What is science if not experimentation persevering. We're going to have two experiments today.
The first is a format of piece that I want to try out, which I'm calling 'Research in Focus'. Papers and presentations (the formats that big research tends to be delivered in) are designed for fellow practitioners, which makes sense. But they can be hard for everyone else to get into.
These 'Research in Focus' pieces should be something that almost anyone can read, and something that even people who read the academic papers can use to get a good overview of the work. I may tinker with the format as time goes on, but the aim will remain the same.
Research in Focus: Karun Singh's Expected Threat
This piece will also be available later as a standalone post later
Why it's worth your time
There are precious few pieces of work that have both an immediate and a long-lasting impact like this one. People still talk about Expected Threat, and the blog is still linked to by others, from The Athletic to academic research papers.
What it says
The aim of the piece is to work on the age-old question of how to value actions on the pitch, beyond just shots (which have xG).
Assuming you don't lose the ball, there are two things you can do when you have it: move it somewhere, or shoot. So then the calculation is about:
- What is the chance of you shooting from where you are vs moving it
- If you shoot, what's the chance of scoring
- If you move it, what are the chances of you (or a teammate) scoring if you shoot from there (which will also depend on where 'there' is)
This is the basic logic behind xT. But there are layers to it, so we don't just ask those questions once, we ask them for a chain of five potential actions. To quote the caption that goes alongside an interactive example in the post, "when the team has the ball in the highlighted zone, they will score in the next 5 actions 9.4% of the time".
What's cool about it
There had been various ways of applying expected goals to non-shooting actions (i.e. passes, dribbles) before this, but compared to xT they were quite simple. Metrics called xGChain and xGBuildup took the expected goals value of a shot and credited every player who'd been involved in the buildup with that figure. Other work in the early and mid-2010s simply valued a pass based on what the xG value would have been if a shot had been taken from where the pass ended.
The basic attraction of xT is that it, intuitively, says "this pass might not set up a goal directly, but a goal might happen in a couple of seconds time because of it".
It's also, and I say this without any criticism at all, very simple both to calculate and understand. If you need some type of 'possession value' model as part of whatever work you're doing, just whip up xT. It'll get you 70% of the way you need to go, in 10% of the time it'd take to get you 90% (metaphorically speaking, these aren't real numbers).
Finally, the post is fabulously presented. The top of the page has a progress bar (which all blogs and reading-based websites should have as standard in my opinion). It has five pieces of interactivity. It opens with a gif.
It also has mathematical equations, helping to make the method reproducible. It has little note boxes, telling you about scalar fields or Markov chains or just letting you know you can skip the maths sections to get to some more visualisations.
It's rare that the presentation of analytics work is as well-thought through as the analytics work itself. This blog post did that, and I think it's a key part in its instant, and four-year-long-and-counting impact.
The second experiment is, in some ways, more traditional. You can now give Get Goalside money.
This isn't going to be one of those paywalls that hides everything. In fact, these are really more 'supporter plans' than 'subscriptions', which is why 'supporters' is the language I'll be using for it.
Ideally this would be a 'pay what you want' set-up, but the newsletter service I use doesn't currently have that option. Instead, I've set up two 'tiers'. One will be £2 per month (just over $2.50USD), the other will be £6 per month (just under $8USD). They both get you the exact same thing, so it's just a case of what value you get out of Get Goalside and what you can afford.
Supporters will get priority on their questions in mailbag newsletters I do in future. If mailbag issues are popular, I might do supporter-only ones (which I'd intend to make freely available at a later date).
I also plan to keep doing 'Research in Focus' pieces, like the one at the start of this newsletter. These will be the bulk of the incentive for supporters: once a number of these have been done (five maybe), they'll go behind the paywall and be for supporters-only when new ones go out. However, I'll make a number of them free to everyone on a rotating basis (I plan to start with two, but this'll grow as the overall number does).
You can join one of the supporter tiers by hitting the button below
(If this has been forwarded to you or you're viewing the web version, you can subscribe to the free version below instead if you wish)
As part of this all, I've also made sure that I've tagged newsletters well. There's a good range of categories on the newsletter's homepage if you want to browse a particular subject that Get Goalside covers (e.g. Applying Analytics, Explainers, Nature of Football, Research in Focus will be there too). There's also a Handy Analytics Resources page that I'll endeavour to keep updated as time goes on.
Thank you again to all the readers, subscribers, and for the kind words about hitting the 100 issue mark earlier this month. If you want to read, or read it again, the link is here. I asked eight people in analytics what they'd learned over the past few years and what they hope to learn in future - it's a good, informative, thought-provoking read.
There'll be a post all about last week's Stats Perform Pro Forum coming very soon, some more Research in Focus pieces, and lots of the usual good Get Goalside stuff.
Have a great week.
If you have any questions or problems with the supporter plans, please email firstname.lastname@example.org