I want to thank the many readers who sent me their comments on PrOPS. In response to these suggestions, I made two changes when I recalculated the metric.
- I controlled for the speed of the players using speed scores.
- I broke the metric out into three predictive components (PrAVE, PrOBP, and PrSLG).
To proxy player speed, I took the average of the five speed scores referenced in Speed Scores and Reaching Base on Errors by Dan Levitt. I decided on using the mean after trying several other combinations of the individual speed scores. The improvement was real yet modest, improving a fit of the regression model just slightly. Breaking PrOPS out into it's components is useful for identifying where players are under-performing.
