trex_8063 wrote:Dr Positivity wrote:trex_8063 wrote:
Just to be clear, that was results of regression studies done on WOWY (WOWYR), which accounts for coinciding injuries/absences, etc, and calculates SRS change (not just crude W/L). And I even made a suggestion for how there can be discrepancies between that and RAPM, too.
I would argue that there's few (any?) among us as sharp in basketball analytics as Elgee, and he still finds this information relevant (even in the databall era). idk.....I find that telling (that it maybe shouldn't be cast aside). Maybe that's just me (and Elgee).
I'm not a believer in ElGee any more than I can throw him frankly, but I won't get into that
I see a LOT of problems with WOWY, even if it's attempted to be regressed
- Huge sample size issue - 10-15 game injuries are just too small. Like the Magic started 8-4 and the Pistons 10-3 this year. Weird stretches can happen in that amount of games.
- Energy level - NBA regular season is not one where players give 100% all the time. Which is why playoff gear and Thibs Bulls overperforming in regular season can happen. So that leaves room for fluctuation such as playing harder against the Warriors than the Hornets, or playing hard when the star is injured. Maybe on the other end as well - teams taking opponent without their star too lightly
- Lag in opponents scouting reports/gameplan for the team without a star
And I think the proof of this is actually looking at the results of when a player gets injured. It's rarely a cut and dry, they're exactly as much worse as you'd expect situation. Some recent examples that stand out to me:
- Spurs go 7-1 without Kawhi last year and do well in playoffs
- Warriors (11-2) have better winning % without Curry than with him this year - this makes perfect sense to me under my "energy level" theory. The only reason GSW isn't 51-4 or something right now is they are saving themselves for the playoffs, without Curry they have more reason to play hard
- Clippers went 12-6 without Paul in 2014, and I believe always did decent when it was just Blake. However drop-off in 2018 season (even with some good additions like Lou Will) may be more reflective of the impact we know CP3 had
- Bulls went 18-9 without Rose in 2012... they were good the seasons after, but not that good
- Raptors went 14-6 without Lowry last year after they got their new players when like Paul/Clippers there's a mountain of evidence saying Lowry has been the guy dragging the Raptors to great seasons year after years
You make some fair points here, and I don't mean to derail; but just to pick a few nits......
1) The Warriors do NOT have a better record without Curry this year. He's missed 15 games, they were 11-4 (.733) in those games; they're 31-9 (.775) with him. And at any rate (as mentioned in last post): WOWYR isn't looking at the W/L column, it's looking at the shift in the point differential.
So where this sample of games for Curry is concerned we could note that not only do they have a little worse record without him, but also that the average pt differential without him was +4.6 (vs +8.6 in the 40 games he played). Something like WOWYR isn't fooled by the similar record.
Side-note (I just happened to investigate it): if we wanted to look a touch deeper, we'd notice an even more striking shift........because the +4.0 difference in the pt differential is despite playing 11 of those 15 Curry-less games at home, and also despite playing mostly weak teams (*see below). As a consequence (counting HCA as worth 3 pts, which is fairly standard), we see their SRS in those 15 games was +2.0 (vs +9.82 in the 40 he's played in).
*the best teams they've played without Curry are Cleveland (not exactly elite this rs, currently -0.60 SRS), Portland (+0.88 SRS, though playing without Nurkic in the game Curry missed), Denver (whom they lost to by 15), the Clippers (+0.60 SRS), Detroit (27-27, +0.26 SRS), Milwaukee (30-24, but -0.22 SRS), and the single best SRS team faced was Utah (+1.89 SRS, though only 27-28 by record; also playing without Gobert in the game Curry missed). It was otherwise teams like the Lakers (twice), Memphis, Sacramento, Dallas, Orlando, Charlotte (twice). Didn't have to face anyone like Houston, Boston, Toronto, SA, Minny, or OKC without Curry.
2) The sample size issue you're talking about isn't so severe as only 10-15 games (except for guys like Dan Issel, who was an utter iron-man), because the WOWYR scores aren't based on single-seasons.....it's trends emerging over whole primes/whole careers (as indicated by "prime, career"). For Carmelo Anthony, for example, that career sample is comprised of ~150 missed games.
With looking at whole prime or whole career WOWYR trends, I'm not sure we should expect to see significantly more "questionable" or "noisy" results than we routinely see with single-year NPI RAPM (though I could be mistaken).
3) These same factors you cite (energy level, lag in opponents prepping for team missing the star, etc) will leave an imprint on RAPM too. You realize that, right? Not to the same degree as it will WOWYR (because RAPM's "off" sample will also be tempered by their off-minutes in games they played in); but it's still affecting it.
Fair points about Curry that, was my mistake.
3) Impact of energy level, opponents, etc. on RAPM - yes it's possible that for a player like Kawhi it also hurt his RAPM last year. But with all the lineup data etc. in RAPM I trust the process to filter it out more than just straight up WOWY.
For the rest - most of my comments are about the concept of WOWY itself. If we're talking ElGee's specific WOWYR, that's a specific can of beans. The main problem is there's way too much subjective discretion that could easily be affected by confirmation bias. I don't know how someone could look at the spreadsheet and see all those manually picked and choosed injury samples and treat that like a real stat, or not see how easy it would be for someone to put their thumb on the scales by adjusting for some players being out but not others. Choosing which years are prime also concerns me if someone had subconscious bias... for example 98 Stockton is not listed as a prime season, but seasons like 15 Wade, 72 Oscar, 02 Malone, 01 Robinson, 16 Melo, etc. are. This cutoff makes a huge difference in his value since 98 is when he missed 18 games, or the first year he really has any data. 97 is his last ASG for a few years (98 he probably missed cause of starting year injured, 99 lockout, 2000 he makes it), but if it was just going by ASGs how come Cheeks prime is listed till 1990 when 1988 is his last ASG and when his stats clearly drop off after? Oscar is the #1 guy in the stat, but there's several years you could use as when his prime really ends, or how harsh you overall want to be on players primes ending if you go into it already thinking Oscar is your man. The whole thing is just ripe for manipulation if the person wants the stat to match his opinions
