RealGM 2023 Top 100 Project - #33 (Deadline 5:00AM PST on 10/12/23)

[tsherkin]

Actually, Stockton penetrated more than Nash, less than Magic. Those PnR plays were based on his attacking the basket and he shot more and a higher percentage at the rim than Nash ever did (Nash would pull up short more for little floaters and the like).

Based on what? A lot of the time, Stockton never broke the plane of his defender, just moved laterally while Malone rolled. A little in-out dribble, curl around to the foul line... and then perhaps the best pocket passes we've seen. I wouldn't say he penetrated more than Nash. Also, with respect to percentage at the rim, it's worth noting that he shot less and a lot of his earlier teams were even faster than the SSOL Suns, which clouds those numbers if you're trying to separate out actual penetration. Stockton very regularly didn't get under the extended foul line on his PnRs, so I don't know that it's an effective demonstration of penetration to mention that for him. I don't think that's a BAD thing, because it usually meant Malone was doing demonic things on the interior, which was the point of course. But for the sake of this specific comment, right?

[trex_8063]

With that in mind, I think we can guess where that late-career RAPM is coming from:

1997: 75% of minutes played with Malone
1998: 84% of minutes played with Malone
1999: 89% of minutes played with Malone
2000: 92% of minutes played with Malone
2001: 87% of minutes played with Malone
2002: 85% of minutes played with Malone
2003: 90% of minutes played with Malone

Poor guy!

Objection: conjecture.

The above speculates [or assumes] that:
1) It is Malone that has the greater impact by a comfortable margin (big enough to drag Stockton's up from what would otherwise be sort of pedestrian numbers, apparently, with <85% overlap in some years [<75% in one]).


And 2) RAPM is not able to parse out the credit accurately because too much minute overlap.

However.....
In '04 Kevin Garnett led the league in [J.E.'s] RAPM by an absurd margin over 2nd place (him at +8.6, 2nd place +5.6).
Sam Cassell (a damn fine player in his own right in '04, yes?) played 89.3% of his minutes on the court with Garnett, yet his RAPM is just +3.0 (another source had him at just +2.3 while Garnett is +8.7).
Latrell Sprewell played 86.3% of his minutes alongside Garnett that same year, yet had a completely mediocre RAPM (around zero by both sources).

Troy Hudson played 90% of his '03 minutes on-court with Garnett [again league-leader in RAPM at +5.4], and yet had an RAPM of like -1.6.
In '05 he played 80% of his minutes with Garnett [+4.4, tied for 7th in league], yet had an RAPM of nearly -3.

In '98, Luc Longley played 89.1% of his minutes with Michael Jordan [who had a 4th in the league +6.15 PI RAPM]......and yet his was -0.61.


That's not a huge sample, but let me say I didn't have to search long/hard to find a number of "cherry-picked" examples (there are numerous more I can drudge up). I think it's nonetheless sufficient to suggest that time on-court with a big impact player is no guarantee (especially year over year, as is the case with Stockton) of impressive impact numbers (apparently not even when you're as good as '04 Cassell, and playing nearly 90% next to a BIGGER impact player than Karl Malone ever was).


And if this argument did have some teeth, it's curious it never came up when its authors were championing Scottie Pippen (83.2% of his minutes next to Jordan in '97, 82.5% in '98).


And finally, at the very least I'd have to object on logisitical grounds wrt '97.
In '97....
Stockton played 74.5% of his minutes with Malone. Though on the flip-side, Malone played 72% of his minutes with Stockton.
ADDITIONALLY:
Malone got 82.1% of his minutes on court with Jeff Hornacek (vs only 61.8% for Stockton [-20.3]).
Malone got 78.3% of his minutes next to Bryon Russell [4th-best player] (vs only 60.7% for Stockton [-17.6]).
Malone got 53.1% of his minutes next to Greg Ostertag [starting C] (vs only 45.9% for Stockton [-7.2]).

Also looking at the WORST rotational players on the roster (Morris, Anderson, and Foster).....
Stockton played 29.1% of his minutes next to a rookie Shandon Anderson [not good] (Malone played something less than 21% of his minutes with Anderson [don't know exactly, but it was so infrequent it doesn't register as one of the 20 most common 2-man line-ups]).
Stockton played 26.1% of his minutes next to Chris Morris [who had a godawful down year] (again, something <21% at least for Malone; too infrequent to register).
Stockton played 21.6% of his minutes next to Greg Foster (again, too infrequent to register for Malone).

Are we REALLY going to look at all of this and say with a straight face that Stockton is the one that had the line-up advantage that year?



Anyway, I'm glad as hell he's finally off the table.

[trex_8063]

Actually, Stockton penetrated more than Nash, less than Magic. Those PnR plays were based on his attacking the basket and he shot more and a higher percentage at the rim than Nash ever did (Nash would pull up short more for little floaters and the like).

Based on what? A lot of the time, Stockton never broke the plane of his defender, just moved laterally while Malone rolled. A little in-out dribble, curl around to the foul line... and then perhaps the best pocket passes we've seen. I wouldn't say he penetrated more than Nash. Also, with respect to percentage at the rim, it's worth noting that he shot less and a lot of his earlier teams were even faster than the SSOL Suns, which clouds those numbers if you're trying to separate out actual penetration. Stockton very regularly didn't get under the extended foul line on his PnRs, so I don't know that it's an effective demonstration of penetration to mention that for him. I don't think that's a BAD thing, because it usually meant Malone was doing demonic things on the interior, which was the point of course. But for the sake of this specific comment, right?

"Old man" Stockton ['97 onward] had 34% of his shots coming at the rim (vs just 18.2% for Nash from '05-'11 [peaked at 23.3% in Phoenix]). Even accounting for Nash's generally larger shot volume, that equates to Stockton taking more [in absolute per 100 numbers] shots at the rim than Nash......and not in a fast era (and again, at a point in his career where he no longer had the first step he'd had circa-1990).

[tsherkin]

"Old man" Stockton ['97 onward] had 34% of his shots coming at the rim (vs just 18.2% for Nash from '05-'11 [peaked at 23.3% in Phoenix]). Even accounting for Nash's generally larger shot volume, that equates to Stockton taking more [in absolute per 100 numbers] shots at the rim than Nash......and not in a fast era (and again, at a point in his career where he no longer had the first step he'd had circa-1990).

Right, but Old Man Stockton (97 onward) was also shooting 8.5 FGA/g, so proportions are a little out of whack, yes? Nash from 05-11 was shooting 12.0 FGA/g.

So if you actually look at FGA inside the RA, you're talking about 2.18 for Nash and 2.89 for Stockton. That isn't actually a huge difference. And how many of those were in transition (for both), right?

[trex_8063]

"Old man" Stockton ['97 onward] had 34% of his shots coming at the rim (vs just 18.2% for Nash from '05-'11 [peaked at 23.3% in Phoenix]). Even accounting for Nash's generally larger shot volume, that equates to Stockton taking more [in absolute per 100 numbers] shots at the rim than Nash......and not in a fast era (and again, at a point in his career where he no longer had the first step he'd had circa-1990).

Right, but Old Man Stockton (97 onward) was also shooting 8.5 FGA/g, so proportions are a little out of whack, yes? Nash from 05-11 was shooting 12.0 FGA/g.

So if you actually look at FGA inside the RA, you're talking about 2.18 for Nash and 2.89 for Stockton. That isn't actually a huge difference. And how many of those were in transition (for both), right?

I'll take your figures as accurate, though not necessarily your conclusion: Stockton attained that advantage of 0.71 additional shots at the rim per game despite averaging 30.2 mpg [vs. 34.1 mpg for Nash '05-'11], and playing for an average pace of 89.3 [vs. an average 95.7 for Nash].

If we look at this in attempts per 100 possessions on the court:
Nash is averaging about 3.20 attempts at the rim.
Stockton is averaging 5.13.

Does this still look like an insignificant difference? ('cause I don't think a +60% edge is insignificant)

[DSMok1]

This is all super-helpful. We're lucky to get it right from the source!

Your point specifically about replacement level is an important one, and one I'd expect many haven't really pondered.

Basically: The lower you set the replacement level, the more longevity-oriented the results will tend to come out.

I've always found the frustrating thing with replacement level - which is a really, really good idea - is the inability to have an objective answer to where it should be. Is -2 too low? Maybe...maybe not.

I'm curious what analysis you've done to try to figure out out what's most appropriate, and how it change depending on what you're looking for.

You mention about appropriateness for Top 100, and I'm curious about what precisely you mean. I'll say up front that you could say I effectively do rough things like this as I scale my analysis because I have a tendency to compare superstars to each other primarily based on their prime, which means diminishing the effective weighting of longevity - but this is not the same thing as a formal statistical process.

When thinking about "replacement level" formally, in the regular season, I'd tend to think of it as basically the level of the entirety of player population playing below a certain MPG threshold, which is something you could regress for. You could in principle do the same thing for the playoffs, or even specific rounds of the playoffs, but the sample would naturally become more and more problematic. Does my thinking sound like you think about it? Do you see any issues with it that cause you to differ?

Replacement level is the level at which you should be able to find a veteran minimum salary player to fill an open roster spot. It was found by regression to be around -2. Note that often worse players are played for development purposes, below a -2 level. Rookies in particular.

Now, there are various schools of thought on how a player should be valued. Replacement level is a good starting point (i.e. we know the value of a veteran minimum contract). We can estimate points per 100 team possessions above replacement level, which is VORP. But should we look at things in a linear manner? In the playoffs, the team rotations are shortened. If the goal is a title, shouldn't we pay players based on CORP? Championship probability added versus replacement players? It's an open research area. I have more thoughts on this but I need to assemble them coherently before posting further.

[DSMok1]

Just quoting a few small sections:

Center's assists are not usually "higher value" than a guard. That the stat quite literally penalizes guards for "handling the ball more" when handling the ball produces better offenses...

...Is rather problematic

Just like favoring defensive stats(blocks and steals) from a small because of replacement level because data which doesn't correlate well happens to correlate better than all the other not very correlative box numbers. And that gets us to Stockton being a steal-leader whose assist-count far outpaces his actual creation...
...

You say "even", but I'd imagine it's an easier to be an outlier when Stockton played. My impression is also that Magic was a much better shot creator with the "quality" of his creations being much higher. Ditto with Nash. Stockton didn't really pry or penetrate which is a big limitation for a playmaker. I'd guess stockton's "quality of creation" was actually pretty low:

.....

And on that note...

Going to clear up a few questions about BPM/VORP here:
I would say BPM is fairly balanced guards vs. big men. There are different weights for many of the box score statistics based on the estimated position of the player and the estimated offensive role of the player.

BPM IS biased towards players that produce good box scores. The argument should be primarily about whether the player's box score statistics accurately reflect the player's actual impact.
.

I would say being biased towards players that "produce good box scores" is very naturally going to lead to bias towards smaller ones, especially if we are going to curve up stats like Blocks, which already overrepresent what smalls offer in terms of rim-deterrence.
....

Even if this sort of curve makes the metric more accurate with a league-worth of players(And I'd be interested if you could elaborate on the regressions you ran), for the purposes of this project where a ton of defenders who would be the runaway leaders on their teams in terms of "rim-load" still on the board, and countless more who would significantly outpace any guard, I'm skeptical of this metric's balance.

Everything you say is valid.

The only correction is that in BPM assists are weighted the same for everyone on offense. Assists are also a positive for bigs on defense, as they are a strong indicator of a big's overall awareness. There is an issue right now where this is not corrected for the big's offensive role; hence Jokic's very high defensive BPM with the current setup.

[tsherkin]

I'll take your figures as accurate, though not necessarily your conclusion: Stockton attained that advantage of 0.71 additional shots at the rim per game despite averaging 30.2 mpg [vs. 34.1 mpg for Nash '05-'11], and playing for an average pace of 89.3 [vs. an average 95.7 for Nash].

If we look at this in attempts per 100 possessions on the court:
Nash is averaging about 3.20 attempts at the rim.
Stockton is averaging 5.13.

Does this still look like an insignificant difference? ('cause I don't think a +60% edge is insignificant)

Yeah, I still don't think that means a whole lot because of how easy it easy for Stockton in his limited minutes and shooting volume to retain the proportion of shot types that he was feeding on, to be honest. I see what you're attempting to demonstrate mathematically, I just don't agree with the conclusion. Playing fewer minutes and taking fewer shots changes things about the types of shots you get and the energy level you have, as does the role you fill.

I do wish we had numbers from Stockton's prime. But I've watched a fair number of Utah games and a ton of footage of that PnR. And ultimately, this is something of a semantic argument, because even if I am correct, I wouldn't consider it a dig at Stockton or a boon for Nash, to be honest. Both of them were very efficient low-volume scorers in the RS, with Nash peaking higher in the RS and then generally through the PS. So this is a style question.

My main initial involvement was because the PnR really didn't send Stockton to the rim as often as it sent him left to right towards the foul line so that Malone could roll to the low left block for the J or all the way to the rim. Stockton was extremely selective about his FGAs and the spots he filled on the floor, which generally worked out very, very well against most opponents. That was sort of the base conceit of Utah's offense, really.

[OldSchoolNoBull]

...

Nomination: Rick Barry

Choosing between Kidd and Barry here...I honestly like Kidd better and was going to vote for him before a last minute flip. Rick Barry was the #1 on two championship teams and four Finals teams(including his ABA years here) and it's difficult for me to look past that. Even if I don't look at his ABA years, he won the ring as the #1 in 1975, something Kidd never did. He was not a terribly efficient scorer, particularly for his volume, but he was still more efficient than Kidd while still being a solid playmaker and rebounder. Kidd was the superior playmaker and defender, and has significantly more longevity, but...I might be too swayed by 1975.

Not 100% comfortable with this and part of me still hopes Kidd prevails.

...

Are you counting 1969 when Barry was injured halfway into the season and his team went on and won the championship without him?

I was, and I hadn't realized that Barry was hurt that year. Hmm. That was also the best team he was ever on by SRS, I think. Significant mistake on my part. There's still 1975 though.

Kidd won anyway.

[Doctor MJ]

This is all super-helpful. We're lucky to get it right from the source!

Your point specifically about replacement level is an important one, and one I'd expect many haven't really pondered.

Basically: The lower you set the replacement level, the more longevity-oriented the results will tend to come out.

I've always found the frustrating thing with replacement level - which is a really, really good idea - is the inability to have an objective answer to where it should be. Is -2 too low? Maybe...maybe not.

I'm curious what analysis you've done to try to figure out out what's most appropriate, and how it change depending on what you're looking for.

You mention about appropriateness for Top 100, and I'm curious about what precisely you mean. I'll say up front that you could say I effectively do rough things like this as I scale my analysis because I have a tendency to compare superstars to each other primarily based on their prime, which means diminishing the effective weighting of longevity - but this is not the same thing as a formal statistical process.

When thinking about "replacement level" formally, in the regular season, I'd tend to think of it as basically the level of the entirety of player population playing below a certain MPG threshold, which is something you could regress for. You could in principle do the same thing for the playoffs, or even specific rounds of the playoffs, but the sample would naturally become more and more problematic. Does my thinking sound like you think about it? Do you see any issues with it that cause you to differ?

Replacement level is the level at which you should be able to find a veteran minimum salary player to fill an open roster spot. It was found by regression to be around -2. Note that often worse players are played for development purposes, below a -2 level. Rookies in particular.

Now, there are various schools of thought on how a player should be valued. Replacement level is a good starting point (i.e. we know the value of a veteran minimum contract). We can estimate points per 100 team possessions above replacement level, which is VORP. But should we look at things in a linear manner? In the playoffs, the team rotations are shortened. If the goal is a title, shouldn't we pay players based on CORP? Championship probability added versus replacement players? It's an open research area. I have more thoughts on this but I need to assemble them coherently before posting further.

Ah, so you've regressed using players a veteran minimum as a group and found they rate as a -2? That's awesome, and an improvement on the approach I was thinking.

Re: VORP vs CORP. Good questions, and yes I agree that CORP is the better approach for both Top 100 and contract consideration.

My thought would be though as beautifully simple and object as veteran minimum as definitionally replacement level in the regular season, it loses some of its spot-on-ness for the playoffs because:

1. Replacement level is dependent on competition, competition level goes up with each round of the playoffs, and so whatever level a veteran minimum represents, if it can be hooked to "replacement level" in the regular season, it seems like it shouldn't be good enough in the playoffs.

2. Come playoff time, teams are largely playing their best players the big minutes rather than worrying about development.

This then would have me thinking again of an approach where players below a certain MPG threshold would be the way to dynamically determine base player level at each stage of the playoffs...assuming sufficient sample to damp down the noise to allow us to achieve meaningfully stable measures.

Thoughts?

[tsherkin]

1. Replacement level is dependent on competition, competition level goes up with each round of the playoffs, and so whatever level a veteran minimum represents, if it can be hooked to "replacement level" in the regular season, it seems like it shouldn't be good enough in the playoffs.

2. Come playoff time, teams are largely playing their best players the big minutes rather than worrying about development.

This then would have me thinking again of an approach where players below a certain MPG threshold would be the way to dynamically determine base player level at each stage of the playoffs...assuming sufficient sample to damp down the noise to allow us to achieve meaningfully stable measures.

Thoughts?

Hmm.

Wouldn't it be appropriate then to look at the postseason once its finished and recalculate a replacement level based on the players who got minutes in the PS? And then do your re-evaluation of players based on that new VORP?

[Doctor MJ]

1. Replacement level is dependent on competition, competition level goes up with each round of the playoffs, and so whatever level a veteran minimum represents, if it can be hooked to "replacement level" in the regular season, it seems like it shouldn't be good enough in the playoffs.

2. Come playoff time, teams are largely playing their best players the big minutes rather than worrying about development.

This then would have me thinking again of an approach where players below a certain MPG threshold would be the way to dynamically determine base player level at each stage of the playoffs...assuming sufficient sample to damp down the noise to allow us to achieve meaningfully stable measures.

Thoughts?

Hmm.

Wouldn't it be appropriate then to look at the postseason once its finished and recalculate a replacement level based on the players who got minutes in the PS? And then do your re-evaluation of players based on that new VORP?

I think calculating replacement level based specifically on the context of that round in that season would be something potentially ideal but I worry for sample size. My guess is that you'd need to do this based on a larger epoch in order to get something decently stable, despite the fact that actual changes over the years would get dismissed as noise.

In general I think I'd be looking for an epoch length that allowed me to get similar replacement level values, and thus use a single replacement value estimation when comparing performances in seasons near enough in time that I didn't think the landscape had changed significantly.

I don't think I'd want to replace regular season VORP based on playoff data. Just from a regular season MVP perspective, you achieve that regular season value in a context where the veteran minimum seems like a pretty dang good replacement level definition.

[tsherkin]

I think calculating replacement level based specifically on the context of that round in that season would be something potentially ideal but I worry for sample size. My guess is that you'd need to do this based on a larger epoch in order to get something decently stable, despite the fact that actual changes over the years would get dismissed as noise.

For the round seems too small, I meant more for the entire postseason.

In general I think I'd be looking for an epoch length that allowed me to get similar replacement level values, and thus use a single replacement value estimation when comparing performances in seasons near enough in time that I didn't think the landscape had changed significantly.

Oh, you mean like a multi-year sample of postseasons? That makes plenty of sense.

I don't think I'd want to replace regular season VORP based on playoff data. Just from a regular season MVP perspective, you achieve that regular season value in a context where the veteran minimum seems like a pretty dang good replacement level definition.

Yeah, I follow you, not a replacement, but another tool for a specific context. That's how I read it in the first place.

[penbeast0]

...

Nomination: Rick Barry

Choosing between Kidd and Barry here...I honestly like Kidd better and was going to vote for him before a last minute flip. Rick Barry was the #1 on two championship teams and four Finals teams(including his ABA years here) and it's difficult for me to look past that. Even if I don't look at his ABA years, he won the ring as the #1 in 1975, something Kidd never did. He was not a terribly efficient scorer, particularly for his volume, but he was still more efficient than Kidd while still being a solid playmaker and rebounder. Kidd was the superior playmaker and defender, and has significantly more longevity, but...I might be too swayed by 1975.

Not 100% comfortable with this and part of me still hopes Kidd prevails.

...

Are you counting 1969 when Barry was injured halfway into the season and his team went on and won the championship without him?

I was, and I hadn't realized that Barry was hurt that year. Hmm. That was also the best team he was ever on by SRS, I think. Significant mistake on my part. There's still 1975 though.

Kidd won anyway.

Yes, Barry in 75 had an amazing playoff run (over my Bullets, I might add). I just was never that impressed by him the rest of his career and his personality was pretty toxic.

[OhayoKD]

Just quoting a few small sections:

Center's assists are not usually "higher value" than a guard. That the stat quite literally penalizes guards for "handling the ball more" when handling the ball produces better offenses...

...Is rather problematic

Just like favoring defensive stats(blocks and steals) from a small because of replacement level because data which doesn't correlate well happens to correlate better than all the other not very correlative box numbers. And that gets us to Stockton being a steal-leader whose assist-count far outpaces his actual creation...
...

You say "even", but I'd imagine it's an easier to be an outlier when Stockton played. My impression is also that Magic was a much better shot creator with the "quality" of his creations being much higher. Ditto with Nash. Stockton didn't really pry or penetrate which is a big limitation for a playmaker. I'd guess stockton's "quality of creation" was actually pretty low:

.....

And on that note...

Going to clear up a few questions about BPM/VORP here:
I would say BPM is fairly balanced guards vs. big men. There are different weights for many of the box score statistics based on the estimated position of the player and the estimated offensive role of the player.

BPM IS biased towards players that produce good box scores. The argument should be primarily about whether the player's box score statistics accurately reflect the player's actual impact.
.

I would say being biased towards players that "produce good box scores" is very naturally going to lead to bias towards smaller ones, especially if we are going to curve up stats like Blocks, which already overrepresent what smalls offer in terms of rim-deterrence.
....

Even if this sort of curve makes the metric more accurate with a league-worth of players(And I'd be interested if you could elaborate on the regressions you ran), for the purposes of this project where a ton of defenders who would be the runaway leaders on their teams in terms of "rim-load" still on the board, and countless more who would significantly outpace any guard, I'm skeptical of this metric's balance.

Everything you say is valid.

The only correction is that in BPM assists are weighted the same for everyone on offense. Assists are also a positive for bigs on defense, as they are a strong indicator of a big's overall awareness. There is an issue right now where this is not corrected for the big's offensive role; hence Jokic's very high defensive BPM with the current setup.

My bad then. I got assists being more valuable for bigger players from this:

Assists are worth far more for a post player than for a point guard. Point guards handle the ball a lot and usually generate lower value assists. A center’s assists are usually very high value. In addition, post players that pass well are typically better defenders.

 
Assists as an indicator for defensive awareness is interesting. Have you run a regression where you tested using assists as a defensive boon for everyone as opposed to just bigs(maybe scaling the effect up based on bigger estimated positions)?