Towards a new kind of analytics
(22 Oct 2018) This article was originally published on StatsBomb. Re-reading it today, I cringe at its earnestness and didactic tone, but I still stand by every word and, unfortunately, it would be hard to argue that the situation in the public analytics community has changed materially. Luckily, it looks like the quantum leap from counting stats to large scale dynamical models is being made by several companies in the field, and while this work is not as public or as academic as I would like, progress is being made.
I have been involved in football analytics for four years and doing it for a living since 2014. It has been a wonderful adventure, but there is no denying that the public side of the field has stalled. But this is not really a “crisis of analytics” piece or an indictment of the community. Instead, I want to point out one critical barrier to further advancement and plot a course around it. In short, I want to argue for a more theoretical, concept-driven approach to football analysis, which is in my opinion overdue. It is going to be easy to read this short article as a call to basic, as opposed to applied, research and consequently dismiss the ideas as impractical. Try not to do this. I like applied football analytics and I firmly believe that it has value — even the public variety. But I also believe that we have now reached the point where all obvious work has been done, and to progress we must take a step back and reassess the field as a whole.
I think about football analytics as a bona fide scientific discipline: quantitative study of a particular class of complex systems. Put like this it is not fundamentally different from other sciences like biology or physics or linguistics. It is just much less mature. And in my view we have now reached a point where the entire discipline is held back by a key aspect of this immaturity: the lack of theoretical developments. Established scientific disciplines rely on abstract concepts to organise their discoveries and provide a language in which conjectures can be stated, arguments conducted and findings related to each other. We lack this kind of language for football analytics. We are doing biology without evolution; physics without calculus; linguistics without grammars. As a result, instead of building a coherent and ever-expanding body of knowledge, we collect isolated factoids.
Almost the entire conceptual arsenal that we use today to describe and study football consists of on-the-ball event types, that is to say it maps directly to raw data. We speak of “tackles” and “aerial duels” and “big chances” without pausing to consider whether they are the appropriate unit of analysis. I believe that they are not. That is not to say that the events are not real; but they are merely side effects of a complex and fluid process that is football, and in isolation carry little information about its true nature. To focus on them then is to watch the train passing by looking at the sparks it sets off on the rails. The only established abstract concept in football analytics currently is expected goals. For good reasons it has become central to the field, a framework in its own right. But because of it focuses on the end result (goal probability), all variation without impact on xG is ignored. This focus on the value of a football action or pattern rather than on its nature seriously undercuts our understanding of the fundamental principles of the game. Just like isolated on-the-ball events, expected goals tell us next to nothing about the dynamic properties of football.
Indeed it’s the quantitative dynamics of football that remains the biggest so-far unexplored area of the game. We have very little understanding of how the ball and the players cross time and space in the course of a game, and how their trajectories and actions coalesce into team dynamics and, eventually, produce team outputs including goals. This gap in knowledge casts real doubts on the entirety of quantitative player analysis: since we do not know how individual player actions fit in the team dynamics, how can we claim to be rating the players robustly? And before the obvious objection is raised: these dynamic processes remain unexplored not for the lack of tracking data. The event data that is widely available nowadays contains plenty of dynamic information, but as long as our vocabulary forces us to consider every event in isolation, we cannot but glimpse it.
Luckily, a newer concept is emerging into view and taking a central place: the possession chain (possession for short). A possession is a sequence of consecutive on-the-ball events when the ball is under the effective control of a single team. A football game can then be seen as an (ordered) collection of sequences. It is a very positive development since possessions make much more sense as the fundamental building blocks of the game than events. This is because they are inherently dynamic — they span time and space. I believe that they should be studied for their own sake, and if you only compute them to figure out who should get partial credit for the shot at the end of it, then in my opinion, you are doing analytics wrong – or at least not as well as you could be.
To give an example of such a study and why it is important, consider the question: what makes two possessions similar? To a human brain, trained in pattern recognition for millions of years, it is a relatively easy question. It is however a quite difficult, basic research task to devise a formal similarity measure given the disparate nature of the data that makes up a possession (continuous spatial and time coordinates, discrete events, and their ordering). For the sake of argument, assume that we have a measure that we are happy with. It has an immediate, powerful application: we can now measure the similarity of playing styles of teams and players by measuring the similarity of possessions in which they are involved. This method is bound to be much more precise than the current, purely output-based methods, and as we know, playing style similarity has a wealth of applications in tactics and scouting. But that’s not the end of the story. Our hypothetical measure, having already provided a considerable applied benefit, can now be fed back into basic research. Under a few relatively mild additional assumptions, the measure gives a rich structure to the set of all possible possessions, potentially allowing us to deploy a century of research in general topology and metric spaces to make statements about football. But for all these potential rewards, the subject remains unexplored because of the twin obstacles of inadequate conceptual arsenal and perceived lack of immediate applied benefit.
Based on what I sketched above, my suggestion for everyone involved in the field is to be more ambitious, to think more expansively and to not settle for imperfect investigations of lesser questions just because the data seems to be limiting us in this way. It isn’t, at least not all the time. Instead of counting events in more and more sophisticated ways, let us focus on possessions, ask broader questions and interrogate the data in more creative ways than before. I firmly believe that the payoff of this approach will be far greater than anything we have achieved so far.
I want to thank Colin Trainor (@colintrainor), Ted Knutson (@mixedknuts), James Yorke (@jair1970) and especially Thom Lawrence (@deepxg) for their feedback on earlier versions of this article.