Doctor MJ wrote:70sFan wrote:Doctor MJ wrote:So, seems like you guys are circling around a key thing without explicitly saying it:
Team-wise, the game is 50/50.
This does not mean individual stars on one end of the floor are operating with the same impact on that end of the floor as a counterpart star does on the other end.
Back in the '60s, it's pretty clear that in practice top defensive anchors were able to have bigger impact than top offensive leads, but now I believe the data is pretty clear cut pointing to individual potential for offensive impact being greater than for defensive impact.
It's a reasonable statement, but I haven't seen any reliable estimation of how much more impactful offensive stars are compared to defensive stars. I know it's
opinio comunis among basketball fans, but I'd like to see such data.
Another thing is that in this specific situation, we're talking about the best defender in the league who also happened to be a star offensively vs Curry who doesn't bring any value on defense.
With that said, I believe the overall RAPM data sides with Davis during the regular season over Curry this year, and then that series they played where Davis was the undeniable star of the show looms large.
...which was my original point. A point SpreeS finds questionable.
Okay, so if I go to nbashotcharts, go to single year RAPM for '22-23, downlaod the CSV, and that take the standard deviation for the league based on this data, here's what I get for Offense vs Defense:
Offensive RAPM SD: .766
Defensive RAPM SD: .655
This is what I would point to as evidence that in general players have more capacity for impact on offense than defense. Where players are better able to separate themselves from the peers, we'd expect greater variance, thus a greater SD means greater separation, and hence impact.
Let me know if you have any questions or objections.
I'm kinda motivated to do this for every year they have now, and I'll fully acknowledge that if this year ends up looking like a fluke that would disprove my point, but I don't think that's what we'd see in general based on my prior investigations along these lines - which obviously doesn't go back as far as we'd like due to the unavailability of the requisite data.
Thanks for the data, I appreciate effort. I decided to run the numbers for you for a few more seasons (along with luck adjusted numbers):
2023 Offensive RAPM SD: .766, .600 LA
2023 Defensive RAPM SD: .655, .519 LA
2022 Offensive RAPM SD: .770, .603 LA
2022 Defensive RAPM SD: .677, .513 LA
2021 Offensive RAPM SD: .892, .669 LA
2021 Defensive RAPM SD: .691, .564 LA
2020 Offensive RAPM SD: .772, .687 LA
2020 Defensive RAPM SD: .751, .612 LA
2019 Offensive RAPM SD: .765, .628 LA
2019 Defensive RAPM SD: .655, .536 LA
2018 Offensive RAPM SD: .759, .604 LA
2018 Defensive RAPM SD: .696, .531 LA
2017 Offensive RAPM SD: .862, .737 LA
2017 Defensive RAPM SD: .694, .567 LA
2016 Offensive RAPM SD: .796, .682 LA
2016 Defensive RAPM SD: .727, .611 LA
2015 Offensive RAPM SD: .876, .724 LA
2015 Defensive RAPM SD: .759, .654 LA
2014 Offensive RAPM SD: .804, 1.724 LA
2014 Defensive RAPM SD: .733, 1.514 LA
2013 Offensive RAPM SD: 1.160, 1.041 LA
2013 Defensive RAPM SD: 1.048, .928 LA
2012 Offensive RAPM SD: 1.330, 1.201 LA
2012 Defensive RAPM SD: 1.206, 1.403 LA
2011 Offensive RAPM SD: 1.251, 1.674 LA
2011 Defensive RAPM SD: 1.157, 1.431 LA
2010 Offensive RAPM SD: 1.233, 1.149 LA
2010 Defensive RAPM SD: 1.053, 1.003 LA
Some thoughts:
- it seems that the variance is indeed bigger on offense. To make a more in-depth analysis, I'd have to make histograms to see how much deviation is influenced by bad outliers vs good outliers (but I don't have the time for that now),
- there are many ups and downs in variance year after year, showing how noisy a single season sample is,
- pre-2014 RAPM numbers are way noiser, I don't know the reason for that though (probably older and less accurate possession counting methods?).
The difference between offense and defense indeed exists. Now, it would be interesting to try to translate this difference into something quantifible. Less than 0.2 in RAPM value is almost nothing considering the noisiness of the stat, but we can't ignore it either. Don't have the time now to do anything else, but let me know what you think about it.