THE BOOK--Playing The Percentages In Baseball

I’ve been calculating lacrosse stats for the last few months and been doing so on a possession basis by a formula I found here: http://orange44.blogspot.com/2006/01/of-lacrosse-and-efficiency.html

So, possession-based stats are definitely a viable metric.

I haven’t run any year-to-year studies or anything of that nature; I’ve only got data from 2009 through 2011. But, offensive and defensive efficienies (goals/possession, goals allowed/possession) yield r-squared’s of .594 and .428 for the entire sample.

The rest of the component data in my set and his seem to yield similar results. If we’re looking at certain components, GB% and Possession% (offensive possessions/total possessions) rate the highest at .476 and .448 r-squared’s.

I’m not surprised by the shooting percentage against being unstable; I’m guessing that it’s much like NHL goaltending. The best goalies make an impact, but by and large, the goalie itself doesn’t matter so much as the team in front of him does.

Also, shooting rates (goals per shot attempt, goals allowed per shot attempt) should be regressed. Some teams fire off so few shots (St. Joseph’s in 2011) that they should be regressed 45% towards the mean. Overall, the 50% regression point is 228 shot attempts and the average team only takes 511. Even the teams that take the most shots still are displaying only around 75% observed talent.

Goals allowed per shot attempt is right around the same; 50% regression point at around 215 attempts.

I use these to generate regressed shooting percentages and offensive/defensive efficienies based on this.

Thanks for linking to this, really interesting.

I’ll note that those r-squared numbers are how they relate to raw winning%, by the way.