Pinto’s Probabilistic Model of Range

David Pinto of Baseball Musings has started releasing numbers on his Probabilistic Model of Range, which is a stat to evaluate defense.

2 Responses “Pinto’s Probabilistic Model of Range”

  1. Anonymous says:

    Nice to see someone finally putting together good defensive statistics. I really like the design: actual minus expected outs. Only thing that makes me skeptical of the model is the fact that summed actual outs is different from summed expeted outs. Any exponential model should yield an overall difference of zero, unless there’s a regularizer/weight penalty; I can’t see a reason for a non-exponential model nor a regularizer…

  2. Anonymous says:

    Turns out there was a very good reason for the difference: model was trained on 2002-2004 data, evaluated on 2004 data. Doh!

    Anyway, wouldn’t it be great to do something like this for hitting and pitching too? We all know that there are park, league and divisional factors that make the numbers biased. Pinto’s method does an excellent job of eliminating a large part of that bias. Imagine a batting average regression with variables such as pitcher faced, ballpark and umpire!