Some Historical Plus-Minus

[Djoker]

Are you done or almost done the 1996 Bulls season? Sorry for asking but just curious how it's coming along.

You said that 1996 is your priority and then 1993 IIRC.

[rk2023]

I did Ewing, 21 of the 25 games listed on the sample were full - so I only used those.

Code: Select all

Knicks +/- Per 48: -1.69 (Average MOV on the season was -0.2)

Ewing +/- Per 48: 1.17

NYK +/- Per 48 w/o Ewing: -12.87 

+/- per 48 Swing: 14.04

Despite playing on better teams later on (Knicks were 82-82 in 1990 & 91), these seasons seem to be Ewing's apex (1990 being the better one). Assuming that these samples hold towards being representative of Ewing's lift, he grades out as a solid floor-raiser. This track record parallels what Moses was able to do in Houston and is paraded for. Not insinuating Ewing > Moses, but interesting food for thought ITO ranking two all-timers thought of in the 20s range(s) historically.

[rk2023]

.

If you happen to have any, some Dominique Wilkins data would be greatly appreciated

[Squared2020]

.

[ceiling raiser]

I wonder if Magic's +/- data will continue to be on par with Jordan's as the sample increases.

[rk2023]

Here's some updated 1993 data:

Michael Jordan,CHI Opp Score | OPoss O PTS | DPoss D PTS | P/M
6-Nov-92 CLE 101-96 | 79 83 | 79 87 | -4
9-Nov-92 IND 102-97 | 79 92 | 82 87 | 5
22-Nov-92 PHX 128-111 | 91 114 | 88 94 | 20
21-Dec-92 MIA 86-82 | 77 78 | 78 71 | 7
12-Jan-93 ORL 122-106 | 75 92 | 77 71 | 21
26-Jan-93 DAL 123-88 | 63 74 | 62 36 | 38
3-Feb-93 SAC 107-88 | 81 79 | 81 63 | 16
17-Feb-93 UTA 114-96 | 67 75 | 64 59 | 16
11-Mar-93 MIA 95-97 | 80 78 | 81 79 | -1
20-Mar-93 WSB 126-101 | 67 86 | 66 63 | 23
2-Apr-93 NJN 118-105 | 82 96 | 81 79 | 17
9-Apr-93 ATL 88-87 | 92 85 | 95 78 | 7


Sorry it is isn't nicely formatted. I'm being lazy.

Using the same methods as earlier, I got a +/- Per 48 of 17.64 for Jordan where the Bulls were 13.0 Per 48 in that sample. It's only 12 games, and just to point out something that checks out and is provided on Squared's Website itself - the sample does overstate CHI here compared to their MOV of 6.3 / game on the season and 57-25 record. Nonetheless, the more info the merrier

When I get more time, I plan to look into adding O and D splits for all the Bird/Magic/MJ/Robinson data I've gotten to so far - as well as add opponent info for some sensitivity analysis all to be published to Tableau and/or as part of an article.

[DraymondGold]

.

Loving the 1993 RAPM! Great work as always

Would you happen to have the rotation data for the 76ers with Erving and/or Moses? I'm interested to see how much Erving's slightly inconsistent raw plus/minus from year to year could be contextualized by who he's sharing the court with (e.g. how much is it from playing with his bench players vs just not fitting as well with the starters, etc.).

[colts18]

[quote="Squared2020"][/quote]
I see that you posted lineup stint data on your twitter. Do you have the lineups for both the 85 and 88 Celtics and Lakers? Do you have that data for the 96 Bulls? Is this data available for 1991? I want to compare those elite lineups with today's lineups. It would also be great for showing the MJ numbers without Pippen and vice versa.

[Squared2020]

.

[colts18]

I see that you posted lineup stint data on your twitter. Do you have the lineups for both the 85 and 88 Celtics and Lakers? Do you have that data for the 96 Bulls? Is this data available for 1991? I want to compare those elite lineups with today's lineups. It would also be great for showing the MJ numbers without Pippen and vice versa.

I had posted the Celtics one on Twitter: https://x.com/Squared2020/status/1703474045038244002?s=20

1985 Lakers here: https://x.com/Squared2020/status/1703472106288435679?s=20

Here's 1988 Celtics and Lakers lineup data: https://x.com/Squared2020/status/1707871820098183272?s=20

Here's 1991 and 1996 Bulls lineup data: https://x.com/Squared2020/status/1707872475470860451?s=20

Hopefully you can get them all there. I don't have the time to update the media section of my site to hold those images.

Thanks so much. I made a thread on the Plus/Minus numbers for Jordan based on your data. SPOILER: MJ is really good.

I did a preliminary analysis on the Lakers and Celtics data. I noticed the Celtics on numbers were really bad then checked up to find that the sample for the 1985 Celtics doesn't include 36 games where they went 33-3 . The Lakers data had James Worthy coming out not well. He suffered from the same issues that Rodman/Grant did in the Bulls sample where the 2 main stars played well without the 3rd star (Worthy/Rodman/Grant). I am curious if that is a trend. I don't know how long it takes you to generate this data, but can you provide the 1991 Celtics/Lakers and 1988 Bulls lineup data?

[Squared2020]

.

[ceiling raiser]

Can't find the other recent thread, so wanted to bump this.

There is a lot of sparring around the MJ stuff, but can we take a second to appreciate Magic here? I feel like his all-time stock had fallen in the past few years, but was great to see him second in the 85-96 expanding sample by Squared2020.

[x]

Read on Twitter

[/x]

[DraymondGold]

We have a new post on Squared2020’s full-career RAPM data! This new set includes 20% of total games from 1985, 1987, 1988, 1990, 1991, 1993, and 1996. It’s an astounding amount of work (1535 games tracked!!), and we really are getting to the sample size where we can start to make some higher confidence comparisons with these top players. Here are the links:

Spoiler:


New Historical RAPM: 1985–1996: https://squared2020.com/2025/01/26/historical-rapm-1985-1996/

1970 RAPM (a small sample, not included in multi-year sample above): https://squared2020.com/2021/12/19/1969-1970-nba-rapm/
1980 RAPM (a small sample, not included in multi-year sample above): https://squared2020.com/2021/04/11/1979-80-nba-rapm/

1985 RAPM: https://squared2020.com/1984-85-rapm/
(Squared2020 has 1987 plus minus data, but to my knowledge hasn’t published an RAPM for that year)
1988 RAPM: https://squared2020.com/1987-88-nba-rapm/
1990 RAPM: https://squared2020.com/2025/01/26/1989-1990-nba-rapm/
1991 RAPM: https://squared2020.com/2021/09/11/1990-1991-nba-rapm/
1993 RAPM: https://squared2020.com/2023/08/25/1992-1993-nba-rapm/
1996 RAPM: https://squared2020.com/2022/07/18/1995-1996-nba-rapm/
Note some of the single-year pages have not been fully updated to include some of the latest tracking, which was included in the 1985–1996 Historical sample.

In terms of sample size: Stars play maybe ~65 possessions per game. Assuming they play ~75 games per season, then we have roughly ~6–6.25 seasons worth of data for Jordan, ~3.5-3.75 seasons for Magic, ~1.75 seasons for Robinson, 2.25–2.5 seasons for Ewing, 1.5–1.75 seasons for Kareem, 1.5–1.75 seasons for Hakeem, 2.25–2.5 seasons for Ewing, 1–1.25 seasons for Stockton, 0.75–1 season for Shaq, ~4.5 seasons for Pippen, ~3.25 seasons for Bird.

Starting from the top, Jordan continues to look like a GOAT candidate in the impact data, as expected. He looks like the clear best offensive player in this stretch (clearing Magic -- though that part wasn't necessarily expected), while being one of the best non-big perimeter defenders (he’s nearly tied with Pippen). In overall impact, even his low evaluation would put him above every player’s mean evaluation save Magic and regular season David Robinson. He’s 89% more valuable than Hakeem’s mean evaluation, and 16% more than prime Magic’s. Wow!

Magic looks next best. This sample for him is almost entirely in his best 7-year stretch (~3.5 seasons worth of data from 85-91, plus maybe ~15 games from his 1996 return). He looks 2nd best on offense as well, but is surprisingly positive on defense. He’s also the only player with a sizably high lower-bound (which would rank 9th overall). Where is this defensive value coming from? Steals? Having a high offensive load that either allows defensive teammates to rest or tires out the opponents? Defensive rebounding? James Worthy does not stand out in this data (619th at -0.43 overall, +1.25 offense, -1.68 defense), but on film, Magic does not look better defensively than peak Worthy. So is Magic taking some of Worthy's true defensive value (at least in the 1985 sampling, their minutes overlapped to an extremely large extent, making it hard to determine value appropriately; unclear how much this is a concern in other years), or is Worthy's decline into the early 90s making his overall numbers look worse?

From 3-6, we have a series of centers: Robinson, Kareem, Ewing, Hakeem. Regular-season Robinson continues to look ~all-time in peak impact metrics. This isn’t exactly new — it’s consistent with the raw plus minus data we have from 1994–1996. Although it is interesting that Robinson’s single-season RAPM in 1993 and 1996 rank worse than the data from both years. Robinson looks like the best defender in this sample. Old Kareem continues to impress. Is this a sign of massive per-possession impact in a more specialized, smaller-load role, or would we project Kareem’s impact to be even higher in his prime? Ewing looks quite respectable, and ranks above Barkley despite often being considered slightly worse.

Hakeem continues to look disappointing in impact data (relative to his all-time reputation; still great overall). We have a little over ~1.5 seasons worth of data from him, so there’s still plenty of uncertainty. But his defense doesn’t stand out in the way we might expect if he’s the GOAT modern defender. (Regular Season) Robinson’s defense is rated 49% better than Hakeem’s, Ewing’s is 30% better, Hakeem is 1% better than Mark Eaton, and (per-possession) Manute Bol is 48% better. Hakeem’s clearly in the upper echelon of defenders and overall players — but not quite as high as we’d expect given his reputation.

Is the uneven spread of games hurting Hakeem? Squared2020 hasn’t updated all the single-year pages yet, but from the look of it, the most represented year is 1991 (commonly considered his ~2nd/3rd best defensive year ever, but one of his worst prime offensive years and a down year overall, making up ~27% of his known games). 1996 is next (~21% of known games; obviously this is late prime and thus lower impact), followed by 1993 (~16% of known games; in contention for his best regular season). We have 1990 data (commonly considered his best defensive year), but we don’t know how much. His rookie year, 1985, is the least represented (~9% of known games). Ultimately, there’s just some uncertainty here. If we were to project, I would say this the uneven sampling isn’t dragging down his DRAPM too much. We have data from his two best defensive regular seasons, his ~2nd best defensive regular season likely contains the largest sample of games we have for him, and his ~best overall regular season is the 3rd most represented year here. But better sampling might improve his offensive RAPM or overall RAPM slightly. We are missing 1994 (1st or 2nd best regular season) and 1991 has the largest sample (down offensive year). There’s uncertainty in offensive/defensive splits for RAPM, since the two relate to each other, so it’s also possible Hakeem’s ORAPM could be getting some of the credit that actually belongs to his DRAPM. In sum, Hakeem’s still a bit lower impact than we’d expect. But again, there’s large uncertainty bars. Up to you how reasonable it is for him to improve in the missing data, or in the playoffs.

Barkley is about as expected. He’s below the very best tier of players (or below the top two tiers if you give Jordan and magic their own tier), but definitely in the next group after. He looks great on offense (he’s actually 3rd overall), and poor on defense, again as expected.

Stockton is next, at 8th overall. While Karl Malone is… down at 343rd?!? That’s pretty shocking. We have ~1.25 seasons worth of data for Stockton and Malone. They were known to have an unusually high number of minutes shared, so perhaps that’s giving the regression some trouble? If we have a small sample of Stockton non-Malone minutes and Malone non-Stockton minutes, and if the former end up looking better, maybe that causes much of Malone’s value to be attributed to Stockton? It would be great if Squared2020 hadn’t been bullied into leaving this forum, since we might otherwise just ask him what the off-samples were (inb4 people butt in to argue he wasn’t treated poorly, when he said he was and had numerous posters agreeing with him, or butt in to argue that he received an apology that should have made up for everything, when he made it clear it was a pretty poor apology). Anyways, is this data enough to say younger Malone was significantly less valuable relative to Stockton? Is Stockton more of a driver of the early–mid 90s Jazz than people give him credit for? Probably not yet given the shared minutes concern... I’m particularly interested in getting a larger sample of data for these two, to help this issue.

Pippen and Rodman end up next to each other, at 14th and 15th. Rodman looks barely better than Pippen defensively (they’re effectively tied given the uncertainty). Rodman’s impact metrics look a bit disappointing (relative to the reputation) when he was with the Bulls, so it’s nice to see him looking more respectable. That’s probably the first and last time you’ll hear someone say Rodman and respectable in the same sentence.

Like Hakeem, Larry Bird also disappoints relative to his all-time expectations, at 23rd. Bird’s performance is probably the biggest drop from expectations, after Malone of course. His offense is still quite good — Top 5 in the sample, but clearly below Jordan and Magic. His offense is also clearly better than any DRAPM in the sample. As has been noted before, the best offensive value seems to outpace the best defensive value, even in the late 80s and 90s (at least according to Squared2020’s RAPM). This is a bit surprising. Is it possible some sort of rotation difference can account for this, or did the value of defense really peak in the dead ball era (~2002–04) after the 3 point line was added?

It’s Bird’s defense that drags him down, at -1.29. Going by single-year RAPM, Bird looks best on the Celtics in 1985 (somewhat negative defense), Bird looks 3rd best on the Celtics in 1988 (below McHale, then Parish; somewhat positive defense), 2nd best in 1990 (below McHale, somewhat negative defense), 2nd best in 1991 (below Parish, slight negative defense). Given this, it’s a bit surprising Bird ends up looking better than McHale, and that his defense looks so poor. If you squint, you might convince yourself you can see evidence of Bird’s aging (he’s in the top 10 in single-year RAPM in 80 and 85, but below top 10 in 88, 90, 91). I would love to see some data for 1984/1986, when Bird was really at his peak surrounded by a well-fitting supporting cast. I’m open to suggestions on how to interpret the Bird data (particularly given how the single-year data differs from the multi-year data re: the Celtics’ MVP and Bird’s defense). Ultimately, it seems Bird’s overall value is a fair bit lower than his reputation would suggest, at least in available regular season plus minus data, when he was older.

It’s useful to split things up into top set of players overall, offensively, and defensively:
Top overall players]:

Spoiler:

1. Michael Jordan +9.88
2. Magic Johnson +8.52
3. David Robinson +6.14
4. Kareem Abdul-Jabbar +5.89
5. Patrick Ewing +5.70
6. Hakeem Olajuwon +5.23
7. Charles Barkley +4.99
8. John Stockton +4.96
9. Shaquille O’Neal +4.21
10. Jerome Kersey +4.06
...
14. Dennis Rodman +4.04
15. Scottie Pippen +3.96
...
20. Clyde Drexler +3.58
...
23. Larry Bird +3.54
..
29. Robert Parish +3.14
…
34. Kevin McHale +2.95
...
342. Karl Malone +0.19

Top offensive players:

Spoiler:

1. Michael Jordan +8.51
2. Magic Johnson +7.52
3. Charles Barkley +6.41
4. Dominique Wilkins +4.92
5. Larry Bird +4..83
6. Otis Thorpe +4.34
7. Reggie Miller +4.13
8. John Stockton +4.04
9. Tom Chambers +4.01
10. Kareem Abdul Jabbar (?) +3.87

Top defensive players:

Spoiler:

1. David Robinson +3.31
2. Manute Bol +3.25
3. Scott Williams +3.09
4. Larry Nance +2.99
5. Patrick Ewing +2.88
6. John Salley +2.72
7. Vinny Del Negro +2.60
8. Anthony Mason +2.34
9. Sam Perkins +2.24
10. John Starks +2.24
11. Hakeem Olajuwon +2.22
12. Mark Eaton +2.19

Some statistical next steps / questions:
1. Could we estimate on-off for the usual candidates? We have their on-rating, and we can estimate their off based on the difference between their on-rating and the full-season team performance (which assumes the games are sufficiently sampled and randomly spread that using the full-season team performance is appropriate). The easiest thing to do would be to take a flat average of the team performance in the years included, but we might want to weight team performance by the sample of games from each year (e.g. each team might have a different sample of games in 1985 vs 1988 vs…).

2. How much does uneven sampling from year to year effect the result? If one player has a lot of games in 1996, is their RAPM most influenced by their value in that year, or is there some kind of correction for this applied?

3. Would Squared2020 ever do several-year RAPM, as a bridge between single-year and full-career RAPM? Based on the years he's tracked, he might do e.g. 4-year RAPM in 85-88, 88-91, 90-93, 93-96.

4. Would Squared2020 ever do a full RAPM analysis from 1985 until now, merging all his data and adding all the plus minus data from 1997-24/2025? That would really be the creme de la creme of long-sample RAPM. If so, how would the uneven number of games work? E.g. for Jordan, Robinson, or Hakeem, would their 1997+ seasons end up carrying more weight because we have full-season data compared to the earlier years? What are his thoughts on the age-adjustment that Engelmann does, and would that be a feasible correction to the uneven sample sizes?

[eminence]

Think he's got to set a higher possession bar for the minimum and bucket more guys, there are dudes on there with sub 100 possessions.

[lessthanjake]

This is not a substantive comment—which I think DraymondGold covered in-depth—but I really do like that Squared has put a confidence interval on this. I wish all RAPM measures did that.

[Special_Puppy]

How plausible is it that Bird was actually a slight negative defensively like the RAPM says? The other players with negative DRAPM are Charles Barkley and Alex English who most modern observers think are problematic defenders which does lend some credence to the RAPM. Bird was selected to 3 all defensive teams but it’s possible contemporary observers were just misevaluating his defense and missing something thing (or things) important

[Special_Puppy]

Rodman being top 10 is notable to me. He’s a weird player that box score based metrics struggle evaluating so it’s cool that he looks so good here

[EmpireFalls]

How plausible is it that Bird was actually a slight negative defensively like the RAPM says? The other players with negative DRAPM are Charles Barkley and Alex English who most modern observers think are problematic defenders which does lend some credence to the RAPM. Bird was selected to 3 all defensive teams but it’s possible contemporary observers were just misevaluating his defense and missing something thing (or things) important

Almost a certainty, given the years selected in this RAPM sample (90 and 91 especially). He was a liability outside his absolute physical peak, and declined heavily each year following 1987.

[OhayoKD]

Squared2020 hadn’t been bullied into leaving this forum, since we might otherwise just ask him what the off-samples were (inb4 people butt in to argue he wasn’t treated poorly, when he said he was and had numerous posters agreeing with him, or butt in to argue that he received an apology that should have made up for everything, when he made it clear it was a pretty poor apology).

The "bullying" in question:

Spoiler:

I respect Squared’s work in trying to put that together, but I wish he had held back on sharing any data that was far short of usual sample standards, because this scattering of games has so far pretty much only worsened discourse.

The "numerous posters agreeing with him", yourself included, along with squaredcircle, were the only ones bullying anyone in this exchange.

[lessthanjake]

Squared2020 hadn’t been bullied into leaving this forum, since we might otherwise just ask him what the off-samples were (inb4 people butt in to argue he wasn’t treated poorly, when he said he was and had numerous posters agreeing with him, or butt in to argue that he received an apology that should have made up for everything, when he made it clear it was a pretty poor apology).

The "bullying" in question:

Spoiler:

I respect Squared’s work in trying to put that together, but I wish he had held back on sharing any data that was far short of usual sample standards, because this scattering of games has so far pretty much only worsened discourse.

The "numerous posters agreeing with him", yourself included, along with squaredcircle, were the only ones bullying anyone in this exchange.

I don’t want to wade into that past discussion in any substantive way, but I find it really curious that you write a post specifically aimed at saying that no one bullied Squared, while simultaneously mocking his username by calling him “squaredcircle.” Just feels really odd that you can’t stop yourself even in a post that is specifically aimed at suggesting no one behaved badly towards Squared.