Continuing from prior thread:
AEnigma wrote: That is not a comparative exercise; you are just throwing out rankings as if “all-defence” should be blindly taking the top ten in DRAPM and calling it a day.
And of course you when you take the approach that everything from 2009-13 should be sequestered, you are then primarily taking five-year splits which tend to sunk by one particularly bad year. So NBARAPM has him 21st in 2006/07 DRAPM, 17th in 2007/08 DRAPM, 9th in 2015/16 DRAPM, 2nd in 2019/20 DRAPM, and 6th in 2020/21 DRAPM. Increase it to three years, he is 35th from 2005-07, 28th from 2006-08 (and 33rd from 2005-08), 22nd from 2014-16, 28th from 2015-17, 5th from 2019-21, and 34th from 2020-22 (and 28th from 2019-22).
I really don’t see why you think this is helpful to you. It is well-understood that RAPM is far less reliable with smaller samples due to having high variance. All you’re saying is that in far less reliable data, LeBron’s defense looks better than in more reliable data. It’s obviously not persuasive and is transparently trying to rely on noise.
And while you can say larger timespans can be sunk by one particularly bad year, that’s generally otherwise known as things evening out in terms of variance in larger samples. If LeBron has enough particularly bad DRAPM years that he doesn’t have three-year or five-year spans that are particularly good, then that absolutely tells us something. You know full well that RAPM over relatively small samples is highly flawed due to its high variance, and yet you’re wanting to cherry-pick out smaller samples that do not include the more negative years. Again, it’s transparently trying to rely on as much positive noise as possible, and needless to say is not a serious approach.
And of course you typically glossed over what his career (or most of a career) long stretches all indicate, which is not something you see from pretty much any other “short defensive peak” player — again, as you are fully aware. But no one is arguing he should have been on an all-defensive team in 2018, or even recently, so why the focus on those specific samples.
Career RAPM is only tangentially relevant to this discussion, because it really doesn’t tell us anything about how his defensive impact ebbed and flowed throughout his career. So yeah, I glossed over something that is largely irrelevant. You want to use career RAPM to draw some tortured inference about what you think it suggests about specific timeframes in his career, but we actually have RAPM for those specific timeframes in his career and they do not agree with the inference you’re trying to draw. Obviously data that directly bears on the actual question is more persuasive than an inference you are taking from data that clearly isn’t on point for the discussion.
I would also note that, as you know, career RAPM is very flawed. It has a major problem since players’ quality changes a lot over the course of their career and it is very difficult to account for that in a 28-year RAPM measure. It is something that ends up requiring a prior to try to roughly approximate where players were in their career or just curving players based on their age. This introduces serious issues. There is a sweet spot with RAPM where the timespan is long enough to not have high variance but not so long that it runs into massive problems with player aging. That sweet spot is often considered to be five-year RAPM (which is why, for instance, BPM was tested against five-year RAPM). And that’s what I used. So the data I used was not only far more relevant to the actual discussion, but is better data anyways.
