lessthanjake wrote:OhayoKD wrote:There being a "box-score advantage" at all is just a non-starter etymologically as the numbers you pulled up have not really been justified as more useful in interpreting "production" for jokic and luka than what the person your replied to you used. Even keeping things to the BBR box-score, most people interpret Luka as having better numbers, and obviously you don't have any real argument against any sort of interpretation that values creation more, replacing assists with box-creation, shots created, # of defenders bypassed, or any interpretation that values ball-handling more.
That’s just not really true. Both versions of BPM have been crafted to actually correlate well with impact. So yes, they have generally been justified as more useful in interpreting production than raw box score stats.
Hey jake! Just to provide a little more context for the accuracy of BPM:
Thinking Basketball's BPM, which the most accurate box stat on the market, is significantly closer to APM and PIPM than PER or PPG in accuracy of evaluating current season value, and
actually outperforms single-season APM and PIPM in predicting future value since 1998. It does a better job than other box stats at capturing defense (Russell has the 6th best career in Backpicks VORP, compared to Basketball Reference VORP ranking Russell 20th, despite a dearth of defensive box stats back in the 60s) and playmaking (Magic has the 5th best career in Backpicks VORP, compared to Basketball Reference VORP ranking Magic 26th).
Source:
https://fansided.com/2019/01/08/nylon-calculus-best-advanced-stat/lessthanjake wrote:Pretty much all the "impact" stuff you list can be sorted as "box-score" as well and we again get to you counting outputs which doesn't really prove anything logically.
What is your point here? The only thing that wouldn’t have any box-score component at all would be something like raw RAPM. As far as I know, we don’t have that for the 2023-2024 season, so if you’re arguing that that’s what we should use instead then it’s just a totally null argument. The closest we have to that at this point is simply raw on-off, where Jokic is miles ahead of Luka. And, in any event, there’s good reason that these impact stats include a box component—which is that raw RAPM (or any similar type of measure) is really noisy in single-season samples, such that a box component is generally understood to make it more accurate. So you’re basically arguing for something that doesn’t exist and isn’t better anyways, instead of the better and readily available stuff that on balance shows Jokic ahead.
Yep, adding box stats to a plus minus metric to make intelligently designed hybrid stats definitely increases the accuracy of the stat.
And agreed, Jokic definitely far exceeds Luka in something "pure" like on/off:
-Jokic: +20.0 (one year), +20.9 (two year), +19.4 (three year)
-Luka: +9.4 (one year), +7.3 (two year), +5.0 (three year)
i.e. Jokic is having all-time regular season on/off, approaching GOAT level, while Luka's is concerningly low for a player of his caliber (and worth diving deeper to see what's happening there -- see, e.g., the recent Thinking Basketball video).
One correction though: I've actually recently discovered this website, thebasketballdatabase.com , which has some RAPM numbers and playoff RAPM for the full plus minus era, including the latest season!
You can find the 1 year, 3 year, and 5 year RAPM data for the latest seasons here:
https://thebasketballdatabase.com/2023-24RegularSeasonPlayerRAPMComprehensive.html. As you know, there's some variations in RAPM, based on different uses of priors and lambda (which regresses outlier events, particularly players that have low sample sizes such as bench players with very few minutes, back to the mean). They present their RAPM as a "vanilla" RAPM, with no priors (although they don't compare their lambda vs Engelmann's or Goldsteins). And Jokic has a similar significant advantage vs Luka here too:
(Basketball Database) RAPM ranks:
-Jokic: 2nd (one year), 1st (three years), 5th (five years)
-Luka: 128th (one year), 64th (three years), 49th (five years)
Again, Jokic is miles ahead of Luka, at least in this measure.
Not to say this is the end-all, be-all. Data comes with uncertainty bars, and I'm not sure whether the stats have Jokic beyond the uncertainty range of Luka. There's different ways to evaluate a player, as I'm sure you'd agree. Some people might prefer Luka's box production more, or prefer to do film study over statistics (although usually detailed film-study tends to favor Jokic), or perhaps people have more resilience concerns with Jokic (i.e. is his decrease in playoff on/off actually a signal, e.g. that his lack of perimeter mobility on defense makes him vulnerable to certain matchups). Other people might want to incorporate other contextual factors into MVP... some people actually prefer to have voter fatigue (e.g. to save the 3+ MVP winners for the truly all-time players), which might favor Luka since he hasn't won yet.
Still, I tend to agree with you that the stats pretty clearly favor Jokic (or at least don't favor Luka), whether you're looking at pure impact, or box hybrid, or WOWY.
On another note -- If you're interested, that site I linked also includes playoff RAPM, although it's a little hard to navigate to (Seasons -> *pick a season* -> players -> Postseason -> *pick RAPM*). I'm not sure how useful "vanilla" playoff RAPM is, given the small sample size, and the minimal number of opponents/lineups (e.g. it's hard to tell a player A in the west is vs player B in the east, when the two lost in the first/second round and never faced each other or may not have even had any common opponents).
Personally, my ideal playoff RAPM would use some sort of prior to make up for this (e.g. RS RAPM or box inputs or or whatnot), and might even include the RS play by play data too, as a full-season RAPM. All that to say, if you're going to use the postseason RAPM from that website, I'd heavily encourage taking the number with a healthy grain of salt (i.e. wide uncertainty bars), and really only in the large samples like 5-year PS RAPM. So perhaps not very useful... but it could still be interesting if you're curious!