketball Research at the Milwaukee Bucks) talks about ‘good’ and ‘bad’ stat names. Some examples of bad ones come from the old ice hockey days: Corsi and PDO, both giving very little insight into what the stats mean. ‘Expected goals’ may not be perfect, but it’s good enough.

A bigger problem was just that there was very little content out there that aimed and explained expected goals at a general audience. Joel of MessiSeconds fame was basically the only person seriously doing this.

Part of this, I think, is because the people who were involved in analytics early doors were somewhat wedded to the purity of the numbers and the statistical method (understandably and defensibly). Another part is that xG is a fairly limited tool, through which only a relatively small selection of stories can be told, and they mostly boil down to “this team/player you like? yeah, they’re Actually Bad”.

I exaggerate there for effect, but the dampening of excitement around players or teams on a hot streak was certainly a theme around that period of time, and it understandably pissed people off. Joel/MessiSeconds got a lot of angry West Ham fans in his mentions for his video in October 2015 which (accurately) predicted that the Hammers would drop off from their 3rd-place form. (They finished seventh).

And then the analytics ‘movement’ as a whole caught a hell of a lot of anger for its judgement of Marcus Rashford at the end of the same season…

Rashford/Iheanacho

The Marcus Rashford-vs-Kelechi Iheanacho ‘debate’ can be summed up really well by a single Michael Caley tweet (the tweet itself seems to have vanished into the ether, but I happen to still have it, having used it in an explanatory xG article at the time, in May 2016):

While the surrounding debate featured a lot more snark from both sides, this tweet is the essence of what this whole thing was. It’s well-centred, while giving an impression of why people so took against analytics twitter, particularly the more brash/sarcastic sections.

It should be said that even people who liked Iheanacho by eye have little answer for what’s happened to him since that 2015/16 debut season where he scored 8 goals in the equivalent of 8.5 matches (766 minutes) for Manchester City. He’s now very rarely getting gametime at Leicester City. So while one could say that ‘analytics got Iheanacho wrong’, so did everybody else.

But it was mainly the pissing on the Marcus Rashford bonfire that got people up in arms.

At the time he was vastly overperforming his expected goals (unsurprising considering that he scored in the debut of virtually every competition he played in), and stat-types pointed this out. They probably could have done so more tactfully, but it was a valid thing to note, and Rashford’s goal output hasn’t been anywhere near as high as it was in his early, explosive, debut days.

In that first half-season, he got 0.51 non-penalty goals per 90 minutes (h/t FBref); since then that rate has been 0.32, 0.44, 0.33, and 0.4 in 2019/20. In expected goals terms — and to provide a similar metric to Michael Caley’s original tweet, which had him averaging around 0.33 expected goals + expected goals assisted per 90 — he’s enjoyed a continual rise in the Premier League. For the past three seasons, his rate has been 0.42, 0.5, and 0.57 per 90.

That increase could partly be a natural progression as the player ages, and partly as Rashford has spent more time as a central striker. Crucially, though, that 0.42 from 2017/18 (the earliest season that the website has data) probably isn’t too far off what Rashford might have had in 2015/16 if he hadn’t been hampered by an unusually restrictive manager in Louis van Gaal.

While Rashford/Iheanacho will likely be held against stattos by some, I don’t think it’s wholly fair to use it as an example where analytics got it wrong (and where the predictions were wrong, the analytics ‘movement’ is far more likely to review and reassess methods and predictions than those outside it). What could certainly have been different is the tone.

The need to be loud to be listened to…

This is probably a good time to talk about why early analytics twitter seemed to piss off so many people with its tone. I’ve mentioned already how I think that xG analysis lends itself (too much) to party-pooping, and there’s another turn-off that seems to be baked in too — the tendency to come off as “I’m smarter than you” when using data to back up, or create, arguments.

These are things that I think even media communication experts would struggle to deal with (and god knows we in the analytics-verse are not that). But another element of the disdain some have for analytics twitter is the need to be loud to be listened to when you’re a small group who are trying to disrupt the status quo in one way or another.

It feels crass to compare analytics in football to activists trying to shift the Overton window in political discourse, but it also seems to me to be basically the same phenomenon. Radicals yell about their given cause and make wild claims while more moderate sections quietly go about their business and talk to the unconverted in a slightly less obnoxious way; and then the mass public grows in awareness and understanding of this new cause, and some become radicals themselves and some become moderates; and the cycle continues and continues as the movement grows and/or change occurs.

Naturally, though, the loudness annoyed some people, also being seen as representative of the wider ‘movement’ at times, and, I think, unintentionally resulted in another thing.

‘Don’t cross’, and other prescriptive advice

Some have criticised analytics folks for giving advice like ‘don’t cross’, advice which is, to them, either too rigid or too obvious.

Early analytics research did indeed include the observations that through-balls were more likely to lead to goals than crosses, for example. Also that corners into the box might not be as exciting as in-stadium crowds often appear to think they are (although, somewhat ironically, StatsBomb are now one of the principle cheerleaders of set-pieces).

There are a few things going on in the criticism that the analytics movement offered prescriptive, and very basic, advice, but the main one is that all of this didn’t tend to be advice at all.

More often, it was just research about football, trying to work out what we could learn about football from the data. Sometimes it backed up conventional wisdom (through-balls create higher value chances than crossing); sometimes it changed conventional wisdom (shooting from distance ain’t that great). Practitioners, though, tended to be aware of the limitations of their research.

One other important point is that the simple analysis done in the early days was very useful at quantifying aspects of the sport. It needed to be done, even if it wasn’t riveting and, more importantly, it was useful to know what conclusions were to be found in the data, even if it wasn’t riveting. Data analysis that backs up conventional wisdom is still really important because it helps rule out the possibility that conventional wisdom is wrong.

A final point to note is that this early, simple(r) analysis coincided with the need to be loud to be listened to, which isn’t ideal but was probably an awkwardly necessary part of the journey.

I’ve now written 1300+ words addressing the dissatisfaction and disdain that some people still hold for the analytics ‘movement’, and so I’m going to spend the last part of this focussing on happier things. Things I’d like to see more of/tell past-me to focus more on if I could go back in time 4-5 years.

The happy/advice section

Make it fun/relatable

This isn’t necessarily easy. The point of analytics is to be analytical, and that doesn’t tend to link up well with hot takes. As I mentioned before, a lot of the early analytics content was pouring cold water on the hot takes of the day. But maybe you can use stats to craft hot takes that have a larger chance of standing up over time.

I’ve also always thought that stats guys should just, like, fucking swear more. I think this is part of the popularity of things like the StatsBomb podcast: not that it’s an R-rated f-bomb-a-thon, but that it’s normal people talking about stuff. There’s a tendency in analytics-people writing — whether because of the weddedness to the stats or because of our largely middle class backgrounds — to be pretty high-fallutin’ in our language, and if we spoke like normal people rather than dickheads then people might’ve been more receptive to us.

Spend more time on trivia

By ‘trivia’ I don’t mean ‘who was the Shrewsbury Town captain in 1990/91’. I don’t even mean ‘Virgil van Dijk has made X interceptions this season; no, I mean things like how teams get to the box, which players pass the most on a team, splitting a player’s chances up into really good ones and ok ones and pretty bad ones. In other words, things that are relatively simple, relatively tangible, and relatively important to know.

I’ve been really inspired by the aforementioned Seth Partnow’s work on Twitter and in The Athletic since he left the Bucks, as it’s the kind of thing I’ve always liked messing around with. You don’t need to understand the science of statistical modelling to calculate it, and it’s all a lot less abstract than something like expected goals (which, while a pretty simple concept, is still nonetheless a concept one needs to get one’s head around).

If there’d been more ‘good trivial’ stats (and not just possession percentages) used by stats-folks and the media could that have paved the way better for expected goals? Maybe, idk.

In the end, it doesn’t matter

I’ve typed a hell of a lot of words here about stats and mistakes and things that could have been better, but in the end xG is on Match of the Day, Liverpool are getting adoring press over their analytics department, and The Athletic have been (for whatever reason…) liberally using Opta graphics in their articles*.

(*being from a company that helps media companies make data visualisations, I have opinions about this, but the main one is that the visualisations that we make at Twenty3 are great and The Athletic, like every other media company, should sign up to use our Content Toolbox)

The point being… if this was ever a fight, it’s the analyticos that’ve won it.

Has analytics made Liverpool boring?

Tue, 29 Oct 2019 00:00:00 GMT

Before we start, a quick note that the paper for mine and my Twenty3 colleagues’ StatsBomb conference project on ‘How to break down a set defence’ is available here.

There are a few things that we know for certain about Liverpool Football Club and their men’s first team.

They employ a whole analytics department, headed up by PhD-from-Cambridge-holding Ian Graham and including worked-at-CERN-yes-that-CERN-the-Big-Hadron-Collider-CERN Will Spearman.
The last 18 months or so has seen Liverpool men’s team focus their play through the full-backs to an extraordinary extent. Granted, it’s helped by having two players of the quality of Trent Alexander-Arnold and Andrew Robertson, but to have your two full-backs combine for 23 league assists is quite something.
The analytics folk at Liverpool appear to be actually listened to (not a given in football). There’s the story that opens the afore-linked NY magazine article about Ian Graham wowing Jurgen Klopp, and later in the piece it quotes Klopp as saying, in reference to the analytics crew: “The department there in the back of the building? They’re the reason I’m here.”

And now for a couple of things that we sort of know to be true, and can do a little bit of speculation with:

The centre of the pitch is more dangerous than the flanks, as a general rule (for some sort of detail on this, have a play with the interactive Expected Threat map around halfway through this blog from Karun Singh). It seems likely that conceding turnovers in central midfield is more dangerous than conceding turnovers from advanced full-backs at a similar height up the pitch.
The thing that Ian Graham is “really obsessed about,” he says in a recent Freakonomics podcast, “is the risk-reward payoff of passes.” He was, primarily, talking about attacking passes and the fact that dangerous players can have low pass completion rates...

…However, we know that Liverpool have smart people working for them (people who, in a pre-LFC life, have developed advanced ‘pitch control’ models that calculate the values of areas of the pitch at any given time).

We know that these smart people are listened to (or, at the very least, notably appreciated). We know that the men’s team has two very capable attacking full-backs. We know that the team has started using those full-backs, rather than the central midfield, to progress the ball.

We also know that Jurgen Klopp’s teams were, once upon a time, famed for their exciting and fast-paced style of football; a style that now seems to have disappeared or been diluted.

‘Risk-reward payoff of passes’? Would it be too much to suggest that the analytics department have worked out that progressing the ball through the full-backs is a better risk-reward payoff than doing it through the centre, and that’s what’s behind Liverpool’s change of style of late?

Like the newsletter? Sign up now

Maybe it is.

Despite the — quite frankly, surprising — amount of media coverage the Anfield brain trust have had, we don’t know much about what they do for the club or what actually gets listened to. To take a pure cynic’s view, one could say that Liverpool have been very good at media and public relations of late. Putting their analytics department in the spotlight presents them as modern and forward-looking to fans (and sponsors?, and particularly American ones of both who are more au fait with figures in their sporting fun?).

There’s also the matter of personnel. Early in Klopp’s reign, when the central midfield was more attacking in nature, Adam Lallana, and then Alex Oxlade-Chamberlain, were part of the regular starting XI. Both have had injury trouble, and perhaps Liverpool don’t have players similar enough to play those roles (although most people assumed Naby Keita would be let loose in a similar manner, which it doesn’t seem he has been).

And then there’s the Premier League.

Liverpool — as you may have heard this before — haven’t ever won a Premier League title. Their last top-flight men’s trophy was in 1989/90 and that, coupled with the feud with Manchester United who used that drought to overtake them in number of league titles won, is a great source of pain for the fans.

Perhaps playing more conservatively in the centre is a way of taking some of the self-inflicted danger that an intense press can bring. Perhaps this style, this shift to using the full-backs as ball progression options and leaving the central midfield with more defensive and stabilising responsibilities, is simply a way to give the team as a whole more control in matches. And so perhaps the change has a singular focus, to try and maximise the chances of winning that elusive first. ever. Premier League.

Attacking stats are role-dependent too

Tue, 22 Oct 2019 00:00:00 GMT

A quick note before we start. A couple of weeks ago my Twenty3 colleagues presented some research we’d done on ‘How to break down a set defence’ at the StatsBomb Innovation in Football conference. The full paper is now online here.

For a good couple of years, it’s been accepted that football players’ defensive statistics are extremely dependent on the role that they’re asked to play. It’s not enough to sort a table by the numbers of tackles per 90 to order players as good defenders to bad defenders — that’ll just give you the ‘search and destroy’ DMs at the top and everyone else below them. Or, to be more specific, the names at the top will probably be ‘good’ but being further down the list doesn’t necessarily mean that they’re not good.

In the realm of attacking stats, we’ve traditionally been far readier to take the numbers at face value. A large part of this is just that it seems to work. Forwards with high numbers of expected goals are generally the best forwards. Attacking midfielders with high numbers of expected goals assisted are generally the best attacking midfielders.

But part of it, I think, is that there are fewer ways to skin a cat in attack than defence, and that’s blinded us to the effect a player’s role within a team can have on their stats with forwards. As an attacker, your job is generally to score, or directly contribute to other people scoring. You do the first by shooting and shooting alone; you do the second by passing. That’s two types of output you need to look at — in defence there can be three, four, five, as well as the very large output that isn’t measurable, that of ‘I’m defending this space so well that the opposition isn’t going to attempt going through it at all’.

It’s very rare for players in forward positions to stray from their usual roles too. Roberto Firmino is one of the few examples, a nominal central striker who’s usually outdone in expected goals by the two teammates flanking him. He’s such a notable example, though, that the entire footballing world knows about him and naturally looks to other stats. He’s involved in build-up play, so we look at his passing stats; he presses from the front, so we look at his defensive stats.

We know, though, that there’s greater variety in role as we take one step back from centre-forwards. Players who feature in attacking midfield positions may actually be something like second strikers. Mason Mount, for example, has around 0.36 expected goals for himself and 0.12 expected goals assisted for others, according to StatsBomb numbers. He presses quite a lot too, to add to the confusion about the ‘type’ of AM he is.

Particularly in the last 18 months or so, there has been a rapid development of different types of ‘Expected Possession Value’ models (see: xThreat by Karun Singh, a Possession Value framework from Nils Mackay at Opta Pro, the (slightly) older xGChain from StatsBomb, and various ‘non-shot expected goals’ models have been around for longer than that but I’m struggling to find links right now).

They’ve generally sought to measure how much value players further back in the goal-scoring process (ie, midfielders or defenders) add to the chances of goal-scoring. That makes sense, and I’ll preface all of my critiques about these models by saying that the people creating them are far smarter than me and they deserve immense credit for creating these models in the first place: progress is an iterative process, but it’s far easier to be the one coming in late in the day and saying ‘what about this?’ than it is to be the one taking the first step.

Across the board, these models have shown that passes in midfield have little (if any, really) direct value in scoring goals*. I have some quibbles about the fact that midfield is very dynamic and questions about whether the models capture this, but that concern is most relevant when applying the results of the model; however in terms of the descriptive value of the model I think that it’s valuable, and true-to-life, to say something like ‘by and large, taking an average of events in this area, actions in midfield have little direct impact on scoring goals’.

*In fact, the visualisations in Singh’s post show how quickly xT values rise as the ball gets closer to goal. It seems natural that players whose role dictates that they play close to these areas will be the ones get big ‘boosts’ from contributing to a team’s expected threat. The image on the left is a conventional heatmap of xT values in each zone; the image on the right is a 3D version.

With that in mind, we should think again about what we actually want to use these models for. Do we really want to use them to build a list of midfielders who contribute the most to goalscoring? Or, to put a different spin on that question, is that even the job of most central midfielders?

If the roles of attacking midfielders vary then the roles of central midfielders vary even more. And I’m not even convinced that ‘contribute heavily to increasing chances of the team scoring’ is part of most central midfielders’ roles.

On the offensive side, maybe this is something we need to think more fundamentally about. Is contributing to chances of scoring one continual phase, which runs from the moment that a team gets possession of the ball, or are there multiple phases to it?

I’ve thought for a while that there’s a kind of ‘endgame’ to football possessions. It’s the part where the team is solidly in the final third — or even the final quarter of the pitch — and need to find a way to get through, find a chink in the armour, find a way to manipulate the opponent to create space. (This doesn’t need to be for a shot directly, although it can be, it could be working to make space on the wing that leads to combination play that leads to a cut-back that leads to a shot. But the working of space must start somewhere). If this is a separate phase in itself, then it has implications for how we think about the game.

Is it the job of traditional central midfielders to help their team to this stage, rather than to be responsible for more directly increasing chances of scoring? (Part of their job, of course, will be defensive too, but that’s a whole different subject).

There’s two questions this post raises then. 1) What is the role of central midfielders in contributing to their team’s possession phases? 2) How much of players’ possession-based metrics is role-dependent and how much is indicative of ability.

Characteristically, I don’t have any firm answers (although I think I’ve dangled a couple of theories in this piece already), but I feel like they’re worthwhile questions to be asking.

Enjoy this? Subscribe to the newsletter:

Already subscribed? Spread the word:

Spending a Sunday with pressing stats

Tue, 15 Oct 2019 00:00:00 GMT

This past Friday was the StatsBomb Innovation in Football conference (I think I’ve remembered the official name right…), a fun day at Stamford Bridge listening to people who’ve spent a lot of time with data. There was me and my colleagues from Twenty3 delivering a talk on how to break down a set defence; there was Ajax Head of Sport Science Vosse de Boode delivering a superb presentation about all the stuff they’re doing over there; there was Thom Lawrence from StatsBomb having an, amusing and technical, footballing existential crisis of sorts. All the good stuff.

Michael Caley was also there, and his talk was on pressing. It was a really thorough and interesting overview of the different methods that have been used in the public analytics sphere over the years, and it made me want to get my teeth into the problem. (All of the talks from the conference will go up online, and I highly recommend checking them out when they do. If you’re interested in the talk I was involved with on breaking down a set defence, we have a proper paper that will be going online too).

And so I spent much of my weekend making notes on pressing concepts and data types and going around in circles of investigation before arriving on a method that I’d played around with last month. Let me take you on that journey.

First, let’s tackle the conceptual problems that surround trying to measure pressure in event data (for ‘event data’, think of a row in a spreadsheet for every pass, tackle, shot, etc).

There’s the obvious one that it tells you barely anything about where the defensive team actually is. StatsBomb’s data has ‘pressure’ events as well as things like tackles and interceptions, but that still relies on some kind of action on the part of defensive players.

This is a problem because, like with the famous story of ‘missing’ bullet holes in returning warplanes, teams in possession will pass around the opposing defensive block. And because they pass around them and pass away from potential pressure and risk, there usually isn’t a lot of direct evidence of where defending teams are positioned.

A similar-ish ‘missing data’ problem is that teams who are successful in a high press or high block will stop opponents before they get down the field, and that could have an effect on statistics. A high percentage of a successful high-pressing team’s defensive actions will be high up the field not only because their aim is to force defensive actions in those locations, but also because by doing so they limit the amount of actions they will ever need to make further back.

And then we have a bunch of more footballing conceptual problems.

One is that you need to separate ‘pressing’ from ‘defensive block height’. They’re two different, but related, things and if you’re looking to quantify one of them you should be aware of the other.

One is that you need to separate ‘process’ from ‘result’ (or at least be aware of the interaction between the two). A stat like ‘high turnovers’ is purely about the result, but other metrics may be looking at aspects of both. This will be more of a problem if one is looking for ways to quantify process or intention, as one’ll need to be aware of how much one’s actually just seeing ‘result’ showing up.

And a related problem is that there are two things to want to measure as far as pressing goes: type and efficacy. It’s important to really think about what one’s measuring and what the results may yield, because a metric designed to look at type of press may be being heavily influenced by the efficacy (or lack of it).

That’s a lot of things to be aware of, and it’s likely that any metric that one puts together is going to run into several of these. Nothing’s going to be perfect. But you can try and be as good as possible.

Because of all of these problems, I think that a lot of the existing metrics — which are mostly quite broad proxies — have an advantage stemming from exactly that broadness.

Colin Trainor’s original PPDA (opposing Passes Per Defensive Actions) is really cool, and the concept can be applied to more specific areas of the pitch. It kind of measures process as well as result (harrying the opponent high up the field — process; but it relies on making defensive actions — result), but it’s more than fine as a broad look.

Opposition pass completion percentage is similar, and is something that I’ve used before, and taking it a step further by only looking at passes going forwards is something that Paul Riley has done. He limited the area he looked at to the ‘second box’ (the 18 yards forwards from the 18-yard box), thereby looking at a more specific type of press.

Anyway. Inspired by Caley’s talk, I wanted to tackle ‘pressing’ more systematically, thinking about the best metrics to use for each of the problems that I’ve talked about so far.

I figured that StatsBomb’s data, with their pressure events, would help. I reasoned that a defensive block could perhaps be identified by players receiving the ball under pressure.

Ah, but wait. Surely it would have to be passes towards goal? Players being pressured when receiving backwards passes would surely be being ‘pressed’ or closed down in a more active sense rather than being put under pressure by a defensive block.

Ok, so let’s do that.

Ah, but wait.

Below is a plot of under pressure moments following a player receiving a pass that was going towards goal. It shows Chelsea pressure from a Chelsea vs Manchester City WSL match in 2018/19, but the point is that massive area of nothing in the middle of the field (Chelsea are defending the goal to the right).

That doesn’t mean that Manchester City were playing freakishly directly that match or that Chelsea’s midfield was a sieve, but that this is the nature of the sport.

So that was my initial ‘defensive block height’ idea out of the window. I realised, after several hours of work, that part of the way that a defensive block manifests itself is in the way it forces teams to try and pass around it. (Now that I type that, it seems very obvious).

Back to the drawing board…

Fortunately, I had another idea stowed away which, thinking about it, is pretty similar to Riley’s.

About a month ago, I’d wanted to check out the idea of ‘buffer zones’ — the heights of the pitch where teams struggled to complete passes towards goal. Flipping that gives some gauge of defensive height.

It looks a little like this*, and once again teams are defending the goal on the right. Red means they’re harder to pass through in that area, green is easier than league average:

*this is a quick mock-up I did when I first looked at this, so the data’s out of date by a few weeks now.

One downside of this is that it doesn’t separate ‘process’ from ‘results’ of the defensive block as much as I would like, nor necessarily ‘pressing’ from ‘block’. A team that attempts a high block but is bad at it will show up as easy to pass through in a similar way to teams that just attempt a low block and cede some of the midfield to their opponents.

However, one part of this that I do like is that it can give a sense of different line heights. Some teams have one band where they’re slowing down their opponents and then another further back. Some are league average everywhere (which, actually, isn’t really what I wanted when I started looking for a gauge of ‘defensive block height’ but ah well).

I guess what I want to get across is some of my thought process about pressing stats that stemmed from Caley’s talk on Friday. The journey did not exactly come to a satisfactory ending. But it was a relatively interesting journey.

Enjoy this? Subscribe to the newsletter:

Already subscribed? Spread the word:

The value of 'certainty'

Tue, 16 Jul 2019 00:00:00 GMT

Welcome to the Get Goalside! football analysis newsletter, glad to have you aboard.

There’s big money being spent in the transfer window. It doesn’t feel like too long since £130m would get you half a team; now, Manchester United look set to spend it on Aaron Wan-Bissaka and Harry Maguire.

Both the £50m spent on a full-back who’s had one season as a starter, and the £80m on the centre-back who no-one is praising for his defending, have raised eyebrows. Wan-Bissaka, at least, is a youngster who is clearly pretty damn good at defending (as highlighted in the middle of a tweet thread about William Saliba on the shiny new Get Goalside! twitter account:)

Having trouble viewing this tweet? Click here to open original tweet.

No-one, as near to ‘literally’ no-one as you can get, is mentioning Harry Maguire’s defending when justifying his price. This says two things: 1) not much is known about players’ defensive qualities 2) football now values qualities other than defending in a central defender to such an extent that defending can be glossed over for Maguire.

Let’s tackle the second thing. What qualities other than actual defending are worth spending that much money on?

Maguire caught attention for his threat from set-pieces during the World Cup last summer, and various people at StatsBomb - the closest thing to a public set-piece evangelist sect that the game has - believe in his value there. Speaking for myself, I can’t say I know much about his value there other than he’s big and he has a big, solid head, but I trust their opinion.

The second is his on-ball ability. He dribbles a lot and progresses the ball. He’s good at it, partially in his bravery to actually do it, partly in his vision, partly in his execution. He’s slightly clumsy while doing it, but it’s probably as much as an aesthetic problem as a meaningful one.

The third factor in Maguire’s favour is the fact he’s ‘a known quantity’. While ‘he knows the league’ became such a cliche and empty buzzword phrase to justify signings or managerial appointments that it became roundly mocked among certain parts of twitter (of which I’m a part), there’s reason to it in certain circumstances. And United’s is one of them.

The Manchester United squad needs such a comprehensive rebuild that taking chances on potential - but not certain - stars is a far riskier proposition than it usually is.

One might be able to get a right-back of a similar quality to Wan-Bissaka for less than £50m, but they’re certainly not around in the Premier League and anyone else would have the question marks of adapting to a country, learning the language, learning the league and its players. Wan-Bissaka not only knows the pace and quirks of the league and its teams, but he knows how to play against many of the players he’s likely to come up against here. That matters. The margin of expected outcomes is smaller.

On the Maguire front, I do not believe that he’s a magnificent defender. He is, at worst, broadly competent. At best he is moderately good. His speed of turning is not great; he is reticent to engage with attackers (not necessarily a bad thing, but a notable stylistic quirk) and can be clumsy when doing so; he has had a surprising amount of communicative issues with Wes Morgan in his time at Leicester. He is not bad positionally. Substitute out player and team names and these could all be applied to Chris Smalling.

Don’t scoff - let’s not forget that Smalling had a period of time when he was truly lauded. Under Louis van Gaal in 2015/16, there was a stretch of a few months where his role and his form aligned and his reputation went through the roof.

I, however, responded with skepticism and wrote a piece on a blog that has since gone offline, but I’ll screenshot part of the conclusion for you as it’s interesting to look back on these things (there are bits of the whole piece I think are flawed, and the comment on Alderweireld here is a little unfair, but by and large it holds up).

I could well be wrong, but defensively I’m not sure if Maguire is really an upgrade on Smalling (although he’s a couple of years younger), and in that sense I can only repeat the last sentence of that extract from the Smalling post: If Manchester United are spending £80m on Maguire then it’s as much of an indictment on the current quality of their squad as it is praise of the defender.

But - assuming that United really do know that the Leicester City man has noticeable defensive holes - the familiarity with Maguire is valuable.

The club have gone through nearly a decade of failing to replace Rio Ferdinand and Nemanja Vidic, putting faith first in the young English talent of Smalling and Phil Jones, and then in a series of ‘mid-range priced, flawed talent’ signings in Marcos Rojo, Eric Bailly, and Victor Lindelof.

Lindelof has, at least, performed well enough that he’s being linked with Barcelona (take that with as large a pinch of salt as you wish), but it took him nearly a whole season to find his feet. United simply can’t afford to have to buy yet another central defender in a year’s time.

Paris Saint-Germain’s seemingly impending purchase of Abdou Diallo for around £25-30m will be used as a juxtaposition to Maguire, and it’s a fair argument. But the point of this post is to make the counter-argument that, even if Diallo has a far higher ceiling than Maguire, United will not want to add to that list of Rojo, Bailly, Lindelof. They need a high basement level, even if that means lowering the ceiling on what they’re getting, so that they can cross some positions off their depth chart as ‘sorted’, for now, to turn their focus to others.

With Wan-Bissaka, they’re crossing off right-back. Even if he costs £50m and he isn’t great offensively, he should be good enough for the next few years. Luke Shaw’s 2018/19 means he’s probably nailed down left-back too.

Splurging £80m on Maguire is an attempt to nail down centre-back. That’s not to defend the spending, per se, but it helps explain and understand it. While his defensive quality might not be too different to what Smalling was offering, the stuff that Maguire brings to the table elsewhere is an upgrade. With a rotating cast of him, Lindelof, Smalling, and Bailly, that positional group should be fine enough for the time being. On their own, none of the four inspire incredible excitement, but on any given day you’ll probably have two who are fit and in some reasonable degree of form.

That just leaves central midfield, right wing, striker, and the post-De Gea era at goalkeeper.

That’s a lot. And that’s why United are spending £50m on Wan-Bissaka and £80m on Maguire.

Enjoy the post? Subscribe and get Get Goalside! in your inbox every Tuesday

NB: If United don’t end up buying Maguire, my thoughts on him still stand. If City buy him, *shrugs*. Maguire would have less defending to do there, so the team would focus on getting the most out of his on-ball and set-piece ability, but *shrugs again*.

Would you have bought Van Dijk from Celtic?

Tue, 07 May 2019 00:00:00 GMT

Hello again. The fun thing about this newsletter going out on a Tuesday means writing it on a Monday, which means waking up and seeing all the disgruntled tweets about Game of Thrones. It sets a nice low bar for me to clear.

A player who’s had to clear a very high bar in the past 18 months is Virgil van Dijk, arriving at Liverpool for the eye-watering price that he did. So, I’m interested in knowing what the Dutchman was like at Celtic.

I’ve seen contrasting reports about it in the past. Some have said that they tried to get Van Dijk signed up straight away (which sounds a lot like an Arsene Wenger-ism), some clearly weren’t sure whether he was ready for a big step up.

Because this is just a newsletter and not an in-depth report, I just watched one game, a Europa League tie at Parkhead against Inter Milan (one including future teammate Xherdan Shaqiri, in an amusing turn).

There were a couple of things in Van Dijk’s game that night that have been a through-line from that game in February 2015 to his PFA Player of the Year performances in 2019.

One is the so-good-it’s-scary body control he has.

The below situation actually comes from an uncharacteristic piece of miscontrol from Van Dijk that gifts Inter possession.

As the man on the ball drives centrally and the man between him and Van Dijk pulls to the left, pinning Van Dijk’s RCB partner Jason Denayer, it forces Van Dijk into a decision.

He picks a good moment to step up towards the man on the ball, but I’m interested in the way that he does it. Look at the little steps he takes as he turns. The fact he’s never off-balance.

It’s hard to do a firm comparison, but if most other defenders are required to turn like that, they’ll tend to brace their legs as they put their feet down, the lack of bend in their leg being the thing that slows them down (regular disclaimer here that I’m not a kinaesthetics expert).

This tends to work for taking away pace that a player’s using, but it’s not so good for balance. Van Dijk, on the other hand, takes all that speed away in his knees and thighs, every little step acting like a shock absorber. He takes about five or six steps as he turns almost a complete 180, and only one - one of the first - is a ‘bracing’ step.

Subscribe now

Again, not a kinaesthetics expert, but I think this is part of what enables Van Dijk to be so good. The skill in one-on-ones or runs in behind defences is catching the defending players off-balance (generally physically, sometimes mentally, though the latter tends to lead to the former). If you’re turn like Van Dijk, it’s going to be incredibly difficult to catch you out.

Van Dijk coupled this physical skill with a good reading of the game.

Inter have managed to get the ball into space in midfield. There is no-one between the man on the ball and Van Dijk. WHAT WILL OUR INTREPID HERO DO NOW?!

It turns out that Van Dijk stays exactly where he is. He just moves his body slightly, De Gea-esque, to block a potential pass.

I think that Van Dijk reads the player on the ball’s intention to make the pass in behind Van Dijk, which is why the Dutchman doesn’t feel he needs to do anything more than this (and I’ll come back to this later too).

The man on the ball delays for a moment, waiting (or hoping) for his teammate to get into some space. Van Dijk shifts a step to his right, in exactly the same position.

And intercepts the pass.

Subscribe now

Now, I said I’d come back to why Van Dijk stays more or less exactly where he was when the player initially received the ball. One reason might just be incredible foresight. As good as the centre-back is, though, I’m not giving him credit for being psychic.

This is the third through-line between Van Dijk’s Celtic days and now — he’s a defender who naturally drops off. I’ve mentioned this previously in Get Goalside!, when Van Dijk was on international duty alongside Matthijs de Ligt.

[Leroy] Sane was pushing up quite a lot throughout the match. In the first half, that made Van Dijk [who was on his side] retreat a little, rather than try and play him offside throughout large periods of the game. Whether the Liverpool centre-back didn’t communicate this or De Ligt didn’t notice it, the 19-year-old from Ajax stayed keeping a high line.

In the second half, [the centre-backs switched sides so that van Dijk was no longer paired against Sane and De Ligt was the one dropping off to deal with him]. Whether through a natural inclination to drop off or because he was more aware of where his centre-back partner was, Van Dijk dropped off level with De Ligt. It helped.

I think that Van Dijk’s natural inclination is to drop a touch deeper than other centre-backs. It was the case at Celtic, probably a factor in Van Dijk staying where he was in the previous screenshots, and the case now.

The body control and reading of the game are positives, the inclination to drop deep is value-neutral — a stylistic thing — but there were a couple of definite negatives in the game too.

The first goal, scored by Shaqiri, comes from the Powercube himself making a run behind Van Dijk. The defender’s feet are pretty planted to the ground. He doesn’t take a look over his shoulder.

I would say that this is uncharacteristic, but there were a couple of other times during the match where a similar thing happened with Van Dijk looking like he was trying to keep an offside line and trying to watch what the player on the ball was doing.

Too late, Van Dijk checks his shoulder and Shaqiri has a run on him. Van Dijk’s feet have only just started moving.

The goal ends up being a little fortuitous, but the distance between the two of them is pretty clear when the pass is in flight.

That instance captures a lack of awareness* and lack of foot movement, two things that might put a scout off a player. They’re also things that have probably improved since 2015.

*His standing-still interception kind of plays into this. Without better quality images I can’t be 100% certain, but I’m pretty sure Van Dijk doesn’t look over his shoulder at the run in behind him. He’s basing his actions on where the player on the ball is looking. That’s fine, and is often intelligent defending, but I get wary of players who rely on it too much, which I think the centre-back probably was in the lead-up to Shaqiri’s goal.

Subscribe now

I don’t think these things are necessarily things that have been completely eradicated from Van Dijk’s game, but I’d say they’re less of an issue now.

A sidenote: This is a useful exercise to do with players to get a gauge on how general player progression goes. From the outside, it’s hard to know how a player is going to progress as time goes on (from the inside, it’s probably difficult to tell too). But you can get an idea of how likely it is that a player will pick up or improve a certain skill.

You’re not able to get that just from one match, but y’know, hypothetically.

Thanks for reading. Let me know any thoughts, either on Twitter @EveryTeam_Mark or, if you’re viewing this as an email, just hit ‘reply’. If you’re not a subscriber, though, now’s a great time to do it. Get the weekly Get Goalside! football analysis newsletter to your inbox totally free.

Other stuff

I wrote an article this week over at Football Whispers about whether players defend differently after they’ve been booked. I think it’s pretty interesting. Read it here.

I also wrote a blog post about trying to come up with a data visualisation for it, if you’re interested in that kind of thing. That’s here.

Thanks for reading. If you’ve enjoyed this or found it interesting, it’d be great to share this around. ‘Til next week :)

'Possession adjusting': An essay.

Tue, 23 Apr 2019 00:00:00 GMT

Hello everyone, and I hope you’ve had a great Easter weekend if you celebrate it (religiously or culturally (yum, chocolate)), or just a great week if you don’t. I hear it’s Passover too! In England, the weather’s nice and M&S pre-mixed Mojitos are flying off the shelves!

Um, anyway.

The following is on ‘possession adjusting’. A basic definition from mine and Ashwin Raman’s football analytics glossary:

Often used with defensive statistics, these are metrics that have been adjusted to reflect the amount of possession a team has had.

Now we begin…

1.

In the beginning, there were no stats.

Then there were shot stats, and it turned out that they roughly matched who the good strikers were.

And then… there was the wilderness.

Around 2013-2014, WhoScored and Squawka sprang up and stats gradually hit the mainstream. People did not know what the stats meant. Even the smart people. And most of us, and I include myself in this, are not born smart.

From shots, sprung expected goals. As the tips of the tree of football, the attacking goalmouth, was fed the sun of statistical enlightenment, its rays gradually worked their way further down the ecosystem. Shot assists, and then expected goals assisted, and then ball progression.

But defensive stats stayed difficult. After resisting the idea for too long, believing that Opta’s figures contained some magic bullet, I accepted that the best way to work with (the currently available) defensive stats was to use them to build an idea of a player’s style or role within their team’s system.[1]

There are numerous problems with using defensive stats as a sign of good defending. The fact that some players are instructed to be more defensively proactive than others, for one. Are large numbers a sign of good execution, or just tactical role?

Another is that some players just don’t get the opportunity to do much active defending. When I was starting out blogging, it was around the 2014/2015 season. John Terry and Gary Cahill were in a soon-to-be-title-winning Jose Mourinho team and Burnley were getting their first taste of the Premier League. Some centre-backs – the defensive position I was focussed on – were just told to back off and sit behind their midfield.

An idea that I toyed with back then, and which Ted Knutson has continued[2], was about adjusting the defensive statistics in order to level the playing field. If Terry and Cahill were bunkered behind a midfield, surely it’s not fair to judge their (to come up with random figures) 1 tackle per game with the 3 tackles per game that other centre-backs were averaging.

‘Possession adjusting’ was born.[3] The concept is simple. Players on teams who spent most of the game defending (well, without the ball) get their defensive numbers dampened a little, while players on teams who have a lot of the ball get their numbers boosted.

2.

Recently, I’ve been thinking a lot about this.

Somebody (and I can’t remember who, to my annoyance) pointed out to me several years ago that both teams in a match are naturally going to have a similar number of turnovers. It’s just the maths behind one object, the ball, being tossed between two parties. Quirks of refereeing and data collection mean the figures are unlikely to match exactly, but they should be pretty damn close.

This came back to me recently, and I started thinking about how this concept applied to adjusting defensive stats. Let’s call it ‘turnover adjusting’, compared to ‘possession adjusting’.

Here are the theoretical arguments underpinning those two approaches:

‘Possession adjustment’: Making defensive actions, like tackles or interceptions, are partly due to role and partly due to opportunity. If a team spends most of the match without the ball, their players have more opportunity/more time in which they can win the ball back.

‘Turnover adjustment’: There are only a finite amount of times that the ball can be turned over in a match, and the number of turnovers will broadly match up between the two teams. The team who dominates possession, therefore, will naturally benefit from as many turnovers as the team they’re dominating. The turnovers that they do benefit from, then, are just condensed into a smaller period of time. Some matches are more bitty, therefore have more turnovers, therefore have more chances to ‘earn’ defensive action stats than others.

3.

I’ll be honest, I find both theories fairly convincing. They’ve been circling around in my head and, just when I think I’ve settled on one being correct, the other muscles back in.

I should briefly talk about why adjusting defensive stats might be worthwhile.

Adjusting defensive stats isn’t about seeing which are the ‘good’ defenders, but if you’re using the defensive stats to determine a player’s role, then you need to isolate that role as much as possible. In other words, you have three things that can affect a player’s stats. Their execution, their tactical role, and things outside their control like the quality of their or their opponents team and, from that, how much of the ball either side sees.[4]

On the most basic level, it’s kinda useful/interesting to know who the ‘high-volume’ defensive players are in football. Who the players who are just really damn active at breaking up their opponents’ play are.

You want to know this because, as a team, you might be looking for a player to play that kind of role, and you want to know who’s doing it at the moment. You don’t want your list of potential candidates skewed in a way that means the player’s role on the team they’re currently on might not translate to the team that you’re on.

(Or you’re in the media and you want to remove some easy-to-remove caveats around using statistics if, indeed, this particular caveat is easy to remove).

4.

There’s an argument, though, that there’s no point in adjusting defensive stats at all. We don’t adjust attacking stats like shots or expected goals just because a striker plays for a good team, for example.[5]

But why should you use possession or the amount of turnovers in a game to adjust defensive stats? What are you looking for in the numbers? Do you get closer to this with that adjustment?

Why am I asking all these questions?

The last one is one I can actually answer. I’m asking because I don’t have a firm opinion.

I think I come down on the side against possession adjusting. If it’s supposed to be a way to even things out between similar roles on different teams, I’m not sure it does it. The theory behind ‘turnover adjustment’ seems to me to hold true. Players on high-possession teams will be getting rewarded just because their team keeps the ball (I think).

Possession adjustment also assumes that possession is the determining factor in opportunity for proactive defending across all positions on the pitch. I think this is flawed.

It also depends on the defensive action itself. Even with my doubts about possession adjusting, it makes more sense for tackles and interceptions than clearances, the majority of which come from clearing crosses or long balls.

Statsbomb have a stat called ‘pressures’. It does what it says on the tin – it counts the amount of times a player pressures the person on the ball. I mention this because, crucially, it isn’t a ‘turnover’ stat like tackles or interceptions. If there’s any defensive stat that makes sense to possession adjust – at least as I theorise it – it’s these pressures.

so yeah. I am probably ‘against possession adjusting’, but I’m not sure I’ve decided what, if anything, I am ‘for’. Thank you for listening.

If you want to get in touch on this, you can find me on Twitter @EveryTeam_Mark or, if you’re a subscriber, reply to this email!

Footnotes:

[1]The same, really, is true of shots. There are forwards, like Roberto Firmino, whose job is only partly to be a direct goal threat. They’re rare though, which is why the primitive measures like shot volume tended to match up with good strikers. Tactical variety among defensive players is far more common.

[2]I believe that I came to the idea independently of him, but I can’t exactly guarantee that I hadn’t seen him or someone else talking about it and just can’t remember it.

[3]Although I did something slightly different, combining possession with the share of shots on target and blocked shots that a team conceded. The idea for me was that those shot numbers would give a sense of what the midfield was letting through.

[4]You could also argue that a player’s role should be taken across all facets of the game. A ‘search-and-destroy’ type of defensive role on a high-possession team won’t be the same as a ‘search-and-destroy’ type of role on a low-possession team, I think. This complicates things slightly, but is another argument against possession-adjusting defensive stats (I think?). i.e, you can’t just transplant one abstract notion of a particular player role from one type of team to another, the act of having more or less possession will change the role enough that they are no longer the same role.

[5]Maybe we should! Who knows! There is an argument that, instead of splitting stats up on a ‘per 90 minute’ basis, we should do it on a ‘per X possessions’ (usually per 100 possessions) basis. Thinking about it, this incorporates both of the ‘possession adjustment’ and ‘turnover adjustment’ theories. However, it’s far less intuitive than normalising stats on a ‘per game’ basis, and you also don’t necessarily solve the problem of knowing to what extent normalising the stats is even a worthwhile thing. (For example, the point raised in the previous footnote still applies).

Get Goalside

Extra time for in-play time

Notes

That's bait

A whole new world model

Press 1 for rest defence. Then 2, 7, 4, then 1 again.

Crabs, the everything app

What's the worldview of your data?

Football is a simple game...

Friendly Neighbourhood Thanos

What *would* Football 2.0 look like?

The 'Just About Managers'

Square one

State of play - football

State of play - accessing data

State of play - accessing insight

Post-script

The Grand Rewrite

Finding the bench to mark against

'Analytics history' - there is so much to write about and share

Are footballers artists or accountants?

Congratulations, you have been acquired

Time on the clock

It's a grab bag

Leading from the front

Seeing it through their eyes

When the Saints go hiring in

The bag holding the thoughts together

Spotify for 'successful line breaks'

Play smarter, not better?

Answering the hardest question in football

Tough on Moneyball, tough on the causes of Moneyball

Cutting down the nuance

The tiers

Vocabulary for chaos

Some reading, some vocabulary

Get Goalside 2024-2025

Fishing for principles

Looking for stability

Building capacity and breaking lines

The path to now

Notes

References/Further Reading

The wobbly chair of sports software

Who will win the processing war?

System stability

Possession adjusting: Part Two

Here be numbers

So what now

No, so really now what

Possession is not nine-tenths

Data tables are in the eye of the beholder

More than meets the eye

This is where the fun begins

Tying the threads of the Get Goalside cinematic universe

So what now?

Lessons in genAI

Tools for tools

Context is king

Content is king too

The ‘pool of knowledge’ problem

The end

Bias for bravery

Video is all you need

What's your research question?

It's the incentives, stupid: when expected goals are the problem

Positional Play and manager metrics

Why is the impact of football analytics in question?

Football is ruthlessly efficient

Other sports are uniquely inefficient

Football should actually be much different

The data isn’t better than (most) coaches… yet

The smart people are quiet (or, vice versa)

We’re looking in the wrong place

Just run some more

The four quadrants of football

A fifth birthday

Defensive stats

‘What is football’

Conference-watch

What would Football 2.0 look like?