Doctor MJ wrote: It isn't true that there isn't inverse correlation. Fine to say it's not as big as it superficially looks, but the reality is that both Stockton's raw assists and assists per 100 went down from peak as the Jazz became contenders, and this happened while Malone's respective numbers went up.
I do think you can make the case that this shouldn't be used against Stockton in evaluation, but it's always noteworthy when a player's team success doesn't peak at the same time as his individual data peaks.
Players usually do more the worst their teams are so i would expect stockton to reduce his role as more talent came to his team, this is not always the case as sometimes the teams adjust as they add talent around their top offense players (think phil jackson reducing jordan minutes but maintaining his shot rate for the most part)
In this case the noteworthy element is that the more stockton reliant versions of the jazz failed to floor raise to a high level offense which would point to the same limitations stockton had as a top 2 scorer on a team
Brilliant team player and adds value in lots of ways but whether it was limitations or reluctance, was not a consistent source of self created scoring/creation which is one of the thinghs that players like paul, magic or nash have he didnt
Uh, but Malone did MORE as the team did better. Surely you're not looking to suggest that that's a sign of him being worse than Stockton.
My point was not really about malone but about stockton specifically, i am actually agreeing with your ideas here for the most part
So, this is the thing that Engelmann has always been prone to do that makes the stuff he likes best less useful.
All of us looking at the original spreadsheet can get a sense of how things like longevity are playing in here and mentally adjust in our heads, but once he incorporates aging in the black box it becomes much harder to parse out what's happening.
Similarly RAPM just doesn't work with coaches. You can't judge a coach game to game based on whether he's on the sideline or not, let alone possession by possession.
I have worked with Jerry on many flavors of RAPM since he first started publishing them on APBR. Each flavor of RAPM has its own advantages and disadvantages, its own use and things it should not be used for.
Personally I prefer having a unique, custom Bayesian prior for each player-season, rather than a constant aging for every player. (That sort of prior handles aging implicitly). I also personally am not at all a fan of coaching in the regression. On the other hand, I very much do prefer the rubber band effect.
I was going to totally change all my views based on this, but then I remembered a new number will come out next month which shows something different, and that I'd prefer to use logic and reasoning more than relying on numbers that can't fully account for context.
Warspite wrote:Billups was a horrible scorer who could only score with an open corner 3 or a FT.
Doctor MJ wrote: Uh, but Malone did MORE as the team did better. Surely you're not looking to suggest that that's a sign of him being worse than Stockton.
Did he? Or did he do more things that were tracked by the box-score.
Malone's minutes fell more from earlier years through '97 than Stocktons did (obviously post injury onwards Stockton fell off considerably more).
Hmm. I think the point about box score vs total action is interesting in general, but when we're talking about a guy playing the most minutes on his team it's hard for me to make a case like "Sure he takes the shots and getting more assists than before, but others are doing the really hard work."
At the core, I directionally agree with you. Stocktons role got a bit smaller and Malones a bit bigger when Hornacek got added (probably). But I find emphasizing it strange as I think it's a very small change for each (like going out of ones way to say a player going from 89% from the line to 90% improved - technically true, but very whatever). Myself, I'm not certain who had a bigger role in either period. Both certainly had a lot to do before and after Hornacek.
Imo the very clear driver in jumping to contender is Hornacek, not the very small changes in Stockton/Malone roles.
I think the fact that there was one set of numbers and then he layered on some additional adjustments (that there’s serious disagreement as to the validity of adding) and the numbers changed fairly significantly is indicative of the fact that this is an area where you can kind of just massage these numbers to make them say a lot of different things. And this shouldn’t be surprising to anyone with any background in statistical analysis. Which is why my attitude on this stuff is that all of it is flawed and the best thing to do is probably to take every measure we’re aware of and consider all of it. Even that’s not a super great solution, since there certainly could be (and probably are) errors that correlate across most/all measures. But I think it’s the best approach we have. Which is all to say that it’s good to have another data point that we can throw into the pile.
OhayoKD wrote:Lebron contributes more to all the phases of play than Messi does. And he is of course a defensive anchor unlike messi.
lessthanjake wrote:I think the fact that there was one set of numbers and then he layered on some additional adjustments (that there’s serious disagreement as to the validity of adding) and the numbers changed fairly significantly is indicative of the fact that this is an area where you can kind of just massage these numbers to make them say a lot of different things. And this shouldn’t be surprising to anyone with any background in statistical analysis. Which is why my attitude on this stuff is that all of it is flawed and the best thing to do is probably to take every measure we’re aware of and consider all of it. Even that’s not a super great solution, since there certainly could be (and probably are) errors that correlate across most/all measures. But I think it’s the best approach we have. Which is all to say that it’s good to have another data point that we can throw into the pile.
lessthanjake wrote:I think the fact that there was one set of numbers and then he layered on some additional adjustments (that there’s serious disagreement as to the validity of adding) and the numbers changed fairly significantly is indicative of the fact that this is an area where you can kind of just massage these numbers to make them say a lot of different things. And this shouldn’t be surprising to anyone with any background in statistical analysis. Which is why my attitude on this stuff is that all of it is flawed and the best thing to do is probably to take every measure we’re aware of and consider all of it. Even that’s not a super great solution, since there certainly could be (and probably are) errors that correlate across most/all measures. But I think it’s the best approach we have. Which is all to say that it’s good to have another data point that we can throw into the pile.
I think a good takeaway is that LeBron, Garnett, CP3, Curry, Jokic, and Duncan are monsters no matter how you slice the data.
lessthanjake wrote:I think the fact that there was one set of numbers and then he layered on some additional adjustments (that there’s serious disagreement as to the validity of adding) and the numbers changed fairly significantly is indicative of the fact that this is an area where you can kind of just massage these numbers to make them say a lot of different things. And this shouldn’t be surprising to anyone with any background in statistical analysis. Which is why my attitude on this stuff is that all of it is flawed and the best thing to do is probably to take every measure we’re aware of and consider all of it. Even that’s not a super great solution, since there certainly could be (and probably are) errors that correlate across most/all measures. But I think it’s the best approach we have. Which is all to say that it’s good to have another data point that we can throw into the pile.
101 in how not to use statistics
What’s the better approach? In theory, it would be better to identify which measure actually is the best and most accurate and use that and ignore everything else. But we don’t have much idea what methodological differences make things more or less accurate, not to mention in most cases we don’t even know the precise methodology used by a given measure (which obviously makes identification of how good the measure is impossible). So there’s no way to identify the best measure with any degree of confidence, which means that attempts to do it anyways will almost certainly just result in people taking the measure with the output they most like and coming up with a rationale for why it’s the best measure. And that is surely a less accurate method than just considering everything (for various obvious reasons, including that at least considering everything will make a lot of errors cancel out). It’s pretty akin to discussing whether one is better off looking at one specific political poll that one thinks is the best or looking at an aggregation like 538. Nothing’s perfect but the latter is generally going to be the better approach.
OhayoKD wrote:Lebron contributes more to all the phases of play than Messi does. And he is of course a defensive anchor unlike messi.
lessthanjake wrote:I think the fact that there was one set of numbers and then he layered on some additional adjustments (that there’s serious disagreement as to the validity of adding) and the numbers changed fairly significantly is indicative of the fact that this is an area where you can kind of just massage these numbers to make them say a lot of different things. And this shouldn’t be surprising to anyone with any background in statistical analysis. Which is why my attitude on this stuff is that all of it is flawed and the best thing to do is probably to take every measure we’re aware of and consider all of it. Even that’s not a super great solution, since there certainly could be (and probably are) errors that correlate across most/all measures. But I think it’s the best approach we have. Which is all to say that it’s good to have another data point that we can throw into the pile.
I think a good takeaway is that LeBron, Garnett, CP3, Curry, Jokic, and Duncan are monsters no matter how you slice the data.
Yeah, in essence the error in these measures is virtually never going to be big enough to make guys like that not look great. But it will be enough to change how guys like that look relative to each other.
OhayoKD wrote:Lebron contributes more to all the phases of play than Messi does. And he is of course a defensive anchor unlike messi.
A summary on the +/- data we have available for the duo (as this is kind of the thread for that): -Broadly Malone holds a moderate edge in the era where they are both carrying extreme minutes ('94'-'97). -'98 to '00 the per possession data points towards Stockton, but Malone is maintaining the big minutes while Stockton is stepping back (relative to Malone, still a fair minutes load). -'98 with Stockton missing time is our only real prime WOWY sample for either, and it looks reasonably good for each. They go 11-7 without Stockton, but jump to an elite 51-13 with him. -From '01 onwards is where the per possession data seems pretty overwhelmingly in Stockton's favor. Malone maintains the minutes edge.
Other noteworthy things: -Malone was the one thought of as better in the league starting from Layden's time and lasting through the end of their careers. -Broadly the Jazz were good but non-contending teams in the pre +/- era (shoutout to giving the Lakers a good series in '88, and '92 generally) -They entered into clear contender status from '95 to '99. -Declined from there.
My thoughts, kind of chronological order: -I think Malone carrying heavy minutes a bit sooner relatively helped him in the early awards voting. -'88 is a year I count as prime Stockton, but view as pre-prime for Malone. The coaching change to Sloan really pushed the team more towards centering around Malone (and towards that iconic PnR). Both ran variations on Motta's offense, but Layden was much less structured about it than Jerry was. Along with Malone himself more clearly improving through his career than Stockton did. -Malone improves up and past Stockton under Sloan, while Stockton maintains his level. I have Malone peaking in the fringe MVP range, and Stockton in the back half of the top 10, All-NBA 2nd team range. Malone was better when both were at their bests. -Stockton maintained impact well from '98 onwards and the minutes drop is generally overstated (I see a lot of parallels with Robinson coming back from injury). -Malone hits a more clear tipping point, and from '01 onwards Stocktons impact edge was outweighing Malones minutes edge for overall impact imo.
Overall: -I find both accurately rated in their primes (in awards and general perception), Malone maybe slightly overrated by MVPs, but it was a weaker era at the very top and I'm not upset about him winning an MVP award. -Stockton's longevity I think gets a bit underrated and I see it as a half step above Malone. Malone being a bit overrated on that end (both are obviously excellent). -So overall I think I'm a tad higher on Stockton/lower on Malone than average, but still favor Malone in the direct comparison.
DraymondGold wrote:-Some surprisingly lower ranks: -Kawhi Leonard at #19 overall and #22 offensively is still pretty good, but a defensive rank of 173 is way lower than his reputation. -Giannis Antetokounmpo at 38th overall is definitely lower than expected, particularly since his impact metrics in 19/20 were so good and we're also mid peak/prime for him too. Both his defense and offense are out of the top 50. -Anthony Davis and Luka Doncic at 88th and 95th overall are also a bit low. Doncic's reputation of having worse plus/minus numbers continues. -Dwyane Wade and Kobe Bryant are out of the top 100 overall. Wow! Kobe's 22nd offensively, right in line with some of the other Top 20 overall players, Wade's a bit worse at 42nd offensively. It's their defense that's disappointing... Wade's neutral and Kobe's a negative defender. One wonders how much non-prime years are lowering their ranking here. Kobe's reputation as having slightly worse impact metrics continues. For the Kobe fans: Kobe has an unusual number of non-prime games in his career (e.g. 266 RS games pre-age 22 when Wade was drafted, 325 RS games at age 32+ in 2011–2016, 50 + 22 PS games in those age ranges, for 663 total. For comparison, Wade has 0 games pre-age 22, 389 games after age 32+ in 2014–2019 RS, 45 PS games in those age ranges, for 434 total. Based on this, and considering their overall rank is right near each other, there is an argument that Kobe might end up looking better than Wade in say ~10 year prime RAPM, though neither are probably valuable enough in RAPM to crack the top tier based on how low their total career is).
I'm only marginally surprised by Kawhi's defensive rank: he's been mailing it in all through the rs from '17 onward. I would guess if we looked only at '12-'16, it would look pretty awesome.
Regarding the Kobe/Wade discussion: I don't think looking at pre-22/post-32 is an effective or accurate means of distinguishing a player's prime, nor is looking any sort of best 10 years sample. Some [most?] players have a prime that is shorter than 10 years; some even have full careers that are shorter. Meanwhile some guys have primes that are clearly LONGER than 10 years (e.g. Karl Malone or LeBron or KAJ).
I definitely do NOT consider rookie Wade to be prime Wade (but it is considered so by this methodology). Nor would I consider '11 (and arguably '12-'13 and maybe '00 [it's hard for me to call a 27/6/6 season for a winning team (and 8th-rated offense) "non-prime", at least definitively]) non-prime for Kobe.
Personally, I'd gauge Wade to have probably 550 (of 1,054) rs games and 56 [of 177] playoff games that are non-prime--->for a total of 606. And Kobe I'd gauge to have maybe 373 (of 1,346) rs games and 50 [of 220] playoff games that are non-prime--->total of 423.
Oh for sure! I have no problem with anyone saying rookie Wade wasn’t prime Wade. The exact years were less important to me than the general point that Kobe’s RAPM in this database might be pulled down more by non-prime seasons that some other traditional top 15/20 candidates. Kobe has a lot of non-prime games, more than Wade if their prime timespans are equal in this comparison (since Kobe had a longer career). This issue is worsened if Kobe’s non-prime years were less valuable per possession than stronger longevity players like Garnett or Duncan… that would pull his average down and exaggerate the difference in value between prime Kobe and the primes of other top 15/20 candidates who were higher in this database.
I ended up setting age 22 as the boundary in this case because for a fair apples to applies comparison, we often want to compare similar time spans, even if players’ primes may not be equal in length. E.g. if we go by your timespan (00–13 at most), we give Kobe 14 prime years, while 04–15 seems clearly too long for prime Wade.
Philosophically, I think part of the issue is that we’re trying to apply binary terminology (prime vs non-prime) to a thing that’s inherently a gradient (e.g. adding +6 value in one season vs +7 vs +8 the next). So people might even agree about the value of every individual season, but end up disagreeing in a discussion like this just because they set the threshold for ‘prime’ differently as a player slowly improves from their pre-prime into their prime or from their prime into their post-prime. I try to use language like ‘extended prime’ or ‘early/late-prime’ to be clear what I’m referring to, but I’m far from perfect.
Take Kobe. I think most people would agree 00 is clearly a step up from the previous year, but also clearly a step down from the following year. And the on/off (and presumably the RAPM) would agree with that. So do we consider 2000 to be prime Kobe? Early/extended prime? Maybe, maybe not. But even if you do consider it to be prime, if you’re doing a 10-year prime like I was, you still might not end up including it, if you think 2000 is worse than the year at the other end of the 10-year timespan (e.g. I have 2010 Kobe > 2000 Kobe, so I have 2001–2010 Kobe > 2000–2009 Kobe).
Regardless of the specifics, my general point was that Kobe (and perhaps Wade too) has a lot of non-prime games in his career that were a larger step down from his prime value, compared to some of the other Top 15/20 GOAT candidates (e.g. Garnett/Duncan per above), and that this might drop his average value more in a database like this. If so, a database might like this might exaggerate the difference between prime Kobe and the primes of other Top 15/20 candidates, even if it would still say Kobe is slightly less valuable in a prime to prime comparison.
DraymondGold wrote:-Old Stockton (12 overall) >> Old Malone (224 overall) -Old Ewing and old Hakeem are disappointing, both out of the Top 400 overall (69th/70th defensively), at least compared to old Mutombo and Mourning and Barkley and Robinson.
Stockton is indeed a monster. I believe him to be seriously underrated in MOST circles. Old Ewing and old Hakeem were just that: old. This likely contains only one season for each which might be called [late] prime.
That said, I recall Hakeem's rs on/off (and associated AuPM) for '94-'96 to be disappointing as well. This is part of why I think his greatness gets overinflated at times, based upon his playoff heroics and narrative.......but he never strung together a full season that's remotely close to GOAT-tier. I don't know if that means his playoff samples might be a pinch flukey, or if he was deliberately mailing it in all rs (why would you do that, saddling yourself with a #6-seed when you have a chance to contend?), but anyway.
Agreed about Stockton!
Agreed about Hakeem and Ewing, they were definitely both old. I’d consider 1997 to be the last (extended) prime year for Hakeem.
Still, my point was it’s a disappointing performance for them, even after correcting for the fact that they’re old (although being disappointing when you're old isn't the be-all and end-all). I just found an age-corrected career RAPM (see below). But before that, let’s look at this simpler career RAPM (1997–2024). I’ve looked at most (all?) of the players who had several all-star appearances that were drafted in the 80s and had careers going into this database. In order of their RAPM rank, Hakeem ranked *significantly* behind: -Stockton -Jordan (>40% of his sample is from Wizards years) -(HM: Mutombo, drafted in 1991 and had lots of old seasons, he would rank here) -David Robinson -Reggie Miller -Clyde Drexler -Mitch Richmond -Barkley -Pippen -(HM: Gary Payton, drafted in 1990, would rank here) -Karl Malone. Hakeem ranks slightly behind -Horace Grant -Patrick Ewing, and only barely above Shawn Kemp and Dennis Rodman.
Obviously there’s a lot of uncertainty; players’ career curves are different, this includes injured and post-prime years for Hakeem, etc., but it also includes injured and post-prime years for everyone else, and it includes one of Hakeem’s late prime years in 1997 (with one of his better playoffs). Hakeem looks like one of the single least-valuable guys drafted in the 80s who made it into this sample.
Note: we also have age-adjusted career RAPM (1997-2022), where we lose 2 years, but adjust for unequal ages, I think (?) by comparing player seasons only with those at the same age (https://docs.google.com/spreadsheets/d/1OzfLtHanVmSCPy8Y3cvCj5uFG9k7cPbDO9sQq9JgbuU/edit#gid=0), although unfortunately twitter is somewhat low on methodology details. The age adjustment might not be perfect, but by eye, this gets closer to our intuition, boosting the older players for whom we really only have post-prime seasons, while dropping the younger players for whom we an unusual higher percentage of their better seasons. This would allow us to adjust for only having old Hakeem by only comparing him to other players when they were old.
And adjusting for age, Hakeem ranks 220th all time, still behind the same players that were drafted in the 80s I mentioned above. Patrick Ewing ranks 116th.
Maybe Hakeem’s drop off post-prime was bigger than the drop off for everyone else. Fully possible. Size and defense are generally things that drop off less with age, but on film, Hakeem's more athletically drive defense seemed to drop off more with age than players like Mutombo, Duncan, Garnett. While there’s still enough missing data to argue, with this high uncertainty, Hakeem could have a top ~7 peak/prime, this is another case of surprisingly low value for Hakeem, making it less likely to interpolate that Hakeem was quite so high on the GOAT RAPM ranking, at least in my opinion! Though there's still plenty of uncertainty and obviously people can have different valid criteria, so if you value impact stats less or creation less or defense on film more, that could change things.
For context, here’s some of the other oddly low impact stats for Hakeem:
Spoiler:
Historical Data: We have ~50 Prime Hakeem games from Squared2020 historical sample (in 85, 88, 91, 93 which is often considered Hakeem’s RS peak, and maybe 96). Still plenty of missing data, but certainly enough that we’re starting to get a signal. -Historical Prime On/off: Prime Hakeem’s available on/off is ~ +6, which is below every typical Top 15 candidate we have data for except Kobe (way below even the *career* average of Curry, LeBron, Garnett who are all 10+ career on/off, and clearly under available Jordan/Magic data who seem to be on pace for 10+ career on/off, still below the *career* average of Shaq and Duncan who are 8+ career on/off). The Rockets underperformed in this sample of games, but if Hakeem’s underperformance was proportional to the team’s, his prime average would still be under all these other career averages. -Historical Prime RAPM: In the same sample, we have multi-year RAPM. Prime Hakeem’s available RAPM would put him significantly behind Jordan, Magic, old Kareem, Patrick Ewing, and David Robinson, slightly behind Pippen, Stockton, Robert Parish, Dennis Rodman, and old Bird. If Hakeem’s underperformance was proportional to the team’s underperformance, then Hakeem’s true RAPM would still be clearly behind Jordan, Magic, old Kareem, Patrick Ewing, and David Robinson. In a unified sample, unless the 80s/90s RAPM were systematically higher, Hakeem would likely look below a lot of the better career guys here, like LeBron, Garnett, Curry, Duncan, and Shaq. -> so obviously an incomplete sample with plenty of uncertainty, but Hakeem’s on pace to be one of the least valuable Top 15 candidates.
Peak: Full regular season 1994 Data: From 1994 to 1997, Hakeem’s on/off is +14.5, +11.9, +10.3, +7.4. Which is pretty good! Most people have 1994 Hakeem’s best year and a top 2 regular season for him, and most box stats put it as a top 2 regular season for him. How would this compare to the other modern players? Peak On/off: Hakeem’s 1994 RS run (a Top ~2 run ever for him) would be the 7th best RS on/off for LeBron, 6th best for Curry, 6th best for Garnett, 4th best for Duncan, and 3rd best for Shaq. Peak AuPM: Hakeem’s 1994 RS run (a Top ~2 run ever for him) would be the 6th best RS AuPM for LeBron, 6th best for Curry, 4th best for Shaq, tied 34d–4th best for Duncan, and 3rd best for Garnett. Hakeem’s on/off and AuPM in the following years (95/96/97) would also be around equal or worse in a similar analysis compared to the modern Top 15 candidates. Hakeem only really looks better than Kobe in regular season on/off and AuPM. -> so obviously an incomplete sample with plenty of uncertainty, but Hakeem’s on pace to be one of the least valuable Top 15 candidates.
Late-prime/post-prime Playoff Data: We have 1997+ playoff data. 1997 is considered one of Hakeem’s better runs. It’s a top 4 run across most box stats; Most people here have the full 1997 RS + PS in a group of his ~6th-9th best runs, while having it be the near-worst Prime regular season for him yet, so they presumably have his postseason ranked higher than 6th–9th (see Realgm Prime project). Let's say it's his ~5th best PS . 1997 Playoff On/off: Hakeem’s 1997 PS run (a Top ~5 run ever for him) would be LeBron’s 11th best on/off run, Jordan’s ~10th best run (when I last checked, when he was missing some years e.g. 1987), Shaq’s 9th best run, Garnett’s 9th best run, Kobe’s 8th best run, Duncan’s 8th best run, and Curry’s 6th best run. 1997 Playoff AuPM: Hakeem’s 1997 PS run (a top ~5 run ever for him) would be LeBron’s 12th best AuPM run, Shaq’s 7th best run, Garnett’s 6th best run, Duncan’s 6th best run, Curry’s 6th best run. Later Playoffs: All three playoff runs we have for Hakeem after 1997 have a *negative* on/off, for an average of -16.2. (14 games, 399 on minutes, 273 off minutes). Obviously a small sample of playoff runs, but it's another oddly low-impact datapoint for Hakeem. -> so obviously an incomplete sample with plenty of uncertainty, but Hakeem’s on pace to be one of the least valuable Top 15 candidates.
Late-prime/post-prime data: We have post-prime on/off and RAPM data (e.g. in this thread). And even adjusting for age, Hakeem’s post-prime RAPM in RS + PS (1997–retirement) would put him as one of the least valuable typical top 15 candidates with available data.
Full career WOWY data: We have 5 standard WOWY stats: 1) Prime WOWY, 2) multi-year team-change WOWY from e.g. retirements/trades, and 3 adjusted WOWY stats that adjust for teammates/opponents and increase the sample size by incorporating data points from those other players (WOWYR, GPM, and Moonbeam’s RWOWY). Out of typical top 15 candidates: LeBron, Shaq, West, Oscar, and Curry are all are better in every available WOWY stat. Jordan, Kareem, Russell, Magic, Garnett, and Bird are better in 4/5 available stats. Everyone mentioned prior, in addition to Duncan, Wilt, and Kobe, are better in a majority of the adjusted WOWY metrics. Hakeem’s best stat (raw prime WOWY) would put him 11th all time (if we include Curry), sneaking ahead of Russell, Wilt, Kareem, and Jordan, but all those players have contextual factors for why raw prime WOWY might underrate them. Prime Russell only missed 23 games and so would have lots of noise; ~95% of Wilt’s ‘without’ sample came from 1965 when there was a lineup change replacing Wilt and Wilt was playing injured for much of that season; ~77% of Kareem’s ‘without’ sample came from those broken hand seasons when he came back likely before he was fully healthy; Jordan has basically no missed games in 1987–1993 (his best 7-year stretch), ~92% of Jordan’s non-94 without sample comes from 95 when he came back before he was fully in-shape, and 94 has the obvious contextual factors that have been discussed to death… massive bumps in Pippen/Grant on/off and box stats, and other lineup changes replacing Jordan. Bird’s only behind in WOWYR, but his WOWYR has the highest uncertainty of anyone in the stat due to some uncommon lineup/minute changes, and Bird looks better in the other 4 WOWY stats. Meanwhile, Hakeem missed over 60 games (64 missed games in 1985-94, 74 in 1986–95), so it’s not like Hakeem has a tiny ‘without’ sample that would increase his uncertainty in these stats. -> so obviously plenty of uncertainty in WOWY stats, but Hakeem’s on pace to be one of the least valuable Top 15 candidates.
Ironically, the most favorable type of stat are box stats, which have a reputation of underrating defense (but also often underrate creation, which is Hakeem’s weakest area). The three most accurate box stats are (in order) Backpicks BPM, PIPM, and RAPTOR. Among the available data for the typical top 15 players... Peak: Backpicks BPM and PIPM give Hakeem a bottom-third regular season peak and middle-third playoff peak, while RAPTOR gives Hakeem bottom-third regular season and playoff peak among the typical Top 15. Career: BPM and PIPM I think (from memory) give Hakeem a middle-third total career value, while RAPTOR still gives Hakeem a bottom-third career value.
Anyway, that was a bit of an aside. All that to say, I tend to agree Hakeem's value gets overrated at times on this board, at least compared to our statistics, perhaps due to his playoff heroics and narrative. I also agree that if the playoff sample wasn't partially fluky/noise, that would require an outlier level of coasting in the regular season... which presumably wouldn't be the best way to help your team win, although I guess you could debate how much it mattered. Interesting stuff either way, and it's nice to have another stat to evaluate these players!
DraymondGold wrote:Other Current players: these are some other guys for whom we only have young/prime seasons. We might expect their ranks to drop over time if they're mid-peak now and haven't had any post-prime seasons to drag them down. -Tatum at 7th overall is obviously great, propped up by being Top 15 offensively -Embiid 10th overall is also great. He seems much more valuable defensively.
Chris Paul is legit beastly too.
Indeed he is! It's a shame certain players are so injury prone in the playoffs. Playoff-specific injuries make it so much more difficult to judge the resilience of players at a reasonable sample size.
lessthanjake wrote:I think the fact that there was one set of numbers and then he layered on some additional adjustments (that there’s serious disagreement as to the validity of adding) and the numbers changed fairly significantly is indicative of the fact that this is an area where you can kind of just massage these numbers to make them say a lot of different things. And this shouldn’t be surprising to anyone with any background in statistical analysis. Which is why my attitude on this stuff is that all of it is flawed and the best thing to do is probably to take every measure we’re aware of and consider all of it. Even that’s not a super great solution, since there certainly could be (and probably are) errors that correlate across most/all measures. But I think it’s the best approach we have. Which is all to say that it’s good to have another data point that we can throw into the pile.
101 in how not to use statistics
What’s the better approach? In theory,
Weighing and utilizing numbers based on what they measure, how well they measure it, and what relevance it has to whatever specifically you are claiming? Of course that requires actually understanding what you're using and being comfortable with uncertainty...
If the "blackbox" matters to you then weighing transparent methodologies higher is a much more logical step than pretending everything is fundamentally equal...
It’s pretty akin to discussing whether one is better off looking at one specific political poll that one thinks is the best or looking at an aggregation like 538.
But 538 does not just consider everything.
538 ignores certain polls that don't meet certain standards, differentiates in weights the polls that they do accept based on various factors(track-record being one) and then also weigh a bunch of historic factors before fine-tuning the formula based on predictivity...
And that is surely a less accurate method than just considering everything
"Considering everything" sounds good until you realize "everything" includes infinite possible variations of stuff whose outputs or process don't seem reasonable to you as well as whatever you do. There are infinite potential variations of measures where your best player ever pick, Jokic, wouldn't rank top 10 in the current league. There are infinite potential variations of measures where curry would not be top 15.
You are not considering everything. You are considering everything that is convenient. If you are not willing to discriminate between data, then there's not much point using it in the first place.
trex_8063 wrote: That said, I recall Hakeem's rs on/off (and associated AuPM) for '94-'96 to be disappointing as well. This is part of why I think his greatness gets overinflated at times, based upon his playoff heroics and narrative.......but he never strung together a full season that's remotely close to GOAT-tier. I don't know if that means his playoff samples might be a pinch flukey, or if he was deliberately mailing it in all rs (why would you do that, saddling yourself with a #6-seed when you have a chance to contend?), but anyway.
Alternatively, maybe the off over a few rs minutes per game without him is also fluky:
Spoiler:
lebron 09-21 wowy 656-263 with lebron 0.714% win rate 37-73 without lebron 0.336% win rate net rating with lebron +6.49 (59 win pace level) net rating without lebron -5.50 (25 win pace level) +8.6 ortg difference -3.68 drtg difference +12 total swing
duncan 01-07 396-144 (73.3% win rate) with duncan 17-17 (50% win rate) without duncan +8.5 net rating with duncan (64 win pace level) -0.4 net rating without duncan (40 win pace level) +4.3 ortg change -4.6 drtg change +8.9 overall change
magic 84-91 454-149 75.3% win rate with 29-24 54.7% win rate +7.4 net rating with (61 win pace level) +0.2 net rating without (42 win pace level) +4.9 ortg difference -2.3 drtg difference +7.2 overall diffrence
hakeem 91-97 333-177 65.3% win rate with hakeem 26-38 40.6% win rate without hakeem +3.5 net rating with hakeem (51 win pace level) -2.7 net rating without hakeem (33 win pace level +1.1 ortg change -5.1 drtg change +6.2 overall change
jordan 88-98 bulls with MJ 490-176 (73.6% win rate) bulls without MJ 90-64 (58.4% win rate) net rating with MJ +7.7 (62 win pace level) net rating without MJ +3.6 (52 win pace level) +5.1 ortg difference +1.1 drtg difference +4 total swing
You've described his case as theoretical, but rs impact in line with his best contemporaries over large samples -> best playoff riser in terms of team-wide results and internal box-scaling -> longetivity advantage over pretty much all his contemporaries doesn't seem the least bit theoretical to me.
Theoretical considerations solidify it(not optimized by coaches until his 30's, ect) but Hakeem has had an argument as a most-valuable-guy-in-the-league candidate based on what demonstrably happened as early as his 2nd year: https://forums.realgm.com/boards/viewtopic.php?t=2312865
Weighing and utilizing numbers based on what they measure, how well they measure it, and what relevance it has to whatever specifically you are claiming? Of course that requires actually understanding what you're using and being comfortable with uncertainty...
If the "blackbox" matters to you then weighing transparent methodologies higher is a much more logical step than pretending everything is fundamentally equal...
Where did I say everything must be treated “fundamentally equal”? I simply said we should consider everything. You’ve read in things I didn’t say and then started arguing against the concept you read in. If there are genuinely objective reasons to weigh certain measures more highly than others and we can identify which methods are better and worse in terms of that factor, then that might provide a good reason to weigh different measures differently. That said, I’d caution against trying to go too deep into that, since there’s a fine line between identifying genuinely objective reasons in good faith and coming up with rationales to justify preferencing output one likes more.
It’s pretty akin to discussing whether one is better off looking at one specific political poll that one thinks is the best or looking at an aggregation like 538.
But 538 does not just consider everything.
538 ignores certain polls that don't meet certain standards, differentiates in weights the polls that they do accept based on various factors(track-record being one) and then also weigh a bunch of historic factors before fine-tuning the formula based on predictivity...
538 has generally ignored almost nothing (other poll aggregation sites have generally ignored more and have criticized 538 for their permissiveness—though 538 is now run by a different person who takes a bit of a different view on this). Anyways, again, I’m not suggesting we can’t weight things differently if there’s a genuinely objective reason to do so, so that is not inconsistent with what I wrote. That said, it’s worth acknowledging that it’s a lot easier for 538 to weight things based on somewhat objective factors, since in that case there’s an actual real end result that one can measure track records against. There is no real measure of player impact that we can look at and use to tell us which impact measures have proven most accurate in the past. If there were, we wouldn’t need to use impact measures at all! The best thing we can do is to look at what measures would’ve predicted raw RAPM the best in out-of-sample years, but that doesn’t actually tell us much since raw RAPM is itself a very flawed measure rather than being an objective end result. A measure can correlate better with raw RAPM simply because its errors most closely correlate with raw RAPM’s errors, rather than because it is actually more accurate in measuring impact. Actual election results are not like raw RAPM. So, for our purposes, we have much less means of reliably identifying what factors make things more accurate or what measures have proven to be more accurate. Indeed, we have very little way of doing that at all, because we do not have some measure of player impact that we know is accurate that we can compare against (whereas for polling, we have actual election results).
Anyways, you’re missing the point of the analogy by trying to draw fine distinctions that aren’t really relevant to the reason I made the analogy. The point here is that when faced with a bunch of competing data points that we know inherently are subject to significant error and where we have no actually good way of identifying exactly how big those errors are and what direction they go, it is generally a better approach to consider all the data points rather than just picking and choosing whatever individual data point most fits our priors/preferences.
And that is surely a less accurate method than just considering everything
"Considering everything" sounds good until you realize "everything" includes infinite possible variations of stuff whose outputs or process don't seem reasonable to you as well as whatever you do. There are infinite potential variations of measures where your best player ever pick, Jokic, wouldn't rank top 10 in the current league. There are infinite potential variations of measures where curry would not be top 15.
You are not considering everything. You are considering everything that is convenient. If you are not willing to discriminate between data, then there's not much point using it in the first place.
I think you ultimately need to just accept that there is no genuinely great way to use this sort of data. It’s all flawed, and we have no real way of figuring out how flawed any particular measure is. Considering everything does inherently involve considering measures that are worse than other measures. Of course, that’s not ideal. But trying to dig deep into “discriminat[ing] between data” when we have very little objective criteria with which to actually do that is generally going to be worse, since it will virtually always just be an excuse for someone to come up with reasons to consider output they like and throw away output they don’t like. I’m okay with weighing some measures more than others if there’s a genuinely objective reason to do so, but we should be very careful and stringent about doing that, because there’s almost no genuinely objective reasons in an area where we have no objective facts to compare factors or track record against. So, for example, I think I’m probably comfortable with weighing measures a bit more highly if the methodology is more transparent (since I think there’s a general objective reason to think transparency in methodology probably correlates to at least some degree with accuracy), but it’s hard for me to think of much more that would be justified.
OhayoKD wrote:Lebron contributes more to all the phases of play than Messi does. And he is of course a defensive anchor unlike messi.
eminence wrote:Imo the very clear driver in jumping to contender is Hornacek, not the very small changes in Stockton/Malone roles.
I know this thread is about the new RAPM, but this jumped out at me.
I would point out that in 1992, before Hornacek was there, the Jazz won 55 games with a 5.70 SRS and a +6.6 Net Rtg and got two games away from the Finals.
Further, in 1993-94 - when Hornacek was acquired in trade, only played 27 games, and only started in 9 of them - the Jazz went 53-29 with a 4.10 SRS and a +4.5 SRS and got within three games of the Finals. They went 36-19(.655) prior to the trade, and then 17-10(.630) with him post-trade. I calculated their average MOV in the 27 games Hornacek played, and it was 4.67, compared to their full-season SRS of 4.10. I don't take too much from these splits other than that there wasn't some huge jump in team performance and that their season numbers would've been comparable either way.
I suppose it depends on how you define contender, but it seems like they were in the mix without Hornacek.
Now, I like Hornacek, and I do agree that they got better with him, but I just disagree with the quoted characterization.
eminence wrote:Imo the very clear driver in jumping to contender is Hornacek, not the very small changes in Stockton/Malone roles.
I know this thread is about the new RAPM, but this jumped out at me.
I would point out that in 1992, before Hornacek was there, the Jazz won 55 games with a 5.70 SRS and a +6.6 Net Rtg and got two games away from the Finals.
Further, in 1993-94 - when Hornacek was acquired in trade, only played 27 games, and only started in 9 of them - the Jazz went 53-29 with a 4.10 SRS and a +4.5 SRS and got within three games of the Finals. They went 36-19(.655) prior to the trade, and then 17-10(.630) with him post-trade. I calculated their average MOV in the 27 games Hornacek played, and it was 4.67, compared to their full-season SRS of 4.10. I don't take too much from these splits other than that there wasn't some huge jump in team performance and that their season numbers would've been comparable either way.
I suppose it depends on how you define contender, but it seems like they were in the mix without Hornacek.
Now, I like Hornacek, and I do agree that they got better with him, but I just disagree with the quoted characterization.
The Jazz made a jump in 1995 in their first full season with Hornacek. The Jazz were legit contenders from 95-99. I don't think their team was contenders before that.
One_and_Done wrote:I was going to totally change all my views based on this, but then I remembered a new number will come out next month which shows something different, and that I'd prefer to use logic and reasoning more than relying on numbers that can't fully account for context.
I think we are kindred souls on this point. I read each release with an open mind, but time and time again, beyond obvious correlations at the very top of the data set I have absolutely no idea what to take from any of this, and never have with RAPM.
A formula with a complete lack of transparency in inputs/adjustments (i.e. trust us 'ridge regression') that results in significant and repeated outliers - to be fair those limitations are acknowledged.
The question becomes noting the fundamental limitations/problems why persist with it at all? We are trying to drive a rickety car with 2 flat balding tyres and pretending that just because we appear to be moving forward its getting us to our destination.
Having multiple data points to assess is nice, but in reality it also muddies the waters, in this case providing little tangible, explainable value and creating another outlet for quantifiable bias to creep in. Sorry I dont mean to crap on anyone's enjoyment of new information, and of course we need people out there pushing the analytics envelope, but RAPM just seems to trigger all the wrong things for me.
So, this is the thing that Engelmann has always been prone to do that makes the stuff he likes best less useful.
All of us looking at the original spreadsheet can get a sense of how things like longevity are playing in here and mentally adjust in our heads, but once he incorporates aging in the black box it becomes much harder to parse out what's happening.
Similarly RAPM just doesn't work with coaches. You can't judge a coach game to game based on whether he's on the sideline or not, let alone possession by possession.
I have worked with Jerry on many flavors of RAPM since he first started publishing them on APBR. Each flavor of RAPM has its own advantages and disadvantages, its own use and things it should not be used for.
Personally I prefer having a unique, custom Bayesian prior for each player-season, rather than a constant aging for every player. (That sort of prior handles aging implicitly). I also personally am not at all a fan of coaching in the regression. On the other hand, I very much do prefer the rubber band effect.
I'm glad Jerry is back in the long-term RAPM business; he had data sourcing issues for a while.
Your post makes me chagrin. I should be more respectful, particularly about someone whose tools I have used for free.
Thing is though, the public has not had access to a consistent selection of clearly identified variations of APM/RAPM. I don't begrudge the existence of any stat, but when we know that the emphasis on a particular variation will often kill off their cousins, I believe it is appropriate to comment on the nature of what is being done, why its better, and why its worse.
So with the note up front that what I'd really like is more complete and organized statistical results, what I'd say is that some statisticians - and Englemann is one of them - don't seem to be concerned about the black box they create for the basketball analyst when they throw everything into the stew, and to me this is a perspective that should at least be as prominent as any other when considering what optimization directions are worth pursuing.
On your "unique, custom Bayesian prior for each player-season", it'd be interesting to learn more. I have a lot of confidence in your expertise, but the word "custom" here disturbs me a bit. Can you elaborate on that?
Re: rubber band effect adjustment. Would you mind explaining what this entails?
Thanks for sharing your latest work. I'm appreciative you're still active in the public space, both in terms of your stats, and your explanations.
eminence wrote:Imo the very clear driver in jumping to contender is Hornacek, not the very small changes in Stockton/Malone roles.
I know this thread is about the new RAPM, but this jumped out at me.
I would point out that in 1992, before Hornacek was there, the Jazz won 55 games with a 5.70 SRS and a +6.6 Net Rtg and got two games away from the Finals.
Further, in 1993-94 - when Hornacek was acquired in trade, only played 27 games, and only started in 9 of them - the Jazz went 53-29 with a 4.10 SRS and a +4.5 SRS and got within three games of the Finals. They went 36-19(.655) prior to the trade, and then 17-10(.630) with him post-trade. I calculated their average MOV in the 27 games Hornacek played, and it was 4.67, compared to their full-season SRS of 4.10. I don't take too much from these splits other than that there wasn't some huge jump in team performance and that their season numbers would've been comparable either way.
I suppose it depends on how you define contender, but it seems like they were in the mix without Hornacek.
Now, I like Hornacek, and I do agree that they got better with him, but I just disagree with the quoted characterization.
I had another post that mentions '92 in particular as deserving of a shout and I have no issues with labeling that team as a contender. But it's the clear top year prior to Hornacek, and just another year compared to the Jazz run with Hornacek.
OhayoKD wrote:If the "blackbox" matters to you then weighing transparent methodologies higher is a much more logical step than pretending everything is fundamentally equal...
We went through this in philosophy, it was called "relativism," and it was very bad. After we got over it, we all sufficiently chastised ourselves. And no harm was done (except all the harm it did).
