Offensive GOAT - Michael Jordan Playoff ON-OFF

[Owly]

Wow, is 1992 Jordan underrated compared to his earlier seasons?

Going through all these seasons, watching the games in entirety, and applying the knowledge I've learned from ~15 years in the league: Jordan's best year is BY FAR 1992. There may be statistical quirks that suggest otherwise, but just the gravity, the asserted dominance on the court, and the ability to trust his teammates even more to leverage that gravity is by far the best of his career (between 1985 and 1996). He also became a more measured defender than in previous seasons.

This is the primary reason that Horace Grant was more dominant in this sample compared to Scottie Pippen. There's several games where the commentators even acknowledge Grant was having a better year (1992) than Pippen. Pippen's not going to catch Horace if more data is added -- unless there are games where Horace is -20 and Pippen is +20.

Jordan's impact data in 1992 looks pretty ridiculous.

The Horace > Pippen thing I do find surprising. I've always been under the impression that 1992 was one of if not Pippen's best season (along with the 1994-1996 stretch).

Otoh thoughts
1) on-off is noisy
2) Grant was really good and has a pretty nice impact profile, especially once you throw in the 94-96 Pollack stuff (otoh).
3) Bach, their defensive co-ordinator/assistant (Winter being the offense guy) I believe had a quote (though I'm now struggling to find it so ... take it with a pinch of salt, I guess) saying something like Grant was the lynchpin (or anchor or most important) of the "Dobermans"
4) Looking at the '92 Reference regular season box-all-in-ones they have Pippen and Grant in the same vicinity (Scottie ahead in PER, BPM, Grant with a slightly larger lead in WS/48). Pippen opens up a bit of a lead in the playoffs (Grant's usage goes further down ... though he's still ahead in WS/48).
5) This is a team marginally above 10 SRS and the rest of the squad are probably mostly "not harmful in this context" more than they are actually "above average" or "good" (and maybe some are harmful) so there's a fair amount of credit to be spread around the top end guys.

[lessthanjake]

The flip side of the regular season impact stuff in 1992 looking above the other first-three-peat years is that the box stuff looks less good in that year than those other years. With impact data being very noisy in single-year samples, I’d generally look at that box and impact data info and think it was a bit of a wash in terms of what years look best. However, what I find really interesting here is that Squared feels strongly about 1992 after watching such a huge number of Jordan games. Very interesting stuff, and I think that moves my view more than the actual impact data does.

I will say that, having not recently watched nearly as many Jordan games as Squared has, I’ve been really impressed by what I saw from some Jordan 1993 playoffs games I saw. I’d always kind of thought of 1993 as not being as good as the prior years, but man he looked ridiculously good in the games I recently watched.

[Squared2020]

Do you mean the regular season or also including the playoffs?

And what would you say is Jordan's best playoffs?

That's a really good question to clarify my comment. I mean the regular season only.

And I wish I could give you a grounded answer for the playoffs question. I've been so focused on regular season data, that I could not give you a fair assessment of Jordan's playoff.

[TheGOATRises007]

I always believed the 92 Bulls were the best Bulls team from the dynasty.

Interesting data and observations as usual from you Squared. We always appreciate your input on here.

[Squared2020]

Otoh thoughts
1) on-off is noisy
2) Grant was really good and has a pretty nice impact profile, especially once you throw in the 94-96 Pollack stuff (otoh).
3) Bach, their defensive co-ordinator/assistant (Winter being the offense guy) I believe had a quote (though I'm now struggling to find it so ... take it with a pinch of salt, I guess) saying something like Grant was the lynchpin (or anchor or most important) of the "Dobermans"
4) Looking at the '92 Reference regular season box-all-in-ones they have Pippen and Grant in the same vicinity (Scottie ahead in PER, BPM, Grant with a slightly larger lead in WS/48). Pippen opens up a bit of a lead in the playoffs (Grant's usage goes further down ... though he's still ahead in WS/48).
5) This is a team marginally above 10 SRS and the rest of the squad are probably mostly "not harmful in this context" more than they are actually "above average" or "good" (and maybe some are harmful) so there's a fair amount of credit to be spread around the top end guys.

I like your assessment here with the fifth comment. The 1992 team had consistency issues in the post with Will Perdue, Scott Williams, and Stacey King. They were more underwhelming than moderately impactful -- even leading to times when Cliff Levingston would have to become the PF due to the bigs lacking a bit.

Grant's midrange improved in 1992, something I don't recall seeing in 1991 nor 1993. Maybe this is an anomaly in his stats, maybe this is an anomaly in a 70 game sample -- he could have "meh" midrange games in the other 12 games. However, it was his rebounding -- both obtaining and creating -- that markedly improved. That doesn't get captured well in PER or BPM. It would be interesting if someone has the time to do a quick REB% calculation for Grant ON/OFF to verify what I'm seeing -- or identify it as 'Faces in Clouds.'

Grant's 1985-1996 RAPM really takes off with 1993 and 1996 data. So it's good to see some of the Pollack stuff align there.

I'm curious how Grant will get impacted overall as I add in 1989 and 1994 data. I can see Scottie's results holding fairly firm.

[homecourtloss]

Otoh thoughts
1) on-off is noisy
2) Grant was really good and has a pretty nice impact profile, especially once you throw in the 94-96 Pollack stuff (otoh).
3) Bach, their defensive co-ordinator/assistant (Winter being the offense guy) I believe had a quote (though I'm now struggling to find it so ... take it with a pinch of salt, I guess) saying something like Grant was the lynchpin (or anchor or most important) of the "Dobermans"
4) Looking at the '92 Reference regular season box-all-in-ones they have Pippen and Grant in the same vicinity (Scottie ahead in PER, BPM, Grant with a slightly larger lead in WS/48). Pippen opens up a bit of a lead in the playoffs (Grant's usage goes further down ... though he's still ahead in WS/48).
5) This is a team marginally above 10 SRS and the rest of the squad are probably mostly "not harmful in this context" more than they are actually "above average" or "good" (and maybe some are harmful) so there's a fair amount of credit to be spread around the top end guys.

I like your assessment here with the fifth comment. The 1992 team had consistency issues in the post with Will Perdue, Scott Williams, and Stacey King. They were more underwhelming than moderately impactful -- even leading to times when Cliff Levingston would have to become the PF due to the bigs lacking a bit.

Grant's midrange improved in 1992, something I don't recall seeing in 1991 nor 1993. Maybe this is an anomaly in his stats, maybe this is an anomaly in a 70 game sample -- he could have "meh" midrange games in the other 12 games. However, it was his rebounding -- both obtaining and creating -- that markedly improved. That doesn't get captured well in PER or BPM. It would be interesting if someone has the time to do a quick REB% calculation for Grant ON/OFF to verify what I'm seeing -- or identify it as 'Faces in Clouds.'

Grant's 1985-1996 RAPM really takes off with 1993 and 1996 data. So it's good to see some of the Pollack stuff align there.

I'm curious how Grant will get impacted overall as I add in 1989 and 1994 data. I can see Scottie's results holding fairly firm.

Grant has to be one of the most underrated players ever. His archetype, i.e., high motor, transition offense and defense, unselfish, no bad shots, defends well on the peri tee and i see is almost always a high plus player.

[Djoker]

Here is all the latest RAPM from the 1985-1996 period that we have thus far thanks to Squared2020. Figured I'd put them all in one post as a summary. In the brackets is the % of all teams' games that are sampled for that season.

1985 (16.4%)

Spoiler:

1986 (8.1%)

Spoiler:

1987 (16.0%)

Spoiler:

1988 (20.0%)

Spoiler:

1989 (Not Available Yet)

Spoiler:

1990 (16.2%)

Spoiler:

1991 (26.6%)

Spoiler:

1992 (19.7%)

Spoiler:

1993 (25.5%)

Spoiler:

1994 (Not Available Yet)

Spoiler:

1995 (Not Available Yet)

Spoiler:

1996 (20.0%)

Spoiler:

1985-1996 (13.7%)

Spoiler:

Michael Jordan RAPM Finishes (Bulls Games Sampled)

1985: 6th (28 games)
1986: 31st (10 games)
1987: 4th (40 games)
1988: 1st (43 games)
1989: N/A
1990: 2nd (56 games)
1991: 1st (57 games)
1992: 1st (71 games)
1993: 1st (79 games)
1995: N/A
1996: 1st (73 games) -- last updated before all 82 Bulls games logged

1985-1996: 1st (402 games) -- last updated before adding 71 games from 1992; includes 5 non-Jordan Bulls games from 1994 and 1995

[f4p]

Here is all the latest RAPM from the 1985-1996 period that we have thus far thanks to Squared2020. Figured I'd put them all in one post as a summary. In the brackets is the % of all teams' games that are sampled for that season.

1985 (16.4%)

Spoiler:

1986 (8.1%)

Spoiler:

1987 (16.0%)

Spoiler:

1988 (20.0%)

Spoiler:

1989 (Not Available Yet)

Spoiler:

1990 (16.2%)

Spoiler:

1991 (26.6%)

Spoiler:

1992 (19.7%)

Spoiler:

1993 (25.5%)

Spoiler:

1994 (Not Available Yet)

Spoiler:

1995 (Not Available Yet)

Spoiler:

1996 (20.0%)

Spoiler:

1985-1996 (13.7%)

Spoiler:

Michael Jordan RAPM Finishes (Bulls Games Sampled)

1985: 6th (28 games)
1986: 31st (10 games)
1987: 4th (40 games)
1988: 1st (43 games)
1989: N/A
1990: 2nd (56 games)
1991: 1st (57 games)
1992: 1st (71 games)
1993: 1st (79 games)
1995: N/A
1996: 1st (73 games) -- last updated before all 82 Bulls games logged

1985-1996: 1st (402 games) -- last updated before adding 71 games from 1992; includes 5 non-Jordan Bulls games from 1994 and 1995

hmm, what am i missing here with Hakeem? in the overall 1985-1996 sample, he finished 4th. but on a year by year basis, he doesn't finish in the top 50 in 1987 (hmm), 1990 (double hmm), or 1993 (!!). and he has 2 other years with a 47th and 48th place finish and then a 35th and 27th. nothing higher than 16th (weirdly in 1996, although i think the rockets were 2-10 without him). seems hard to figure out how he ends up 4th overall while apparently not really being a top 50 player in the nba for half his career and maybe being about 25th in the other half.

[Djoker]

Here is all the latest RAPM from the 1985-1996 period that we have thus far thanks to Squared2020. Figured I'd put them all in one post as a summary. In the brackets is the % of all teams' games that are sampled for that season.

1985 (16.4%)

Spoiler:

1986 (8.1%)

Spoiler:

1987 (16.0%)

Spoiler:

1988 (20.0%)

Spoiler:

1989 (Not Available Yet)

Spoiler:

1990 (16.2%)

Spoiler:

1991 (26.6%)

Spoiler:

1992 (19.7%)

Spoiler:

1993 (25.5%)

Spoiler:

1994 (Not Available Yet)

Spoiler:

1995 (Not Available Yet)

Spoiler:

1996 (20.0%)

Spoiler:

1985-1996 (13.7%)

Spoiler:

Michael Jordan RAPM Finishes (Bulls Games Sampled)

1985: 6th (28 games)
1986: 31st (10 games)
1987: 4th (40 games)
1988: 1st (43 games)
1989: N/A
1990: 2nd (56 games)
1991: 1st (57 games)
1992: 1st (71 games)
1993: 1st (79 games)
1995: N/A
1996: 1st (73 games) -- last updated before all 82 Bulls games logged

1985-1996: 1st (402 games) -- last updated before adding 71 games from 1992; includes 5 non-Jordan Bulls games from 1994 and 1995

hmm, what am i missing here with Hakeem? in the overall 1985-1996 sample, he finished 4th. but on a year by year basis, he doesn't finish in the top 50 in 1987 (hmm), 1990 (double hmm), or 1993 (!!). and he has 2 other years with a 47th and 48th place finish and then a 35th and 27th. nothing higher than 16th (weirdly in 1996, although i think the rockets were 2-10 without him). seems hard to figure out how he ends up 4th overall while apparently not really being a top 50 player in the nba for half his career and maybe being about 25th in the other half.

Good point. That is weird. I'm not sure which Hakeem games are logged. It's possible that there are some from 1994 and 1995 that are in the sample even though Squared2020 didn't tackle those seasons full on yet.

[lessthanjake]

Here is all the latest RAPM from the 1985-1996 period that we have thus far thanks to Squared2020. Figured I'd put them all in one post as a summary. In the brackets is the % of all teams' games that are sampled for that season.

1985 (16.4%)

Spoiler:

1986 (8.1%)

Spoiler:

1987 (16.0%)

Spoiler:

1988 (20.0%)

Spoiler:

1989 (Not Available Yet)

Spoiler:

1990 (16.2%)

Spoiler:

1991 (26.6%)

Spoiler:

1992 (19.7%)

Spoiler:

1993 (25.5%)

Spoiler:

1994 (Not Available Yet)

Spoiler:

1995 (Not Available Yet)

Spoiler:

1996 (20.0%)

Spoiler:

1985-1996 (13.7%)

Spoiler:

Michael Jordan RAPM Finishes (Bulls Games Sampled)

1985: 6th (28 games)
1986: 31st (10 games)
1987: 4th (40 games)
1988: 1st (43 games)
1989: N/A
1990: 2nd (56 games)
1991: 1st (57 games)
1992: 1st (71 games)
1993: 1st (79 games)
1995: N/A
1996: 1st (73 games) -- last updated before all 82 Bulls games logged

1985-1996: 1st (402 games) -- last updated before adding 71 games from 1992; includes 5 non-Jordan Bulls games from 1994 and 1995

hmm, what am i missing here with Hakeem? in the overall 1985-1996 sample, he finished 4th. but on a year by year basis, he doesn't finish in the top 50 in 1987 (hmm), 1990 (double hmm), or 1993 (!!). and he has 2 other years with a 47th and 48th place finish and then a 35th and 27th. nothing higher than 16th (weirdly in 1996, although i think the rockets were 2-10 without him). seems hard to figure out how he ends up 4th overall while apparently not really being a top 50 player in the nba for half his career and maybe being about 25th in the other half.

Good point. That is weird. I'm not sure which Hakeem games are logged. It's possible that there are some from 1994 and 1995 that are in the sample even though Squared2020 didn't tackle those seasons full on yet.

Couldn’t it also just be that the larger-timescale has more data about what Hakeem’s teammates did elsewhere and that results in the model giving more credit to Hakeem for what happened with those players than he gets just from single-year RAPM where the models don’t know anything about his teammates’ other exploits (or lack thereof)?

[Owly]

On Hakeem ...

Looking at the average of the yearly on-offs 94-96 from memory he was joint second or maybe very narrowly third (with or marginally behind Blaylock) though Robinson was nearly lapping the field (Hakeem circa +12, Robinson circa +19). Fwiw, I think some people have done some tracking on here around his apex including playoffs that was giving a very positive impression of him.


I haven't seen the data talked about here. My impression is in theory as adjustments on teammates have a knock-on effect it could be that. It could be a question of where the big samples are, perhaps. That said if no individual years are looking noteworthy and the lineup adjustment is doing substantial lifting ... you're putting a lot of faith in the process at a technical level (and we're already talking constructing RAPM off incomplete datasets) and in the outcomes being "right" in estimating these players to be a big drag.

[Djoker]

I updated the post above with the latest 1993 RAPM. The image that was there before was very outdated when only a few games were sampled. The current one has 25.5% sampled.

These are Hakeem's placements thus far along with the number of Rockets' games sampled the years that I can figure it out:

1985: 48th (9 games)
1986: 35th
1987: not top 50
1988: 47th (14 games)
1989: N/A
1990: 144th (10 games)
1991: 17th (25 games)
1992: 27th
1993: 7th (15 games)
1996: 16th (19 games)

1985-1996: 4th (126 games)

For 1986, 1987 and 1992, I can't figure out the exact number of games sampled but I doubt it's much bigger than about 15 games each.

[Squared2020]

I updated the post above with the latest 1993 RAPM. The image that was there before was very outdated when only a few games were sampled. The current one has 25.5% sampled.

These are Hakeem's placements thus far along with the number of Rockets' games sampled the years that I can figure it out:

1985: 48th (9 games)
1986: 35th
1987: not top 50
1988: 47th (14 games)
1989: N/A
1990: 144th (10 games)
1991: 17th (25 games)
1992: 27th
1993: 7th (15 games)
1996: 16th (19 games)

1985-1996: 4th (126 games)

For 1986, 1987 and 1992, I can't figure out the exact number of games sampled but I doubt it's much bigger than about 15 games each.

There’s an easy explanation for this: RAPM is inherently noisy, and the sampling frame is sparse. When you estimate it season-by-season, the data for each player is limited. So their ratings can swing a lot from year to year, as the confidence bands show. But when you combine seasons, that sampling sparsity drops and redundancy increases, which helps average out the noise and reduce variance. That’s why a player who bounces around in single-year models (like Hakeem) can still emerge near the top in the aggregate analysis.

It’s also worth noting that the yearly rankings aren’t directly additive or linear. In other words, you can’t combine a player’s 1986 and 1991 ranks to predict their aggregate rank. This is because each season’s RAPM is estimated independently, with its own scaling and noise structure. The aggregate model re-estimates everyone jointly across all data, which changes the relative positioning and stabilizes the results.

[lessthanjake]

I do want to note that this gets back to something I’ve noted before about Jordan—which is that him looking consistently great in small-sample RAPM (Squared’s partial samples in various years, single-season RAPM in 1997 and 1998, as well as single-playoff RAPM in 1997 and 1998) is really abnormal. Really small samples are really noisy, so we’d usually expect that even a guy who would comfortably be #1 in a large sample would really bounce around a lot when the samples get really small. For someone to always do so well in small samples suggests that he’s so impactful that even the huge amount of noise in small samples can’t actually bring him down.

Just to put a point on this, even with someone as dominant in terms of RAPM as LeBron, we see prime years where he’s ranked really far from the top of the league in single-year RAPM (i.e. things like 45th in the league). In the data we have so far, Jordan really does look very consistently great even in small samples, which is pretty abnormal and I think is highly unlikely to happen unless a guy is a real impact outlier. Of course, we also have the Squared aggregate RAPM data telling us that, yeah, Jordan and Magic Johnson were miles above everyone else, so this does check out.

[Djoker]

I updated the post above with the latest 1993 RAPM. The image that was there before was very outdated when only a few games were sampled. The current one has 25.5% sampled.

These are Hakeem's placements thus far along with the number of Rockets' games sampled the years that I can figure it out:

1985: 48th (9 games)
1986: 35th
1987: not top 50
1988: 47th (14 games)
1989: N/A
1990: 144th (10 games)
1991: 17th (25 games)
1992: 27th
1993: 7th (15 games)
1996: 16th (19 games)

1985-1996: 4th (126 games)

For 1986, 1987 and 1992, I can't figure out the exact number of games sampled but I doubt it's much bigger than about 15 games each.

There’s an easy explanation for this: RAPM is inherently noisy, and the sampling frame is sparse. When you estimate it season-by-season, the data for each player is limited. So their ratings can swing a lot from year to year, as the confidence bands show. But when you combine seasons, that sampling sparsity drops and redundancy increases, which helps average out the noise and reduce variance. That’s why a player who bounces around in single-year models (like Hakeem) can still emerge near the top in the aggregate analysis.

It’s also worth noting that the yearly rankings aren’t directly additive or linear. In other words, you can’t combine a player’s 1986 and 1991 ranks to predict their aggregate rank. This is because each season’s RAPM is estimated independently, with its own scaling and noise structure. The aggregate model re-estimates everyone jointly across all data, which changes the relative positioning and stabilizes the results.

Thank you for explaining. That's basically how I see it although I could never verbalize it as eloquently.

[Top10alltime]

When I did the Greatest Offensive Legends ON Court here a few months ago, I didn't include MJ because we didn't have the complete data for a few postseason runs but I've since filled all the gaps.

I have a question, do you have any other guys offensive on-court in that time period?

Guys like Magic, Bird, Kareem, KM, Chuck, Barry, Stockton, Drexler, KJ, Reggie, Moses, McHale, Worthy, Dantley, English, IT, King, etc. I'm just asking, do you have them? If not, ok, I just asked.....

[Djoker]

When I did the Greatest Offensive Legends ON Court here a few months ago, I didn't include MJ because we didn't have the complete data for a few postseason runs but I've since filled all the gaps.

I have a question, do you have any other guys offensive on-court in that time period?

Guys like Magic, Bird, Kareem, KM, Chuck, Barry, Stockton, Drexler, KJ, Reggie, Moses, McHale, Worthy, Dantley, English, IT, King, etc. I'm just asking, do you have them? If not, ok, I just asked.....

There is ON Court rORtg for some series of Magic and Bird in this thread:

https://forums.realgm.com/boards/viewtopic.php?f=64&t=2402176

I didn't post it for Kareem but I could as I'm doing him and Magic together.

[Top10alltime]

When I did the Greatest Offensive Legends ON Court here a few months ago, I didn't include MJ because we didn't have the complete data for a few postseason runs but I've since filled all the gaps.

I have a question, do you have any other guys offensive on-court in that time period?

Guys like Magic, Bird, Kareem, KM, Chuck, Barry, Stockton, Drexler, KJ, Reggie, Moses, McHale, Worthy, Dantley, English, IT, King, etc. I'm just asking, do you have them? If not, ok, I just asked.....

There is ON Court rORtg for some series of Magic and Bird in this thread:

https://forums.realgm.com/boards/viewtopic.php?f=64&t=2402176

I didn't post it for Kareem but I could as I'm doing him and Magic together.

Damn it, not the full guys. Guess I'll go ahead and do some of these guys!