Hardballtimes: Forecasting Pitcher Platoon Splits.

[weams]

http://www.hardballtimes.com/forecasting-pitcher-platoon-splits/

This model as our "prior" when projecting pitcher platoon splits, regressing each pitcher's observed platoon splits toward the model's predictions instead of toward an average split. A mixed effects model suggests that we should regress a right-handed pitcher's observed platoon splits by adding in roughly 1,670 harmonic plate appearances, or about 800 innings pitched of his prior. For left-handed pitchers, for whom our prior carries less information, we use only 570 harmonic plate appearances, which a left-handed starter might reach in 380 innings and a left-handed reliever (who likely faces more lefties) might reach in only 280 innings. Using observed career level splits and these regression levels we can estimate career-level wOBA splits. The distributions of priors based on the ridge regression model as well as distributions of estimated career-level splits for left-handed and right-handed pitchers are shown below.

[Frobby]

Geeky.

[weams]

Geeky.

[DrungoHazewood]

http://www.hardballtimes.com/forecasting-pitcher-platoon-splits/

Thank you for both of these posts. I've taken two major things away from them: regression is your friend when projecting, and you must regress to the right thing (in this case arm-angle adjusted baselines). And Tina Fey wow.