RealGM 2023 Top 100 Project - #42 (Anthony Davis)

[Owly]

Really suprised that Drexler fell outside the top 30. Excellent two-way player. Great Advanced Stats. 21st in career VORP

I would guess I lean slightly sympathetic on Drexler though I can't say I have a great read on where we are at ...

Still this reads as quite aggressively bullish ...

I would be curious at the idea that he "fell" outside the top 30" (I'm reading something into phrasing ["fell"] here, based on overall perception of Drexler in the post, it might, I grant, not be intended). It's been rare to see him inside the top 30. I've seen it twice in published lists, both some time ago, both quite some time ago (Keith Thompson, 2005: 28th ... he also has Heinsohn ahead of Robertson so ...; Athlon, '98: 29th ... neither of which would equate to top 30 in real-terms now). Similarly whilst he was in the top thirty in 2006 and 2008 here and has fared solidly since (thrice in the 33-32 range before dropping to 38 last time) there would be little reason to expect him in the top 30.

I would query the impression of "excellent" defense.

On "great" advanced stats that would probably depend somewhat on where one draws the line on such terms and one's stats of choice but I would imagine by this point he does look strong versus much of the remaining field.

[Doctor MJ]

Really suprised that Drexler fell outside the top 30. Excellent two-way player. Great Advanced Stats. 21st in career VORP

I would guess I lean slightly sympathetic on Drexler though I can't say I have a great read on where we are at ...

Still this reads as quite aggressively bullish ...

I would be curious at the idea that he "fell" outside the top 30" (I'm reading something into phrasing ["fell"] here, based on overall perception of Drexler in the post, it might, I grant, not be intended). It's been rare to see him inside the top 30. I've seen it twice in published lists, both some time ago, both quite some time ago (Keith Thompson, 2005: 28th ... he also has Heinsohn ahead of Robertson so ...; Athlon, '98: 29th ... neither of which would equate to top 30 in real-terms now). Similarly whilst he was in the top thirty in 2006 and 2008 here and has fared solidly since (thrice in the 33-32 range before dropping to 38 last time) there would be little reason to expect him in the top 30.

I would query the impression of "excellent" defense.

On "great" advanced stats that would probably depend somewhat on where one draws the line on such terms and one's stats of choice but I would imagine by this point he does look strong versus much of the remaining field.

I’ll echo that last part:

I would say Drexler had a pretty poor rep as a defender, leader, and in terms of being coachable, to the point where Buck Williams was brought in to set the culture that Drexler failed to be able to instill.

I would say a bit like Avery Johnson being more the Spurs’ leader than Robinson or Duncan, except Robinson and Duncan had impeccable work ethics and their coaches loved them.


Sent from my iPhone using Tapatalk

[AEnigma]

I continue to not see a real case for Drexler over Pierce aside from Drexler having far more success as his team’s best player and accordingly receiving more attention. Which, you know, not nothing, but that has not exactly been extended to most other good not great guards/wings thriving in periods of weak perimetre play.

I have Pierce as the better scorer, better shooter and spacer, better ballhandler, similar passer and playmaker (prefer Pierce and think Drexler box score edge was more schematic but will not profess the gap to be clear), similar defender (prefer Drexler as a man and transition defender but think Pierce was smarter rotating and providing help), better length of time spent as a starter / meaningful rotation piece… I understand why Drexler established more of a legacy, but year for year I think Pierce typically had an individual advantage.

To be clear, Drexler is in this area for me. I had Pierce in this area several rounds ago though.

[f4p]

So first thing here, I think this was brought up by another poster and I don't want to rehash it, but the existence of a guy like Andre Drummond (the other "AD") being able to trounce old-timers by this stat as well really makes clear how limited it is as a tool.

I think Davis is better than Baylor & Schayes and will be continuing to vote for Davis above them, but would say that Drummond is DRASTICALLY worse than all 3 of these player to such a degree that the difference between the other 3 is minor in comparison, and so if a stat thinks Drummond is better than old Hall of Famers when the reality is he's not even good enough to play for a contender as more than a guy to fill out the bench, I can't take it as much of an argument.

why are you making this drummond argument against PER again? i just don't get it. first, who is he trouncing? he peaks at a 23.4 PER in the regular season and 19.2 in the postseason. peak, not average. that isn't trouncing anybody. and certainly not baylor or schayes. second, is there a single stat in existence that doesn't have a drummond-level outlier (or much worse)? unless we should never quote any stat because they all have outliers, then i'm not sure of the point (and "i don't just look at one stat" is not the answer i'm going for, i'm talking about negating/removing a given stat simply because of an outlier). if not, i see no reason why andre drummond somehow invalidates the fairly large differences Mogspan brings up between the 3 relevant players. considering he also specifically pointed out that a baylor archetype is basically the exact box-score stuffing, higher-volume/lower-efficiency player that PER typically loves, without the dominant defense of AD to boot. of course, baylor actually does look very good by PER...until 1963. then he just falls off a cliff. that doesn't mean all analysis should end at PER, just that it most certainly should be treated as evidence as much as basically any other stat.

i brought it up the last time you said this, but we can look at RAPM and find much, much weirder results. such as anthony davis finishing like 163rd in the last 25 years. steve nash 30 spots behind paul george. iman shumpert and pablo prigioni over dwight howard. and a bunch of other results. but when these happen, then, because RAPM is supposed to be so much better, everybody supplies endless context (i believe you supplied some in the previous thread) and there are reasons why these results do, or at least could, make sense. but PER, despite tracking this project quite well and arguably avoiding outliers as well as any stat there is (at least at the top of the list), gets one strike and it's out. it's a strange argument.

I'm bringing Drummond up again because PER is getting used very, very similarly to how it was used last time...which I alluded to in my first sentence of this post. I'm not bringing up PER out of the blue as a punching bag, I'm responding to its use, just like I did last time.

Re: Drummond not trouncing Baylor or Schayes. Fair point. "trounce" was not the right word here. Someone taking PER super-seriously would likely conclude that Baylor & Schayes were slightly more effective in their time than Drummond was in his, and that's the misconception I'm looking to address.

and they would conclude pablo prigioni was better than dwight howard by other stats that are very commonly quoted with no similar objections by this board.

As I said in my post, the gap between Davis & Baylor/Schayes in effectiveness is small compared to Baylor/Schayes & Drummond, and so using a stat that indicates otherwise is problematic.

Re: Should we never use stats that have "outliers"? This is a good question to ask. I'm actually fond of stating that stats shouldn't be dismissed simply because you don't like a particular player outcome. Instead what you should do is ask why the stat became so misguided in this specific circumstance.

i agree.

In the case of Drummond, the answer is that use of box score-based stats can allow a big man to fill up the box score pretty dang well without him knowing really knowing what he's doing out there, and this can lead to a fool's gold situation where you think a player is your franchise player - complete with max contracts, re-designed systems around him, and even all-star appearances - when really he was never capable of being a starter-level player against playoff-level competition.

We then take what we learn from the data point and apply it back to the other players. Good performance in the metric obviously is not a damning thing, but if the metric cannot tell the difference between a superstar and and a benchwarmer, then you have to ask what information you can expect to glean by looking at the stat for new players.

like if RPM said 2023 franz wagner was better than giannis and curry? or if RAPM said trent forrest is better than anthony davis and dwade? i don't see how a one-off in PER, taking basically the most extreme "fill up the box score without accomplishing anything" guy in the world is any different than these other cases.

Now, why did I put the word "outlier" in quotes? Because the use of the term generally conveys more than just strange performance by a particular stat when there are volitional agents involved. It's a statement about the volitional agent himself. When we refer to human beings that are "outliers", we generally mean people whose performance diverges from norms...like the types of players we talk about in a Top 100 project. "outlier" is used as a synonym for greatness or terribleness. And that's not what Drummond was. Drummond was perpetually a guy who people got excited about and eventually disappointed on the next level because he lacked feel for the game. He wasn't an outlier, he was a regression to the mean. A flaming mediocrity.

i have to say i don't remember that much excitement about drummond, especially since detroit always sucked, but i guess someone gave him those large contracts. but they always felt like "what are you guys doing?" contracts to me. like there was a segment of people who were constantly excited about wall and beal in washington, but i never got it and feel like the more in-tune crowd knew they were fool's gold.

And if you wanted to identify the big problem with all-in-one box score stats, you might say that they are effectively feel-blind.

yes. and yet they represent this project quite well. in other words, the signal from strong production in how it filters into other areas/stats is apparently pretty strong. you can run around finding all the context, other stats, winning you want and yet you don't really end up that far from something like PER. because the best players tend to put up the big numbers. not because you have to put up pretty numbers to win, but because it's very difficult to keep the truly great players from doing lots of stuff. and essentially hard for the great players themselves to avoid doing lots of stuff when they are always given the ball, get constantly double-teamed and have to throw the ball to open teammates, easily grab rebounds due to height/strength, or are always guarding the basket.

Now, clearly folks want to come back and say "But Davis has feel so that's not an issue here", but that assumes feel is a binary thing and nothing could be further from the case. Feel is more complicated than the box score, not less, so using reasoning like this is essentially saying "Let's assume these guys are all the same at the stuff we can't quantify, and just go by the stuff we can quantify." And this, I'm saying, is problematic for analysis.

okay, but who's to say people haven't complicatedly decided davis has plenty of feel. that they've seen him in the playoffs. that they know he is elite in something (defense) that would be, on the whole, underrepresented by such a stat while the person to whom he is being compared is probably fully or possibly over-represented by the stat. if he then still handily outpaces the other person, it seems a strong point. context has been applied and, even in a favorable light, has not shown the other person to be an equal. and down below you say we should use it as a first pass. what else is a first pass if not a filter? if your first pass puts people in different tiers and then you say "that doesn't look right, let's ignore that and look at everything else", then you didn't really use it as a first pass. which is fine, but you can't say you're using it as a first pass.

Re: What about RAPM? Last time this came up someone (maybe you, I don't remember) asked this same question, and I believe I responded point by point to the cases they mentioned and no one responded back to me. My memory might be foggy, and I'm willing to get more into it if people will remind the context, but this is not something that I can accept being thrown in my face as if I've never considered such concerns, when I've written about such concerns more than almost anyone.

i didn't say you didn't respond to those. i said you DID respond to those. that was my point. when PER fails, we just wash our hands of it and that's that. what can we do, it said andre drummond is decent in the playoffs. when RAPM gives obviously wrong answers, there is now context to explain why these obviously wrong answers shouldn't invalidate RAPM. collinearity, or it shows value in that specific role, or some bad seasons were counted and dragged the numbers down, etc. the results are just as outlier and wrong as drummond looking good, but now RAPM gets the benefit of the doubt that it must still be a good stat overall.

In a nutshell though:

1. RAPM has limitations and I wouldn't advise anyone to rely solely upon it.

i wouldn't advise relying solely on anything. but i also wouldn't advise ignoring something with a strong signal just because people have decided it's a stat we have to move on from despite the signal.

2. The same is true for +/- in general.
3. But +/- in general is based on scoreboard outcome that allows us to ask "How did that happen?", which is where the really useful analysis comes in.
4. I don't object to the use of PER as a first pass analysis, but I know that the numbers come about by simply combining box score columns, so there's no real mystery to what's going on there.
5. The rub is that when you identify that PER is not capturing a player's ability to impact winning, you're basically left at a dead end. "There must be something beyond the box score!"...which we already knew.

In general my advise would be to use all complex stats - whether PER or RAPM - as first-pass analysis tools only, and then to use them along with other tools to drill down into the actual "what happened for the team".

Hence my aversion to using PER in a project like this to try to categorically separate players of different types from different eras as if that's all that needs doing.

As I've said, I intend none of this as a knock against Davis, who I'll be voting for over Baylor & Schayes. I'm just speaking to process.

given the relative lack of information that's pretty much always going to exist for older players, using some of the only widely available stats to differentiate would seem to make a lot of sense. it would seem to make even more sense if the stat is not biased against the guys who look worse by that stat. and even more if the only team data we really have says baylor did not seem to win in a way that would suggest he is undersold by any of these numbers (things which i suspect tend to be implicit in anyone's evaluation even if they only mention something like PER, because we've all had a chance for history to seep into our brains and "baylor not a winner" is kind of baked into any of his evaluations).

[Doctor MJ]

why are you making this drummond argument against PER again? i just don't get it. first, who is he trouncing? he peaks at a 23.4 PER in the regular season and 19.2 in the postseason. peak, not average. that isn't trouncing anybody. and certainly not baylor or schayes. second, is there a single stat in existence that doesn't have a drummond-level outlier (or much worse)? unless we should never quote any stat because they all have outliers, then i'm not sure of the point (and "i don't just look at one stat" is not the answer i'm going for, i'm talking about negating/removing a given stat simply because of an outlier). if not, i see no reason why andre drummond somehow invalidates the fairly large differences Mogspan brings up between the 3 relevant players. considering he also specifically pointed out that a baylor archetype is basically the exact box-score stuffing, higher-volume/lower-efficiency player that PER typically loves, without the dominant defense of AD to boot. of course, baylor actually does look very good by PER...until 1963. then he just falls off a cliff. that doesn't mean all analysis should end at PER, just that it most certainly should be treated as evidence as much as basically any other stat.

i brought it up the last time you said this, but we can look at RAPM and find much, much weirder results. such as anthony davis finishing like 163rd in the last 25 years. steve nash 30 spots behind paul george. iman shumpert and pablo prigioni over dwight howard. and a bunch of other results. but when these happen, then, because RAPM is supposed to be so much better, everybody supplies endless context (i believe you supplied some in the previous thread) and there are reasons why these results do, or at least could, make sense. but PER, despite tracking this project quite well and arguably avoiding outliers as well as any stat there is (at least at the top of the list), gets one strike and it's out. it's a strange argument.

I'm bringing Drummond up again because PER is getting used very, very similarly to how it was used last time...which I alluded to in my first sentence of this post. I'm not bringing up PER out of the blue as a punching bag, I'm responding to its use, just like I did last time.

Re: Drummond not trouncing Baylor or Schayes. Fair point. "trounce" was not the right word here. Someone taking PER super-seriously would likely conclude that Baylor & Schayes were slightly more effective in their time than Drummond was in his, and that's the misconception I'm looking to address.

and they would conclude pablo prigioni was better than dwight howard by other stats that are very commonly quoted with no similar objections by this board.

As I said in my post, the gap between Davis & Baylor/Schayes in effectiveness is small compared to Baylor/Schayes & Drummond, and so using a stat that indicates otherwise is problematic.

Re: Should we never use stats that have "outliers"? This is a good question to ask. I'm actually fond of stating that stats shouldn't be dismissed simply because you don't like a particular player outcome. Instead what you should do is ask why the stat became so misguided in this specific circumstance.

i agree.

In the case of Drummond, the answer is that use of box score-based stats can allow a big man to fill up the box score pretty dang well without him knowing really knowing what he's doing out there, and this can lead to a fool's gold situation where you think a player is your franchise player - complete with max contracts, re-designed systems around him, and even all-star appearances - when really he was never capable of being a starter-level player against playoff-level competition.

We then take what we learn from the data point and apply it back to the other players. Good performance in the metric obviously is not a damning thing, but if the metric cannot tell the difference between a superstar and and a benchwarmer, then you have to ask what information you can expect to glean by looking at the stat for new players.

like if RPM said 2023 franz wagner was better than giannis and curry? or if RAPM said trent forrest is better than anthony davis and dwade? i don't see how a one-off in PER, taking basically the most extreme "fill up the box score without accomplishing anything" guy in the world is any different than these other cases.

Now, why did I put the word "outlier" in quotes? Because the use of the term generally conveys more than just strange performance by a particular stat when there are volitional agents involved. It's a statement about the volitional agent himself. When we refer to human beings that are "outliers", we generally mean people whose performance diverges from norms...like the types of players we talk about in a Top 100 project. "outlier" is used as a synonym for greatness or terribleness. And that's not what Drummond was. Drummond was perpetually a guy who people got excited about and eventually disappointed on the next level because he lacked feel for the game. He wasn't an outlier, he was a regression to the mean. A flaming mediocrity.

i have to say i don't remember that much excitement about drummond, especially since detroit always sucked, but i guess someone gave him those large contracts. but they always felt like "what are you guys doing?" contracts to me. like there was a segment of people who were constantly excited about wall and beal in washington, but i never got it and feel like the more in-tune crowd knew they were fool's gold.

And if you wanted to identify the big problem with all-in-one box score stats, you might say that they are effectively feel-blind.

yes. and yet they represent this project quite well. in other words, the signal from strong production in how it filters into other areas/stats is apparently pretty strong. you can run around finding all the context, other stats, winning you want and yet you don't really end up that far from something like PER. because the best players tend to put up the big numbers. not because you have to put up pretty numbers to win, but because it's very difficult to keep the truly great players from doing lots of stuff. and essentially hard for the great players themselves to avoid doing lots of stuff when they are always given the ball, get constantly double-teamed and have to throw the ball to open teammates, easily grab rebounds due to height/strength, or are always guarding the basket.

Now, clearly folks want to come back and say "But Davis has feel so that's not an issue here", but that assumes feel is a binary thing and nothing could be further from the case. Feel is more complicated than the box score, not less, so using reasoning like this is essentially saying "Let's assume these guys are all the same at the stuff we can't quantify, and just go by the stuff we can quantify." And this, I'm saying, is problematic for analysis.

okay, but who's to say people haven't complicatedly decided davis has plenty of feel. that they've seen him in the playoffs. that they know he is elite in something (defense) that would be, on the whole, underrepresented by such a stat while the person to whom he is being compared is probably fully or possibly over-represented by the stat. if he then still handily outpaces the other person, it seems a strong point. context has been applied and, even in a favorable light, has not shown the other person to be an equal. and down below you say we should use it as a first pass. what else is a first pass if not a filter? if your first pass puts people in different tiers and then you say "that doesn't look right, let's ignore that and look at everything else", then you didn't really use it as a first pass. which is fine, but you can't say you're using it as a first pass.

Re: What about RAPM? Last time this came up someone (maybe you, I don't remember) asked this same question, and I believe I responded point by point to the cases they mentioned and no one responded back to me. My memory might be foggy, and I'm willing to get more into it if people will remind the context, but this is not something that I can accept being thrown in my face as if I've never considered such concerns, when I've written about such concerns more than almost anyone.

i didn't say you didn't respond to those. i said you DID respond to those. that was my point. when PER fails, we just wash our hands of it and that's that. what can we do, it said andre drummond is decent in the playoffs. when RAPM gives obviously wrong answers, there is now context to explain why these obviously wrong answers shouldn't invalidate RAPM. collinearity, or it shows value in that specific role, or some bad seasons were counted and dragged the numbers down, etc. the results are just as outlier and wrong as drummond looking good, but now RAPM gets the benefit of the doubt that it must still be a good stat overall.

In a nutshell though:

1. RAPM has limitations and I wouldn't advise anyone to rely solely upon it.

i wouldn't advise relying solely on anything. but i also wouldn't advise ignoring something with a strong signal just because people have decided it's a stat we have to move on from despite the signal.

2. The same is true for +/- in general.
3. But +/- in general is based on scoreboard outcome that allows us to ask "How did that happen?", which is where the really useful analysis comes in.
4. I don't object to the use of PER as a first pass analysis, but I know that the numbers come about by simply combining box score columns, so there's no real mystery to what's going on there.
5. The rub is that when you identify that PER is not capturing a player's ability to impact winning, you're basically left at a dead end. "There must be something beyond the box score!"...which we already knew.

In general my advise would be to use all complex stats - whether PER or RAPM - as first-pass analysis tools only, and then to use them along with other tools to drill down into the actual "what happened for the team".

Hence my aversion to using PER in a project like this to try to categorically separate players of different types from different eras as if that's all that needs doing.

As I've said, I intend none of this as a knock against Davis, who I'll be voting for over Baylor & Schayes. I'm just speaking to process.

given the relative lack of information that's pretty much always going to exist for older players, using some of the only widely available stats to differentiate would seem to make a lot of sense. it would seem to make even more sense if the stat is not biased against the guys who look worse by that stat. and even more if the only team data we really have says baylor did not seem to win in a way that would suggest he is undersold by any of these numbers (things which i suspect tend to be implicit in anyone's evaluation even if they only mention something like PER, because we've all had a chance for history to seep into our brains and "baylor not a winner" is kind of baked into any of his evaluations).

Okay:

First, you continuing to go "whatabout" with the problematic RAPM values as riposte to problematic PER values is getting annoying. Focus on the responses that have already been given or just leave it alone.

To try to focus us on the actual issue here:

You see folks being inconsistent about their behavior toward PER compared to RAPM.
You correctly see this as a possible source of epistemic bias.
You believe that this inconsistency is utterly unjustifiable.
I have previously given elaborations to try to make clear the difference, which I'll now scroll to see if you addressed in a way I feel compelled to respond to, but the way you start all this by pointing to the same thing again, makes me think we're not going to get much further here.

Re: Don't remember hype around Drummond. The community that was particularly associated it with was the Wins Produced community. That stat was a PER-type state that rated offensive rebounding extremely highly.

That site - belonging to academic Dave Berri, who was hyped by Malcolm Gladwell - had a cultlike community that was convinced Berry had the perfect scientific stat. The site no longer exists, but here's an ESPN article that references it:

ESPN: Drummond is best teenage rookie in his draft class

Referenced quote:

It's not even close. Even if Drummond doesn't suit up for the rest of the season, it would take an insane performance for any other rookie in the race to even match him. Drummond is THE rookie of the year. And he deserves this award not just because of totals, but also because of the historic nature of his performance.

Note that this article is listed as being posted in early February.

So yeah, Andre Drummond is long established as the poster boy for how PER-type stats can drastically distort things.

Now, I know you're going to say that PER isn't Wins Produced so why should the latter be used to damn the former? And it shouldn't be used to say PER is as bad as Wins Produced...but PER still made people think Drummond was an all-star. And of course, officially Drummond was an all-star, but in reality he wasn't even a starter level player.

Okay, so then back to the PER vs RAPM comparison, the issue here is that you're conflating precision and accuracy as if they are the same thing.

The issue with RAPM is one of variance or noise - aka limited precision.
The issue with PER is one of accuracy limited by bias.

To use the classic visual example:



PER is the image in the top right, RAPM is the bottom left. Because we have nothing in the bottom right, we need to use some of both the top-right and the bottom-left in our analysis, but that doesn't mean we treat their respective weaknesses identically.

I'm saying there's literally a bias in PER toward big men, and you're countering with noise examples of RAPM. They just aren't the same thing.

[Doctor MJ]

Induction Vote 1: Anthony Davis



Induction Vote 2: Draymond Green

Debated between these two and went back & forth. May change my mind if both are around next thread.

The fact that I do think Davis has the stronger overall peak was something that kept coming back to mind. Doesn't mean he needs to be ahead of Green, but the reality is that I can't really see giving Green the pick based on longevity. The two are close enough on that front that Green's average level just has to be higher than Davis...

and I struggle to assert that.

Nomination Vote 1: Kevin McHale



I think McHale really was an absolutely incredible two-way player. Often arguably better both as as scorer and a defender than teammate Larry Bird. I think he's clearly a worthy candidate here, just a question of whether he comes out on top. I find myself specifically considering McHale with two other Celtics - Dave Cowens & Robert Parish.

With Cowens, we have him as an MVP where McHale never seriously got that kind of shine. I'm not super-skeptical on Cowens' impact, but I look at how effective McHale is as a scorer, and I respect McHale's defense so much, it's hard for me to really advocate that Cowens was better.

With Parish, McHale generally gets the nod for prime with Parish the longevity argument. I'm inclined to agree with this, but respect Parish enough that I feel like McHale's prime edge may really not be enough to really justify him as the guy who did more for his teams.

Nomination Vote 2: Paul Arizin

So, I'm a bit sheepish about this one. Here I am continuing to not vote for Schayes and finding a way to vote for Arizin even after I use a criteria that hurts Arizin. Well, I do want to be clear that Schayes is above Arizin on my list of candidates even if I am low on Schayes and high on Arizin. The broken up nature of Arizin's career just hurts keeps him from doing all that Schayes did.

[HeartBreakKid]

Vote is for Anthony Davis - Anthony Davis has an amazing combination of efficiency and defense. It took a few years for his defense to catch up to his scoring but it is still quite monstrous and one of the best of his generation. He is a rather diverse player for someone who is primarily an off ball guy. He can pull up from far away, play off the catch, pick and pop, or finish at the rim at an incredible rate. His career is disjointed with injuries as well as an unfortunate sit out situation due to wanting to leave NO. I think he is simply a higher level players than most of the guys left.

My alternate vote is for Elgin Baylor – I used to be incredibly harsh on Baylor. Perhaps because his reputation had him as arguably the best in the world for a long time, and when I saw how inefficient he was relative to the big 4 of his era I said "pffft". I'd say in the past 1.5 years he has went up on my ratings quite a bit. He has a lot more playoff heroic performances than he is given credit for. He was a serious volume scorer on a scale that possibly the other great players left can't match.


My nomination is for Willis Reed - Arguably just as good as Frazier albeit his career feels even shorter.

My alternate nomination is for Kevin McHale – A lot of what I said with Davis applies to McHale. Great combination of hyper efficient scoring and good defense. His defense has less data to back up and he doesn’t seem like he is a true anchor, so that is why I haven’t vote for him earlier. His scoring is quite legendary, but he suffers from a similar situation as Manu in that he didn’t really have his own team for most of his career, and maybe the one season where he did his efficiency was still insane but his volume wasn’t much better. I don’t believe he was a blackhole, just he was so good at scoring there wasn’t much reason to pass, but lack of playmaking comparably does make him seem one dimensional albeit his game has a lot of nuance to it.

[OldSchoolNoBull]

Induction Vote #1: Dolph Schayes

It looks like it's coming down to AD and Schayes, and I'm going with Schayes on the basis of longevity. He has marginally more longevity in the RS, having played 15 seasons while AD just started his 12th, but he's just played a ton more meaningful playoff basketball - even with the playoffs being much shorter in those days, Schayes still played 97 playoff games over 14 seasons, while AD has played just 55 games over his 11 full seasons.

He went to three Finals as his team's best player, while AD has gone to one Finals where it's debatable who was the best between him and LeBron.

And there's these statistical things mentioned by trelos:

Probably the second best player of the 50's. Schayes had 8 seasons over 10 WS, 7 seasons over .200 for WS/48. Compare that with Jason Kidd's 2 and 0 seasons, Stocktons 13 and 14, Miller's 11 and 5. Suggests he's in the ball park. Ultimately, I have him at 2 weak MVP level seasons, 8 All NBA Seasons, 12 All Star seasons. His peak 3yr PS was 25.3 pp75 on +7 rTS%, and regular season he was around 17-18 pp75 on +5-6 rTS%.

He was one of the more efficient scorers in the game in his era, as well as an elite rebounder, and as trelos said above he was constantly at or near the top on his team in WS/48.

I just think Schayes was more impactful for longer in his era.

Nomination Vote #1: Clyde Drexler

Nomination Vote #2: Paul Pierce

I was deciding between Drexler and Pierce here. Pierce has good impact numbers(on/off, RAPM), and so does Drexler for the last two years of his career that he we have the data for(3.77 and 3.27 RAPM for his last two seasons, suggesting perhaps higher numbers in his younger days). Part of me wants to go with Pierce - better shooter, maybe better leader - but what I'm looking at now is that Drexler got his team to the Finals as #1 twice, and Pierce never got there until KG was on his team. Maybe that's not fair because Pierce's teams sucked(though they did get to the ECF in 2002), but the two players are close for me so that's where I'm looking to break the tie.

It looks like Westbrook is going to get it, but I'm pretty skeptical of him.

It is remarkable that he averaged a triple double four times, but his rTS in those seasons was +0.2, -3.2, -5.9, and -6.3 - three big negatives and one barely positive. And people have long suggested the big men in OKC were letting him get rebounds.

What sticks out to me is that, in those three years in OKC after KD left, the Westbrook-led Thunder lost in the first round all three times(even with Paul George and Melo onboard), and Westbrook's RAPM dropped from the 4.81, 4.54, and 5.78 he had recorded in his last three years with KD(when he wasn't averaging a triple double) to 2.91, 2.54, and 2.3 in the following three seasons without KD when he was averaging a triple double.

Post-OKC, the Wizards lost in the first round, and his tenures with the Rockets and the Lakers were both pretty bad.

In J.E.'s 1997-2022 RAPM - which I don't like to rely on too much, but the gap seems worth mention here - Pierce comes in at #25(3.2 O, -1.8 D, 5.0 Total), Drexler comes in at #48(2.9 O,-1.1 D, 4.0 Total), and Westbrook comes in at #168(2.3 O, 0.3 D, 2.0 Total).

I just don't see a lot of evidence of actual impact on winning.

[iggymcfrack]

Induction Vote #1: Dolph Schayes

It looks like it's coming down to AD and Schayes, and I'm going with Schayes on the basis of longevity. He has marginally more longevity in the RS, having played 15 seasons while AD just started his 12th, but he's just played a ton more meaningful playoff basketball - even with the playoffs being much shorter in those days, Schayes still played 97 playoff games over 14 seasons, while AD has played just 55 games over his 11 full seasons.

He went to three Finals as his team's best player, while AD has gone to one Finals where it's debatable who was the best between him and LeBron.

And there's these statistical things mentioned by trelos:

Probably the second best player of the 50's. Schayes had 8 seasons over 10 WS, 7 seasons over .200 for WS/48. Compare that with Jason Kidd's 2 and 0 seasons, Stocktons 13 and 14, Miller's 11 and 5. Suggests he's in the ball park. Ultimately, I have him at 2 weak MVP level seasons, 8 All NBA Seasons, 12 All Star seasons. His peak 3yr PS was 25.3 pp75 on +7 rTS%, and regular season he was around 17-18 pp75 on +5-6 rTS%.

He was one of the more efficient scorers in the game in his era, as well as an elite rebounder, and as trelos said above he was constantly at or near the top on his team in WS/48.

I just think Schayes was more impactful for longer in his era.

Nomination Vote #1: Clyde Drexler

Nomination Vote #2: Paul Pierce

I was deciding between Drexler and Pierce here. Pierce has good impact numbers(on/off, RAPM), and so does Drexler for the last two years of his career that he we have the data for(3.77 and 3.27 RAPM for his last two seasons, suggesting perhaps higher numbers in his younger days). Part of me wants to go with Pierce - better shooter, maybe better leader - but what I'm looking at now is that Drexler got his team to the Finals as #1 twice, and Pierce never got there until KG was on his team. Maybe that's not fair because Pierce's teams sucked(though they did get to the ECF in 2002), but the two players are close for me so that's where I'm looking to break the tie.

It looks like Westbrook is going to get it, but I'm pretty skeptical of him.

It is remarkable that he averaged a triple double four times, but his rTS in those seasons was +0.2, -3.2, -5.9, and -6.3 - three big negatives and one barely positive. And people have long suggested the big men in OKC were letting him get rebounds.

What sticks out to me is that, in those three years in OKC after KD left, the Westbrook-led Thunder lost in the first round all three times(even with Paul George and Melo onboard), and Westbrook's RAPM dropped from the 4.81, 4.54, and 5.78 he had recorded in his last three years with KD(when he wasn't averaging a triple double) to 2.91, 2.54, and 2.3 in the following three seasons without KD when he was averaging a triple double.

Post-OKC, the Wizards lost in the first round, and his tenures with the Rockets and the Lakers were both pretty bad.

In J.E.'s 1997-2022 RAPM - which I don't like to rely on too much, but the gap seems worth mention here - Pierce comes in at #25(3.2 O, -1.8 D, 5.0 Total), Drexler comes in at #48(2.9 O,-1.1 D, 4.0 Total), and Westbrook comes in at #168(2.3 O, 0.3 D, 2.0 Total).

I just don't see a lot of evidence of actual impact on winning.

Let’s not pretend that “going to the Finals” in 1954 is any more of an accomplishment than “going to the second round” in 2018. Dolph went 42-30 and then beat another 42-30 team and then was in the Finals because it was a 9-team league. Davis went 48-34 and then beat a 49-33 team in Portland. And then there’s 1959 where Schayes gets credit for playoff experience by going 35-37 and beating out the 32-40 Warriors for 3rd in the 4 team East.

If you’re looking at team success, the bottom line is Davis won 1 ring against the toughest competition imaginable and Schayes also won one ring… against the easiest competition imaginable. Even looking at era relative performance, if you swap them, Schayes gets one ring MAX on the teams Davis played on and Davis probably gets 1-2 rings MINIMUM on the teams Schayes played on.

[AEnigma]

VOTE: Dolph Schayes
Alternate: Elgin Baylor
NOMINATE: Paul Pierce
AltNom: Kevin McHale

Putting my thumb on the scale a bit for the retired players. Said my piece on Pierce and Drexler a few comments up. I thought LA Bird made a reasonably convincing case for Schayes, and while I respect Baylor’s peak more, the bulk of his career was a significant step down from how he started.

[Doctor MJ]

Induction Vote 1:

Davis - 5 (beast, iggy, Ohayo, Doc, HBK)
Baylor - 3 (trex, Rishkar, Samurai)
Schayes - 5 (Clyde, Dutch, LA Bird, OSNB, AEnigma)
Howard - 1 (trelos)
Green - 1 (hcl)

No majority. Going to Vote 2 between Davis & Schayes:

Davis - 2 (trex, Samurai)
Schayes - 2 (Rishkar, trelos)
neither - 1 (hcl)

Anthony Davis 7, Dolph Schayes 7. Looks like we'll need a run-off here.

Nomination Vote 1:

McHale - 2 (beast, Doc)
Drexler - 4 (trex, Clyde, LA Bird, OSNB)
Westbrook - 4 (Rishkar, Samurai, iggy, Ohayo)
Gervin - 1 (Dutch)
Embiid - 1 (trelos)
Pierce - 2 (hcl, AEnigma)
Reed - 1 (HBK)

No majority. Going to Vote 2 between Drexler & Westbrook:

Drexler - 0 (none)
Westbrook - 3 (beast, trelos, hcl)
neither - 4 (Doc, Dutch, AEnigma, HBK)

Russell Westbrook 7, Clyde Drexler 4

Russell Westbrook will be added to Nominee list.

[Doctor MJ]

One more day for a runoff between Anthony Davis & Dolph Schayes. Using the Batsignal:

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

[Ambrose]

Between those two, Davis.

[trex_8063]

...

Okay:

First, you continuing to go "whatabout" with the problematic RAPM values as riposte to problematic PER values is getting annoying. Focus on the responses that have already been given or just leave it alone.

To try to focus us on the actual issue here:

You see folks being inconsistent about their behavior toward PER compared to RAPM.
You correctly see this as a possible source of epistemic bias.
You believe that this inconsistency is utterly unjustifiable.
I have previously given elaborations to try to make clear the difference, which I'll now scroll to see if you addressed in a way I feel compelled to respond to, but the way you start all this by pointing to the same thing again, makes me think we're not going to get much further here.

I know it can get frustrating when not seeing eye-to-eye with someone; we've been there between the two of us, right? (if you agree with nothing else I say in this post, I'm sure you'd agree with at least that much! )
But for my part, I think f4p has comported himself very well in this discussion, and I actually don't think he's gone far off point (at least, not much more than you have yourself). More than anything, I think you're both just kinda shooting past each other [as is often the case]; and to be fair to f4p, there has been [to my eye] some mixed messaging on your part (I'll elaborate below).

If I may stick my big nose into things and offer my own observations (however welcome they may or may not be).....


I gather that in some prior discussion [which I missed] you gave extensive elaboration/explanation for how various bizarre ["outlier"] RAPM values for certain players in certain years came about (in essence: explaining away the weirdness).
And you HAVE [to some degree, perhaps in more vague/broad terms] done the same for how/why Andre Drummond's PER result is wrong (that is: explain why it pegs him very high/good, when he's actually not).

It's where you [and perhaps a number of others] go from there [conclusions on how to then use the stats] that differs dramatically; which is basically what f4p has pointed out (which he can't very well do without drawing the comparison ["whatabout"]).


With RAPM, the tone and tenor of how to accept explanations for outlier/bizarre results was something like [from what I gather of this prior discussion]: "This is how that may have happened, and that makes sense. Stat still very useful, just gotta be concscientious of the context."

But f4p perceives that PER does not receive the same treatment.
He perceives [I think] that despite some adroit observations on your part for why PER gives a misleading result for Drummond ("explaining away the weirdness", perhaps similar(ish) to what you did for weird RAPM results??), the take-away impression on the stat is still something like: "Look how badly is misrepresented him. It's not a stat that should be trusted or given much of any weight in an argument."

I'm not sure if that was your intent, but I must admit that f4p is not the only one for whom that was the perceived insinuation.
Again, there is arguably some mixed messaging.

While you [in posts after the inciting reply] said this.....

In general my advise would be to use all complex stats - whether PER or RAPM - as first-pass analysis tools only, and then to use them along with other tools to drill down into the actual "what happened for the team".

And this....

I'm actually fond of stating that stats shouldn't be dismissed simply because you don't like a particular player outcome. Instead what you should do is ask why the stat became so misguided in this specific circumstance.

......that actually does contradict what you had said previously.

In the initial reply you had stated:

the existence of a guy like Andre Drummond (the other "AD") being able to trounce old-timers by this stat as well really makes clear how limited it is as a tool.

And....

so if a stat thinks Drummond is better than old Hall of Famers when the reality is he's not even good enough to play for a contender as more than a guy to fill out the bench, I can't take it [PER] as much of an argument.

This has the feel of a broad, summary judgement of the metric.
And it seems [at least as worded here] that PER is damnable and of marginal or negligible value simply for the mere existence of Drummond's result.

You also said----right on the heels of suggesting no stat should be dismissed because you don't like a result---the following:

In the case of Drummond, the answer is that use of box score-based stats can allow a big man to fill up the box score pretty dang well without him knowing really knowing what he's doing out there, and this can lead to a fool's gold situation where you think a player is your franchise player - complete with max contracts, re-designed systems around him, and even all-star appearances - when really he was never capable of being a starter-level player against playoff-level competition.

We then take what we learn from the data point and apply it back to the other players. Good performance in the metric obviously is not a damning thing, but if the metric cannot tell the difference between a superstar and and a benchwarmer, then you have to ask what information you can expect to glean by looking at the stat for new players.

This one is less direct (it's more about tone [and tone is SO hard to interpret in text])......but the bolded could be construed as saying "there's little-to-nothing of value to be gleaned from this stat, so why bother with it?"


So, you might have to forgive some of us readers a bit of confusion. Because in a number of ways it sort of feels like you're saying: "Never dismiss any stat out-of-hand because of a weird/bad result (except PER....it's crap)".......which is a contradiction in itself.


Now to be fair, you did also state that your low opinion of it is also based upon a belief [paraphrased] that the things credited by PER do not, in and of themselves, carry much value (that is: toward winning; impact; whereas, otoh, "impact" is EXACTLY what RAPM attempts to identify and credit).

fwiw, I'm not sure I entirely agree (about PER, I mean). I think a few things PER places value on generally [or at least frequently] positively impact the outcome, even if it gives equal credit to things that are not always equal (e.g. steals accrued at the expense of excessive gambling will be counted the same as steals that come within the context of sound team defense; not all assists are equal in value; not all blocks are equal in value; shooting efficiency is credited by PER but not nearly enough [imo], and thus points for points sake looms excessively large; etc).
What's more (and f4p pointed this out): PER often "gets it right", even if it does so sort of by coincidence---or rather by accident, if you will: because the really good players---in process of doing those things that tend to produce profound positive impact---frequently tend to also fill up the stat-sheet with those things which PER values.

I think there's a philosophical axe to grind here.
On the one hand, you might smirk and decide: "I'm not going to place any weight in a stat that only gets things right by accident!"
But on the other hand, one might suggest: "If it's getting things right often enough to pass the smell test, does the manner in which it got there really matter overmuch?"

idk......that's a philosophical debate that I'm not sure I'm all that interested in having, though I will comment a little further below regarding your visual aids.

But (fwiw) as f4p's pointed out repeatedly: PER rankings do actually dove-tail pretty well with the list results so far.
I might also point that the top 10 players by best single-season scaled rs PER (recall I'd done a scaled version to account for variations in league parity) are:

Kareem Abdul-Jabbar
Michael Jordan
Wilt Chamberlain
Stephen Curry ('16)
LeBron James
Julius Erving ('76 [ABA])
Bob Pettit
David Robinson
Shaquille O'Neal
(*Nikola Jokic)

The 10 best in single-season playoff scaled PER are:

Kareem Abdul-Jabbar
Hakeem Olajuwon
LeBron James
Julius Erving ('76 [ABA])
George Gervin
Michael Jordan
George Mikan
Shaquille O'Neal
Tim Duncan
(*Nikola Jokic)

*I've only updated my sheets as far as '20, but I'm almost certain '22 Jokic would make the cut in both rs and playoffs.

For a stat that isn't measuring anything important, it seems like it does a reasonable job of identifying who the best players are (at least at the highest levels). Obviously some "glitches" and "outliers", but that's why we cut it with OTHER indicators a pieces of information. If those other indicators more or less fall in line.....

And with regards to older players, it's a little bit of an "any port in a storm" situation: there's so much less information to go on. So can we really afford to be dismissive of a metric that generally does this^^ well at the high levels? (tbf, I do think there is a point where there are "diminishing returns" on PER's accuracy in gauging player quality, as we get further below the super-elite).


To be fair, at one other point you mentioned it was more an objection to **the manner in which PER was being used that you were speaking to. (**in a manner that basically says, "Look how high he ranks in PER. He must be super-good and belongs in the conversation here.")

However, f4p pointed out [correctly, I might add] that similar use is applied to other metrics in this project. Even by you:

I think another perspective to take is that our career playoff +/- leaderboard for the data we have goes:

1. LeBron
2. Duncan
3. Green

But similar use of these other metrics often doesn't seem to garner the same blowback (which again, is what f4p has pointed out).

Okay, so then back to the PER vs RAPM comparison, the issue here is that you're conflating precision and accuracy as if they are the same thing.

The issue with RAPM is one of variance or noise - aka limited precision.
The issue with PER is one of accuracy limited by bias.

To use the classic visual example:



PER is the image in the top right, RAPM is the bottom left. Because we have nothing in the bottom right, we need to use some of both the top-right and the bottom-left in our analysis, but that doesn't mean we treat their respective weaknesses identically.

This ties back into the philosophical axe to grind.

RAPM is aiming at the right target, the target you're supposed to be aiming at (the bull's eye), but does a sort of crappy job of consistently hitting it.

But PER isn't even aiming at the right target (it's shooting somewhere off the side), though it hits what it's aiming at pretty well.

I suppose I more or less agree, though I'd wager we differ in degrees.
For example, I suspect I don't think this "wrong" target that PER aims at is quite as far away from the bull's eye as you. And since it usually hits what it's aiming at, it thus tends to be "kinda close" to that bull's eye very often.......which makes it of some value, imo (even if it is wrong-headed [aiming for the wrong goal] in theory).

I'm saying there's literally a bias in PER toward big men,

Or is it simply that big men are "more important" in the game of basketball (generally and historically speaking)? This is perhaps another philosophical axe. As I'd posted regarding the demographics of our top 35: big men are "over-represented" (I think because it is, quite simply, a big man's game).

Anyway, I'll stop there I think.

[f4p]

I'm bringing Drummond up again because PER is getting used very, very similarly to how it was used last time...which I alluded to in my first sentence of this post. I'm not bringing up PER out of the blue as a punching bag, I'm responding to its use, just like I did last time.

Re: Drummond not trouncing Baylor or Schayes. Fair point. "trounce" was not the right word here. Someone taking PER super-seriously would likely conclude that Baylor & Schayes were slightly more effective in their time than Drummond was in his, and that's the misconception I'm looking to address.

and they would conclude pablo prigioni was better than dwight howard by other stats that are very commonly quoted with no similar objections by this board.

As I said in my post, the gap between Davis & Baylor/Schayes in effectiveness is small compared to Baylor/Schayes & Drummond, and so using a stat that indicates otherwise is problematic.

Re: Should we never use stats that have "outliers"? This is a good question to ask. I'm actually fond of stating that stats shouldn't be dismissed simply because you don't like a particular player outcome. Instead what you should do is ask why the stat became so misguided in this specific circumstance.

i agree.

In the case of Drummond, the answer is that use of box score-based stats can allow a big man to fill up the box score pretty dang well without him knowing really knowing what he's doing out there, and this can lead to a fool's gold situation where you think a player is your franchise player - complete with max contracts, re-designed systems around him, and even all-star appearances - when really he was never capable of being a starter-level player against playoff-level competition.

We then take what we learn from the data point and apply it back to the other players. Good performance in the metric obviously is not a damning thing, but if the metric cannot tell the difference between a superstar and and a benchwarmer, then you have to ask what information you can expect to glean by looking at the stat for new players.

like if RPM said 2023 franz wagner was better than giannis and curry? or if RAPM said trent forrest is better than anthony davis and dwade? i don't see how a one-off in PER, taking basically the most extreme "fill up the box score without accomplishing anything" guy in the world is any different than these other cases.

Now, why did I put the word "outlier" in quotes? Because the use of the term generally conveys more than just strange performance by a particular stat when there are volitional agents involved. It's a statement about the volitional agent himself. When we refer to human beings that are "outliers", we generally mean people whose performance diverges from norms...like the types of players we talk about in a Top 100 project. "outlier" is used as a synonym for greatness or terribleness. And that's not what Drummond was. Drummond was perpetually a guy who people got excited about and eventually disappointed on the next level because he lacked feel for the game. He wasn't an outlier, he was a regression to the mean. A flaming mediocrity.

i have to say i don't remember that much excitement about drummond, especially since detroit always sucked, but i guess someone gave him those large contracts. but they always felt like "what are you guys doing?" contracts to me. like there was a segment of people who were constantly excited about wall and beal in washington, but i never got it and feel like the more in-tune crowd knew they were fool's gold.

And if you wanted to identify the big problem with all-in-one box score stats, you might say that they are effectively feel-blind.

yes. and yet they represent this project quite well. in other words, the signal from strong production in how it filters into other areas/stats is apparently pretty strong. you can run around finding all the context, other stats, winning you want and yet you don't really end up that far from something like PER. because the best players tend to put up the big numbers. not because you have to put up pretty numbers to win, but because it's very difficult to keep the truly great players from doing lots of stuff. and essentially hard for the great players themselves to avoid doing lots of stuff when they are always given the ball, get constantly double-teamed and have to throw the ball to open teammates, easily grab rebounds due to height/strength, or are always guarding the basket.

Now, clearly folks want to come back and say "But Davis has feel so that's not an issue here", but that assumes feel is a binary thing and nothing could be further from the case. Feel is more complicated than the box score, not less, so using reasoning like this is essentially saying "Let's assume these guys are all the same at the stuff we can't quantify, and just go by the stuff we can quantify." And this, I'm saying, is problematic for analysis.

okay, but who's to say people haven't complicatedly decided davis has plenty of feel. that they've seen him in the playoffs. that they know he is elite in something (defense) that would be, on the whole, underrepresented by such a stat while the person to whom he is being compared is probably fully or possibly over-represented by the stat. if he then still handily outpaces the other person, it seems a strong point. context has been applied and, even in a favorable light, has not shown the other person to be an equal. and down below you say we should use it as a first pass. what else is a first pass if not a filter? if your first pass puts people in different tiers and then you say "that doesn't look right, let's ignore that and look at everything else", then you didn't really use it as a first pass. which is fine, but you can't say you're using it as a first pass.

Re: What about RAPM? Last time this came up someone (maybe you, I don't remember) asked this same question, and I believe I responded point by point to the cases they mentioned and no one responded back to me. My memory might be foggy, and I'm willing to get more into it if people will remind the context, but this is not something that I can accept being thrown in my face as if I've never considered such concerns, when I've written about such concerns more than almost anyone.

i didn't say you didn't respond to those. i said you DID respond to those. that was my point. when PER fails, we just wash our hands of it and that's that. what can we do, it said andre drummond is decent in the playoffs. when RAPM gives obviously wrong answers, there is now context to explain why these obviously wrong answers shouldn't invalidate RAPM. collinearity, or it shows value in that specific role, or some bad seasons were counted and dragged the numbers down, etc. the results are just as outlier and wrong as drummond looking good, but now RAPM gets the benefit of the doubt that it must still be a good stat overall.

In a nutshell though:

1. RAPM has limitations and I wouldn't advise anyone to rely solely upon it.

i wouldn't advise relying solely on anything. but i also wouldn't advise ignoring something with a strong signal just because people have decided it's a stat we have to move on from despite the signal.

2. The same is true for +/- in general.
3. But +/- in general is based on scoreboard outcome that allows us to ask "How did that happen?", which is where the really useful analysis comes in.
4. I don't object to the use of PER as a first pass analysis, but I know that the numbers come about by simply combining box score columns, so there's no real mystery to what's going on there.
5. The rub is that when you identify that PER is not capturing a player's ability to impact winning, you're basically left at a dead end. "There must be something beyond the box score!"...which we already knew.

In general my advise would be to use all complex stats - whether PER or RAPM - as first-pass analysis tools only, and then to use them along with other tools to drill down into the actual "what happened for the team".

Hence my aversion to using PER in a project like this to try to categorically separate players of different types from different eras as if that's all that needs doing.

As I've said, I intend none of this as a knock against Davis, who I'll be voting for over Baylor & Schayes. I'm just speaking to process.

given the relative lack of information that's pretty much always going to exist for older players, using some of the only widely available stats to differentiate would seem to make a lot of sense. it would seem to make even more sense if the stat is not biased against the guys who look worse by that stat. and even more if the only team data we really have says baylor did not seem to win in a way that would suggest he is undersold by any of these numbers (things which i suspect tend to be implicit in anyone's evaluation even if they only mention something like PER, because we've all had a chance for history to seep into our brains and "baylor not a winner" is kind of baked into any of his evaluations)

Okay:

First, you continuing to go "whatabout" with the problematic RAPM values as riposte to problematic PER values is getting annoying. Focus on the responses that have already been given or just leave it alone.

ok, now put yourself in the shoes of people like me. instead of 2 replies with the same thing, make it more like 1000 and over years and years and you can see what it's like having to constantly defend even the most basic deductions from the box score. like saying, in the middle of a larger post, someone had a 25 PER, 0.250 WS48, and 8 BPM as a quick summary of why it was a good postseason run and getting "not everything is the box score" responses like you were supposed to stop and write 500 words about that postseason run because we can't just look at the very obviously elite box numbers and conclude it was an elite postseason. that's what i'm used to and it gets exhausting after a while.

and i only brought up my examples twice because the very same drummond example was brought up.

To try to focus us on the actual issue here:

You see folks being inconsistent about their behavior toward PER compared to RAPM.
You correctly see this as a possible source of epistemic bias.
You believe that this inconsistency is utterly unjustifiable.
I have previously given elaborations to try to make clear the difference, which I'll now scroll to see if you addressed in a way I feel compelled to respond to, but the way you start all this by pointing to the same thing again, makes me think we're not going to get much further here.

i have addressed it. i have said that no stat is perfect and there are usually good reasons for outlier results. but whereas many people on this board (i'm not just talking about you or anything) will explain away "weird" results in the other stats, the box score seemingly always needs to bat 1.000. because it essentially starts with a presumption of guilt. i suspect this is because people like this board (me included) got tired of hearing that carmelo was great because he score 28 ppg or allen iverson must be a top 20 all-time player because he scored a lot. it all got equated to people being fixated on the box score being wrong. but i suspect it was mostly "raw points = great offense" and "raw steals = great defense" that were probably the biggest culprits. and not accounting for pace and minutes. but many of those things are largely fixed with your PER's, WS48's, and BPM's. perfect? no. possibly less good than some other stats? maybe. so bad they mean nothing? definitely not. even a fairly "volume favorable" stat like PER doesn't love carmelo or iverson. and when we extend the box score conversation to the BBR Big 3, i'd say we have a fairly robust set of stats that are hard to game, even for the most egregious "high production, low impact" players. so when large differences are pointed out, the reflexive "we can't use this, it's the box score" is problematic.

Re: Don't remember hype around Drummond. The community that was particularly associated it with was the Wins Produced community. That stat was a PER-type state that rated offensive rebounding extremely highly.

That site - belonging to academic Dave Berri, who was hyped by Malcolm Gladwell - had a cultlike community that was convinced Berry had the perfect scientific stat. The site no longer exists, but here's an ESPN article that references it:

ESPN: Drummond is best teenage rookie in his draft class

Referenced quote:

It's not even close. Even if Drummond doesn't suit up for the rest of the season, it would take an insane performance for any other rookie in the race to even match him. Drummond is THE rookie of the year. And he deserves this award not just because of totals, but also because of the historic nature of his performance.

Note that this article is listed as being posted in early February.

So yeah, Andre Drummond is long established as the poster boy for how PER-type stats can drastically distort things.

Now, I know you're going to say that PER isn't Wins Produced so why should the latter be used to damn the former? And it shouldn't be used to say PER is as bad as Wins Produced...but PER still made people think Drummond was an all-star. And of course, officially Drummond was an all-star, but in reality he wasn't even a starter level player.

ok, but i feel like Wins Produced got left in the past a long, long time ago, precisely because it said such crazy things. i guess you could use the Ohayo argument that it's no less arbitrary, but i would just say it seemed to produce wildly out of line results a lot. the BBRef stats don't. which is why the latter persist. survival of the fittest and what not. they usually pass the smell test (for stars, not necessarily role players, but the Top 100 project isn't a role player focused project). and mimic this project fairly well. even Ohayo kept posting an article showing that BPM outdoes RAPM is predictive power and WS48 was literally tied with RAPM. and the difference between PER and the others was significantly smaller than the error bars themselves.

Okay, so then back to the PER vs RAPM comparison, the issue here is that you're conflating precision and accuracy as if they are the same thing.

The issue with RAPM is one of variance or noise - aka limited precision.
The issue with PER is one of accuracy limited by bias.

To use the classic visual example:



PER is the image in the top right, RAPM is the bottom left. Because we have nothing in the bottom right, we need to use some of both the top-right and the bottom-left in our analysis, but that doesn't mean we treat their respective weaknesses identically.

I'm saying there's literally a bias in PER toward big men, and you're countering with noise examples of RAPM. They just aren't the same thing.

are we sure PER is in the top right? are we sure RAPM doesn't rely too much on prior informed numbers (which is basically just saying "this answer is wrong, let's tell the formula what we think is the right answer beforehand and see if it spits out the right answer") or can't sort people out from their teammates and opponents in a small enough sample to always be useful? i like RAPM, it certainly rates a lot of the right people highly. but it also swings and misses from time to time. in the spirit of "we shouldn't use any one stat", i don't see it being so accurate that it can be confidently used and PER somehow so inaccurate that we can just dismiss wide differences without some massive archetypal difference in the players (like if we were comparing paul pierce to ben wallace or something).

[Owly]

I'm saying there's literally a bias in PER toward big men,

Or is it simply that big men are "more important" in the game of basketball (generally and historically speaking)? This is perhaps another philosophical axe. As I'd posted regarding the demographics of our top 35: big men are "over-represented" (I think because it is, quite simply, a big man's game).

So I sometimes hear this and I wonder ... granting certainly in the initial phrasing it is phrased as a question.

... what do you mean by this?

Because I could easily get to "more important" on average without thinking that on average big men should look better.

In short (I'm sure others have said it but it stuck with their phrasing, and why not get another Dave Berri mention in) it's "the short supply of tall people".

Bigs could be be a wider ranging group, making them more important without the average being better. It depends on if you think you need really, really tall people but most of the major league era game has at least been played on the assumption that you do. The gap between 1st and 30th or 60th say ... five-eleven to six-four guy is likely to be smaller than 1st to 30th or 60th in a much smaller pool, say six-nine and above).

That would lead to correctly placing more bigs at the very top end of a ranking list. But would not justify box composites (and I will here note PER is not gen1 or indeed any variant of Win Produced) tilting towards bigs on average.

And given the limited supply it's possible that bad bigs with physical tools and/or solid production who may be very weak in non-box areas and are net doing significant harm might tend to get more opportunities. It might be harder for a smaller guy to get to box solid whilst being bad (a big who actually doesn't help team defensive rebound or deter/alter rim attempts could get defensive rebounds and blocks and just not box out and miss rotations and let guys sail by ... a point guard who doesn't do point guard stuff ... the box isn't perfect but I think they'll typically look bad in the box-aggregates).

I'm just postulating stuff here rather than committing hard to a position but I certainly see a route to "bigs are [on average] more 'important'" and simultaneously that we shouldn't see bigs averaging higher statistically.

[Doctor MJ]

ok, now put yourself in the shoes of people like me. instead of 2 replies with the same thing, make it more like 1000 and over years and years and you can see what it's like having to constantly defend even the most basic deductions from the box score. like saying, in the middle of a larger post, someone had a 25 PER, 0.250 WS48, and 8 BPM as a quick summary of why it was a good postseason run and getting "not everything is the box score" responses like you were supposed to stop and write 500 words about that postseason run because we can't just look at the very obviously elite box numbers and conclude it was an elite postseason. that's what i'm used to and it gets exhausting after a while.

and i only brought up my examples twice because the very same drummond example was brought up.

So let me first apologize for my impatience. Clearly we're both dealing with situations where we feel like we're hitting the same thing over and over, and me letting out my frustration doesn't help with the situation.

To try to focus us on the actual issue here:

You see folks being inconsistent about their behavior toward PER compared to RAPM.
You correctly see this as a possible source of epistemic bias.
You believe that this inconsistency is utterly unjustifiable.
I have previously given elaborations to try to make clear the difference, which I'll now scroll to see if you addressed in a way I feel compelled to respond to, but the way you start all this by pointing to the same thing again, makes me think we're not going to get much further here.

i have addressed it. i have said that no stat is perfect and there are usually good reasons for outlier results. but whereas many people on this board (i'm not just talking about you or anything) will explain away "weird" results in the other stats, the box score seemingly always needs to bat 1.000. because it essentially starts with a presumption of guilt. i suspect this is because people like this board (me included) got tired of hearing that carmelo was great because he score 28 ppg or allen iverson must be a top 20 all-time player because he scored a lot. it all got equated to people being fixated on the box score being wrong. but i suspect it was mostly "raw points = great offense" and "raw steals = great defense" that were probably the biggest culprits. and not accounting for pace and minutes. but many of those things are largely fixed with your PER's, WS48's, and BPM's. perfect? no. possibly less good than some other stats? maybe. so bad they mean nothing? definitely not. even a fairly "volume favorable" stat like PER doesn't love carmelo or iverson. and when we extend the box score conversation to the BBR Big 3, i'd say we have a fairly robust set of stats that are hard to game, even for the most egregious "high production, low impact" players. so when large differences are pointed out, the reflexive "we can't use this, it's the box score" is problematic.

Okay, noted.

Re: Don't remember hype around Drummond. The community that was particularly associated it with was the Wins Produced community. That stat was a PER-type state that rated offensive rebounding extremely highly.

That site - belonging to academic Dave Berri, who was hyped by Malcolm Gladwell - had a cultlike community that was convinced Berry had the perfect scientific stat. The site no longer exists, but here's an ESPN article that references it:

ESPN: Drummond is best teenage rookie in his draft class

Referenced quote:

It's not even close. Even if Drummond doesn't suit up for the rest of the season, it would take an insane performance for any other rookie in the race to even match him. Drummond is THE rookie of the year. And he deserves this award not just because of totals, but also because of the historic nature of his performance.

Note that this article is listed as being posted in early February.

So yeah, Andre Drummond is long established as the poster boy for how PER-type stats can drastically distort things.

Now, I know you're going to say that PER isn't Wins Produced so why should the latter be used to damn the former? And it shouldn't be used to say PER is as bad as Wins Produced...but PER still made people think Drummond was an all-star. And of course, officially Drummond was an all-star, but in reality he wasn't even a starter level player.

ok, but i feel like Wins Produced got left in the past a long, long time ago, precisely because it said such crazy things. i guess you could use the Ohayo argument that it's no less arbitrary, but i would just say it seemed to produce wildly out of line results a lot. the BBRef stats don't. which is why the latter persist. survival of the fittest and what not. they usually pass the smell test (for stars, not necessarily role players, but the Top 100 project isn't a role player focused project). and mimic this project fairly well. even Ohayo kept posting an article showing that BPM outdoes RAPM is predictive power and WS48 was literally tied with RAPM. and the difference between PER and the others was significantly smaller than the error bars themselves.

Sure, but the thing is that PER absolutely has the same types of issues as WP, just less egregiously so, so when I'm trying to direct people to understand the type of problem, it's all relevant together. Yes WP had worse Drummond issues...but PER has always had Drummond issues too.

Okay, so then back to the PER vs RAPM comparison, the issue here is that you're conflating precision and accuracy as if they are the same thing.

The issue with RAPM is one of variance or noise - aka limited precision.
The issue with PER is one of accuracy limited by bias.

To use the classic visual example:



PER is the image in the top right, RAPM is the bottom left. Because we have nothing in the bottom right, we need to use some of both the top-right and the bottom-left in our analysis, but that doesn't mean we treat their respective weaknesses identically.

I'm saying there's literally a bias in PER toward big men, and you're countering with noise examples of RAPM. They just aren't the same thing.

are we sure PER is in the top right? are we sure RAPM doesn't rely too much on prior informed numbers (which is basically just saying "this answer is wrong, let's tell the formula what we think is the right answer beforehand and see if it spits out the right answer") or can't sort people out from their teammates and opponents in a small enough sample to always be useful? i like RAPM, it certainly rates a lot of the right people highly. but it also swings and misses from time to time. in the spirit of "we shouldn't use any one stat", i don't see it being so accurate that it can be confidently used and PER somehow so inaccurate that we can just dismiss wide differences without some massive archetypal difference in the players (like if we were comparing paul pierce to ben wallace or something).

Okay, so these are some good question that allude to the more complex n-dimensional shape of the phenomenon we're attempting to study. Let me set context this way:

What I'm referring to are stats that have a bias based on player archetype. In PER, big men who get rebounds and put backs tend to look more valuable than they actually are. It's not the only issue, but it's something I see as a known thing from way back. It doesn't mean that every such big man is overrated by PER, but if he's not overrated by it, it's because the player in question is delivering more subtle impact that doesn't appear in the box score...which PER has no concept of, and thus underrates the player in question along those lines. The fact that those overrating and underrating aspects can conceivably cancel out, doesn't mean that the metric understands an Anthony Davis better than it does an Andre Drummond. It overrates Drummond, and understands Davis even less...but like a broken clock, it might still end up being right sometimes.

But the fact that Davis' PER "looks right" doesn't give the stat credibility when comparing Davis to other players, and particularly when comparing Davis to other types of players.

By contrast, I would argue that +/- stats don't have a player archetype bias. It has flaws, but not that kind of flaw.

To be clear: When I use the term "archetype" here, I'm not referring to things like how many minutes a guy plays. We know that there's a danger in a player's RAPM making him look better than he actually is when the coach plays him only in specific situations he's well-suited for. That's a flaw, and it's not unreasonable to call it a "bias", but it's not a flaw/bias that's grounded in what kinds of box score columns the player tends to fill out.

And of course the bigger danger with +/- stats, which can be argued to encapsulate the above issue, is one of limited sample - aka, noise. With any kind of +/- stat we get, we have to ask ourselves about noise. Even if we're talking about a 25 year long study, I still ask those questions.

But issues of noise are pretty straight forward in the effect they have on our analysis: They decrease precision, aka increase uncertainty. If I see some +/- data and I have enough concerns about uncertainty, I may not feel comfortable adding that data into my analysis of the player, but it has no broader impact on my use of +/-. There's never a time where I would say, "Well RAPM says Player X looks better than I think he was, so since Player Y has similar height, weight, and box score shape, RAPM must overrate that player archetype."

Whereas when it comes to basically any box score-based metric, this isn't just something I do, it's an essential part of the process of gaining skill in using the metric.

Now, a natural response to me saying that would be, "But that's what I also do that when I'm using stats like PER, so why are you saying that's not okay?" And there what I'd say is that if you're comparing players of different archetypes, it's hard for me to have confidence in it, and I would say that in the case of guys like Davis, Baylor & Schayes, they really aren't the same archetypes.

All of that might seem quibbling over details given that I literally rank Davis ahead of these other guys and thus agree with PER in this case, but it's the general concern that made me respond.

[Doctor MJ]

Spoiler:

...

Okay:

First, you continuing to go "whatabout" with the problematic RAPM values as riposte to problematic PER values is getting annoying. Focus on the responses that have already been given or just leave it alone.

To try to focus us on the actual issue here:

You see folks being inconsistent about their behavior toward PER compared to RAPM.
You correctly see this as a possible source of epistemic bias.
You believe that this inconsistency is utterly unjustifiable.
I have previously given elaborations to try to make clear the difference, which I'll now scroll to see if you addressed in a way I feel compelled to respond to, but the way you start all this by pointing to the same thing again, makes me think we're not going to get much further here.

I know it can get frustrating when not seeing eye-to-eye with someone; we've been there between the two of us, right? (if you agree with nothing else I say in this post, I'm sure you'd agree with at least that much! )
But for my part, I think f4p has comported himself very well in this discussion, and I actually don't think he's gone far off point (at least, not much more than you have yourself). More than anything, I think you're both just kinda shooting past each other [as is often the case]; and to be fair to f4p, there has been [to my eye] some mixed messaging on your part (I'll elaborate below).

If I may stick my big nose into things and offer my own observations (however welcome they may or may not be).....


I gather that in some prior discussion [which I missed] you gave extensive elaboration/explanation for how various bizarre ["outlier"] RAPM values for certain players in certain years came about (in essence: explaining away the weirdness).
And you HAVE [to some degree, perhaps in more vague/broad terms] done the same for how/why Andre Drummond's PER result is wrong (that is: explain why it pegs him very high/good, when he's actually not).

It's where you [and perhaps a number of others] go from there [conclusions on how to then use the stats] that differs dramatically; which is basically what f4p has pointed out (which he can't very well do without drawing the comparison ["whatabout"]).


With RAPM, the tone and tenor of how to accept explanations for outlier/bizarre results was something like [from what I gather of this prior discussion]: "This is how that may have happened, and that makes sense. Stat still very useful, just gotta be concscientious of the context."

But f4p perceives that PER does not receive the same treatment.
He perceives [I think] that despite some adroit observations on your part for why PER gives a misleading result for Drummond ("explaining away the weirdness", perhaps similar(ish) to what you did for weird RAPM results??), the take-away impression on the stat is still something like: "Look how badly is misrepresented him. It's not a stat that should be trusted or given much of any weight in an argument."

I'm not sure if that was your intent, but I must admit that f4p is not the only one for whom that was the perceived insinuation.
Again, there is arguably some mixed messaging.

While you [in posts after the inciting reply] said this.....

In general my advise would be to use all complex stats - whether PER or RAPM - as first-pass analysis tools only, and then to use them along with other tools to drill down into the actual "what happened for the team".

And this....

I'm actually fond of stating that stats shouldn't be dismissed simply because you don't like a particular player outcome. Instead what you should do is ask why the stat became so misguided in this specific circumstance.

......that actually does contradict what you had said previously.

In the initial reply you had stated:

the existence of a guy like Andre Drummond (the other "AD") being able to trounce old-timers by this stat as well really makes clear how limited it is as a tool.

And....

so if a stat thinks Drummond is better than old Hall of Famers when the reality is he's not even good enough to play for a contender as more than a guy to fill out the bench, I can't take it [PER] as much of an argument.

This has the feel of a broad, summary judgement of the metric.
And it seems [at least as worded here] that PER is damnable and of marginal or negligible value simply for the mere existence of Drummond's result.

You also said----right on the heels of suggesting no stat should be dismissed because you don't like a result---the following:

In the case of Drummond, the answer is that use of box score-based stats can allow a big man to fill up the box score pretty dang well without him knowing really knowing what he's doing out there, and this can lead to a fool's gold situation where you think a player is your franchise player - complete with max contracts, re-designed systems around him, and even all-star appearances - when really he was never capable of being a starter-level player against playoff-level competition.

We then take what we learn from the data point and apply it back to the other players. Good performance in the metric obviously is not a damning thing, but if the metric cannot tell the difference between a superstar and and a benchwarmer, then you have to ask what information you can expect to glean by looking at the stat for new players.

This one is less direct (it's more about tone [and tone is SO hard to interpret in text])......but the bolded could be construed as saying "there's little-to-nothing of value to be gleaned from this stat, so why bother with it?"


So, you might have to forgive some of us readers a bit of confusion. Because in a number of ways it sort of feels like you're saying: "Never dismiss any stat out-of-hand because of a weird/bad result (except PER....it's crap)".......which is a contradiction in itself.


Now to be fair, you did also state that your low opinion of it is also based upon a belief [paraphrased] that the things credited by PER do not, in and of themselves, carry much value (that is: toward winning; impact; whereas, otoh, "impact" is EXACTLY what RAPM attempts to identify and credit).

fwiw, I'm not sure I entirely agree (about PER, I mean). I think a few things PER places value on generally [or at least frequently] positively impact the outcome, even if it gives equal credit to things that are not always equal (e.g. steals accrued at the expense of excessive gambling will be counted the same as steals that come within the context of sound team defense; not all assists are equal in value; not all blocks are equal in value; shooting efficiency is credited by PER but not nearly enough [imo], and thus points for points sake looms excessively large; etc).
What's more (and f4p pointed this out): PER often "gets it right", even if it does so sort of by coincidence---or rather by accident, if you will: because the really good players---in process of doing those things that tend to produce profound positive impact---frequently tend to also fill up the stat-sheet with those things which PER values.

I think there's a philosophical axe to grind here.
On the one hand, you might smirk and decide: "I'm not going to place any weight in a stat that only gets things right by accident!"
But on the other hand, one might suggest: "If it's getting things right often enough to pass the smell test, does the manner in which it got there really matter overmuch?"

idk......that's a philosophical debate that I'm not sure I'm all that interested in having, though I will comment a little further below regarding your visual aids.

But (fwiw) as f4p's pointed out repeatedly: PER rankings do actually dove-tail pretty well with the list results so far.
I might also point that the top 10 players by best single-season scaled rs PER (recall I'd done a scaled version to account for variations in league parity) are:

Kareem Abdul-Jabbar
Michael Jordan
Wilt Chamberlain
Stephen Curry ('16)
LeBron James
Julius Erving ('76 [ABA])
Bob Pettit
David Robinson
Shaquille O'Neal
(*Nikola Jokic)

The 10 best in single-season playoff scaled PER are:

Kareem Abdul-Jabbar
Hakeem Olajuwon
LeBron James
Julius Erving ('76 [ABA])
George Gervin
Michael Jordan
George Mikan
Shaquille O'Neal
Tim Duncan
(*Nikola Jokic)

*I've only updated my sheets as far as '20, but I'm almost certain '22 Jokic would make the cut in both rs and playoffs.

For a stat that isn't measuring anything important, it seems like it does a reasonable job of identifying who the best players are (at least at the highest levels). Obviously some "glitches" and "outliers", but that's why we cut it with OTHER indicators a pieces of information. If those other indicators more or less fall in line.....

And with regards to older players, it's a little bit of an "any port in a storm" situation: there's so much less information to go on. So can we really afford to be dismissive of a metric that generally does this^^ well at the high levels? (tbf, I do think there is a point where there are "diminishing returns" on PER's accuracy in gauging player quality, as we get further below the super-elite).


To be fair, at one other point you mentioned it was more an objection to **the manner in which PER was being used that you were speaking to. (**in a manner that basically says, "Look how high he ranks in PER. He must be super-good and belongs in the conversation here.")

However, f4p pointed out [correctly, I might add] that similar use is applied to other metrics in this project. Even by you:

I think another perspective to take is that our career playoff +/- leaderboard for the data we have goes:

1. LeBron
2. Duncan
3. Green

But similar use of these other metrics often doesn't seem to garner the same blowback (which again, is what f4p has pointed out).

.

Alright so, there's good stuff above, but I don't think I have anything really specific to say on it that isn't covered by my response to f4p or the comments I make below.

I do apologize again for my impatience with others.

Okay, so then back to the PER vs RAPM comparison, the issue here is that you're conflating precision and accuracy as if they are the same thing.

The issue with RAPM is one of variance or noise - aka limited precision.
The issue with PER is one of accuracy limited by bias.

To use the classic visual example:



PER is the image in the top right, RAPM is the bottom left. Because we have nothing in the bottom right, we need to use some of both the top-right and the bottom-left in our analysis, but that doesn't mean we treat their respective weaknesses identically.

This ties back into the philosophical axe to grind.

RAPM is aiming at the right target, the target you're supposed to be aiming at (the bull's eye), but does a sort of crappy job of consistently hitting it.

But PER isn't even aiming at the right target (it's shooting somewhere off the side), though it hits what it's aiming at pretty well.

I suppose I more or less agree, though I'd wager we differ in degrees.
For example, I suspect I don't think this "wrong" target that PER aims at is quite as far away from the bull's eye as you. And since it usually hits what it's aiming at, it thus tends to be "kinda close" to that bull's eye very often.......which makes it of some value, imo (even if it is wrong-headed [aiming for the wrong goal] in theory).

I'm saying there's literally a bias in PER toward big men,

Or is it simply that big men are "more important" in the game of basketball (generally and historically speaking)? This is perhaps another philosophical axe. As I'd posted regarding the demographics of our top 35: big men are "over-represented" (I think because it is, quite simply, a big man's game).

Anyway, I'll stop there I think.

[/quote]

So, the reason I keep pointing to Drummond is because his existence really makes clear that big men get overrated by the stat. It's not just a case of big men being more valuable because Drummond was not actually valuable, yet still looked like an all-star by box score all-in-ones, which led him to get a max contract and all-star appearances despite never developing the on-court awareness to be someone a contender wants on the floor when things actually matter.

Let me extend this out beyond PER. On bkref, there are 3 main box score all-in-ones that get used:

1. PER
2. Win Shares
3. BPM/VORP

Of those 3, I'd say that PER is the most biased toward Drummond-types and BPM is the least biased...but it is still biased because it still ranks Drummond as having more career value-add than Middleton, Crowder & Barnes, 3 players that I would consider to be not just better basketball players, but drastically so. I think it's correct that neither Crowder or Barnes were all-stars, nevertheless, they are far closer to all-star level players than Drummond ever should have been considered.

None of this is meant to imply I don't use these stats, but I'm cautious in using them, and more cautious in using them to make definitive statements the greater the difference in player type being compared.

[Fundamentals21]

Vote: Anthony Davis

I had to think some of this through, but really if you look at AD’s popularity, it went off after the Laker run, when people realized he could play at a top 5 level when really asked to.

A lot of AD's defense and timing, in the Laker years looks really impressive. If you watch the most recent tapes of him in the playoffs,on his chess matches with Joker and Curry, you'll see this guy really shining. The Lakers went on good D-rating runs, and while the Lakers have some good roleplayers like Austin Reaves, none of them can anchor a defense like AD. If you watch his actual talents like vertical, running, rebounding, or leaping, he’s certainly a legend of the game. His actual level of play in the playoffs is insane, relatively at a 30/10 level when asked to carry the load.

Defensive BPM
VORP
Rebounding
Blocks

All rate him pretty highly. AD's defense is top notch. He can end up in the Top 50 with some of these numbers.

Thinking through some more offense…

AD is reliable offensively with his scoring, sort of like in the impressive KG category. He anchored the NOP offense really well, before ending up as an elite #2 in Lakers. I see AD as scoring a reliable 25/12 on a top seed team. He’s a lot to handle because he really has it down with any big man in the league.

AD in 2018 really was shining, with some Cousins and Holiday action to a top 10 O-rating scoring 28/12. He pretty much destroyed GSW before being eliminated in the playoffs.I would say in this kind of setting, he’s like the lesser version of guys like KG or Duncan.

AD certainly won a title with older version of LeBron, so he has good winning capacity.

[OldSchoolNoBull]

ok, now put yourself in the shoes of people like me. instead of 2 replies with the same thing, make it more like 1000 and over years and years and you can see what it's like having to constantly defend even the most basic deductions from the box score. like saying, in the middle of a larger post, someone had a 25 PER, 0.250 WS48, and 8 BPM as a quick summary of why it was a good postseason run and getting "not everything is the box score" responses like you were supposed to stop and write 500 words about that postseason run because we can't just look at the very obviously elite box numbers and conclude it was an elite postseason. that's what i'm used to and it gets exhausting after a while.

Spoiler:

Okay:

First, you continuing to go "whatabout" with the problematic RAPM values as riposte to problematic PER values is getting annoying. Focus on the responses that have already been given or just leave it alone.

To try to focus us on the actual issue here:

You see folks being inconsistent about their behavior toward PER compared to RAPM.
You correctly see this as a possible source of epistemic bias.
You believe that this inconsistency is utterly unjustifiable.
I have previously given elaborations to try to make clear the difference, which I'll now scroll to see if you addressed in a way I feel compelled to respond to, but the way you start all this by pointing to the same thing again, makes me think we're not going to get much further here.

I know it can get frustrating when not seeing eye-to-eye with someone; we've been there between the two of us, right? (if you agree with nothing else I say in this post, I'm sure you'd agree with at least that much! )
But for my part, I think f4p has comported himself very well in this discussion, and I actually don't think he's gone far off point (at least, not much more than you have yourself). More than anything, I think you're both just kinda shooting past each other [as is often the case]; and to be fair to f4p, there has been [to my eye] some mixed messaging on your part (I'll elaborate below).

If I may stick my big nose into things and offer my own observations (however welcome they may or may not be).....


I gather that in some prior discussion [which I missed] you gave extensive elaboration/explanation for how various bizarre ["outlier"] RAPM values for certain players in certain years came about (in essence: explaining away the weirdness).
And you HAVE [to some degree, perhaps in more vague/broad terms] done the same for how/why Andre Drummond's PER result is wrong (that is: explain why it pegs him very high/good, when he's actually not).

It's where you [and perhaps a number of others] go from there [conclusions on how to then use the stats] that differs dramatically; which is basically what f4p has pointed out (which he can't very well do without drawing the comparison ["whatabout"]).


With RAPM, the tone and tenor of how to accept explanations for outlier/bizarre results was something like [from what I gather of this prior discussion]: "This is how that may have happened, and that makes sense. Stat still very useful, just gotta be concscientious of the context."

But f4p perceives that PER does not receive the same treatment.
He perceives [I think] that despite some adroit observations on your part for why PER gives a misleading result for Drummond ("explaining away the weirdness", perhaps similar(ish) to what you did for weird RAPM results??), the take-away impression on the stat is still something like: "Look how badly is misrepresented him. It's not a stat that should be trusted or given much of any weight in an argument."

I'm not sure if that was your intent, but I must admit that f4p is not the only one for whom that was the perceived insinuation.
Again, there is arguably some mixed messaging.

While you [in posts after the inciting reply] said this.....

In general my advise would be to use all complex stats - whether PER or RAPM - as first-pass analysis tools only, and then to use them along with other tools to drill down into the actual "what happened for the team".

And this....

I'm actually fond of stating that stats shouldn't be dismissed simply because you don't like a particular player outcome. Instead what you should do is ask why the stat became so misguided in this specific circumstance.

......that actually does contradict what you had said previously.

In the initial reply you had stated:

the existence of a guy like Andre Drummond (the other "AD") being able to trounce old-timers by this stat as well really makes clear how limited it is as a tool.

And....

so if a stat thinks Drummond is better than old Hall of Famers when the reality is he's not even good enough to play for a contender as more than a guy to fill out the bench, I can't take it [PER] as much of an argument.

This has the feel of a broad, summary judgement of the metric.
And it seems [at least as worded here] that PER is damnable and of marginal or negligible value simply for the mere existence of Drummond's result.

You also said----right on the heels of suggesting no stat should be dismissed because you don't like a result---the following:

In the case of Drummond, the answer is that use of box score-based stats can allow a big man to fill up the box score pretty dang well without him knowing really knowing what he's doing out there, and this can lead to a fool's gold situation where you think a player is your franchise player - complete with max contracts, re-designed systems around him, and even all-star appearances - when really he was never capable of being a starter-level player against playoff-level competition.

We then take what we learn from the data point and apply it back to the other players. Good performance in the metric obviously is not a damning thing, but if the metric cannot tell the difference between a superstar and and a benchwarmer, then you have to ask what information you can expect to glean by looking at the stat for new players.

This one is less direct (it's more about tone [and tone is SO hard to interpret in text])......but the bolded could be construed as saying "there's little-to-nothing of value to be gleaned from this stat, so why bother with it?"


So, you might have to forgive some of us readers a bit of confusion. Because in a number of ways it sort of feels like you're saying: "Never dismiss any stat out-of-hand because of a weird/bad result (except PER....it's crap)".......which is a contradiction in itself.


Now to be fair, you did also state that your low opinion of it is also based upon a belief [paraphrased] that the things credited by PER do not, in and of themselves, carry much value (that is: toward winning; impact; whereas, otoh, "impact" is EXACTLY what RAPM attempts to identify and credit).

fwiw, I'm not sure I entirely agree (about PER, I mean). I think a few things PER places value on generally [or at least frequently] positively impact the outcome, even if it gives equal credit to things that are not always equal (e.g. steals accrued at the expense of excessive gambling will be counted the same as steals that come within the context of sound team defense; not all assists are equal in value; not all blocks are equal in value; shooting efficiency is credited by PER but not nearly enough [imo], and thus points for points sake looms excessively large; etc).
What's more (and f4p pointed this out): PER often "gets it right", even if it does so sort of by coincidence---or rather by accident, if you will: because the really good players---in process of doing those things that tend to produce profound positive impact---frequently tend to also fill up the stat-sheet with those things which PER values.

I think there's a philosophical axe to grind here.
On the one hand, you might smirk and decide: "I'm not going to place any weight in a stat that only gets things right by accident!"
But on the other hand, one might suggest: "If it's getting things right often enough to pass the smell test, does the manner in which it got there really matter overmuch?"

idk......that's a philosophical debate that I'm not sure I'm all that interested in having, though I will comment a little further below regarding your visual aids.

But (fwiw) as f4p's pointed out repeatedly: PER rankings do actually dove-tail pretty well with the list results so far.
I might also point that the top 10 players by best single-season scaled rs PER (recall I'd done a scaled version to account for variations in league parity) are:

Kareem Abdul-Jabbar
Michael Jordan
Wilt Chamberlain
Stephen Curry ('16)
LeBron James
Julius Erving ('76 [ABA])
Bob Pettit
David Robinson
Shaquille O'Neal
(*Nikola Jokic)

The 10 best in single-season playoff scaled PER are:

Kareem Abdul-Jabbar
Hakeem Olajuwon
LeBron James
Julius Erving ('76 [ABA])
George Gervin
Michael Jordan
George Mikan
Shaquille O'Neal
Tim Duncan
(*Nikola Jokic)

*I've only updated my sheets as far as '20, but I'm almost certain '22 Jokic would make the cut in both rs and playoffs.

For a stat that isn't measuring anything important, it seems like it does a reasonable job of identifying who the best players are (at least at the highest levels). Obviously some "glitches" and "outliers", but that's why we cut it with OTHER indicators a pieces of information. If those other indicators more or less fall in line.....

And with regards to older players, it's a little bit of an "any port in a storm" situation: there's so much less information to go on. So can we really afford to be dismissive of a metric that generally does this^^ well at the high levels? (tbf, I do think there is a point where there are "diminishing returns" on PER's accuracy in gauging player quality, as we get further below the super-elite).


To be fair, at one other point you mentioned it was more an objection to **the manner in which PER was being used that you were speaking to. (**in a manner that basically says, "Look how high he ranks in PER. He must be super-good and belongs in the conversation here.")

However, f4p pointed out [correctly, I might add] that similar use is applied to other metrics in this project. Even by you:

I think another perspective to take is that our career playoff +/- leaderboard for the data we have goes:

1. LeBron
2. Duncan
3. Green

But similar use of these other metrics often doesn't seem to garner the same blowback (which again, is what f4p has pointed out).

.

Alright so, there's good stuff above, but I don't think I have anything really specific to say on it that isn't covered by my response to f4p or the comments I make below.

I do apologize again for my impatience with others.

Okay, so then back to the PER vs RAPM comparison, the issue here is that you're conflating precision and accuracy as if they are the same thing.

The issue with RAPM is one of variance or noise - aka limited precision.
The issue with PER is one of accuracy limited by bias.

To use the classic visual example:



PER is the image in the top right, RAPM is the bottom left. Because we have nothing in the bottom right, we need to use some of both the top-right and the bottom-left in our analysis, but that doesn't mean we treat their respective weaknesses identically.

This ties back into the philosophical axe to grind.

RAPM is aiming at the right target, the target you're supposed to be aiming at (the bull's eye), but does a sort of crappy job of consistently hitting it.

But PER isn't even aiming at the right target (it's shooting somewhere off the side), though it hits what it's aiming at pretty well.

I suppose I more or less agree, though I'd wager we differ in degrees.
For example, I suspect I don't think this "wrong" target that PER aims at is quite as far away from the bull's eye as you. And since it usually hits what it's aiming at, it thus tends to be "kinda close" to that bull's eye very often.......which makes it of some value, imo (even if it is wrong-headed [aiming for the wrong goal] in theory).

I'm saying there's literally a bias in PER toward big men,

Or is it simply that big men are "more important" in the game of basketball (generally and historically speaking)? This is perhaps another philosophical axe. As I'd posted regarding the demographics of our top 35: big men are "over-represented" (I think because it is, quite simply, a big man's game).

Anyway, I'll stop there I think.

So, the reason I keep pointing to Drummond is because his existence really makes clear that big men get overrated by the stat. It's not just a case of big men being more valuable because Drummond was not actually valuable, yet still looked like an all-star by box score all-in-ones, which led him to get a max contract and all-star appearances despite never developing the on-court awareness to be someone a contender wants on the floor when things actually matter.

Let me extend this out beyond PER. On bkref, there are 3 main box score all-in-ones that get used:

1. PER
2. Win Shares
3. BPM/VORP

Of those 3, I'd say that PER is the most biased toward Drummond-types and BPM is the least biased...but it is still biased because it still ranks Drummond as having more career value-add than Middleton, Crowder & Barnes, 3 players that I would consider to be not just better basketball players, but drastically so. I think it's correct that neither Crowder or Barnes were all-stars, nevertheless, they are far closer to all-star level players than Drummond ever should have been considered.

None of this is meant to imply I don't use these stats, but I'm cautious in using them, and more cautious in using them to make definitive statements the greater the difference in player type being compared.

So, I felt compelled to add my two cents into this debate, as I do see it from both sides.

I get where f4p is coming from because I've had those same frustrations from time to time. There are certain impact metric zealots who are, imo, overly and unfairly dismissive of box stats. Like f4p said, you'll present a bunch of impressive box stats for a player and be met with, essentially, "that's just box stuff, blah blah blah", as though it doesn't matter. I don't really see that much of this in this project specifically, but it is there.

Look, in the most simplistic terms, box stats directly measure a player's individual production, and impact metrics measure a team's performance with said player vs without. Both are necessary. I don't agree with the sort of snobbish dismissal of box stats that I see in proponents of impact metrics, and I also don't agree with the few who are too dismissive of impact metrics because said metrics rate certain players high enough to invite skepticism(we had threads upon threads of debate about Stockton, we're talking about Draymond now).

If a player's box and impact metrics agree with each other, then it's an easy analysis, whichever way it goes.

If a player is filling up the box sheet, but it's not reflected in his impact metrics, then you have to ask why his production isn't helping his team as much as you would expect it to and acknowledge that those box stats are maybe misleading(to varying degrees depending on the player and context), as Doc is doing with Drummond.

If a player's impact metrics are off the charts but his box numbers are pedestrian, then you have to ask what the player is doing that isn't showing up in the box stats(usually defense or playmaking that doesn't show up as assists).

But you need all of it to get the full picture and to dismiss any of it is wrong.

Now, while I do defend the value and necessity of box stats, I will say on the specific issue of PER that's being debated...I don't have much use for it, and the only time I really use it at all is when I'm looking at guys from the 50s where there is really not much available statistically. Further, in all my years on RealGM, I've always gotten the sense that PER is the most derided of the box stats.

I appreciate all of the philosophical arguments against it that have been made by Doc and others(I really like that dartboard chart, it illustrates that point really well), but for me, the argument against is more practical. I just don't think it tells me anything specific. What I mean by that is this:

TS(and it's associated stats - rTS, TS Add, TS+, etc) measures scoring efficiency. It is extremely valuable because it is often the difference between the great scorers and the chuckers.

WS/48(which I prefer to raw WS) tells me how important a player was to his team's success(whatever that success may be).

BPM tells me how much better(or worse) a player is than a replacement-level player.

On/Off tells me how much better(or worse) the team performed with the player on the floor.

RAPM tells me the same-ish, but with offensive and defensive components.

What does PER tell me, besides that the player recorded a bunch of counting stats? I just don't feel like it's telling me anything I can't get more clearly/accurately from other stats.