THE BOOK--Playing The Percentages In Baseball

Thursday, June 26, 2008

Real change in talent levels

Dave describes the changes in fastball speed for Verlander.  Yesterday, Josh shows us the standard deviation of a pitcher’s fastball speed:

I thought the average would have been 1.0, but it is actually 1.35.  That’s on a pitch-by-pitch basis.  On a start-by-start basis, the spread will likely be less than 0.5 standard deviations.  If Verlander is really throwing 3 MPH different in two separate months, this is an enormous piece of information to have.  Basically, we’ve got two Verlanders.

The question to ask: how transient is the lower fastball speed?  Is it like a weather pattern, like a tornado, that would basically be random, and therefore, we don’t expect for the tornado to reappear at the same spot.  Or say like lightning hitting the same spot twice.  Or, could this possible recur, and therefore, we need to give say 90% weight to Verlander-we-know and 10% for the Verlander-who-sucks. 

In similar spirit, but not as dramatic, is the drop in fastball speed of Barry Zito between 2005 and 2007: 2.8 mph in two years.  That’s fairly sizeable.  So, when you see his FIP go from 4.34 to 4.82, that is more “real” than someone else who had the same drop in FIP, but who did not have a change in fastball speeds.

Remember, all performance data really is a manifestation of the context and the talent level of the player.  We are inferring the talent level of the player after understanding the environment in which he plays.  We are presuming that the change in talent level of any given player follows the same pattern as any other player, if we don’t know any better.

But, we now know better.  We know that something drastic changed in Verlander and Zito.  We don’t know why it changed.  We don’t know how persistent or transient it is.  We don’t know if it’s a shift like global warming, or a blip like a bolt of lightning.  All of these numbers we have, the tools numbers, like fastball speed, and curveball movements, are themselves just a manifestation of the core to the player: how strong, smart, and coordinated he is.  That is, we infer from the fastball speed and curve movements certain things about the player. 

It all comes down to tools.  If we were god-like, we wouldn’t need to know about any performance results.  That’s what scouting is all about: how god-like can we be in establishing a person’s core true talent, and how much do we need to infer based on the performance results, either at the toolsy-level (fastball speed, bat speed) or at the results-level (BB/PA, BABIP)?

The inference at the results-level gets us most of the way there.  But, in some cases, like Verlander and Zito, it may fail us.  So, we need to get down to the toolsy-level.  And in even smaller cases, we need to get down to the core level (like Ankiel).