Why intuition need not be a dirty word in football analytics
The incomparable Mike Goodman wrote an excellent summary of the state of analytics *ahem* over at Grantland last week. In giving us a snapshot of the current advances in football analytics, Goodman makes the crucial distinction between data and analytics. As a rule, the former tells you what happened, while the latter helps tell you what, to varying degrees, will likely happen.
Goodman is right that football analytics is in its infancy. It will probably take more public involvement of an ever-growing array of eager analysts to experiment to try and add to the predictive metrics we currently have at our disposal, and others to further help us understand how team staff might use them to run a better football club.
However there was one paragraph which stuck out, on James Grayson’s statistic Total Shots Ratio (which Grayson himself has since modified into a more totalistic and more predictive “team rating”). Goodman writes:
TSR is incredibly helpful when it comes to figuring out what teams might do well in any given season (though, of course, some dumb writers will ignore it and pick Manchester United to finish third), but it isn’t all that helpful when it comes to actually managing a team.
What Goodman means of course is that TSR is a team metric; it doesn’t tell a manager which individual players are duds and which are winners and the kind of footballers the club will need to recruit and what kind of tactics they’ll need to employ, etc. etc.
But I disagree that basic predictive metrics like TSR aren’t “all that helpful to actually managing a team.” Here’s a very simple but illustrative example which I cited in the debut Trends in Analytics column.
Let’s say a Premier League team is rocketing up the table with one draw and three wins to start the season but are posting a TSR of .458. On its own, this information isn’t very helpful.But perhaps we might compare this low TSR to the team’s PDO, a number representing their shot percentage plus their save percentage. Because this metric tends to regress to the mean very quickly, a high or low PDO can be instructive in revealing the influence of good or bad luck. Well this particular team is rocking a league high PDO of 1489. We can break that number down into a shot percentage of 60% and a save percentage of 88.9%. These are, on their own, extremely unsustainable numbers, particularly the shooting percentage. So with a low TSR and a high PDO, the simple conclusion to draw is that the team is riding their luck.
If you’re a betting person, you can make quick use of this information. It practically screams, “Continue to bet against this team regardless of recent results!” But if you’re a performance analyst at a club, what do you do with this exactly?
On the one hand, you might concede the information is incomplete. But it’s still something.
For example, if I’m a performance analyst, it would be valuable to know this and other data before I go into the video room to take apart last week’s fixture. If I know that a low TSR is generally going to mean a lower points total as the season progresses, I might look for clues as why my team is, as a rule, conceding more shots than they’re taking. With such a high PDO, I would want to know the tactics my team is relying on to win, whether there is a method in this madness of high save and conversion rates. I would want to know how long my team spends protecting a lead, whether they try to exploit Game States–the shift in behaviour teams tend to exhibit when leading or losing by an x margin of goals, and if they clearly value taking shots with a higher Expected Goal value or are conceding shots with high ExpG value. And I would also as a rule want to know my team’s Expected Goals ratio, too.
Figuring this out is a process that involves both analytics and intuition based on tactical understanding, experience, viewings and reviewings of old football matches. ‘Intuition’ of course need not be a dirty word. The problem I think with boosters and detractors of sports analytics is they expect predictive metrics to do all or nothing. So if TSR can’t tell you on its own which formation to use or which starting XI to put out on a given Saturday, it’s “useless.” Yet simply knowing there is a fundamental problem within a club despite recent positive results can mean the difference between a manager telling a team to “keep up the good work,” and a manager carefully working with their staff and players to find constructive, tactically-grounded ways to help improve overall performance.
Football analytics, even in their imperfect, infantile stage, can still provide clubs a vital tool in the impossible task of improving each and every week. We shouldn’t put more on the numbers than they can bear on their own, but clubs should embrace new ways of achieving competitive edge. It’s not at all clear they are. Yet.