Yesterday I spent a chunk of the afternoon re-acquainting myself with Bill Gerrard’s 2007 Leeds University paper “Is the Moneyball Approach Transferable to Complex Invasion Team Sports?”. Gerrard identifies three primary ‘measurement problems’ in any detailed single player analysis in “complex invasion team sports” like football:
“The tracking problem, the attribution problem and the weighting problem. The tracking problem refers to the identification, categorization and enumeration of different types of player actions in and out of possession and including spatial coordinates of where the actions have taken place on the field of play. The attribution problem is the problem of how to allocate the individual contributions to joint and interdependent actions. The weighting problem is concerned with determining the significance of each individual action to the overall match outcome.”
Addressing all three is a tall order. For some, including many critics of the use of analytics in football, it is a practical, if not literal, impossibility. Football’s inherent complexity means analytics will never yield anything of value in the game. Even if you were able to account for the individual player’s contribution to the team as a whole, would it even tell us anything we don’t already know through sound tactics and good scouting?
Let’s cut out some of the noise and simplify what Dr. Gerrard is specifically after here: a statistical means to isolate the value of a single player to any given team that maintains predictive accuracy over as long a period as possible. In theory, a team’s performance analyst, having discovered this magical metric, will for a time enjoy a market advantage over their big-spending rivals who pay out big for proven stars, that is until everyone else figures it out.
I think however that Gerrard’s analysis of the level of knowledge, both empirical and statistical, required to make definitive statements about a footballer’s ‘absolute’ value—an analysis which I believe broadly reflects a lot of conventional thinking on this subject—suffers from a slight problem of overdetermination.
Overdetermination is when an effect has more than one definitive cause at once, more than needed to draw a reasonable conclusion. To graphically explain what I mean in a footballing context, let’s look at Gerrard’s take on the so-called “tracking problem.”
Gerrard is certainly right that technology has yielded basic single match data that unheard of even a decade ago. Successful final third passes, interceptions and where they occur, and yards-run per footballer can be found fairly easily for anyone interested in them, often in real-time. These data are interesting and can often underline our tactical impressions of a single match; however, at the statistical level their worth is more uncertain, particularly if these isolated match events are randomly distributed and quickly regress to the mean.
The broad point after all of statistics in sports is to isolate exactly what determines one team’s ability to win more games than another team, and if this team has some sort of control over this affect, i.e. if it isn’t just luck or ephemeral skill.
We know with some certainty at the team level for example that a side that consistently takes more shots than they concede will finish higher up on the league table over a full-season, a statistical correlation that you’ll note does not require high-definition cameras and computer-tracking technology to determine. Just as you can broadly predict future performance based on nothing more complex than shots ratios, you may not need insanely precise measurements to discover that exceptional players do certain things more on average than regular players.
The TSR stat though poses a host of questions, among which the most intriguing is how does an individual player or manager affect this ratio? In my experience it’s best (and most fun) to go from the obvious and whittle down to the unknown particulars. So let’s try it.
Why does a team manage to take more shots than it concedes over an entire season? Perhaps it’s because the club spends more money on proven players than their opponents. What then makes these players “proven”? Is it because, if they’re a striker, they’ve scored a lot of goals? Perhaps, but then many strikers have great seasons at one club and then tank at another. Is it because if they’re a defender, they make a lot of tackles? Well, as Sir Alex Ferguson famously discovered with Japp Stam, ‘tackling’ isn’t really useful as an isolated metric in determining the skill of a defender. Often it’s reflected in their positional sense, how they cut off space for their opponents and create it for their team-mates.
Moreover, are the inherent, absolute talents of individual players enhanced by a particular set of talents in their team-mates? Or are they enhanced by a particular tactical approach or managerial style? Do exceptionally talented players all tend to do the same things in the same areas of the pitch?
These are interesting questions, but I think it’s important to stress they may not all require answers before we can begin to investigate the relationship between the skill of an individual player and the skill of a team. Some scouts are already moving ahead with their own bag of tricks. A couple of weeks ago, Adam Bate spoke with Hamburg’s new chief technical scout, Steven Housten. Housten explained part of his methodology in identifying prospective players:
“It’s no longer a case of saying a player has scored X amount of goals or a midfielder has created X amount of assists,” he adds. “You only have to look at something simple like a goal. There are so many types of goals – the difficulty of the goal, the quality of the goal. And with passes there are passes and then passes in the final third. We are now able to break down into it. The hardest thing is to work out what is important and what isn’t important – at a team level but also for individual players.
“For example, I think if you just looked at the players with the highest pass completion you would just be getting defensive midfielders like (John Obi) Mikel (at Chelsea) who tend to make shorter and less incisive passes. Passes in the final third are much more difficult to make and through-balls are passes that create higher quality chances for forwards rather than, say, a cross or something like that.
Housten’s methods are likely far from definitive, but for the purposes of finding interesting ways to isolate individual talent from team ability, they can at least provide some interesting clues. A hypothetical study could for example look to a team with a high TSR, and then to their average starting XI. From there, they could look at their playing histories prior to joining the high TSR club, and, based on their position, look to some basic in-game measurements—passes attempted/completed, interceptions, successful final third passes—as a percentage of the team total over one or more seasons. Perhaps these percentages fluctuate wildly from game to game and hold no predictive weight; then you can look to other, perhaps more precise measurements to see if they hold any predictive power. Or maybe they do.
From there one could (admittedly arbitrarily) introduce a weight based on UEFA co-effecient and average league position, and you kind of have an imprecise means to make judgments on individual player ability.
Maybe all of this has been tried before and failed (if so, would love a link or two!), but it seems to me that good teams are good in part because they have individually good players, and that many (if perhaps not all) good players come with an ability that can be isolated from their team and their manager. Inherent, individual talent, as complicatedly intertwined with the talent of their team and their manager and the talent of their opponents as it may be, will not necessarily be discovered with the help of better technology or more precise measurements (I doubt that a tendency to make left-footed long passes in the centre circle is an indication of individual brilliance, for example). It might be, but analytics would do well to first exhaust all the existing avenues available, no matter how overtly ‘simple’ they may appear.