Post#28 » by Bad Gatorade » Fri Jun 23, 2023 3:10 pm
A few thoughts -
Has the data been adjusted for age at all? I know that Engelmann does so, and some of the "longer" term samples aren't as heavily impacted as a result (in fact, some players such as Stockton look amazing in his data sets because they follow a different ageing curve to the typical player).
I think that 5 year blocks are much better. Large samples can capture "too much" information - e.g. even in older samples that captured, say, 8 years of Durant's career, the vast improvement Durant made in his 3rd year in the league was captured by other players, and so he appeared highly underwhelming in RAPM for quite a while.
Even 5 year samples are imperfect - IIRC, Kyle Korver's 2012-2016 sample looked awesome, but it also looked very clearly better than 2011-2015 and 2013-2017 - even one year can influence things a lot. Honestly, we should be looking at as many non-baffling samples as possible.
I'm sceptical of postseason data in general being vastly different from regular season data, because we've already seen that single year splits (82 games) can yield many, many weird results... and 82 games is often over half of a playoff career. That's a huge chunk, not to mention limited lineup data on account of fewer total equations to work off (fewer players) plus shorter "off" stints and I think that there's a limit to how much we can ascertain without a prior. I think we can see this in the comparison between regular season vs playoffs vs combined data - some players (e.g. Harden) are closer to their playoff data in their "combined" splits, whereas players such as CP3 (with huge negative playoff splits relative to the regular season) don't seem to be heavily impacted when you combine the data.
The lack of a "complete" dataset also doesn't really allow us to troubleshoot results that make no sense - in a larger RAPM sample such as 1998-2019 playoff RAPM, to use CP3 as an example again, some of the players that have played alongside him (Bledsoe, Tucker, Jared Dudley) rank very highly even though they've got tiny samples. A holi stic sample, even one encompassing a large range, is still going to yield "suspicious" results. By the same token, one must consider when players in the playoffs played together in particular. Just... too many variables for me to take too seriously, IMO.
I'd probably be a tad hesitant to reward possessions in the playoffs on the virtue of circumstance - e.g. KG not having as many postseason possessions in his prime Minnesota years shouldn't go against him.
This isn't to say that postseason analysis should play zero part at all, but rather, I'd definitely hedge against a playoff only sample. I'm more than fine with, say, a combined sample that includes increased postseason weighting, or a postseason sample operating off a regular season prior. I've seen the latter on APBRmetrics and I agree quite strongly with the conclusions - you will get players like post-2013 LeBron, Draymond etc that lift their game in the playoffs, and this order of magnitude is larger than the decreases (which are occasionally notable, but generally nothing too crazy at all). I'm more inclined to believe that, as there are players not willing to show all of their tricks, increased effort etc that takes place in the playoffs.
Now, since Kobe has been a huge discussion point, I do think there's potential for his larger RAPM samples to be impacted by his early career results - IIRC, Kobe had a few seasons in the early 2000s missing games, with the Lakers barely missing a beat (almost shockingly so), and his early 2000s RAPM samples don't look as impressive as what we saw during his 2006-10 run. I don't think having Shaq + resilient team structure/GOAT level coaching should go against him as a player, although it goes both ways too with regard to "5 rangzzzz" or whatever justification we choose to use.
I use a lot of parentheses when I post (it's a bad habit)
