In a few previous posts, I looked at the true talent distribution, focusing on looking at the tails and seeing if I could find any evidence that they were “fat”. In doing so I developed some code to find the maximum-likelihood values for the mean and standard deviation of a true talent distribution, assuming the distribution is Gaussian. Read more here.

I have applied this method to platoon splits, in two different ways. Specifically, I looked at the wOBA true-talent distribution for batters, for the 4 possible cases; lefty-faces-lefty, lefty-faces-righty, rightly-faces-left, and righty-faces-righty. For this analysis, I set a limit of at least 85 PA, which was chosen more or less arbitrarily.

Here are the results,

I then used the same methodology to look at platoon splits for individuals, i.e., instead of treating a righty-faces-lefty as independent from when the same righty faces a righty, I took the difference for each batter from when they have the advantage compared to when they don’t. For this, I used the geometric mean of the RL/RR and LR/LL encounters as the effective number of PAs, and set a limit of at least 135 PA.

Here are the results,

This says take the observed difference in wOBA, and regress to the mean by somewhere between 0.25/0.0420**2 ~ 142 PA and 0.25/0.0389**2 ~ 165 PA, which as far as I understand is in stark contradiction to previous research, for example as described here,

http://www.fangraphs.com/blogs/estimating-hitter-platoon-skill/

I don’t currently have an explanation for the discrepancy.

### Like this:

Like Loading...

*Related*