Stats and fielding ability
Stacey Brook notes with approval the arrival of location-specific data for MLB games. Alan Schwarz discussed this in the NY Times about a week ago, stating that “all these pixels and bits will almost certainly revolutionize the analysis of baseball glovework.” That is, Schwarz suggests that knowing the speed and location of every batted ball, placement of the fielders, etc, will enable fielding skill to be more accurately measured. Now, Alan Schwarz certainly knows what he’s talking about, and Stacey may be right that this data will enable economists to better measure “marginal product.”
But, I’m a skeptic on this one. My prediction is that this new data will perhaps be an aid in player development — i.e. to improve positioning in the field, baserunning skill, etc. But I don’t expect it will do much to isolate differences in “baseball glovework,” and what little it does achieve on this score won’t matter very much.
A number of years ago, Jahn Hakes and I decided to enter the quest to measure the missing elements of baseball productivity. We initially focused on fielding ability, and like most others, we failed in our quest. Why? Some people believe that statistical measures of fielding are very poor, and are of little help in distinguishing excellence from competence in this skill. I think that’s right. But there is another factor: as much as baseball aficionados (like me) appreciate the fielding skills displayed by MLB’s best fielders, differences in fielding ability are not a big factor in determining who wins and loses a baseball game.
Bill James’ approach to Win Shares provides a useful benchmark for assessing the value of this new data with regard to the issue of fielding. James asserted that 1/2 the game is offense, 1/2 is defense, and of the defense part, 2/3 is determined by pitching. Now this was merely asserted and not analyzed, but various analyses by others are consistent with the emphasis on pitching as the dominant defensive factor. Taking the win share allocation as given, then all of the effort that goes in to measuring fielding ability can at best capture 16.7% (.5/3) of the variation in game outcomes. And how much of the variation around average fielding ability (the competent ballplayer) cannot be discerned with the naked eye? My hunch is that the scouts will beat the statheads on this one, even with newly improved data.
Despite my skepticism, I recommend a trip to Schwarz’ article, if only to the view video clip which shows a replay with locational data overlaid on the field. It’s pretty cool stuff.