Just so no one misses this who followed the brilliant comments thread in our earlier discussion of faulty statistics in football: We heard today from Bill Gerrard, the professor from Leeds University who’s working with Billy Beane to devise new statistical metrics for football. The background: I’d wondered, not having heard about it for a while, whether the work was continuing. And there was a general feeling among commenters that its proprietary nature was disappointing, because it reserves advances in understanding the game for the teams that are paying for the research rather than making them available to fans.
Here’s what Gerrard says:
A belated follow-up to your discussion of applying statistics to football. Unlike baseball and other striking-and-fielding games in which human observation and the paper-and-pencil method can go a long way to recording player activity, football and other invasion team sports need a certain level of technology to gather even a minimum amount of player action data. And if you want data on activity off the ball you need a tracking system. This is expensive. Combine the cost with the obvious concerns of teams over maintaining confidentiality and not surprisingly little player performance data is in the public domain. At least not in a form that allows much in the way of systematic analysis. So most of my own work over the last five years has not been published since I’ve had to sign confidentiality clauses with teams to get access to their data. (My 2007 article uses some historic Opta data that was published in a series of four yearbooks.)
My collaboration with Billy Beane continues but the research remains highly developmental and is not being used in the day-to-day operations of the A’s soccer team, the San Jose Earthquakes.
It’s fascinating to hear about this from the perspective of someone who’s working on it up close. It still seems too bad to me that so much of Gerrard’s work has been left unpublished—I understand that confidentiality is in the sponsoring teams’ self-interest, I just don’t care enough about what’s best for them not to mind. But the point about expensive technology being required to gather player action data is really interesting to me, and I wonder if anyone (Prof. Gerrard?) can shed some light on the kind of technology involved.
In any case, it’s good to know the work is still progressing, even if it’s progressing in a way that will keep it out of the sight of fans for the foreseeable future. In some ways it’s a shame that we can’t get some big media company to sponsor it—they could afford the technological investment but would then be motivated to share the results of it.
The consolation, I think, is that the complexity that makes football hard to understand and “measure” is also what creates the conditions that allow us to experience its specific kind of beauty. Without having statistics in mind at all, I wrote about this in September:
So what does football “mean”? One way to answer the question is to note that the relative complexity of the game—a large number of players moving simultaneously on a large field with very few stoppages of play—leads to a high degree of confusion. Compared to other popular sports, there are many forces interacting simultaneously and their interactions are remarkably sustained. We might think of this as imposing a high degree of resistance, both to understanding (in comparison to, say, tennis, soccer is hard for a novice fan to follow) and to the emergence of clear intention (what players mean to happen is seldom entirely clear, and even more seldom what actually happens). Another way to describe this would be to say that where many other sports attempt to create an arena in which the randomness, flux, and contingency of life—the forces that work against our own everyday intentions—are largely excluded, football conspicuously lets them in, in such a way that when a clear intention does emerge, when a player scores an improbable goal or completes a visionary pass, it often carries with it the special exhilaration of seeing an affirmation of the potential of the human will.
Obviously there’s more to the aesthetic aspect of the game than equations sprawling into incomprehensible suburbs. But it’s useful to be reminded that the various ways in which we try to understand the game are often surprisingly linked, and linked as well to the various ways it moves us.
Read More: Football as Philosophy, Statistics
by Brian Phillips · January 27, 2009
[contact-form 5 'Email form']
Brian good article (and site as well, being a recent visitor):
“. But it’s useful to be reminded that the various ways in which we try to understand the game are often surprisingly linked…”
Well, yeah. Team sports, especially fluid team sports like football (and basketball, hockey, etc), are pretty hard to generate meaningful statistics at all much less glean anything worthwhile from them. I.E. we know who scored the goal, but how do we (as box-scorers) place a numerical value on the tackle (or strip, or check) that originally gained possession to allow the goal to be scored in the first place?
So, technology is needed to some extent surely just on the basis that a metric-ton of data can be collected. And I think it will take a bit of time to get really meaningful ‘MoneyBall’ type of players out of whatever analysis Mr. Beane and Mr. Gerrard get on the presumption that you’re going to have to:
-One: find ‘undervalued’ players
-Two: have the stats to back that ‘undervalued’ label up.
I guess that’s were we’ll find out where Opta or whoever will be going next for some data ; )
Football and the other complex invasion sports (what you call “fluid” sports) can only be partially understood by looking at the numbers. In part because the numbers tell you mostly about the quantity of player actions and often little about the quality of these actions in terms of their tactical significance. An incomplete pass that gives second ball possession could well be described as a successful pass in tactical terms but would show up in most tracking/video analysis systems as incomplete/unsuccessful. Quality necessarily needs to be judged from a tactical perspective and hence is always going to be subjective and team-dependent. That’s why I dispute the notion that there is some Holy Grail of the ultimate performance measure or set of formulae out there in football waiting to be discovered. Relevant micro performance measures (i.e. those below the level of match outcomes and numbers of shots and goals) will vary between teams depending on their style of play. And of course so much of what a player contributes in football is his/her movement off the ball which is difficult to convert into stats beyond direction, distance and speed. And off the ball movement is not captured by player action stats.
I should say that I’m a qualified UEFA coach so I do try to bring a combination of statistical and coaching knowledge to my analysis. Ultimately I am trying to find ways to extract useful information from the performance data to help enhance evidence-based management in football. As I said the numbers will never tell the whole story of the beautiful game but they can tell part of the story if interpreted properly and hence can add real value to coaching decisions. At the very least the numbers can help ask better questions and offset the inevitable biases of human observation.
“And off the ball movement is not captured by player action stats.”
Really? Although thinking about it for a second, I’m not finding a meaningful way to quantify it that’s relevant to football.
Appreciate your presence on here Mr. Gerrard, and hope to see your work published someday.