THE BOOK--Playing The Percentages In Baseball

Friday, August 07, 2009

Regression of Groundball Rates

Since the number of groundballs a pitcher gives up is one of the most stable performance rates around, when you see a huge change in that metric, this points very strong to a true change in talent level.  Dave Allen discusses Joel Pineiro.

This is how you think about it.  If what you observe has little random variation to it, you do two things:
1. You weight the more recent numbers alot more than the older numbers
2. You regress alot less.

So, for something like baBIP (batting average on balls in park), you might weight the current season at “1”, last year at “.9”, and the year before at “.8”.  But, for gbCON (ground balls per contacted ball), you might have the weights as “1”, “.5”, “.25”.  (Just making numbers up.)

Furthermore, while for baBIP, you might add 1000 BIP of league average performance, for gbCON, you might add just 100 CON of league average performance. (Just making numbers up.)

The question you are always asking yourself is: how much of what I see is real, and how much of what I see is a change from the past.  The answer is ALWAYS greater than 0% and less than 100%.  Our job is to figure out the numbers in-between.