Thoughts on the Actim Index

Having statistics is nice. I like to know how many passes a player has, or how many tackles they made, how often they scored, etc. However, simply having that information isn't the end of the story. Thanks to super-reader RocketsAstros, I came across a link showing that Florent Malouda was the #2 player in the Premier League this season (he has since risen to #1 in the six hours since I first came across those rankings despite not playing a game). Apparently, these rankings are numerically divined based on the numerous factors, including goals, shots, assists, fouls, tackles, offsides, etc.

Great! We have an objective player grading system. Or do we?

As a quick aside, I've worked (fairly successfully) on statistical analysis in sports for several years, mostly in a much easier sport than soccer to numericise - baseball. I've been fortunate enough to have been able to contribute in a very minor way to the field, but more than anything else I've picked up a sense of how things ought to be done in analysis of any given sport, and the sort of mistakes that get made along the way. Obviously, the individual aspects of baseball can't help us with football analysis, but the big-picture ideas certainly can. Here's the biggie, as far as the Actim Index goes.

Statistics in anything cannot be taken as raw numbers alone. Yes, we know that more complete passes, more goals, more tackles are presumably better than less, across the board. However, we have no idea - absolutely none - of how much value a complete pass provides compared to a goal. Is a pass 1/100th of a goal? More? Less? You tell me. It gets even more difficult with defending, when you have to assess how likely the other team scoring was before the defender mopped up the situation.

These questions, right now, are impossible to answer in any meaningful way, at least according to my understanding of how much data is on hand. We may be able to use regression analysis to come up with approximations, but if working with baseball statistics has shown me anything it's that using a regression analysis - a data driven approach - is a recipe for making ludicrous errors in building your statistics. The biggest giveaway here is that the ranking doesn't actually scale to anything worth measuring in a soccer match. It's just a number!

We don't want to know how many Actim points a player's been worth - we want to know how many actual points a player's been worth. We need a Rosetta Stone of soccer statistics, because right now we're at the stage where we can count numbers but cannot translate them into a meaningful language. That Rosetta Stone is the game state, something that will be fiendishly difficult to derive (I've touched on it during earlier sketches about my thoughts on soccer statistics), but to me, well worth the effort. Once we get that, we'll have statistics that are more than simply numerical constructs.

Florent Malouda probably hasn't been the best player in England this year, but that doesn't mean that the entire field of statistics is non-valid in some way. As far as I can tell, this is simply a matter of over-exuberance from the statistics guys, the sort of thing that in ten years time we'll look back on and laugh about. We're not nearly done yet!