Wednesday, February 04, 2009

BABIP IV: The Bad Sequel That Goes Straight to Cable

I'd like to close the book on this topic, because I think I'm boring everyone except Brett, Jason, and Eugene to tears.

Hardball Times' theory is useful, and the research they've done is tremendous. But a careful reading of their work on the subject reveals a mere "59% correlation on BABIP for the players in their sample." Another way of putting this is that slightly more than two out of five hitters in their sample don't correlate with their hypothesis.

As they point out, their work is a vast improvement over previous work on the subject, which produced a mere 18% correlation. However, you cannot state with any kind of authority that "A will happen because of B" based on a 59% correlation.

A scientist realizing a 59% correlation and a 41% improvement over previous work would rightfully be praised. But that scientist would also admit that there was a lot of work to be done to improve the model.

What makes me nervous about these kind of studies isn't the studies themselves, but the absolute authority with which everyone speaks after they read these studies. The vast majority of comments on this subject haven't used statements like "in all probability" or "more likely than not", but have spoken in absolute terms on a matter that - by the admission of the study's authors - is far from the absolute and final say on the matter.

Getting back to Rotisserie, some of the worst players I've competed against have also been some of the smartest: players who know all of the ins and outs of the inner numbers of the game but can't cope with the fact that some players don't hew to the statistical models. This happens with even the most basic of numbers like OBP. These owners will put far too low a bid limit on a player with a bad OBP or poor batting eye and - as a result - savvier owners will swoop in and buy those players.

Let's continue to read these articles, discuss them at great length, and admire the excellent work the authors at Hardball Times, Baseball Prospectus, and other web sites of their ilk have compiled. But let's not pretend that their conclusions are absolute and final when the authors themselves do not even make this claim.


Toz said...

Mike and I spent quite a bit of time the other day discussing this issue.

Ultimately, I look at this from a predictive value standpoint. If I were a scientist, 58-59% success rate is not considered predictive value. In scientific terms, it is the failure of the hypothesis. There can be any number of reasons for this...incomplete formulaic data, wrong categorical data, improper control group, etc. No matter how you slice it, however, it fails.

Also keep in mind that the development of the theory is based upon a perceived flaw...working backwards from the perceived conclusion usually results in a self-proving theorum, in which the 41% non-predictive results are just dismissed as anomolies.

There is value to the analysis, but query what it adds to the fantasy baseball analysis. We know intuitively that red flags pop up with a drastic drop in average combined with a spike upward in strikeouts and/or a spike downward in walks. We know that an improved K/BB in light of a dropping average is not as big a red flag and could be a sign of potential upside value. The point is that xBABIP does not add to the analysis.

Formulas for sports are flawed because they can never account for certain factors: concentration; psyche; impending free agency; home life; vision problems; injuries; and luck. Too many variables in my opinion.

Eugene Freedman said...

I made a comment about Alex's site under Mauer about how Rotisserie's worst aspect is that it reinforces mistaken beliefs about what the most important stats are.

The inverse is true to. People who know the most about baseball stats may draft the best team for real baseball, but a bad team for rotisserie because it's really a game of categories not a game of baseball management.

Joe Mauer is a much better baseball player than he is a rotisserie player. The inverse is true about guys like Scott Podsednik, Juan Pierre, etc., who are specialists, but are actually bad players when the total game is added up. Yet, they appear valuable because of the reliance of the specific markers of rotisserie.