No such stat as defensive +/-
Really?
Kanter scored 17 points and grabbed 12 rebounds in just 20 minutes of work Tuesday night. He is averaging 14.4 points and 10.6 rebounds per game, but he ranks 60th in defensive real plus-minus amid the 61 qualified centers.
Yeah, my mistake,
Kanter's 61 out of 61 now.
http://www.espn.com/nba/statistics/rpm/_/page/2/sort/DRPM/position/9
So the thing about RPM is that it realises that regular +/- is highly susceptible to small sample size and all kinds of uncontrollable factors (did an opponent have an unusually strong or weak game that had nothing to do with you and skews your stats and provides little predictable value?). There was just too much variance within years and between years for +/- to be too useful a stat.
So what they've done is tried to reduce standard errors by "regularising". This means subjecting the data to stuff like similarity scores, looking at three years data rather than just year, a player's age, the score of the game, even a player's height (which is why taller players have better DRPMs...the model presumes taller players are better defenders). The model makes assumptions on what parameters should be set as limits for a players contributions and uses them as inputs for the stat.
RPM suffers from two major flaws: 1) there's an awful lot they haven't ever explained (like how much each factor is worth), and 2) like all +/- stats, it cannot completely isolate a player's contribution as opposed to who the player tends to play with (collinearity or multicollinearity issues). It still relies on loads of presumptions (a short, aging point guard who only plays garbage time minutes and has struggled for the past couple of years can only have x impact on a game and his stats should be massaged accordingly).
Most stats guys begrudgingly say it's probably as good a +/- stat as any (begrudgingly, I'm guessing, because it's proprietary and also they stole most of the model and refurbished it). But it ain't gospel.
Poll
