What can we make of Squared2020's partial 84-85, 87-88, 90-91 RAPM?

[ceiling raiser]

Sources for those not familiar:

84-85 RAPM
87-88 RAPM
90-91 RAPM

There are also partial RAPMs for 69-70 and 79-80, albeit with a much lower number of sampled games.

EDIT: Removed some specific observations about splits given sample issues.

Thoughts: Impressed by Jordan, Kareem, and Magic; Bird seems to underwhelm. Your thoughts? These are all between 12%-20% of the yearly samples of games.

[Dr Positivity]

Interesting idea but probably needs more games. More evidence Laimbeer might have been Bad Boy Pistons best.

[capfan33]

Would be awesome if we could get this data from the start of Magic/Birds career as I think the NBA has most of these games locked away. And yea, we need a much bigger sample to really draw any conclusions from the data.

[eminence]

In an evenly distributed sample you'll start to see real stabilization around the 20-30 game mark in a season in my experience, so some of the larger team samples in the 3 main years might have some real value, but I'm not certain what somewhat randomly removing teams from the sample does to it. Off the top of my head I can't think of any clear directional effects (unless players had clearly bad matchups or something of that sort that were either over or under sampled).

So looking at teams that'd be in that area of sample size:
'85 Celtics - 46 game sample, unfortunately at first glance it doesn't look very representative, as they were 33-3 in non sampled games, so I'd certainly have reservations here.
'85 Sixers - 31 game sample, this one looks okay
'85 Bulls - 32 game sample, also looks okay
'85 Pistons - 21 game sample, imbalanced in the opposite direction, 7-14 in sample, 39-22 out of it, also fairly small
'85 Lakers - 41 game sample, looks alright

'88 Celtics - 43 game sample, looks more balanced this season
'88 Bulls - 43 game sample, looks decent
'88 Lakers - 54 game sample, looks good

'91 Bulls - 38 game sample, maybe a bit imbalanced, but I think useable
'91 Lakers - 21 game sample, a bit small

So I'd mainly be looking at high possession players from the '85 Sixers (we have years of Sixers +/- data to compare this to), the '85 Bulls, '85 Lakers, '88 Celtics, '88 Bulls, '88 Lakers, and '91 Bulls for takeaways.

I would try to avoid much of anything on players not from those teams, as it winds up being very much just a small sample measure of their matchups specifically against the Celtics/Bulls/Lakers. I wouldn't look too much into splitting offense/defense at this size of sample.

Bulls ('85/'88/'91) - MJ - 5.03, 7.47, 6.82 - hey, this guy might be alright. I kid I kid, though on first glance his rookie year is a bit higher than I would've anticipated, so that's a tiny positive nudge for him. Grant/Pippen both looking solid.

Lakers ('85/'88/'91) - Kareems '88 result is significantly higher than I would've guessed. Interesting, something to chew on for sure. Magic looks close to on par with MJ over the period with results of 8.92, 6.62, 4.63.

Celtics ('88) - McHale and to a lesser extent Parish the ones coming out looking impressive instead of Bird.

[jalengreen]

According to squared2020 here https://forums.realgm.com/boards/viewtopic.php?f=64&t=2123285&p=93982301#p93982301,

The '91 RAPM total estimates have an error of 5.2 to 7.9 - "about triple compared to a full season."

And even for a full season as we know, error is quite high so too much stock can't be put into them. For these small samples, that's even more true

Goes without saying that it's very awesome work from squared2020 though of course. I just don't think we can draw many conclusions from the data yet

[Owly]

First up really appreciate anyone doing this kind of work. Seems like an awful lot of work if this is indeed a one person project. So huge respect first up.

In terms of value ... at this point ... I would guess probably very little. Take '91. Average pace is 96.6. So ten full games worth of data (if the player stayed on court 75% of the time) would be 724.5 possessions at each end. And bar Bulls, Pistons, Celtics, I guess Spurs I'm not seeing many much over that. And unless I'm being forced too because the playoffs is soo important that we need to analyze it (and even then tbh) I'm not going to put a whole lot of stock in such small samples.

I also don't know about the work aspect of it but my impression is the value of RAPM is in having a good gauge of how good the other guys on the court are (and adjusting accordingly) so the value probably increases (especially versus just on-off) something like exponentially? Whilst all the reads are really fuzzy I might be inclined just to give the raw on-off data. But I don't really know what I'm talking about here (and it looks like someone who does has now already responded).

I am very curious about those most televised teams. If you've got, say, over 75% of their games for a multiple year stretch ... again, not a mathematician, nor an expert, nor anywhere close ... that feels like you could start getting something of real value (less noise, maybe something where you can start to compare with the databall era?). But that feels like a project that would take a lot of work and expertise and skilled game trackers and coordination etc.

[Doctor MJ]

Yeah, I'm reluctant to use this much at this point, but look forward to it getting to a point where I feel I can.

I'll also say that I'm aware of other projects people are doing to get better +/- type data views of earlier eras, so yeah, I expect that eventually we'll be able to use data from the past we don't currently have to help us better understand the impact from that era, just as it was the case when we started getting '90s data.

[homecourtloss]

According to squared2020 here https://forums.realgm.com/boards/viewtopic.php?f=64&t=2123285&p=93982301#p93982301,

The '91 RAPM total estimates have an error of 5.2 to 7.9 - "about triple compared to a full season."

And even for a full season as we know, error is quite high so too much stock can't be put into them. For these small samples, that's even more true

Goes without saying that it's very awesome work from squared2020 though of course. I just don't think we can draw many conclusions from the data yet

I’ve said it before, but I hope somewhere there’s a big enough basketball fan out there who’s willing to fund a project such as this. I’d love to see the data results.

You’re right about the use of this data though @eminence above does establish some good parameters for usability.

Jordan looks good though peak season RAPM a bit underwhelming. Laimbeer looks great. Magic and Kareem look like the monster combo they were though 1980 Kareem looks off. Old Kareem Bird a bit underwhelming though his defensive fall out seems corroborated by this data. Hakeem a bit underwhelming.

[ceiling raiser]

I’m pretty sure the NBA has to have physical PBP documents somewhere. When I spoke to Pollack (RIP) a few years ago he mentioned that it’s possible the league still has them around, but that he accessed them following each season and wasn’t sure if they were thrown out by the league the following year.

If the league does have physical PBP, maybe we can crowdfund and purchase somehow.

[Squared2020]

.

[colts18]

Once again, Look who is near the top of the RAPM leaderboards. John Stockton. He has a +20 Plus/Minus in both the 1988 and 1991 samples. The wild part is that the sample is a disproportionately bad one for the Jazz. They have a 15-22 record in the sample (86-41 record outside of the sample) and Stockton's On Court numbers is still a positive at +3.

Every type of Plus/Minus data we have concurs with the impact that Stockton's Box Score implied.

-5th place in the 1997-2001 RAPM
-Top 15 in 2002 and 2003 RAPM at age 40-41
-#1 RAPM in 2001 at age 38-39. #1 PIPM too
-Top 15 in each of the 1994-1996 seasons
-Amazing impact in the 1988 sample
-Amazing impact in the 1991 sample
-#1 in Fpliii's sample of 76ers opponents from 1980-1994

[trex_8063]

It's interesting stuff (I've bookmarked this thread, so I can follow his progress), and I'm really glad someone is taking the time to do this.

The results so far?
Well, some are fairly intuitive and certainly pass the "smell test" [e.g. Jordan being at the top of the league].
In other instances there is obviously grossly insufficient sample and a lot of line-up noise [e.g. in the '88 dataset, Mark Eaton has a REALLY solid positive ORAPM and a marginally negative DRAPM ].

Line-up noise feels evident elsewhere, like how strong rookie Horace Grant looks. I mean, he was a solid contributor off the bat [quicker than Pippen was, actually]; but I suspect his numbers are sort of "riding Jordan's coattails" at that point.
Probably ditto old-man Kareem [with Magic].

I'm a little skeptical Bird's defensive imprint in '88 could be THAT bad, or that his overall impact is so pedestrian.......'88 has a good case as his BEST offensive season, imo; just can't see it sitting down around 70th league-wide.

But yeah; really cool stuff. It's definitely something I'll check in on from time-to-time (I assume he'll be updating those sheets as he logs more games).

[wojoaderge]

I'd like to see how many 1984-85 Utah Jazz games he watched to put The Whopper at #20. I'm thinking not a whole lot and one of them was definitely Game 5 against Houston.

[Owly]

I'd like to see how many 1984-85 Utah Jazz games he watched to put The Whopper at #20. I'm thinking not a whole lot and one of them was definitely Game 5 against Houston.

At the bottom, 6 Jazz games watched (3 wins, 3 losses), unclear how many featured Paultz.
Based on what is stated on the site regarding possible games (943) (and comparison with full RS records) it would seem not to feature any playoff games.

[wojoaderge]

I'd like to see how many 1984-85 Utah Jazz games he watched to put The Whopper at #20. I'm thinking not a whole lot and one of them was definitely Game 5 against Houston.

At the bottom, 6 Jazz games watched (3 wins, 3 losses), unclear how many featured Paultz.
Based on what is stated on the site regarding possible games (943) (and comparison with full RS records) it would seem not to feature any playoff games.

Ah, I see, he works for the NBA. He must have access the entire library. From memory, I remember/know what 5 out of 6 games are:

12/9/84 vs. Kings
12/16/84 @Bucks
12/22/84 vs. Mavs
1/22/85 vs. Nets
3/9/85 @Bulls