THE BOOK--Playing The Percentages In Baseball

Wednesday, November 05, 2008

UZR and the 2008 Gold Glove awards

The NL Gold Glove awards came out today.  Before we even see the winners, we can expect that the “analysts” will be screaming bloody murder at some of them, right?

Well, without any screaming, I give you each of their UZR for 08 (and their UZR the previous 3 years).  The reason the prior 3 years’ UZR is relevant to this year’s awards is the following:

UZR does NOT necessarily tell you how “well” a player played this year.  It only tells you how well the “data” thought that player played given the very imperfect nature and limitations of the data.  For example, the data and resultant UZR might “think” that a particular ball or balls was hard to field based on the written record of where it was hit and how hard the “stringer” thought it was hit.  In reality, the ball may have been easy to field for a number of reasons. And vice versa of course. And easy example of the latter would be a player playing in an atypical position and a slowly hit ball takes a tricky bounce.  UZR thinks it is an easy ball to field (given its speed and the usual position of the fielder) when in fact, it is difficult if not impossible to field.

What does that have to do with prior years?  Let’s say that a player is +10 in UZR, suggesting a very good fielder.  As I said, that could mean that he truly made lots of difficult plays or it could mean that the UZR data was “wrong.” Obviously the former is more likely, on the average. However, the better he was in the past, the more likely it was that UZR was “right” this year.  That’s it.

There is one other thing to keep in mind when comparing subjective evaluations (like the GG’s) to objective ones.  And I am not talking about the biases and inaccuracies (and irrationality) that go into these awards, which are plenty of course (all they essentially do is ask the managers, ”Hey, do you remember who was really good on defense this year?” Literally, their answers could be based on one play, or it could be based on what they read in the papers, or it could be based on what they saw last year, or it could be based on how many errors the player had this year – in fact, most winners have few errors, etc.).  UZR and the other objective metrics are context neutral – they don’t care who was on base (other than how that affects the baseline percentages), how many runs were actually saved or not on a particular play made or not made, the score, etc.  I would think that if a manager sees a player make a great play with no one on base in a 10-run blowout and then he sees a player make the same great play with the bases loaded to save the game, that he gives more weight to the latter, either consciously or subconsciously.

Bottom line when comparing, say UZR (or Dewan or ZR or whatever), to the real awards, just because UZR may “say” something different, does not mean that the award was “wrong” (although it does suggest that it is).

Anyway, here we go:

Read More