96-97 (NPI) RAPM is out!

[Durins Baynes]

I see many posters who do nothing (it seems) but quote APM. That's literally their only argument. There's no context (except to look at the APM of other players on the team). It's quite frustrating. I don't mind it being used in a general way to say "well, it seems to suggest he was pretty good that year", but the idea of using it to claim one guy is better than another guy (especially another guy with a similar APM anyway) is silly. Yet it is constantly done in posts I see. I'm consistent in that I don't care much for either, but I do find it convenient that KG fans seem to have embraced APM so long as it supported their claims, then dismissed NPI APM when it came out as irrelevant (since the latter suggests Shaq and Duncan were even better than APM suggested- and of course, better than KG even with his go to stat).

[E-Balla]

I guess people evaluated talent better in the 90's because those rankings seem super clean.

Odd to me that you say this while all the other negative comments here are based on singling out high marks from minor players just like always happens with +/- data.

Well I usually think separating the results for good players and role players is a good idea so I'm mainly looking at the stars vs other stars and bench players vs other bench players.

[acrossthecourt]

I see many posters who do nothing (it seems) but quote APM. That's literally their only argument. There's no context (except to look at the APM of other players on the team). It's quite frustrating. I don't mind it being used in a general way to say "well, it seems to suggest he was pretty good that year", but the idea of using it to claim one guy is better than another guy (especially another guy with a similar APM anyway) is silly. Yet it is constantly done in posts I see. I'm consistent in that I don't care much for either, but I do find it convenient that KG fans seem to have embraced APM so long as it supported their claims, then dismissed NPI APM when it came out as irrelevant (since the latter suggests Shaq and Duncan were even better than APM suggested- and of course, better than KG even with his go to stat).

NPI RAPM is an improvement over standard APM but still very buggy and noisy. It doesn't have a good degree of separation from other metrics, in terms of predictive power. Prior-informed RAPM and statistical +/- hybrids are a lot better though.

[mysticbb]

NPI RAPM is an improvement over standard APM but still very buggy and noisy. It doesn't have a good degree of separation from other metrics, in terms of predictive power. Prior-informed RAPM and statistical +/- hybrids are a lot better though.

I tried to explain the differences to him specifically, and he still writes the same nonsense.

Given the fact that you are the one creating those NPI RAPM values, would you mind running it again without increasing the value of the playoffs possessions as well as using a gaussian distribution?

Sending the matchupfile would probably too much to ask for?

[NO-KG-AI]

It is pretty easy to pretend that the only thing KG had going for him statistically is adjusted plus/minus... lol. Really, the only thing that's ever gone against KG is the "winner" narrative, and once that got tossed, the goal posts shifted. You don't need to invent stats to show that KG's peak was outstanding, there are plenty of individual and team based numbers that show it.

[Durins Baynes]

I rate KG 12th all time, so I don't think I can be accused of underrating him. But he was not comparable to guys like Shaq and Duncan, I don't care what variety of APM says so.

[NO-KG-AI]

That's a wide spectrum though. He wasn't as good as the best of Shaq, but neither was Duncan.

[Durins Baynes]

Peak to peak no. Career wise Duncan was probably better, though I can see arguments both ways. Even peak wise, the gap between Shaq and Duncan is smaller than the gap between Duncan and KG. Shaq's peak in a lot of ways is theoretical outside 2000 because most of his years he just didn't play up to it even remotely consistently.

[Ryoga Hibiki]

I guess people evaluated talent better in the 90's because those rankings seem super clean.

Odd to me that you say this while all the other negative comments here are based on singling out high marks from minor players just like always happens with +/- data.

Well I usually think separating the results for good players and role players is a good idea so I'm mainly looking at the stars vs other stars and bench players vs other bench players.

when you compare people who played similar minutes is not that bad.
Anyway, you can find a role player with great impact, but it can easily be that he's used only in the situations where he would be effective the most. Doesn't mean he was a batter player overall.

[NO-KG-AI]

Odd to me that you say this while all the other negative comments here are based on singling out high marks from minor players just like always happens with +/- data.

Well I usually think separating the results for good players and role players is a good idea so I'm mainly looking at the stars vs other stars and bench players vs other bench players.

when you compare people who played similar minutes is not that bad.
Anyway, you can find a role player with great impact, but it can easily be that he's used only in the situations where he would be effective the most. Doesn't mean he was a batter player overall.

What? Logic? Putting things into context? this can not be RealGM.

[lilojmayo]

It is great to see someone having taken the time to clear up the mess the NBA.com pbp provide for the most part. Seeing such numbers gives some hope to get the matchupfiles for 1997 to 2000 at one point in the future. Anyway, I took the liberty to merge those numbers with my SPM by calculating the appropiate coefficients via a regression on the team's game by game performances.
Here is the result for the Top50 players of that season:

Code: Select all

Name                  Tm    Rat
Michael Jordan        CHI   9.2
Karl Malone           UTA   8.6
Shaquille O’Neal      LAL   6.6
Grant Hill            DET   6.2
Tim Hardaway          MIA   6.2
Scottie Pippen        CHI   5.7
Patrick Ewing         NYK   5.5
Mookie Blaylock       ATL   5.4
Christian Laettner    ATL   5.3
John Stockton         UTA   5.2
Terry Mills           DET   5.2
Alonzo Mourning       MIA   5.1
Jeff Hornacek         UTA   5.0
Gary Payton           SEA   5.0
Hakeem Olajuwon       HOU   4.8
Charles Barkley       HOU   4.3
Arvydas Sabonis       POR   4.2
Toni Kukoc            CHI   4.2
Chris Webber          WSB   4.1
Clyde Drexler         HOU   4.0
Detlef Schrempf       SEA   3.9
Hersey Hawkins        SEA   3.6
Anfernee Hardaway     ORL   3.5
Vlade Divac           CHH   3.3
John Starks           NYK   3.0
Rasheed Wallace       POR   3.0
Kenny Anderson        POR   2.9
Ron Harper            CHI   2.8
Dikembe Mutombo       ATL   2.8
Tyrone Hill           CLE   2.8
Shawn Kemp            SEA   2.8
Greg Ostertag         UTA   2.7
Kevin Garnett         MIN   2.7
Reggie Miller         IND   2.6
Terrell Brandon       CLE   2.5
Bo Outlaw             LAC   2.5
Joe Dumars            DET   2.4
Byron Scott           LAL   2.3
Kendall Gill          NJN   2.3
Horace Grant          ORL   2.3
Mitch Richmond        SAC   2.3
Nate McMillan         SEA   2.3
Kevin Johnson         PHO   2.2
Gheorghe Muresan      WSB   2.2
Glen Rice             CHH   2.1
Darrell Armstrong     ORL   2.0
Chris Mills           CLE   1.9
Chris Gatling         TOT   1.8
Vin Baker             MIL   1.8
Danny Manning         PHO   1.7


To put those numbers into perspective: From 2001 to 2013 the highest value was achieved by LeBron James in 2010 with 9.2 as well. He has 9.1 in 2009 and 2013. Kevin Garnett finished with 8.5 in 2004. In turn, Jordan 1997 scored in that metric as well as peak-level James, Karl Malone anno 1997 scored as well as peak-level Kevin Garnett. BUT, the amount of players with +5 or +2 is unusually high, usually a season has about 5 to 6 players with +5 or more, and about 35 players with +2 or more.
On the other end of the scale I get Roy Rogers with -5.1 and Brian Shaw with -6.0 being worse than any other player with enough minutes played in the dataset from 2001 to 2013. Before, Trenton Hassell in 2010 finished with -5.0.
So, not quite sure what to make of it ...

97 Jordan is. Thebestoverallversion of Jordan when you factor in the complete package. So it makes sense it rivals lebron

[lorak]

1998 RAPM was published: http://ascreamingcomesacrossthecourt.bl ... ormed.html

Thanks again acrossthecourt!

Some observations:
- Hakeem very bad on offense
- another piece of evidence how great were Stockton and Schrempf
- as a rookie Duncan already was as good (or even slightly better) as DRob
- Pippen was better on offense than defense (I told you so!)

[ceiling raiser]

1998 RAPM was published: http://ascreamingcomesacrossthecourt.bl ... ormed.html

Thanks again acrossthecourt!

Some observations:
- Hakeem very bad on offense
- another piece of evidence how great were Stockton and Schrempf
- as a rookie Duncan already was as good (or even slightly better) as DRob
- Pippen was better on offense than defense (I told you so!)

Agreed, thanks ATC!

The NPI results aren't perfect, but I think this well illustrates the bigs vs smalls dichotomy. By-and-large we're likely almost always underrating defensive contributions/overrating offensive contributions for bigs, and underrating offensive contributions/overrating defensive contributions for smalls.

[ronnymac2]

Why does Hakeem Olajuwon rank so poorly on offense in 1998?

16.4 points, 3 assists, 2.5 offensive rebounds, 2.7 turnovers. He shot 48 percent from the field and 75 percent from the line on 4.5 attempts. He had a true shooting percentage of 53 percent.

[acrossthecourt]

I think there's a weird interaction between Barkley's offense and Olajuwon's. Something like their offense didn't work well with both guys on the court or with only Olajuwon, but they were a lot better with only Barkley. (That's just a guess.) There are a number of explanations for that. Perhaps there's another player mixed in there confounding things. A prior-informed version might be able to clear that out.

I'm actually re-running the '97 dataset (added some games that were missing previously and rewrote the code that signifies the end of a lineup stint so it's more accurate now), and then I'll do a prior-informed pure-RAPM version of '98. I'll also include versions where a statistical plus/minus is used as a prior for both seasons.

Also, I'm not sure how JE calculated his data, what program he used, what lambda he had, etc. It seems like his lambda was harsher than the ones I've been using (I stick with the lambda.min found by the glmnet package.) So I wouldn't say there's a direct one-to-one comparison between the sets.

[Ryoga Hibiki]

On Olajuwon '98, I actually remember how ineffective he was looking vs the past. Much slower in the post, he was actually scoring only with his mid range jumper. I remember myself thinking he was done as a top player and the faster everyone understands it the better it will be.

Sent from my Nokia 3210 using Tapatalk

[Gregoire]

I think there's a weird interaction between Barkley's offense and Olajuwon's. Something like their offense didn't work well with both guys on the court or with only Olajuwon, but they were a lot better with only Barkley. (That's just a guess.) There are a number of explanations for that. Perhaps there's another player mixed in there confounding things. A prior-informed version might be able to clear that out.

I'm actually re-running the '97 dataset (added some games that were missing previously and rewrote the code that signifies the end of a lineup stint so it's more accurate now), and then I'll do a prior-informed pure-RAPM version of '98. I'll also include versions where a statistical plus/minus is used as a prior for both seasons.

Also, I'm not sure how JE calculated his data, what program he used, what lambda he had, etc. It seems like his lambda was harsher than the ones I've been using (I stick with the lambda.min found by the glmnet package.) So I wouldn't say there's a direct one-to-one comparison between the sets.

Does this data from 98 and 97 seasons include the playoffs?

[acrossthecourt]

I think there's a weird interaction between Barkley's offense and Olajuwon's. Something like their offense didn't work well with both guys on the court or with only Olajuwon, but they were a lot better with only Barkley. (That's just a guess.) There are a number of explanations for that. Perhaps there's another player mixed in there confounding things. A prior-informed version might be able to clear that out.

I'm actually re-running the '97 dataset (added some games that were missing previously and rewrote the code that signifies the end of a lineup stint so it's more accurate now), and then I'll do a prior-informed pure-RAPM version of '98. I'll also include versions where a statistical plus/minus is used as a prior for both seasons.

Also, I'm not sure how JE calculated his data, what program he used, what lambda he had, etc. It seems like his lambda was harsher than the ones I've been using (I stick with the lambda.min found by the glmnet package.) So I wouldn't say there's a direct one-to-one comparison between the sets.

Does this data from 98 and 97 seasons include the playoffs?

Yes, it's there in the google doc.