Fishing for principles

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….