Last time out I explained, in detail, how I calculate Usage Adjusted Rating (a usage adjusted version of Alternate Win Score). I liked the results, but I thought that they could be better. In order to try to better value defense, I decided to try to include a weight to factor in minutes per game played. I made this decision based on the idea that coaches, generally, won’t play someone a lot of minutes if he’s got shaky counting stats- which basic UAR covers- unless he’s providing other value. So I added a factor that gives a slight boost to players who play 20 minutes or more per game and gives a slight negative to players who play under 20 minutes a game.
In addition, I took the UAR with the minutes per game adjustment (70%) and blended it with non-adjusted +/- per pace adjusted 48 minutes (20)% and added a zero-weight to regress it to the mean (10%), as this was the blend that best correlated with xRAPM. Then I made the metric 100 possessions, instead of per 48 minutes pace adjusted. I call this new metric UARPM100, which is a bit of a mouthful, but it conveys the information contained within the metric, so I’m sticking with it.
I ran a correlation of UARPM100 over past years against xRAPM from Jeremias Englemann at stats-for-the-nba.appspot.com. The r-squared for UARPM100 against xRAPM was roughly .67. The r-squared becomes much, much stronger if prior year xRAPM is blended with UARPM100. The r-squared for blended prior year xRAPM and UARPM100 is roughly .82 with in-year xRAPM, which is obviously very strong. Given my belief that xRAPM is probably the best one-number metric in the public domain, I feel pretty good about UARPM100’s results. Here are the results for UARPM100 through the December 16, 2013 games (minimum 120 minutes played):
Going forward, I will be updating UARPM100 as close to daily as possible. Periodically, I will also post UARPM100 that’s prior informed by 2012–13 xRAPM. Hope you enjoy!