Originally Posted by
Perry Mason
Right. Adjusted on/off numbers are meaningless for injured players working their way back. I also expect most posters relying on these ridiculous Bayesian concoctions do not understand the myriad shortcomings with these numbers. I expect TP to have terrible 2015 season numbers thus far. He came back too early from injury and played half-speed in many others.
If I had the time, which I do not, I could provide horrible RAPM numbers for great players during a period in which they struggled with injuries. I could also provide RAPM's that are very high for players most would recognize as middling or only decent. For example, RAPM can overrate glue guys that put a good lineup over the top. This is because no matter what RAPM will always be calculated using data for a player in a lineup and against specific matchups. It is inescapable and "correcting" for it only happens if there was mixing and matching of a player in different lineups to improve the data (which does not apply for most teams, Spurs excepted more than anything), and then by extrapolating.
Globally, RAPM is somewhat meaningful, but on the same global basis, it isn't that much different in insight (and will have a similar number of misfires) from using raw stats + human judgment, which may over-rate inefficient players and poor defenders, or PER which overrates offense.
RAPM is often called the most accurate predictive single stat. But this is damning praise given the absence of any really comparable stat.