E-Balla wrote:dhsilv2 wrote:E-Balla wrote:There's a difference between actually calculating it and knowing how it's calculated but nice try. If I wanted to take the time to gather lineup data and create a python program to run a ridge regression on the data I could. I know what the strengths and weaknesses of ridge regression are. I know what data is used in calculating RAPM. I know none of that about RPM. I only know what Englemann claims and no statistician will tell you their stat is flawed.
The only pissing piece in RPM is the prior and outside of plain vanilla RAPM, most of the RAPM posted on the internet doesn't come with a complete guide for exactly how the person building it setup the prior.
True but most include the variables they put in the prior. Things like age, previous RAPM results, etc. They might be used in different ways and given different weights but using the boxscore for a prior then adds in the question of how good his boxscore prior is on it's own. Let's say someone came up with a Wins Produced prior version of RAPM. WP has good predictive power but is it good on the individual level? Not really. And I'm personally guessing his boxscore prior is just an updated version of xRAPM (sidebar but iirc xRAPM had better predictive results than RPM and anything else and iirc it's all boxscore) but I don't know at all and that's the problem. Plus like I said it's not even meant to determine past performance but is instead meant for future predictions.
The tools get measured for their predictive power as that's the best "test" for these. The purpose however in these is to show the value of a player in the role they are in. When we have players who have similar roles (high usage ball dominate guys for example) then there is some fairness is saying one does that role better than another. Even still it isn't that simple, and box score metrics aren't that simple. Wade was made better by having shaq, even old shaq, on his team. Just like a guy like Porter makes others on his current team better with his range and ability to take tougher defensive assignments.
That said we know enough and we have enough years of data to have a pretty good feeling on what RPM is and is not telling us. Anyone who has even a light understanding of what's going on with that stat should be able to understand the information and use it. Claiming otherwise is just silly, we can't all be experts on every single mathematical solution give to us. We can if we're well informed make reasonable judgments as to the value of what metric.
As for RAPM, given the scale doesn't even seem consistent, I think there's a lot of blackholes. I've heard of people using 0 priors, or setting rookies to -1 as a prior. I'm not sure if the prior is the prior year's vanilla RAPM or if they used the same RAPM I'm seeing as the prior and the chains back to their rookie year where again they used some plug for that prior. I haven't spent much time with it, but anecdotally many people have put our RAPM calculations and for the same years with the same data some players seem to move a lot more than others.
