A couple of quickies this week:
Judging the Value of Goals
To borrow a canned expression from my old music teacher, my preferred approach to analytics involves the KISS method: Keep It Simple Stupid.
This isn’t to ward off trying something ambitious with complex data sets and counter-intutive key performance indicators and whatnot, but more to encourage those outside the proprietary data wall to just sit and stare at what’s right in front of their noses.
I thought of this when I was directed to 11tegen11′s article on De Zestien this last week on a means to rate a league’s “most valuable scorer.” The article is here, but author Sander IJtsma has been using this approach for a while.
The explanation of his method can be found on his old blog, but it runs on the idea that not all goals are created equal. Obviously a goal scored in extra-time against a superior opponent away is going to be more valuable than the sixth goal in a 6-0 rout against minnows at home. IJtsma however figured out a lovely, simple way of expressing that value, based on match outcome probabilities provided by bookmakers, who adjust their odds mid-match depending on the scoreline:
At any given moment during the match, an expected value can be computed for the amount of points any team wins from the match. Simply multiply the chance of a win by 3 and the chance of a draw by 1. Should a team at any point during the match have a 30% (or 0.3) chance of winning and a 35% (or 0.35) chance of drawing the match, the expected value for the amount of points won from that match would be 3 * 0.3 + 1 * 0.35 = 1.25.
The value of scoring a goal at that point in the match can simply be computed by taking the difference between the expected value for the amount of points won from the match just before the goal and immediately after the goal.
Based on this calculation, IJtsma was able to compile a list of Eredividie’s MVPs for the season (Jozy Altidore ranks 4th). Now some immediate things jump out here. For one, goals are fairly randomly distributed (Poisson curve), so it’s not clear whether players exercise much control over this value, ie when they score, against which type of opponents, and in which particular game state (0, -1, +1, etcetera). It shouldn’t be too difficult though to look into whether certain players have scored high on this list consistently, without quick mean regression.
Then there’s the element of chance creation as well; no player scores in isolation, and the value of a particular player to score in a valuable situation in terms of expected points could be a function of team as much as player.
Still, for what it is, it’s pretty interesting. And, as with all good things in the analytics field, it opens itself up to more scrutiny, which could uncover some pretty cool things eventually.
Last night during the Sunderland Stoke match, ESPN colour guy and former Liverpool player Steven McManaman started banging on about how players needed experience in a relegation battle to better succeed in a relegation battle.
I’m a little tired of this “experience” thing being bandied about with abandon from pundits and fans alike as if it’s a necessity, like skill. For one, it’s hard to hate on Alan Hansen for telling the world “You can’t win anything with kids,” and then in the next breath talk about the importance of “experience” at the end of the season, or in a cup final.
This isn’t to discount the value of experience per se; it’s surely a good thing to have a better idea of what to expect in a high pressure situation, calmed nerves, insight into how to wind down a game or defend a 1-0 lead etc. But this presumes that all players learn from their experiences equally. Moreoever, it also presumes that the value of things like better confidence and the ability to better handle stress in a high pressure situation makes a meaningful difference in a game determined by luck, differences in talent, tactical preparation, etcetera.
This is one of those problems that doesn’t necessarily have a simple solution, either. You could look at a team’s progression in the Champions League based on the number of players with experience in the competition, but then this could just as easily be explained by their innate talent. As in good players tend to have Champions League experience because they’re good players.
There are exceptions of course; Dortmund have a whole slate of players in the big show for the first time. The annoying thing is though is that should Dortmund lose against Bayern, lack of experience may be cited as a “factor.” If they win, it could be “despite their inexperience.” Which is essentially the same thing as saying, “Lack of experience can get a team to the CL final, just not allow them to win it.” This isn’t to say experience can’t influence how a team performs, but it’s very difficult to isolate from other factors. In fact, I think it’s pointless to make the attempt.
That’s because “experience” as a means to explain a particular result is a self-fulfilling theory that cannot be verified. That doesn’t mean experience isn’t important, but it does mean that you can’t say beyond a doubt it was the determining factor in the outcome of a match. But this is exactly what McMananam is inferring when he says that a team “needs” experience in a relegation battle to perform better in that situation. Unless you can definitively show that when two teams of equal talent play each other the more experienced team will win at a statistically meaningful rate, you’re not saying anything at all.