Why is the impact of football analytics a question?

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 paragrah:

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.

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:

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.