RealGM 2023 Top 100 Project - #4 (Bill Russell)

[OhayoKD]

Induction vote:

Russell - 12 (AEnigma, f4p, rk, ltj, Samurai, hcl, trelos, beast, eminence, ljspeelman, ShaqA, Doc)
Duncan - 7 (trex, OaD, Dr P, ceoofk, iggy, Dooley, Ambrose)

Russell wins.

Nomination vote:

Garnett - 5 (AEnigma, trelos, iggy, eminence, ljspeelman)
Magic - 8 (trex, f4p, rk, OaD, Samurai, beast, Doc, Ambrose)
Curry - 3 (ltj, Dr P, Dooley)
Kobe - 1 (ceoofk)
Mikan - 1 (ShaqA)

Magic wins.

Will get new thread going and then pretty this up later.

Not that it changes much, but I did cast a voting post with the header "this will be my voting post". Suppose it's my bad for not putting the votes at the top like I did with the first two rounds.

[lessthanjake]

Following on with some more information, this time from the GitHub RAPM data (which ends at 2018-2019, which is why I don’t include years beyond that in the timeframes):

Average Regular Season GitHub RAPM in Prime Years

- Stephen Curry (2013-2014 to 2018-2019): 5.482
- Michael Jordan (1996-1997 to 1997-1998): 4.335
- Kevin Garnett (1999-2000 to 2007-2008): 4.096
- LeBron James (2008-2009 to 2018-2019): 3.881
- Tim Duncan (1998-1999 to 2006-2007): 3.530
- Shaquille O’Neal (1997-1998 to 2004-2005): 2.365

There’s also playoff RAPM data. The playoff RAPM appears to be on a different scale for whatever reason, so the numbers are all lower:

Average Playoff GitHub RAPM in Prime Years

- Michael Jordan (1996-1997 to 1997-1998): 2.527
- Stephen Curry (2013-2014 to 2018-2019): 2.132
- LeBron James (2008-2009 to 2017-2018): 1.713
- Tim Duncan (1998-1999 to 2006-2007): 0.925
- Shaquille O’Neal (1997-1998 to 2004-2005): 0.621
- Kevin Garnett (1999-2000 to 2007-2008): 0.612

This version of RAPM is a bit wonky and I know some don’t like it, but it’s one of the RAPM measures we have and it seems worth posting how the above players come out in their primes. As with the AuPM/g and RPM measures I posted earlier, Steph Curry comes off looking extremely good.

(Note: Again, I didn’t include Hakeem here, since he wasn’t really in his prime by the 1996-1997 season, which is when the data starts. But in case there’s any curiosity, Hakeem had a regular season average of 2.395 in 1996-1997 to 1997-1998, and an average playoff RAPM of 0.170 in that timeframe).

Why are you so pushing Curry at this stage of voting? The best case scenario for Curry is 10-11th. More likely will be around 12th-14th.

I’m pushing Curry at this stage because I think he should get into the nominee pool at this stage (and be voted in soon too). I get that he probably won’t end up as high as I personally would put him, since my sense is that there’s some voters here that are less high on Curry than I am. But I think he has a good case to be highly ranked, and indeed that that case is the type of case that people on this particular sub-forum would generally appreciate. And I think that to not make that case because I think a lot of people here won’t agree with it would basically be to consciously succumb to groupthink. So I’m arguing what I think is correct, without regard to what my perception is about whether others would agree (or whether their minds could be changed by points I might make).

I wonder, where would you rank Hakeem then? How does he compare to the rest of the top bigs left?

To me, Hakeem is fairly clearly behind Duncan (I explained that a little bit here: viewtopic.php?p=107594616#p107594616). I have him behind Shaq as well, though that’s starting to get closer than Duncan. There’s more complicated reasoning than this, but basically I’d take prime Shaq even over prime Hakeem, and I’m probably a little higher on Shaq’s earlier years than I am on Hakeem’s earlier years. I’d probably ultimately have Wilt ahead of Hakeem too, but that’s one I’m less sure of—in part because I waffle a lot on how to rank Wilt (there’s just a lot of uncertainty for me with Wilt, due to lack of footage and whatnot). To me, Hakeem is more a peer with someone like Kevin Garnett. I’d *probably* have Hakeem over Garnett, but I think it’s close. In terms of ranking overall, he’d be somewhere between 10th and 13th to me, I think (though that’s a preliminary assessment—obviously discussion like this project is aimed at an assumption that people’s preliminary assessments can be changed).

The only way that theory survives even the slightest scrutiny is if you assume Hakeem took a massive leap around 1992/93 right as the rest of the Rockets roster collapsed.

1985-91 Rockets with Hakeem: +2 net rating
1984-91 Rockets without Hakeem: -2.1 net rating

1992-97 Rockets with Hakeem: +3.5 net rating
1992-97 Rockets without Hakeem: -6.3 net rating

People who understand player development rightly recognise that is in fact not what happened, but anyone who sincerely believes Hakeem went from a +4 regular season player to a +10 one in a span of an offseason, before factoring in playoff elevation, should probably also be pushing Hakeem as a top three or two or one sustained peak, and on that basis alone he would merit reentry into this fringe top five discussion.

For whatever reason, that is not what is happening here.

I don’t think it’s at all implausible that Hakeem took a significant leap in that time period. It’s not uncommon for players to take a significant leap at some point—whether that’s due to peaking physically or just getting things together better mentally. I think your objection would probably be that Hakeem wouldn’t have been peaking physically at ages 30-32 or whatever. And that’s probably true (though 30-32 is not generally a *bad* age physically). But the mental aspect of the game is really enormous, and I don’t find it at all implausible that someone might get it together more mentally at that age. And there’s some anecdotal reasons to think that that might’ve happened with Hakeem, as well as the obvious fact that he did actually peak in those years in terms of box-score stats.

As for the precise +4 —> +10 thing, there’s obviously some statistical noise there, especially with the +10 number, which is based on just a 26-game off sample (not the worst sample size, but not ideal either). So I don’t know that it makes much sense to put much value on the *exact* change in these sorts of WOWY numbers. What we can say is that these numbers indicate it’s likely he had a good bit more impact in those years than he did in the prior years. And that’s perfectly consistent with him also being a statistically superior player in those years in terms of box score.

Overall, I just don’t think it makes sense to take Hakeem in the several years starting at 1992-1993 and assume he must’ve been just as good in the prior years but just brought down by his team or something. Essentially all available indicators—the available impact data, box-score stats, and his team’s overall success without wildly different talent-level teams—tell us that he was *not* as good in the prior years.

For MVP-voting, sure. In terms of actual "value" or "goodness", no. It does not need to be "much" higher, which is part of why using mvp-voting as a proxy for player-goodness questionable here. Not sure it matters too much how your interpretation of his "production" stacks up to offensively slanted contemporaries when the rockets defense spikes by 2-points with rookie Hakeem as the team jumps from bad to good. Hakeem's "production" looks fine compared to players of a similar profile(which happens to have the best track-record of overall success and "lift" across nba-history), and frankly without questionably justified priors, nearing 5-point lift on a decent team marks a player who easily sorts into the top-10. It actually was enough for Ralph Sampson to get 10th despite being on the 84 rockets and minimally shifting his raw "production" and Jordan didn't even need to have a winning-record to finish 6th. In-fact he would also win less in 1987 despite Hakeem's team coking up and his co-star(previously voted higher than Hakeem) missing half-season and end up finishing 2nd on the MVP ballot. Apply that theory of "you need to be worth much more on a worse team" and the votes here look even worse.

The rest is you not using stats properly imo but we've covered this before so we can table it.

I think you’re assuming that MVP-voting just follows a logic that is completely divorced from player-goodness. To some degree that’s not totally wrong, but do think it is also internalized in MVP voting that it’s more impressive from the standpoint of player-goodness to lift a good team as opposed to a mediocre one. That’s a good deal of why team record matters so much in MVP voting!

As for the idea that “nearing 5-point lift on a decent team marks a player who easily sorts into the top-10,” I’m assuming from the context that you mean top 10 in the league, not top 10 all time, since you mentioning Sampson being 10th. And I’d agree with that! I’m not saying Hakeem wasn’t a top 10 player in the league. I’m saying his MVP voting was roughly in line with where he stood in the league, and his MVP voting was typically putting him in the top 10! As for your comparison to Jordan, the difference is that box-score stats matter when assessing how good a player is, and so Jordan was finishing highly in MVP voting because his box-score stats were off the charts (for instance, in that 1987 season, he led the league in PER, Win Shares, BPM, VORP, etc., and had the highest PER anyone had had in 15 years). Hakeem’s box-score stats were good, but weren’t providing a reason to have him ranked any higher, like Jordan’s were. It’s perfectly reasonable to conclude that when two guys are on pretty mediocre teams and one puts up historic box-score numbers and the other doesn’t that the latter player isn’t as good.

[lessthanjake]

This is why people create CORP or VORP evaluations that take into account longevity and value of each season. This methodology can be used as a baseline to get the idea of what comparison is reasonable and what is not.

The results I have seen (along with the ones I provided) suggest that Kobe and Curry is a legit comparison, Stockton and Robinson is probably slightly less close and the ones you mentioned are pointless indeed. Now, if you have a different method of evaluating longevity or a different weighing procedure, I am all ears. If your analysis is only about gut feeling, then I am not interested in that.

Isn’t CORP basically not really a metric at all, and basically just layering subjective evaluations onto a philosophical heuristic? Like, in theory, CORP is actually a valid way to think about things that balances weighing peak/prime and longevity, so I do tend to think it’s a very helpful heuristic for thinking about things. But what that method spits out is going to just be dependent on pretty subjective valuations of how much a player increases the chances to win a championship. It’s not a precise metric, but rather just a way of thinking about things. For example, I could very easily come to a conclusion that Steph Curry has already added more by a CORP methodology than Kobe did—in fact, I do think that that’s the case. But, of course, you could use the same heuristic and come to the opposite conclusion.

The inputs are. The formula(where the longetvity/peak weightings come from) theoretically should not be(srs-to championship studies were the basis though I am not well versed on the specifics).

Yeah I get that—though even those weightings you’re referring to are a subjective weighting. As I see it, CORP is just a helpful heuristic to help think about career value/greatness. To my way of thinking, I can and do apply a CORP-style analysis and put Steph Curry above essentially everyone that is left, because my assessment of what peak Curry was adding to championship chances is through-the-roof high, based on what I see in his impact metrics (which included basically having half a decade where he took a team that was only okay overall without him, and made them a 68-win 10+ SRS teams with him. 68-win, 10+ SRS teams have extremely high title-likelihood!).

Relatedly, I also did a quasi-CORP-style analysis yesterday, which I didn’t post since it was fairly back-of-the-napkin and I’d done it for my own amusement. But, what I did was I looked at AuPM/g each season, and simply measured how much a player’s AuPM/g was above 3.0 (which is the line for about top 15 in the league) each season and added that together for a player’s entire career. My reasoning was essentially that championship chances are basically driven almost entirely by players that are above that sort of threshold and are of course driven disproportionately by players being a lot above that threshold. So it was an exercise that was basically informed by a CORP-style thought process. And by that measure, Steph Curry has already had more AuPM/g-above-3.0 in his career than Duncan, Garnett, Shaq, Kobe, etc, and actually by a decent margin over all of them. It’s a back-of-the-napkin measure that I’m not holding up as super meaningful, but it just goes to show how it’s far from clearly the case that CORP analysis would leave Curry behind the players being discussed now.

[AEnigma]

The dispute is not that he improved. The dispute is that you want to use small samples to make one point and then not carry through with other similar samples. If the 1992-97 without numbers oversell him then the similar assumption should be that the 1985-91 numbers undersell him.

Everyone can identify Hakeem’s career arc. It reflects most career arcs, albeit with a later true peak than average. So we take the career samples (or extended prime samples I guess), and those extended samples on balance have him as a top ~ten regular season candidate with more robust playoff maintenance/elevation and a longer extended prime than basically everyone ahead of him. Quantifying that precisely leaves a range, sure, but everyone can see what happens when the priority seems to be to take the lowest estimate at every opportunity because well we use box production as a filter and strangely it seems as though most box aggregates tend to disproportionately skew against all-time defenders…

[lessthanjake]

The dispute is not that he improved. The dispute is that you want to use small samples to make one point and then not carry through with other similar samples. If the 1992-97 without numbers oversell him then the similar assumption should be that the 1985-91 numbers undersell him.

Everyone can identify Hakeem’s career arc. It reflects most career arcs, albeit with a later true peak than average. So we take the career samples (or extended prime samples I guess), and those extended samples on balance have him as a top ~ten regular season candidate with more robust playoff maintenance/elevation and a longer extended prime than basically everyone ahead of him. Quantifying that precisely leaves a range, sure, but everyone can see what happens when the priority seems to be to take the lowest estimate at every opportunity.

I don’t know why those two assumptions would have to go hand in hand. The 1985-1992 “off” sample is actually almost 2.5x as large as the later “off” sample. So it’s perfectly reasonable to assume that the earlier numbers are probably more accurate than the later numbers. (Though, obviously, none of the sample sizes here are enormous, so there’s some uncertainty on both sides of the equation). And that’s especially the case when the earlier numbers are broadly consistent with the limited snippets of RAPM data we have, while the later numbers aren’t so much (though we don’t have anything from 1993-1995, so it’s just RAPM from the tail end of Hakeem’s peak).

And, notably, we are actually having a discussion *about the first 8 years of Hakeem’s career.* Using data from those actual years, Hakeem really doesn’t look all *that* great. Your point seems to basically be that we should look at what happened after that (i.e. at his peak) and essentially smooth it out such that we end up taking a better view of the earlier years of Hakeem’s career than the actual data from those years warrants. It’s not an entirely unreasonable view, I suppose. But nor is it unreasonable to say that we should measure how good Hakeem was in the various time periods of his career based on what actually happened in that time period and the data we have on it, rather than giving him credit in his earlier years for his later improvement. For what it’s worth, I rate Hakeem’s peak very highly. But I don’t think we need to use that to act like Hakeem was something he wasn’t in his earlier years.

[70sFan]

When 2 players are comparable, then longevity matters. When players are not in the same tier, longevity is largely irrelevant. I would never rank Stockton over D.Rob for example. If you want an even more obvious example, imagine comparing Jamison or Shareef to Bill Walton or Penny based on longevity.

This is why people create CORP or VORP evaluations that take into account longevity and value of each season. This methodology can be used as a baseline to get the idea of what comparison is reasonable and what is not.

The results I have seen (along with the ones I provided) suggest that Kobe and Curry is a legit comparison, Stockton and Robinson is probably slightly less close and the ones you mentioned are pointless indeed. Now, if you have a different method of evaluating longevity or a different weighing procedure, I am all ears. If your analysis is only about gut feeling, then I am not interested in that.

Isn’t CORP basically not really a metric at all, and basically just layering subjective evaluations onto a philosophical heuristic? Like, in theory, CORP is actually a valid way to think about things that balances weighing peak/prime and longevity, so I do tend to think it’s a very helpful heuristic for thinking about things. But what that method spits out is going to just be dependent on pretty subjective valuations of how much a player increases the chances to win a championship. It’s not a precise metric, but rather just a way of thinking about things. For example, I could very easily come to a conclusion that Steph Curry has already added more by a CORP methodology than Kobe did—in fact, I do think that that’s the case. But, of course, you could use the same heuristic and come to the opposite conclusion.

To some degree, you are right - but weighing criteria can't be completely arbitrary.

[70sFan]

For what it’s worth, I rate Hakeem’s peak very highly. But I don’t think we need to use that to act like Hakeem was something he wasn’t in his earlier years.

What do you think separates so much older Hakeem from the younger one then? Do you see any substantial improvements in his game?

[AEnigma]

The dispute is not that he improved. The dispute is that you want to use small samples to make one point and then not carry through with other similar samples. If the 1992-97 without numbers oversell him then the similar assumption should be that the 1985-91 numbers undersell him.

Everyone can identify Hakeem’s career arc. It reflects most career arcs, albeit with a later true peak than average. So we take the career samples (or extended prime samples I guess), and those extended samples on balance have him as a top ~ten regular season candidate with more robust playoff maintenance/elevation and a longer extended prime than basically everyone ahead of him. Quantifying that precisely leaves a range, sure, but everyone can see what happens when the priority seems to be to take the lowest estimate at every opportunity.

I don’t know why those two assumptions would have to go hand in hand. The 1985-1992 “off” sample is actually almost 2.5x as large as the later “off” sample. So it’s perfectly reasonable to assume that the earlier numbers are probably more accurate than the later numbers. (Though, obviously, none of the sample sizes here are enormous, so there’s some uncertainty on both sides of the equation).

Yes, and that larger sample is why when you look at the career mark as a whole, it is closer to that early sample. That is how averages work.

And that’s especially the case when the earlier numbers are broadly consistent with the limited snippets of RAPM data we have, while the later numbers aren’t so much (though we don’t have anything from 1993-1995, so it’s just RAPM from the tail end of Hakeem’s peak).

I do not find them remotely consistent, and until recently most people have known not to use RAPM built around a scattershot of games.

And, notably, we are actually having a discussion *about the first 8 years of Hakeem’s career.* Using data from those actual years, Hakeem really doesn’t look all *that* great. Your point seems to basically be that we should look at what happened after that (i.e. at his peak) and essentially smooth it out such that we end up taking a better view of the earlier years of Hakeem’s career than the actual data from those years warrants. It’s not an entirely unreasonable view, I suppose. But nor is it unreasonable to say that we should measure how good Hakeem was in the various time periods of his career based on what actually happened in that time period and the data we have on it, rather than giving him credit in his earlier years for his later improvement. For what it’s worth, I rate Hakeem’s peak very highly. But I don’t think we need to use that to act like Hakeem was something he wasn’t in his earlier years.

Which brings us back to: if Hakeem is “only” a +4 player at the start of his career, then those same indicators should tell you that he is in the ballpark of a +10 player one offseason later. Maybe that strikes you as a rational way to look at the sport. For most people, it is an indicator of why it is better to expand samples and use some critical thinking about what we see in isolated snippets of data. Or if we are not going to expand samples, we could look at the year to year variance instead and notice that, weirdly, the bulk of the samples you use from the start of his career are from two seasons specifically where the team performed at a level inconsistent with what we saw in all surrounding seasons. And on that note, you also have liked to push the idea that Hakeem’s impact is itself comparatively inflated on a bad team — without looking at that 1986 sample which sees him also posting +4 regular lift on a better regular season team.

Yet again, for whatever reason, none of those clarifications have been of interest to you.

[lessthanjake]

Yes, and that larger sample is why when you look at the career mark as a whole, it is closer to that early sample. That is how averages work.

I’d appreciate you cutting down a bit on the sarcasm/mocking (i.e. “That is how averages work.”).

Anyways, I understand that you can throw that in a career average and say what the career average says. But a career average doesn’t tell us how good Hakeem was in his first 8 seasons. It tells use how good he was on average across his career. I feel like you’re wanting to give Hakeem full credit for how good he was in his peak years, while simultaneously giving him credit for being at his career-average level in his non-peak years. That basically double counts the greatness of his peak level. He was great at his peak. But a career average that includes that great peak is going to overestimate how good he was in other years. Unless you want to *also* discount his peak as a result of the data from his earlier years, then the approach doesn’t make sense (or at least is fundamentally biased in Hakeem’s favor).

And that’s especially the case when the earlier numbers are broadly consistent with the limited snippets of RAPM data we have, while the later numbers aren’t so much (though we don’t have anything from 1993-1995, so it’s just RAPM from the tail end of Hakeem’s peak).

I do not find them remotely consistent, and until recently most people have known not to use RAPM built around a scattershot of games.

RAPM with a low sample size is noisy, but at the very least it can be helpful as a check on other data that also has a low sample size. Here, for Hakeem’s earlier years, we have WOWY data that suggests his impact wasn’t super high. We also have limited RAPM data that suggests his impact wasn’t super high. Both data points represent limited data, but when they both point in the same direction, it can make us more confident in what we’re seeing, as compared to just having one limited data point. Meanwhile, the WOWY data for Hakeem in his peak years is very good, but the limited RAPM data we have isn’t as good, so that should naturally lower our confidence in it as compared to what we have in the earlier years. Basically, at a fundamental level I agree with you that limited snippets of RAPM are a noisy indicator, but it’s also not completely meaningless, and can help give us a bit more or less confidence in other data.

Which brings us back to: if Hakeem is “only” a +4 player at the start of his career, then those same indicators should tell you that he is in the ballpark of a +10 player one offseason later. Maybe that strikes you as a rational way to look at the sport. For most people, it is an indicator of why it is better to expand samples and use some critical thinking about what we see in isolated snippets of data. Or if we are not going to expand samples, we could look at the year to year variance instead and notice that, weirdly, the bulk of the samples you use from the start of his career are from two seasons specifically where the team performed at a level inconsistent with what we saw in all surrounding seasons. And on that note, you also have liked to push the idea that Hakeem’s impact is itself comparatively inflated on a bad team — without looking at that 1986 sample which sees him also posting +4 regular lift on a better regular season team.

Yet again, for whatever reason, none of those clarifications have been of interest to you.

Again, I’d appreciate if you could shift to a friendlier tone (i.e. not things like “Yet again, for whatever reason, none of those clarifications have been of interest to you.”).

On the substance of this, I get the point you’re making. Yes, sample sizes aren’t big for this sort of data, so perhaps some amount of data smoothing could make sense. But at the same time, the underlying data points aren’t actually all coming from the same underlying full sample, or even necessarily a similar one.* Hakeem in his peak years is not the same player as Hakeem in his earlier years. So it’s not just a question of throwing all the data into an average. It’s a question of assessing how much of the difference in the data points is Hakeem having improved as a player and how much of it is just statistical randomness (the latter of which could actually go in either direction). We have very good reason to think that Hakeem had a lot more impact in his peak years—I don’t think anyone would say his teams were really a whole lot better than before, and yet he won a lot more games with them while being superior statistically at an individual level!

At its core, you’re ultimately just trying to use data from Hakeem’s later years to inform how well he played in his earlier years. And, as I noted above, unless you’re going to *also* use the earlier data to inform how good he was at his peak (which would require lowering our opinion of his peak), then I don’t think your approach makes sense.

Also, as a sidenote, the 1985-1986 Rockets were not a particularly good regular season team. They won 51 games, but only had a +2.1 SRS. I don’t think any “lift” compared to his 14 games off the court in that season show the kind of lift to high heights that I’m talking about. Of course, he did get them to the finals, so that was definitely impressive, and is by far the most impressive thing in his first 8 seasons.

_____________

* Of course, you make a similar point that a lot of the “off” data from Hakeem’s earlier years comes from years where the team was a bit better. And to some degree that’s true. But at the same time, a good portion of it also comes from a year they failed to even make the playoffs. And there’s also a giant 54-point loss in the off-sample that actually skews the off sample a lot in Hakeem’s favor. So it goes both ways. It’s also not like the Rockets were wildly different in quality in different years: Their win total hovered between 41 and 52 the entire time period. Just for your reference, if you weight Hakeem’s missed games by the team’s SRS the season he missed the games, it comes out to a weighted average of 1.58 SRS, as compared to an average overall SRS in the time period of 1.02 SRS. Hardly some enormous or highly meaningful difference.

[LukaTheGOAT]

Yes, and that larger sample is why when you look at the career mark as a whole, it is closer to that early sample. That is how averages work.

I’d appreciate you cutting down a bit on the sarcasm/mocking (i.e. “That is how averages work.”).

Anyways, I understand that you can throw that in a career average and say what the career average says. But a career average doesn’t tell us how good Hakeem was in his first 8 seasons. It tells use how good he was on average across his career. I feel like you’re wanting to give Hakeem full credit for how good he was in his peak years, while simultaneously giving him credit for being at his career-average level in his non-peak years. That basically double counts the greatness of his peak level. He was great at his peak. But a career average that includes that great peak is going to overestimate how good he was in other years. Unless you want to *also* discount his peak as a result of the data from his earlier years, then the approach doesn’t make sense (or at least is fundamentally biased in Hakeem’s favor).

And that’s especially the case when the earlier numbers are broadly consistent with the limited snippets of RAPM data we have, while the later numbers aren’t so much (though we don’t have anything from 1993-1995, so it’s just RAPM from the tail end of Hakeem’s peak).

I do not find them remotely consistent, and until recently most people have known not to use RAPM built around a scattershot of games.

RAPM with a low sample size is noisy, but at the very least it can be helpful as a check on other data that also has a low sample size. Here, for Hakeem’s earlier years, we have WOWY data that suggests his impact wasn’t super high. We also have limited RAPM data that suggests his impact wasn’t super high. Both data points represent limited data, but when they both point in the same direction, it can make us more confident in what we’re seeing, as compared to just having one limited data point. Meanwhile, the WOWY data for Hakeem in his peak years is very good, but the limited RAPM data we have isn’t as good, so that should naturally lower our confidence in it as compared to what we have in the earlier years. Basically, at a fundamental level I agree with you that limited snippets of RAPM are a noisy indicator, but it’s also not completely meaningless, and can help give us a bit more or less confidence in other data.

Which brings us back to: if Hakeem is “only” a +4 player at the start of his career, then those same indicators should tell you that he is in the ballpark of a +10 player one offseason later. Maybe that strikes you as a rational way to look at the sport. For most people, it is an indicator of why it is better to expand samples and use some critical thinking about what we see in isolated snippets of data. Or if we are not going to expand samples, we could look at the year to year variance instead and notice that, weirdly, the bulk of the samples you use from the start of his career are from two seasons specifically where the team performed at a level inconsistent with what we saw in all surrounding seasons. And on that note, you also have liked to push the idea that Hakeem’s impact is itself comparatively inflated on a bad team — without looking at that 1986 sample which sees him also posting +4 regular lift on a better regular season team.

Yet again, for whatever reason, none of those clarifications have been of interest to you.

Again, I’d appreciate if you could shift to a friendlier tone (i.e. not things like “Yet again, for whatever reason, none of those clarifications have been of interest to you.”).

On the substance of this, I get the point you’re making. Yes, sample sizes aren’t big for this sort of data, so perhaps some amount of data smoothing could make sense. But at the same time, the underlying data points aren’t actually all coming from the same underlying full sample, or even necessarily a similar one.* Hakeem in his peak years is not the same player as Hakeem in his earlier years. So it’s not just a question of throwing all the data into an average. It’s a question of assessing how much of the difference in the data points is Hakeem having improved as a player and how much of it is just statistical randomness (the latter of which could actually go in either direction). We have very good reason to think that Hakeem had a lot more impact in his peak years—I don’t think anyone would say his teams were really a whole lot better than before, and yet he won a lot more games with them while being superior statistically at an individual level!

At its core, you’re ultimately just trying to use data from Hakeem’s later years to inform how well he played in his earlier years. And, as I noted above, unless you’re going to *also* use the earlier data to inform how good he was at his peak (which would require lowering our opinion of his peak), then I don’t think your approach makes sense.

Also, as a sidenote, the 1985-1986 Rockets were not a particularly good regular season team. They won 51 games, but only had a +2.1 SRS. I don’t think any “lift” compared to his 14 games off the court in that season show the kind of lift to high heights that I’m talking about. Of course, he did get them to the finals, so that was definitely impressive, and is by far the most impressive thing in his first 8 seasons.

_____________

* Of course, you make a similar point that a lot of the “off” data from Hakeem’s earlier years comes from years where the team was a bit better. And to some degree that’s true. But at the same time, a good portion of it also comes from a year they failed to even make the playoffs. And there’s also a giant 54-point loss in the off-sample that actually skews the off sample a lot in Hakeem’s favor. So it goes both ways. It’s also not like the Rockets were wildly different in quality in different years: Their win total hovered between 41 and 52 the entire time period. Just for your reference, if you weight Hakeem’s missed games by the team’s SRS the season he missed the games, it comes out to a weighted average of 1.58 SRS, as compared to an average overall SRS in the time period of 1.02 SRS. Hardly some enormous or highly meaningful difference.

Okay, talking a look at things from a PS lense, what is the argument that Hakeem wasn't an MVP level player in the PS from 86-91 (no PS in 92)?

I've already posted the numbers, so I don't want to keep harping on how good he looks there compared to many years of prime Duncan.

[lessthanjake]

For what it’s worth, I rate Hakeem’s peak very highly. But I don’t think we need to use that to act like Hakeem was something he wasn’t in his earlier years.

What do you think separates so much older Hakeem from the younger one then? Do you see any substantial improvements in his game?

Well, I think the typical answer to this that you’d get from both me and others is that Hakeem became a better and more willing passer—that he was previously a bit of a black hole that shot way too much into double- and triple-teams and that in his peak years he’d rapidly developed into a much more willing passer. And we do see that in his stats, with Hakeem having an immediate jump to previously-unseen assist totals starting in the 1992-1993 season (and I’ll note that this jumped by even greater amounts in the playoffs in those peak years—roughly doubling his prior playoff assist rate, evidencing a definite change in approach IMO), which was also accompanied by an increase in his scoring efficiency. Of course, I’m sure there’s also plenty of subtle things that would be hard to specifically identify—such as mental stuff, preparation for opponents, or really subtle things he’s doing on the court that opponents would’ve recognized but that we can’t easily identify on TV or without a really fine-tooth-combed film analysis of a ton of games (i.e. the type of little things that veterans get better at over time).

[lessthanjake]

Okay, talking a look at things from a PS lense, what is the argument that Hakeem wasn't an MVP level player in the PS from 86-91 (no PS in 92)?

I've already posted the numbers, so I don't want to keep harping on how good he looks there compared to many years of prime Duncan.

I think the argument is that you can’t really look like an MVP-level player just based on the first round of the playoffs, and almost all those years ended in first-round losses. And those included losses against a few pretty mediocre teams that did not have a player nearly as good as Hakeem.

Again, the comparator I give—which you might have strong views on—is Luka Doncic. Have Luka’s playoff performances made him an MVP-level player the last 4 years? They’ve certainly been good, and he did manage to overachieve by upsetting a much better team once, but IMO he’s not really gone consistently far enough in the playoffs for us to really say he’s been a truly MVP-level player based on playoff performance.

IMO, Luka’s a good comparison for more than just that. Luka and early Hakeem both have really good box-score stats but not the best in the league, have relatively mediocre impact data while on middling teams, are statistical playoff risers but have rarely left the first round and have even failed to make the playoffs, and managed to make one overachieving playoff run where they beat a much better team. And in both cases, they typically landed in the bottom half of the top 10 in MVP voting. And I think that’s roughly right for both of them. Though, given your username, perhaps you disagree!

Now, to be clear, I wouldn’t be surprised at all if Luka, like Hakeem, finds another level and *does* win MVP and win titles eventually. He’s young. I just think that so far Luka’s early career is shaking out similarly to Hakeem’s early career, and so I see early Hakeem as analogous to Luka.

[AEnigma]

At its core, you’re ultimately just trying to use data from Hakeem’s later years to inform how well he played in his earlier years. And, as I noted above, unless you’re going to *also* use the earlier data to inform how good he was at his peak (which would require lowering our opinion of his peak), then I don’t think your approach makes sense.

It is extraordinarily difficult for me to not see this as a deliberate misrepresentation of everything I am writing.

This is what I wrote:

The only way that theory survives even the slightest scrutiny is if you assume Hakeem took a massive leap around 1992/93 right as the rest of the Rockets roster collapsed.

1985-91 Rockets with Hakeem: +2 net rating
1984-91 Rockets without Hakeem: -2.1 net rating

1992-97 Rockets with Hakeem: +3.5 net rating
1992-97 Rockets without Hakeem: -6.3 net rating

People who understand player development rightly recognise that is in fact not what happened, but anyone who sincerely believes Hakeem went from a +4 regular season player to a +10 one in a span of an offseason, before factoring in playoff elevation, should probably also be pushing Hakeem as a top three or two or one sustained peak, and on that basis alone he would merit reentry into this fringe top five discussion.

I am bolding the key part to make it sink in: I do not think Hakeem magically made that type of a jump. I alluded to as much with this response:

The dispute is not that he improved. The dispute is that you want to use small samples to make one point and then not carry through with other similar samples. If the 1992-97 without numbers oversell him then the similar assumption should be that the 1985-91 numbers undersell him.

Everyone can identify Hakeem’s career arc. It reflects most career arcs, albeit with a later true peak than average.

If you want to throw out development curves and take those indicators at face value, fine. However, you are not. What you are doing is hammering that early sample as representative, but then glibly cruising past the later sample because it is “noisier” and thus must be some sort of overestimation. The latter sample is noisier, yes, but correcting for that noise takes us back to the curve, where the worst Hakeem samples are not as bad as they appear (nor as they as good). You claim I am taking the good samples outright and then curving up the bad samples, but you do so with absolutely no basis.

In contrast, time and time again, you push the worst interpretations of Hakeem here while trying to curve down his best interpretations. Such as trying to equate a 12-game without sample in 1992 and a single large blowout as a case of “going both ways” with a 14-game without sample featuring the best cast of Hakeem’s pre-Drexler career (which you chide for only being 2 SRS across the season with him missing those games) and with a 26-game without sample where the team performed unusually well.

I twice sought to address the issue with that approach by showing how essentially it was the sole 1991 sample distorting everything, and that too was rejected and ignored:

The 1984-86 Rockets] won 37.5% of their games without [Hakeem] and 61.3% of their games with him, so unless you think he got worse after his sophomore year, that too undersells him.

1986: 64.7% win rate and +3.31 net rating with, 50% win rate and -0.79 net rating without; total difference of 14.7% win rate and 4.1 net rating

1987: 53.3% win rate and +2.12 net rating with, 28.6% win rate and -11 win rate without; total difference of 24.7% win rate and 13.12 net rating

1988: 57% win rate and +1.63 net rating with, 33.3% win rate and -5 net rating without; total difference of 23.7% win rate and 6.63 net rating

1992: 57.1% win rate and -0.2 net rating with, 16.7% win rate and -11.4 net rating without; total difference of 40.4% win rate and 11.2 net rating

For net rating purposes 1984 does not even really matter, because the Rockets were -2.9 that year and the next eight years they were -2.8 without Hakeem, but it sure as hell matters when you build this around win percentage: it changes the sample from “45%” without to 39.6%, right in line with their expected net rating. The 1985-92 Rockets were +1.8 with Hakeem, so just by net rating that is a +4.6 shift roughly taking a 32.5-win team to a 46-win team. Is that incredible regular season lift, no, it is not Lebron or Minnesota Garnett tier, but it is a lot rarer over a sustained pre-peak sample than you seem to be crediting. And as far as the hypothesis that amount of lift precludes him from showcasing high “impact”, well, it is right on par with Lakers Shaq, and that iteration was not exactly struggling to show up atop RAPM leaderboards.

You talk about not curving, but your entire complaint comes from a curve overweighing 1991.

1985-90 Rockets with Hakeem: +1.8
1985-90 Rockets without Hakeem: -4.3

You talk about willingly accepting raw impact indicators even if they might not fit a curve, yet you do not specifically highlight 1991 as some singularly awful year.

And perhaps most crucially, you have been asked multiple times to articulate what exactly Hakeem did to produce this monumental change in his real impact and have generally dodged the question beyond vaguely gesturing at some increase in offensive production and box score aggregates (which are in no way proportional to the “impact” shift you are backing). There needs to be some anchoring in reality here, and it needs to go beyond dismissing people for “groupthink” because a bunch of us see all-time defensive campaigns in 1989/90 that have never really been replicated since.

So the question remains: how is any of this approach remotely balanced?

[homecourtloss]

Following on with some more information, this time from the GitHub RAPM data (which ends at 2018-2019, which is why I don’t include years beyond that in the timeframes):

Average Regular Season GitHub RAPM in Prime Years

- Stephen Curry (2013-2014 to 2018-2019): 5.482
- Michael Jordan (1996-1997 to 1997-1998): 4.335
- Kevin Garnett (1999-2000 to 2007-2008): 4.096
- LeBron James (2008-2009 to 2018-2019): 3.881
- Tim Duncan (1998-1999 to 2006-2007): 3.530
- Shaquille O’Neal (1997-1998 to 2004-2005): 2.365

There’s also playoff RAPM data. The playoff RAPM appears to be on a different scale for whatever reason, so the numbers are all lower:

Average Playoff GitHub RAPM in Prime Years

- Michael Jordan (1996-1997 to 1997-1998): 2.527
- Stephen Curry (2013-2014 to 2018-2019): 2.132
- LeBron James (2008-2009 to 2017-2018): 1.713
- Tim Duncan (1998-1999 to 2006-2007): 0.925
- Shaquille O’Neal (1997-1998 to 2004-2005): 0.621
- Kevin Garnett (1999-2000 to 2007-2008): 0.612

This version of RAPM is a bit wonky and I know some don’t like it, but it’s one of the RAPM measures we have and it seems worth posting how the above players come out in their primes. As with the AuPM/g and RPM measures I posted earlier, Steph Curry comes off looking extremely good.

(Note: Again, I didn’t include Hakeem here, since he wasn’t really in his prime by the 1996-1997 season, which is when the data starts. But in case there’s any curiosity, Hakeem had a regular season average of 2.395 in 1996-1997 to 1997-1998, and an average playoff RAPM of 0.170 in that timeframe).

Even though we might get some similar numbers if we calculated the real RAPM over these prime spans, I wanted to point out that you can’t average out or RAPM numbers because the different amount of possessions, the different priors used in each season, and so on. I’m assuming you didn’t calculate out the true RAPM over these years as that is a massive undertaking and requires a lot of data not readily available unless one is running some sort of script and scrubbing that data.

Regardless, Steph Curry has been an incredible impact engine, and that type of impact is very difficult to ignore, and I am looking at him near the top after the top 12 or so players. I also see his game aging well over the next few years, at which point, it might be impossible to keep them out of the top 10. But I have a question here – whether you look at the GitHub RAPM or shotcharts RAPM (the two you seem to use), Draymond Green comes out looking extraordinarily well. How do you reconcile that with Curry’s impact? I know there has been previous discussion about collinearity and other factors, but Draymond Green‘s primary impact has been on defense, different than Curry’s primary impact on offense. The only other duo that comes close to this type of overall impact over such a long stretch of time it’s probably Kareem and Magic Johnson but we really don’t have data for them. We can look at 2020 and say that Draymond falls off but in many ways that seems like a throwaway season

If we look at Engelmann’s 1997–2022 PI RAPM (RS+PS), Draymond is 12th, most of his monstrous impact coming on defense (basically only Kevin Garnett, Dikembe, Mutombo, and Ben Wallace above him with a few others close by), but being a sizable positive on offense as well. do have some issues with how Draymond would fare without curry, his positive value towards winning margins is generally coming on defense.

If we look at Engelmann’s Single Season 1997–2019 PI RAPM (RS+PS), you have Draymond’s 2016 at number 8, 2015 at 78, 2017 at 195.

If we look at Cheema’s 1997-2022 RAPM 5 year peaks (RS+PS), you have Draymond with seven 5-year stretches in the top 320, one of them at 60, and four from 107 to 118 out of 16,000+, ahead of several Kobe, Steve Nash, Nikola Jokić, Dirk, Kawhi, Leonard, Chris, Paul, Tim Duncan, etc., stretches.

If we want to use the wonky GitLab data, we have Draymond at 22nd/30th in the RS/PS in 2014, 5th/1st in 2015, 1st/1st in 2016, 6th/2nd in 2017, 40th/2nd in 2018, 32nd/6th in 2019.

If we want to use NBA ShotCharts RAPM, we have Dray at 48th in 2014, 3rd in 2015, 1st in 2016, 4th in 2017, 49th in 2018, 23rd in 2019.

Will end this message from LukaTheGOAT that involves some of his AuPM numbers and overall PS rise:

Too little credit on defense-Draymond Green

From 2015-2020, Draymond Green is 2nd in the NBA in Playoffs PIPM
https://www.reddit.com/r/nba/comments/j9xodj/from_20152020_playoffs_the_highest_rated_players/

This is significant because PIPM is too box-score dependent, yet someone like Draymond comes out looking so good in the metric, despite so much of what he does not showing up in the box-score.

It is not just PIPM of box-hybrid models that that are high on Draymond. From 2014-2019, Draymond lead the NBA in PS RAPTOR WAR.

From 15-17, Draymond is 2nd in PS AuPM/G.

And when you consider that Golden State's defense improves from the RS to PS more than almost any dynasty ever, I think it makes sense to look towards Draymond for a lot of Golden State's success.

Watch on YouTube

As a matter of fact, RAPTOR projections considered Draymond to be the NBA player who improved most from the RS to PS in the NBA during that time frame at a whopping 1.4 points per 100 possessions. The next most improved player was Lebron who was at 0.9 pts per 100 possessions. https://fivethirtyeight.com/features/how-our-raptor-metric-works/

According to AuPM/G, which has data going back to 96-97, no player has improved more in from the RS to PS in their career than Draymond Green. As mentioned in the article, "among players with at least five qualifying runs, Green has the largest improvement in AuPM history. And this isn’t from slow-rolling the regular season either. In the seven seasons he’s played in the postseason, Green’s posted a hefty +3.5 AuPM per game in the regular season and then a whopping +4.7 in the playoffs. That’s like going from the sixth-best player in the league to the second." https://backpicks.com/page/6/

Kevin Pelton also wrote an article about how Draymond was statistically the 2nd biggest playoff riser during some specific time period, but I cannot find it

If you want numbers that look at the pure plus-minus side of things (and does not include anything pertaining to the box-score), I should note, Draymond looks arguably better...

Draymond is #1 in 14-18 PS RAPM, and #1 in 15-19 PS RAPM.

If we know GSW's offense declines in the PS, but their defense makes one of the biggest improvements ever, and we know that Draymond has been the captain of those GSW offenses, and all the data we have suggests he is among the biggest improvers in performance come PS time, I will put my money on him

[lessthanjake]

Even though we might get some similar numbers if we calculated the real RAPM over these prime spans, I wanted to point out that you can’t average out or RAPM numbers because the different amount of possessions, the different priors used in each season, and so on. I’m assuming you didn’t calculate out the true RAPM over these years as that is a massive undertaking and requires a lot of data not readily available unless one is running some sort of script and scrubbing that data.

Yeah, I know that, and it’s something I’ve mentioned before and should’ve caveated again here. There’s no true apples-to-apples comparison though, so it’s largely the best we have in this regard, when comparing impact data across eras. And the biggest point I’ve been making is that, *in the same seasons,* prime Steph’s impact data was superior even to prime LeBron James. So it’s not like Steph is lacking in impact data that’s impressive outside of cross-era RAPM comparisons. I’d argue there’s no one who is more impressive in that regard, since prime Steph was regularly outdoing this board’s #1 all time (and my personal #2 all time) in his prime.

Regardless, Steph Curry has been an incredible impact engine, and that type of impact is very difficult to ignore, and I am looking at him near the top after the top 12 or so players. I also see his game aging well over the next few years, at which point, it might be impossible to keep them out of the top 10. But I have a question here – whether you look at the GitHub RAPM or shotcharts RAPM (the two you seem to use), Draymond Green comes out looking extraordinarily well. How do you reconcile that with Curry’s impact? I know there has been previous discussion about collinearity and other factors, but Draymond Green‘s primary impact has been on defense, different than Curry’s primary impact on offense. The only other duo that comes close to this type of overall impact over such a long stretch of time it’s probably Kareem and Magic Johnson but we really don’t have data for them. We can look at 2020 and say that Draymond falls off but in many ways that seems like a throwaway season

If we look at Engelmann’s 1997–2022 PI RAPM (RS+PS), Draymond is 12th, most of his monstrous impact coming on defense (basically only Kevin Garnett, Dikembe, Mutombo, and Ben Wallace above him with a few others close by), but being a sizable positive on offense as well. do have some issues with how Draymond would fare without curry, his positive value towards winning margins is generally coming on defense.

If we look at Engelmann’s Single Season 1997–2019 PI RAPM (RS+PS), you have Draymond’s 2016 at number 8, 2015 at 78, 2017 at 195.

If we look at Cheema’s 1997-2022 RAPM 5 year peaks (RS+PS), you have Draymond with seven 5-year stretches in the top 320, one of them at 60, and four from 107 to 118 out of 16,000+, ahead of several Kobe, Steve Nash, Nikola Jokić, Dirk, Kawhi, Leonard, Chris, Paul, Tim Duncan, etc., stretches.

If we want to use the wonky GitLab data, we have Draymond at 22nd/30th in the RS/PS in 2014, 5th/1st in 2015, 1st/1st in 2016, 6th/2nd in 2017, 40th/2nd in 2018, 32nd/6th in 2019.

If we want to use NBA ShotCharts RAPM, we have Dray at 48th in 2014, 3rd in 2015, 1st in 2016, 4th in 2017, 49th in 2018, 23rd in 2019.

I think the answer to this is that Draymond Green is a really good basketball player, who also had high impact, albeit less than Steph. And that’s not surprising. At their peak, the Warriors averaged a 68-win pace when Steph played, over a 5 year span. As good as Steph is, and as much as they were only fairly average when he was out, you don’t win at a 68-win pace without having anyone else that’s quite impactful. Indeed, even them being a .500 team without Steph is suggestive of having other players that had solid impact! Obviously, for a time that included Durant, but they had two unreal seasons even without Durant. That has to require more than *just* Steph! And, as good as Draymond is (and I suspect he might even push up around #75 in this project), I think we should look at teams that had anything like the kind of success the Warriors had even without Durant (i.e. multiple titles, 67 and 73 wins in two seasons, etc.) and realize that the #2 player on teams like that are usually, frankly, better than Draymond: Guys like Scottie Pippen, Oscar Robertson, Wilt Chamberlain, Kevin McHale, Kareem, etc. Draymond is a great player and fits very well with Steph, but I don’t really think Draymond being great mitigates what Steph has done, because Steph has done stuff that no one in history has done without a player of *at least* Draymond’s level. In other words, I just don’t see “Yeah, but he had Draymond” as a valid counterargument to Steph’s achievements, when those achievements are things like multiple titles and 140 wins in two seasons. As good as Draymond is, most of the top players in history have played with people who are as good or better, and they didn’t do some of the things Steph’s Warriors did! And I don’t think there’s any great player in history—and certainly not those that are left in this project—with whom we can’t squint at certain data and have one of their great teammates come out looking like a huge engine of the team’s success (see, for instance, Ginobili with Duncan).

To the extent there’s an argument that Draymond has *more* impact than Steph (which I don’t think is the point you’re making), I’d say that’s not supported overall by the metrics we have—in which Steph comes out overall as the clear standout of his generation, and Draymond comes out looking like a very impactful player but not at the same level. Relatedly, we have some pretty strong indicators of which one of them had the most impact. As I noted in prior posts, in the dynasty years, in regular season + playoffs, the Warriors’ net rating with Steph on the court and Draymond, Durant, and Klay ALL off the court was better than the Warriors’ net rating with Draymond, Durant, and Klay ALL on the court and Steph off the court. That is remarkable! And since 2013-2014, the Warriors have a +6.83 net rating with Steph on and Draymond off, and just a +0.80 net rating with Draymond on and Steph off (the gap is a bit less if you take out 2019-2020, but is still substantial). When Draymond was the best player on the Warriors in 2019-2020, they were the worst team in the league, and were barely any better when he was on the court. I think it’s overall clear which one is *more* impactful.

Overall, if the point is just that Steph had another really good player on his team, my response is to agree but to invite people to consider whether they truly think Draymond Green is *so* extraordinary as to render it at all sensical that the Warriors achieved the ridiculous level of success they’ve achieved—in both regular season and playoffs—without Steph needing to be one of the top several greatest players of all time.

________

EDIT: One tiny thing I want to address is that I don’t agree with the notion that there’s not collinearity issues when assessing Draymond’s impact because he affects defense and Steph affects offense. Basketball is a fluid game, where both sides of the ball have a significant impact on the other because it is much easier to score off of a missed shot or turnover, and where teams can consciously choose to put more or less defense-minded lineups on the floor, and players have limited energy that they use in a game. If Steph and Draymond are on the floor and the defense is great, how much of this is Draymond being a great defender, and how much of it is Steph improving the offense and thereby making it easier on the defense. How much of it is that the guy guarding Steph is probably a useless husk on offense from running around screens like a maniac trying to stay with Steph? And how much of it is that, with Steph, the team can run lineups that would not normally be playable offensively and still get okay (or even good) offensive results? After all, the Warriors are one of the only teams regularly running multiple non-shooters on the floor at once these days. One of those non-shooters is Draymond. And things like Looney/Draymond lineups are great defensively, but wouldn’t be playable offensively if they didn’t have Steph. If Steph wasn’t so good offensively, such that you couldn’t run that Looney/Draymond lineup, you’d need to have Draymond at the 5 spot to keep him in the game, and that wouldn’t be as good defensively (though of course, when they did the death lineup with Durant it was so good offensively that it didn’t matter). So I don’t think it makes sense to isolate out the team’s offense and defense and say Steph is responsible for one and Draymond for the other. That’s not really how it works, and Steph has a huge effect on the team’s defense. And let’s remember that the Warriors had a bad defense in 2019-2020, even with Draymond on the floor.

[AEnigma]

Draymond Green comes out looking extraordinarily well. How do you reconcile that with Curry’s impact?

2015-23 Postseason Warriors
Draymond without Steph: +8.43, 1144 minutes
Steph without Draymond: +2.47, 847 minutes

Draymond without Steph or Durant: +8.62, 826 minutes
Steph without Draymond or Durant: +1.02, 682 minutes

Draymond without Steph or Klay: +20.99, 122 minutes
Steph without Draymond or Klay: +3.20, 346 minutes

Pretty small samples there so…
Draymond and Klay without Steph: +6.83, 1022 minutes
Steph and Klay without Draymond: +1.93, 502 minutes

It has been rightly pointed out this is partially skewed by easier opponents missed by Steph. We can adjust accordingly:

Draymond without Steph in shared games: +3.05, 699 minutes
Steph without Draymond in shared games: +1.80, 770 minutes

Draymond without Steph or Klay in shared games: +15.93, 92 minutes
Steph without Draymond or Klay in shared games: +1.58, 327 minutes
Draymond and Klay without Steph in shared games: +0.89, 718 minutes
Steph and Klay without Draymond in shared games: +1.79, 441 minutes

1998-2019 Playoff RAPM: Draymond = +7.66 (tied #1), Steph = +4.38 (#10)

Time and time again we see that in the postseason it is the symbiosis of these two that carries through.

[homecourtloss]

Yeah, I know that, and it’s something I’ve mentioned before and should’ve caveated again here. There’s no true apples-to-apples comparison though, so it’s largely the best we have in this regard, when comparing impact data across eras.

If you know that then I don’t think posting a list of “average RAPM during prime” is very helpful, because that “average” does not really exist. What you can do is list average rankings during a year, which, of course, Steph Curry would look very good in as well. And no, it’s not the best we have. Even if we say that there are no purely true apples to apples comparisons, we have the closest for the databall age, and they have been calculated for us in five year segments (RS+PS) in Cheema’s set, and for just the regular season in the NBA Shot Charts data. Why would anyone list out incorrect averages that don’t exist rather than post the multiple five-year intervals that coincide with the players’ respective primes?

I think the answer to this is that Draymond Green is a really good basketball player, who also had high impact, albeit less than Steph. And that’s not surprising. At their peak, the Warriors averaged a 68-win pace when Steph played, over a 5 year span. As good as Steph is, and as much as they were only fairly average when he was out, you don’t win at a 68-win pace without having anyone else that’s quite impactful.

I agree, but the discussion is about just how good Draymond is, and how much of his impact we can tease out from Steph Curry’s; obviously, it can’t be done 100% satisfactorily, but what we can do is look at other players to see if they had a teammate who had the type of impact that Draymond has had, and in the Databall age, nobody has. We can say older David Robinson paired with young Tim Duncan being one, though for a shorter amount of time, and then Tim Duncan with Manu Ginobili being another, but that pairing does not compare to the length along with heights that Draymond reaches with Curry or Dray on his own in 2016 for example. So, if we want to list out these numbers, then we also have to look at the fact that Draymond Green grades out as more impactful over a long period of time than even most players thought to be driving engines for their respective teams’ success, including over noisy playoff data, including in 2016 when he spent the lowest percentage of his playoff minutes paired with Steph Curry but had his best post season along with 2017. Now, this doesn’t mean the Draymond is “better“ than those other players, but with this amount of data and the number of possessions, it’s something to think about. And, if the postseason lineup data is posted, well, Dray grades out, extremely, extremely well.

2015-23 Postseason Warriors
Draymond without Steph: +8.43, 1144 minutes
Steph without Draymond: +2.47, 847 minutes

Draymond without Steph or Durant: +8.62, 826 minutes
Steph without Draymond or Durant: +1.02, 682 minutes

Draymond without Steph or Klay: +20.99, 122 minutes
Steph without Draymond or Klay: +3.20, 346 minutes

Pretty small samples there so…
Draymond and Klay without Steph: +6.83, 1022 minutes
Steph and Klay without Draymond: +1.93, 502 minutes

It has been rightly pointed out this is partially skewed by easier opponents missed by Steph. We can adjust accordingly:

Draymond without Steph in shared games: +3.05, 699 minutes
Steph without Draymond in shared games: +1.80, 770 minutes

Draymond without Steph or Klay in shared games: +15.93, 92 minutes
Steph without Draymond or Klay in shared games: +1.58, 327 minutes
Draymond and Klay without Steph in shared games: +0.89, 718 minutes
Steph and Klay without Draymond in shared games: +1.79, 441 minutes

1998-2019 Playoff RAPM: Draymond = +7.66 (tied #1), Steph = +4.38 (#10)

Time and time again we see that in the postseason it is the symbiosis of these two that carries through.

To the extent there’s an argument that Draymond has *more* impact than Steph (which I don’t think is the point you’re making), I’d say that’s not supported overall by the metrics we have—in which Steph comes out overall as the clear standout of his generation, and Draymond comes out looking like a very impactful player but not at the same level.

I don’t think anyone is really arguing he’s at the same level, but if we are looking at these numbers, then there is something that Draymond is doing that allows him to have this type of impact, which, if we are going to give weight to these numbers, then we also have to give pause before we say that “prime Steph was regularly outdoing this board’s #1 all time (and my personal #2 all time) in his prime,” or that Curry is the “true king of the databall age,” when the other player doesn’t have any teammate remotely close to what Draymond was doing (if we’re using these numbers in our arguments). Also, as mentioned before, Curry wasn’t “outdoing this board’s #1” in Englemann’s set, Cheema’s set or Taylor’s AuPM sets. For whatever reason, you don’t address those sets, but only refer to the GitLab and ShotCharts RAPM.

Relatedly, we have some pretty strong indicators of which one of them had the most impact. As I noted in prior posts, in the dynasty years, in regular season + playoffs, the Warriors’ net rating with Steph on the court and Draymond, Durant, and Klay ALL off the court was better than the Warriors’ net rating with Draymond, Durant, and Klay ALL on the court and Steph off the court. That is remarkable! And since 2013-2014, the Warriors have a +6.83 net rating with Steph on and Draymond off, and just a +0.80 net rating with Draymond on and Steph off (the gap is a bit less if you take out 2019-2020, but is still substantial). When Draymond was the best player on the Warriors in 2019-2020, they were the worst team in the league, and were barely any better when he was on the court. I think it’s overall clear which one is *more* impactful.

No disagreements here — Steph Curry was the driving force and the most impactful player on those teams 2017–2019 teams. But one of the things that we inherently have to accept if we are going to post RAPM numbers in arguments (provided we understand how those numbers are generated) is that RAPM numbers, provided that we have enough data, are superior to raw on/off numbers as far as teasing out where impact is coming from as RAPM numbers will smooth out line up irregularities and lineup noise as much as is possible— raw on/off numbers will not. And, in those RAPM numbers, Draymond shines.

EDIT: One tiny thing I want to address is that I don’t agree with the notion that there’s not collinearity issues when assessing Draymond’s impact because he affects defense and Steph affects offense.

I didn’t say that there were no collinearity issues — we’ve been discussing about these issues for a decade since these statistics first came out. What I did say was that “there has been previous discussion about collinearity and other factors, but Draymond Green‘s primary impact has been on defense, different than Curry’s primary impact on offense.”

Basketball is a fluid game, where both sides of the ball have a significant impact on the other because it is much easier to score off of a missed shot or turnover, and where teams can consciously choose to put more or less defense-minded lineups on the floor, and players have limited energy that they use in a game. If Steph and Draymond are on the floor and the defense is great, how much of this is Draymond being a great defender, and how much of it is Steph improving the offense and thereby making it easier on the defense. How much of it is that the guy guarding Steph is probably a useless husk on offense from running around screens like a maniac trying to stay with Steph? And how much of it is that, with Steph, the team can run lineups that would not normally be playable offensively and still get okay (or even good) offensive results? After all, the Warriors are one of the only teams regularly running multiple non-shooters on the floor at once these days. One of those non-shooters is Draymond. And things like Looney/Draymond lineups are great defensively, but wouldn’t be playable offensively if they didn’t have Steph. If Steph wasn’t so good offensively, such that you couldn’t run that Looney/Draymond lineup, you’d need to have Draymond at the 5 spot to keep him in the game, and that wouldn’t be as good defensively (though of course, when they did the death lineup with Durant it was so good offensively that it didn’t matter). So I don’t think it makes sense to isolate out the team’s offense and defense and say Steph is responsible for one and Draymond for the other. That’s not really how it works, and Steph has a huge effect on the team’s defense. And let’s remember that the Warriors had a bad defense in 2019-2020, even with Draymond on the floor.

First bolded: yes, we know that and they have been posts on this board related to that and how much we can tease out from there.

Second bolded: I’m not saying that’s how it works, which is a strawman argument that you are creating here. We know that there is a relationship between offense and defense. One of the things that RAPM does well if given enough possession data is illustrate impact, both offensive and defensive, isolated from other factors as much as possible, which is not possible to isolate completely since the game is fluid. But if you have two players, in this instance, who have shared a great chunk of their minutes together, and a 10 year span of RAPM data for these two players and has one player with an almost +5 advantage in ORAPM and the other with a +5 vantage in DRAPM, well, even with the given fluidity of the game, even with how multiple factors affect both an offense and a defense (such as you have mentioned above—Curry’s unique skills allow to put defensive parts around him and be successful), there is enough proof with the great deal of certainty that these players have impacted the game in different ways.

[lessthanjake]

I don’t have time to respond to all of this at the moment, so I may come back later to some of it, but for now just a few quick responses:

I don’t think anyone is really arguing he’s at the same level, but if we are looking at these numbers, then there is something that Draymond is doing that allows him to have this type of impact, which, if we are going to give weight to these numbers, then we also have to give pause before we say that “prime Steph was regularly outdoing this board’s #1 all time (and my personal #2 all time) in his prime,”or that Curry is the “true king of the databall age,” when the other player doesn’t have any teammate remotely close to what Draymond was doing (if we’re using these numbers in our arguments). Also, as mentioned before, Curry wasn’t “outdoing this board’s #1” in Englemann’s set, Cheema’s set or Taylor’s AuPM sets. For whatever reason, you don’t address those sets, but only refer to the GitLab and ShotCharts RAPM.

That’s not really true. I’ve actually posted Taylor’s AuPM/g numbers—which have Steph Curry leading the league every season for 6 straight years. And I’ve addressed Cheema’s numbers, which have Curry leading the NBA in two five-year spans during LeBron James’s prime. Not sure what the Englemann numbers are, but the other two support my point!

First bolded: yes, we know that and they have been posts on this board related to that and how much we can tease out from there.

Second bolded: I’m not saying that’s how it works, which is a strawman argument that you are creating here. We know that there is a relationship between offense and defense. One of the things that RAPM does well if given enough possession data is illustrate impact, both offensive and defensive, isolated from other factors as much as possible, which is not possible to isolate completely since the game is fluid. But if you have two players, in this instance, who have shared a great chunk of their minutes together, and a 10 year span of RAPM data for these two players and has one player with an almost +5 advantage in ORAPM and the other with a +5 vantage in DRAPM, well, even with the given fluidity of the game, even with how multiple factors affect both an offense and a defense (such as you have mentioned above—Curry’s unique skills allow to put defensive parts around him and be successful), there is enough proof with the great deal of certainty that these players have impacted the game in different ways.

I’m not sure RAPM actually can properly deal with some of the effects I’m talking about, at least not to the extent of properly figuring out whether an effect should be part of ORAPM or DRAPM. If Steph Curry allows his team to play a more defensive lineup, then I don’t think the effect of that will show up for Curry in terms of DRAPM, because the measure will control for the defensive quality of the players on the court with him. And that defensive quality that is being controlled for will simply be better! RAPM is trying to isolate out what someone does on the court, and is specifically designed to *not* give them credit for what lineups they play with. Of course, the flip side of this that it’s controlling for the defensive lineups being not so great offensively, so it’s not necessarily something that negatively affects the overall RAPM number (which Curry always does great in anyways), but I do think RAPM will naturally underrate a player’s defensive effect when the effect is caused by them opening up their team to play more defensive lineups.

If you know that then I don’t think posting a list of “average RAPM during prime” is very helpful, because that “average” does not really exist. What you can do is list average rankings during a year, which, of course, Steph Curry would look very good in as well. And no, it’s not the best we have. Even if we say that there are no purely true apples to apples comparisons, we have the closest for the databall age, and they have been calculated for us in five year segments (RS+PS) in Cheema’s set, and for just the regular season in the NBA Shot Charts data. Why would anyone list out incorrect averages that don’t exist rather than post the multiple five-year intervals that coincide with the players’ respective primes?

But I’ve also posted about the Cheema and NBAShotCharts stuff—where Steph comes out #1 in the league in multiple five-year intervals in both. And I’ve posted Steph’s rankings in individual years and in 3-year and 5-year spans that we have that data for. It all looks very good for Steph. And indeed, that’s the sort of data I began with—including comparing how he stacked up specifically against his contemporary LeBron James in the various different data sets and time intervals. This is why my argument is in large part centered around Steph having outstripped LeBron in impact metrics—because that’s the most direct comparison point. It gets harder to compare things, though, when comparing against players who didn’t play in the same intervals or years as Steph, because the available data upon which to make an apples-to-apples comparison is much more limited. There’s no real way to properly compare prime-years RAPM for players who played in different eras. If you want to object that taking averages across metrics like RPM is flawed, then I welcome you to do that. It’s a valid criticism! But there’s not a whole lot else to look at for what I was trying to get at—so a flawed measure is what I went with. Perhaps you think it’s so flawed that it’s worse than nothing. Not sure I agree, but I’m fine with you making that assertion.

At the very least, we could go through the same measures I posted and look at players’ league rankings by year and Steph would look equally as good compared to these other players.

[homecourtloss]

I don’t have time to respond to all of this at the moment, so I may come back later to some of it, but for now just a few quick responses:

That’s not really true. I’ve actually posted Taylor’s AuPM/g numbers—which have Steph Curry leading the league every season for 6 straight years. And I’ve addressed Cheema’s numbers, which have Curry leading the NBA in two five-year spans during LeBron James’s prime. Not sure what the Englemann numbers are, but the other two support my point!

Leading in two 5 year spans doesn’t make Curry “the king of the databall era” when the highest 5 year spans belong to Lebron (as well as the highest PS only spans) and include some Curry’s best years, i.e., 2012-2016, and 2013-2017. 2014-2018 Curry was higher than 2014-2018 LeBron, but not higher than three separate 5 year spans by LeBron, i.e., 2006-2010, 2012-2016, 2013-2017. Lebron also has 8 of the top 23 five-year stretches, 10 of the top 30. Curry has 4 of the top 30. Not sure how you can say “regularly” when 2017-21 LeBron is lightly ahead of 2017-2021 Curry and 2016-2020 LeBron is ahead of every Curry stretch other than 2014-2018.

Curry does really well in the AuPM numbers, but James sits atop those numbers so not sure how that supports “king of the databall era.”

Not sure what the Englemann numbers are, but the other two support my point!

They were posted in this very thread 3 posts up in a post you replied to as well as posted other times by other posters that you seem to never respond to.

If we look at Engelmann’s 1997–2022 PI RAPM (RS+PS), Draymond is 12th, most of his monstrous impact coming on defense (basically only Kevin Garnett, Dikembe, Mutombo, and Ben Wallace above him with a few others close by), but being a sizable positive on offense as well. do have some issues with how Draymond would fare without curry, his positive value towards winning margins is generally coming on defense.

If we look at Engelmann’s Single Season 1997–2019 PI RAPM (RS+PS), you have Draymond’s 2016 at number 8, 2015 at 78, 2017 at 195.

I’m not sure RAPM actually can properly deal with some of the effects I’m talking about, at least not to the extent of properly figuring out whether an effect should be part of ORAPM or DRAPM.

Nothing can quantify everything perfectly but when you have hundreds of thousands of possessions and you have such a stark difference between ORAPM or DRAPM, a lot is being properly quantified. If one sits down and tries to calculate a version of RAPM after scrubbing some play-by-play data, one will see this.

I’ve also posted about the Cheema and NBAShotCharts stuff—where Steph comes out #1 in the league in multiple five-year intervals in both. And I’ve posted Steph’s rankings in individual years and in 3-year and 5-year spans that we have that data for. It all looks very good for Steph. And indeed, that’s the sort of data I began with—including comparing how he stacked up specifically against his contemporary LeBron James in the various different data sets and time intervals. This is why my argument is in large part centered around Steph having outstripped LeBron in impact metrics—because that’s the most direct comparison point.

As mentioned to you several times, NBA ShotCharts data doesn’t include the playoffs, but Curry does really well in them, so I’m not sure why you posted “average“ RAPM numbers that do not exist.. Steph Curry’s numbers look great but Cheema’s numbers do not support the position that “Steph Curry has Outstripped LeBron in impact metrics.” I see what you’re saying about a direct comparison point because they’re playing at the same time (same players in the league, etc.) but you cannot be the king of the databall era by having marginally higher numbers in a few spans, when the other players has the best numbers overall and the overall highest 5-year RAPM spans AND 5 year spans in his 30s (i.e., 2016-2020 and 2017-2021) that are higher than the ones in Curry’s prime.[/quote]

If you want to object that taking averages across metrics like RPM is flawed, then I welcome you to do that. It’s a valid criticism! But there’s not a whole lot else to look at for what I was trying to get at—so a flawed measure is what I went with. Perhaps you think it’s so flawed that it’s worse than nothing. Not sure I agree, but I’m fine with you making that assertion.

Again, as I posted, average RPM/average RAPM is nonsensical due to how RAPM is calculated. On a board such as this that values data and nuance, made up data doesn’t further discussion. We have 5 year intervals we can look at. Also, in the post you are responding to, I mentioned that
[quote=“homecourtloss”]What you can do is list average rankings during a year, which, of course, Steph Curry would look very good in[/quote]

[lessthanjake]

I agree, but the discussion is about just how good Draymond is, and how much of his impact we can tease out from Steph Curry’s; obviously, it can’t be done 100% satisfactorily, but what we can do is look at other players to see if they had a teammate who had the type of impact that Draymond has had, and in the Databall age, nobody has. We can say older David Robinson paired with young Tim Duncan being one, though for a shorter amount of time, and then Tim Duncan with Manu Ginobili being another, but that pairing does not compare to the length along with heights that Draymond reaches with Curry or Dray on his own in 2016 for example. So, if we want to list out these numbers, then we also have to look at the fact that Draymond Green grades out as more impactful over a long period of time than even most players thought to be driving engines for their respective teams’ success, including over noisy playoff data, including in 2016 when he spent the lowest percentage of his playoff minutes paired with Steph Curry but had his best post season along with 2017. Now, this doesn’t mean the Draymond is “better“ than those other players, but with this amount of data and the number of possessions, it’s something to think about. And, if the postseason lineup data is posted, well, Dray grades out, extremely, extremely well.

I’m not sure that that’s really right though.

Let’s take the Cheema data. Draymond has had full five-year spans in that data ranking: 7th, 8th, 4th, 7th, and 19th in the given timeframes. Here’s some other examples of players that were not the best player on their team:

- Ginobili’s 5-year spans with Duncan ranked: 12th, 10th, 3rd, 3rd, 3rd, 2nd, 4th, 7th, 5th, 8th, 13th, 4th, 5th, and 12th. This is on its face superior to Draymond’s rankings.

- David Robinson’s 5-year spans with Duncan ranked 4th and 6th.

- John Stockton ranked 3rd, 1st, and 3rd in his final three five-year spans, with Karl Malone (and of course we don’t have data for Stockton before this).

- Russell Westbrook ranked 7th in his last five-year span with Durant.

- Dwyane Wade was still being rated top 5 in RAPM in this data set in the years with LeBron, until that final broken down 2013-2014 year gets added to the mix and the ranking goes down a good deal (though that’s also from taking out his great 2008-2009 year).

- Kyle Lowry played a season with Kawhi, and in the 5-year spans that included that season, Lowry ranked 4th, 5th, 7th, 6th, and 17th.

The NBAshotcharts data is typically similar to the above to the extent the data exists (which isn’t always the case, since it only starts in later years). For instance, Ginobili’s last five-year span with Duncan was 6th in the NBAshotcharts data (compared to 12th in the Cheema data). Cheema’s data doesn’t have the last couple seasons though, so a couple additional notes based on that:

- Jrue Holiday spent the last three years with Giannis, and those years are recent enough that they’re not really in the Cheema data. And in the NBAshotcharts data, Holiday is ranked 4th in the last three years, which is as highly ranked as the best Draymond timespan by this measure.

- Paul George is ranked 4th in the most recent five-year span, again as highly ranked as the best Draymond timespan (and, in case you wonder if it’s biased by that last OKC year still in the data set, he’s ranked 6th in the last three years).

And, of course, while these are a lot of examples, the number of examples that can be found are inherently limited by what data exists. There are very likely other examples from prior eras of the NBA. For instance, Kareem was 2nd and 4th in the Squared RAPM data from 1984-1985 and 1987-1988. Kevin McHale was 3rd in the Squared 1987-1988 data. Pippen was 2nd in the 1995-1996 Squared data. Obviously very limited data, but these are exactly the types of second-best players we’d probably expect to be up there in RAPM, and there they are.

And I should note that Draymond’s impact metrics were not as good in recent timeframes. As noted above, his last five-year timeframe in the Cheema data (2017-2021) was ranked 19th. And his last several three-year timeframes in the NBAshotcharts data (using three-year so we have a timeframe after 2019-2020, as well as one that is between the 2015-2016 season and the 2019-2020 season) were 30th, 61st, 27th, 54th, and 12th. And the Warriors won 3 titles in the years encompassed in those timeframes!

The bottom line is that there’s been other similarly or even more impactful players who played with major stars. Indeed, the guy about to be ranked #5 had guys generally ranked even better in RAPM (i.e. Ginobili and Robinson).

_____________________

I also think that this discussion misses that Steph is a huge enabler of Draymond being impactful, and that that’s a *good* thing. For instance, Draymond gets more offensive impact with Steph, because Steph papers over Draymond’s bad shooting and consistently creates short roll situations where Draymond can best utilize his passing. Draymond also gets more defensive impact with Steph, because Steph’s offensive ability makes it so Draymond can actually be played at PF when it’s advantageous to do so (which is the vast majority of the time), despite Draymond’s lack of shooting.

At a basic level, I think it’s a massive *positive* for a star to be someone that other great players can be hugely impactful alongside! That is what being a great ceiling raiser looks like! You play in a way where your presence doesn’t suck the air out of the room and other great players on your team can maximize their impact too. It’s not a bad thing for Draymond or Ginobili to have been really impactful! It’s a function of playing alongside all-time great guys that can get huge impact while leaving room for others to get huge impact too. This is a feature, not a bug! You concede that Draymond might not be better than other players who have had less impact alongside stars. And I think that’s essentially undeniably right. So why isn’t it a positive for Steph that Draymond could get more impact alongside Steph than plenty of better players have gotten alongside other all-time greats? This is actually a huge part of why I rank Steph so highly!