RAPM: 5-year samples

[LAL1947]

Here is a list of Top 100 players sorted by RAPM and Luck-adjusted RAPM for the last 5 years (as of June 13th, 2022).

Credit to @FloppyMoose for the lists.

The 5-year RAPM sample ranks Patrick Beverly higher than Anthony Davis, Fred VanVleet, Kyle Lowry, Bam Adebayo, Draymond Green, Khris Middleton, Domantas Sabonis, Jaylen Brown and Devin Booker.

Keep in mind, this sample includes 5 years of their data, so anomalies like this should not exist here. What are the PC Board's thoughts about this... does this invalidate the stat for player comparisons?

RAPM

Code: Select all

6.42 Stephen Curry
6.23 Giannis Antetokounmpo
6.09 Chris Paul
5.94 Joel Embiid
5.66 Jayson Tatum
5.55 Paul George
5.35 Kawhi Leonard
5.05 Rudy Gobert
4.91 Kevin Durant
4.82 Nikola Jokic
4.82 Jrue Holiday
4.75 LeBron James
4.68 Jimmy Butler
4.56 Pascal Siakam
4.47 Damian Lillard
4.24 Steven Adams
4.01 Alex Caruso
3.78 James Harden
3.74 Karl-Anthony Towns
3.59 Immanuel Quickley
3.58 Otto Porter Jr.
3.52 Clint Capela
3.47 Patrick Beverley
3.47 George Hill
3.46 Robert Covington
3.24 Danny Green
3.19 Joe Ingles
3.17 Derrick White
3.06 Jusuf Nurkic
2.99 Anthony Davis
2.94 Royce ONeale
2.94 Mike Conley
2.93 Jamal Murray
2.9 Draymond Green
2.89 Fred VanVleet
2.88 Kyle Lowry
2.81 Aaron Gordon
2.72 Gordon Hayward
2.72 Bam Adebayo
2.7 Mikal Bridges
2.7 Brook Lopez
2.67 Al Horford
2.64 Nene
2.55 Marcus Smart
2.42 Kyrie Irving
2.39 Danilo Gallinari
2.38 Reggie Jackson
2.37 Paul Millsap
2.35 Davis Bertans
2.34 Jakob Poeltl
2.34 Montrezl Harrell
2.28 Eric Gordon
2.27 Khris Middleton
2.26 Seth Curry
2.24 Dorian Finney-Smith
2.24 Thaddeus Young
2.19 Kemba Walker
2.18 Darius Garland
2.16 Al-Farouq Aminu
2.15 Desmond Bane
2.11 Nikola Mirotic
2.1 Matisse Thybulle
2.09 Domantas Sabonis
2.08 Jaylen Brown
2.08 Jonas Valanciunas
2.03 Reggie Bullock
2.01 Isaiah Hartenstein
1.98 Jarrett Allen
1.97 Luka Doncic
1.97 Jeremy Lamb
1.96 Devin Booker
1.94 Chris Boucher
1.92 Hassan Whiteside
1.89 Monte Morris
1.89 Ricky Rubio
1.87 Justise Winslow
1.85 Malcolm Brogdon
1.8 Bradley Beal
1.8 OG Anunoby
1.78 Jalen Brunson
1.77 Dwight Powell
1.75 Dejounte Murray
1.75 Kevon Looney
1.75 Michael Porter Jr.
1.74 Andre Roberson
1.73 Duncan Robinson
1.72 Herbert Jones
1.71 Marc Gasol
1.71 Luc Mbah a Moute
1.68 Evan Mobley
1.67 Victor Oladipo
1.66 P.J. Washington
1.66 Jose Alvarado
1.65 Kyle Korver
1.63 Nemanja Bjelica
1.61 LaMarcus Aldridge
1.6 Ersan Ilyasova
1.6 Kristaps Porzingis
1.6 Ben Simmons
1.6 Patty Mills


Luck Adjusted RAPM

Code: Select all

5.85 Giannis Antetokounmpo
5.08 Stephen Curry
5.05 Nikola Jokic
4.6 Rudy Gobert
4.5 Jayson Tatum
4.39 Chris Paul
4.29 Joel Embiid
4.23 Jimmy Butler
4.21 Paul George
4.19 Kawhi Leonard
4.19 LeBron James
3.89 Jrue Holiday
3.84 Damian Lillard
3.44 James Harden
3.43 Steven Adams
3.41 Karl-Anthony Towns
3.4 Brook Lopez
3.27 Alex Caruso
3.25 Kevin Durant
3.11 Jakob Poeltl
2.85 Robert Covington
2.84 Pascal Siakam
2.79 Anthony Davis
2.79 Clint Capela
2.79 Derrick White
2.73 Jusuf Nurkic
2.68 Mike Conley
2.67 Khris Middleton
2.58 Draymond Green
2.56 Bradley Beal
2.53 Danny Green
2.53 Herbert Jones
2.53 George Hill
2.52 Otto Porter Jr.
2.51 Patrick Beverley
2.51 Joe Ingles
2.47 Kyle Lowry
2.45 Aaron Gordon
2.39 Fred VanVleet
2.34 Malcolm Brogdon
2.32 Kemba Walker
2.32 Mikal Bridges
2.32 Immanuel Quickley
2.25 Hassan Whiteside
2.22 Al-Farouq Aminu
2.2 Michael Porter Jr.
2.16 Victor Oladipo
2.15 Zion Williamson
2.13 Al Horford
2.08 Kyle Korver
2 Ersan Ilyasova
1.99 Ricky Rubio
1.94 Dejounte Murray
1.94 Davis Bertans
1.92 Royce ONeale
1.89 Jose Alvarado
1.86 Domantas Sabonis
1.86 Andre Iguodala
1.83 Kelly Olynyk
1.83 Evan Mobley
1.81 Cody Martin
1.79 Kyrie Irving
1.79 Ja Morant
1.78 Kristaps Porzingis
1.76 Marcus Smart
1.75 Jonas Valanciunas
1.69 Cameron Johnson
1.65 Russell Westbrook
1.62 Eric Bledsoe
1.6 CJ McCollum
1.59 Paul Millsap
1.57 Dorian Finney-Smith
1.57 Jarrett Allen
1.57 Duncan Robinson
1.56 Andre Roberson
1.55 Jalen Brunson
1.55 Michael Kidd-Gilchrist
1.53 Tyus Jones
1.53 Richaun Holmes
1.53 Boban Marjanovic
1.52 J.J. Barea
1.5 Nikola Vucevic
1.49 Myles Turner
1.49 Seth Curry
1.49 Thaddeus Young
1.47 Montrezl Harrell
1.45 LaMarcus Aldridge
1.43 Brandon Clarke
1.42 Kevon Looney
1.42 Eric Gordon
1.4 Isaiah Hartenstein
1.37 Bam Adebayo
1.36 Luc Mbah a Moute
1.35 Jamal Murray
1.34 DeAnthony Melton
1.31 Dwight Powell
1.31 John Konchar
1.26 Nemanja Bjelica
1.25 Klay Thompson
1.24 Danilo Gallinari


[MyUniBroDavis]

Here is list of Top 100 players sorted by RAPM and Luck-adjusted RAPM for the last 5 years.

Credit to @FloppyMoose for the lists.

The 5-year RAPM sample ranks Patrick Beverly higher than Anthony Davis, Fred VanVleet, Kyle Lowry, Bam Adebayo, Draymond Green, Khris Middleton, Domantas Sabonis, Jaylen Brown and Devin Booker.

What are the PC Board's thoughts about this... does it invalidate the stat?

RAPM

Code: Select all

6.42 Stephen Curry
6.23 Giannis Antetokounmpo
6.09 Chris Paul
5.94 Joel Embiid
5.66 Jayson Tatum
5.55 Paul George
5.35 Kawhi Leonard
5.05 Rudy Gobert
4.91 Kevin Durant
4.82 Nikola Jokic
4.82 Jrue Holiday
4.75 LeBron James
4.68 Jimmy Butler
4.56 Pascal Siakam
4.47 Damian Lillard
4.24 Steven Adams
4.01 Alex Caruso
3.78 James Harden
3.74 Karl-Anthony Towns
3.59 Immanuel Quickley
3.58 Otto Porter Jr.
3.52 Clint Capela
3.47 Patrick Beverley
3.47 George Hill
3.46 Robert Covington
3.24 Danny Green
3.19 Joe Ingles
3.17 Derrick White
3.06 Jusuf Nurkic
2.99 Anthony Davis
2.94 Royce ONeale
2.94 Mike Conley
2.93 Jamal Murray
2.9 Draymond Green
2.89 Fred VanVleet
2.88 Kyle Lowry
2.81 Aaron Gordon
2.72 Gordon Hayward
2.72 Bam Adebayo
2.7 Mikal Bridges
2.7 Brook Lopez
2.67 Al Horford
2.64 Nene
2.55 Marcus Smart
2.42 Kyrie Irving
2.39 Danilo Gallinari
2.38 Reggie Jackson
2.37 Paul Millsap
2.35 Davis Bertans
2.34 Jakob Poeltl
2.34 Montrezl Harrell
2.28 Eric Gordon
2.27 Khris Middleton
2.26 Seth Curry
2.24 Dorian Finney-Smith
2.24 Thaddeus Young
2.19 Kemba Walker
2.18 Darius Garland
2.16 Al-Farouq Aminu
2.15 Desmond Bane
2.11 Nikola Mirotic
2.1 Matisse Thybulle
2.09 Domantas Sabonis
2.08 Jaylen Brown
2.08 Jonas Valanciunas
2.03 Reggie Bullock
2.01 Isaiah Hartenstein
1.98 Jarrett Allen
1.97 Luka Doncic
1.97 Jeremy Lamb
1.96 Devin Booker
1.94 Chris Boucher
1.92 Hassan Whiteside
1.89 Monte Morris
1.89 Ricky Rubio
1.87 Justise Winslow
1.85 Malcolm Brogdon
1.8 Bradley Beal
1.8 OG Anunoby
1.78 Jalen Brunson
1.77 Dwight Powell
1.75 Dejounte Murray
1.75 Kevon Looney
1.75 Michael Porter Jr.
1.74 Andre Roberson
1.73 Duncan Robinson
1.72 Herbert Jones
1.71 Marc Gasol
1.71 Luc Mbah a Moute
1.68 Evan Mobley
1.67 Victor Oladipo
1.66 P.J. Washington
1.66 Jose Alvarado
1.65 Kyle Korver
1.63 Nemanja Bjelica
1.61 LaMarcus Aldridge
1.6 Ersan Ilyasova
1.6 Kristaps Porzingis
1.6 Ben Simmons
1.6 Patty Mills


Luck Adjusted RAPM

Code: Select all

5.85 Giannis Antetokounmpo
5.08 Stephen Curry
5.05 Nikola Jokic
4.6 Rudy Gobert
4.5 Jayson Tatum
4.39 Chris Paul
4.29 Joel Embiid
4.23 Jimmy Butler
4.21 Paul George
4.19 Kawhi Leonard
4.19 LeBron James
3.89 Jrue Holiday
3.84 Damian Lillard
3.44 James Harden
3.43 Steven Adams
3.41 Karl-Anthony Towns
3.4 Brook Lopez
3.27 Alex Caruso
3.25 Kevin Durant
3.11 Jakob Poeltl
2.85 Robert Covington
2.84 Pascal Siakam
2.79 Anthony Davis
2.79 Clint Capela
2.79 Derrick White
2.73 Jusuf Nurkic
2.68 Mike Conley
2.67 Khris Middleton
2.58 Draymond Green
2.56 Bradley Beal
2.53 Danny Green
2.53 Herbert Jones
2.53 George Hill
2.52 Otto Porter Jr.
2.51 Patrick Beverley
2.51 Joe Ingles
2.47 Kyle Lowry
2.45 Aaron Gordon
2.39 Fred VanVleet
2.34 Malcolm Brogdon
2.32 Kemba Walker
2.32 Mikal Bridges
2.32 Immanuel Quickley
2.25 Hassan Whiteside
2.22 Al-Farouq Aminu
2.2 Michael Porter Jr.
2.16 Victor Oladipo
2.15 Zion Williamson
2.13 Al Horford
2.08 Kyle Korver
2 Ersan Ilyasova
1.99 Ricky Rubio
1.94 Dejounte Murray
1.94 Davis Bertans
1.92 Royce ONeale
1.89 Jose Alvarado
1.86 Domantas Sabonis
1.86 Andre Iguodala
1.83 Kelly Olynyk
1.83 Evan Mobley
1.81 Cody Martin
1.79 Kyrie Irving
1.79 Ja Morant
1.78 Kristaps Porzingis
1.76 Marcus Smart
1.75 Jonas Valanciunas
1.69 Cameron Johnson
1.65 Russell Westbrook
1.62 Eric Bledsoe
1.6 CJ McCollum
1.59 Paul Millsap
1.57 Dorian Finney-Smith
1.57 Jarrett Allen
1.57 Duncan Robinson
1.56 Andre Roberson
1.55 Jalen Brunson
1.55 Michael Kidd-Gilchrist
1.53 Tyus Jones
1.53 Richaun Holmes
1.53 Boban Marjanovic
1.52 J.J. Barea
1.5 Nikola Vucevic
1.49 Myles Turner
1.49 Seth Curry
1.49 Thaddeus Young
1.47 Montrezl Harrell
1.45 LaMarcus Aldridge
1.43 Brandon Clarke
1.42 Kevon Looney
1.42 Eric Gordon
1.4 Isaiah Hartenstein
1.37 Bam Adebayo
1.36 Luc Mbah a Moute
1.35 Jamal Murray
1.34 DeAnthony Melton
1.31 Dwight Powell
1.31 John Konchar
1.26 Nemanja Bjelica
1.25 Klay Thompson
1.24 Danilo Gallinari


This is shotcharts right?

You’ll always expect wonky result or two I think, and there are reasons ADs 5 year rapm wouldn’t be great (I’d know lol).

It’s not a measure of goodness it’s a measure/estimator for impact which inherently has noise. I think their 5 year rapm might just be an averaging but I could be wrong in thag

It’s nice they have a luck adjusted one

[WestGOAT]

Reposting what I posted in the other thread:

.

Source of this RAPM ranking?

What are the actual numbers/values, and associated confidence intervals or standard errors, supporting these rankings? I have seem that the RAPM confidence intervals can be quiet large: https://squared2020.com/2019/10/03/exercising-error-quantifying-statistical-tests-under-rapm-part-iv/amp/

Not that the ranking is worthless without these, but a lot of context is missing.

[LAL1947]

This is shotcharts right?

You’ll always expect wonky result or two I think

Yes, this is Shotcharts... and this sample includes 5 years of their data, so you would expect the list to be smoothed out instead of having anomalies like this one.

and there are reasons ADs 5 year rapm wouldn’t be great (I’d know lol).

I don't see why AD's RAPM should be lower. AD has had time out injured... but he played 9,274 minutes in the past 5 seasons, while Patrick Beverly has played 6,121 minutes.

This RAPM sample also puts Patrick Beverly over Khris Middleton, Bam Adebayo, Jaylen Brown, Devin Booker, Fred VanVleet, Kyle Lowry, Draymond Green and Domantas Sabonis... so let's not forget those guys.

It’s not a measure of goodness it’s a measure/estimator for impact which inherently has noise.

That's a good observation. I think the PC Board should take this into consideration, as they seemingly do their best player rankings using impact and/or efficiency while ignoring roster context (fit + quality + depth), rather than goodness.

[Dutchball97]

It's important to remember RAPM rates how well someone does in their role so guys like Bev might grade out better than some lower-middle end star players due to how well they relatively impacted the game. Since it's 5 years you've also got to account for some of those guys you mentioned barely having a 5 year career yet. Bam has played exactly 5 seasons and he wasn't nearly as good in his first 2 seasons so his average is lower than you'd expect by his play the last 3 years. AD obviously gets downgraded because of many injuries and generally inconsistent form. Then you've got someone like Lowry who has been steadily declining the last few years.

RAPM isn't my favorite stat either and while the multi-year versions are supposed to reduce noise, they might not always give an accurate average because not everyone stays around the same level in a 5-year period. I don't think this means you should invalidate any stat that has some questionable results though. One stat having certain players lower might just be because that specific formula doesn't account as much for something they do well or is too harsh on something they don't excell at but when you see a pattern come up across multiple different stats rating a player lower than his reputation I do think it is wise to take that into account.

[Bad Gatorade]

What's notable about Patrick Beverley's RAPM is that he actually ranks very highly on offence too (55th in the entire league in this span), but also that since this database has been created (going back to the 2012-2016 span), 55th is actually his lowest rank on offence, and his lowest overall rank is 37th.

Given that he frequently misses games and it's over a very large sample... perhaps there's a lot more to him than meets the eye?

I do remember that in non-Harden minutes, he was actually quite a productive point guard. In 2017, the Rockets had an ORTG of 114.1 in the Beverley + no Harden minutes, he averaged 9 assists per 75 possessions to 2.6 turnovers (46.5% of those being "at rim" assists). He's quite a good 3 point shooter and quite a good offensive rebounder for a guard, too (1.5 per 36 for his career is very good for a point guard whose scoring game is primarily 3 point based). He's also a hustle heavy player, making him solid enough at grabbing loose balls on the offensive end/screens from a PG etc.

I think his scores are arguably higher than they should be if we're looking at "goodness", because he's good enough to keep offences afloat against bench lineups whilst having a game that scales well next to ball dominant players. In other words, I don't think he could keep this level of ORAPM up as a pure lead guard, and his foul proneness limits the minutes that he can play in the first place, but yeah, I think that his surprisingly high RAPM scores can be "explained" and that Beverley has consistently been in a "best case" scenario to generate high RAPM scores, if that makes sense.

I think that RAPM definitely has some blind spots -

* Role on the team, where I think Beverley would quite arguably benefit since he has a nice mix of being a nice role player for a high volume player AND being a surprisingly decent backup PG

* Minutes, where Beverley benefits by being foul prone, which not only limits his overall productivity, but also has the byproduct of affecting players after he goes to the bench. He's been a starter most of his career and it's not surprising if he racks up a lot of fouls early, and then gets benched, putting his team in the penalty (which would reduce his team's DRTG, but not when he's on the court)

* Ability to scale up on the offensive end, where Beverley does a solid job as a backup PG, but he's not good enough to rely on large stretches, or to generate buckets in a closer game scenario

* Ability to scale up on the defensive end, where Beverley is a bulldog throughout the game, whereas a guy like LeBron or Kawhi knows when he has to up the defensive intensity during a closer game

* Being able to change an approach based on the current score margin - this one is a bit "harder" to explain, but here I go. One thing I credit LeBron for heavily is that LeBron's propensity to score/pass can change quite a bit during courses of the game, where he has the capacity to play off whatever is needed - if he feels like he needs a consistent bucket, he'll score. If he has room to increase a scoring margin through higher risk, higher reward shots, he's amazing at finding 3 point shooters. Beverley's poor interior scoring game limits this capacity a bit

* Versatility, which is kind of self explanatory

A lot of the guys ranked behind Beverley in RAPM can play bigger minutes, can scale up through better effort, can default to more reliable shooting options, aren't costing their team through foul proneness, and so on. Beverley is still a good player, but also one whose RAPM scores are likely inflated a bit relative to the average player. I'd imagine numerous higher-profile players who are slightly beneath Beverley on RAPM are still providing more in terms of "win probability" or "total value added per game" but aren't in a situation which maximises their RAPM. After all, would you take 25 minutes of Beverley or 38 minutes of FVV? Does Beverley change the outcome of a close game as much as Anthony Davis can?

I don't think that RAPM should be "invalidated" at all, because it provides a perspective that many other stats (e.g. box score stats) just don't provide at all, and helps isolate individuals from team success a bit more. Sometimes, our eye test and our focus fails to capture things RAPM does provide a numeric evaluation for - for example, we'd split hairs endlessly over whether a player is a +5 or +6 offensive player, but we struggle to, or fail to provide any distinction between a -1 and -2 defensive player. It's a very valuable tool on this front. I do think that RAPM can often provide valuable insight that we might otherwise miss, simply because there's a finite amount of basketball we can watch and synthesise accurately.

It does, however, require context (as all stats do, to be fair) and I'd argue that it's also worth looking at many samples for RAPM of differing lengths and time periods - for example, we might have RAPM for 02-06, 07-11 and 02-11, and a player could look notably better in the 02-11 sample rather than the two individual 5 year samples. Anomalies can still exist, even in larger samples, and we should do everything we can in order to assess whether an anomaly is truly just a very weird sampling error, or if it stems from hidden value. Then, of course, we should consider context on top of this. Once we mesh all of this together... RAPM is incredibly valuable. But yes, if we're just using a single sample and drawing sweeping conclusions, it can be highly misused.

[ceiling raiser]

General thought on RAPM -

Do we need to better distinguish between floor raising and ceiling raising efforts?

i.e. take Player A and Player B

Player A has a 5.0 RAPM (+5.25/-0.25 O/D)
Player B also has a 5.0 RAPM (+3.00/+2.00 O/D)

However say Player A has an on-court NetRtg of +11.0, Player B + 2.50.

How would you interpret this breakdown?

[falcolombardi]

General thought on RAPM -

Do we need to better distinguish between floor raising and ceiling raising efforts?

i.e. take Player A and Player B

Player A has a 5.0 RAPM (+5.25/-0.25 O/D)
Player B also has a 5.0 RAPM (+3.00/+2.00 O/D)

However say Player A has an on-court NetRtg of +11.0, Player B + 2.50.

How would you interpret this breakdown?

the way rapm works i think it would be unlikely a player with a 5 RAPM would have a netrtg so low wouldnt it?

edit: by net rating you mean team rating om court or on/off?

[ShotCreator]

Anthony Davis, Fred VanVleet, Kyle Lowry, Bam Adebayo, Draymond Green, Khris Middleton, Domantas Sabonis, Jaylen Brown and Devin Booker.

Not a surprise. None of these guys consistently play at elite levels.

Anthony Davis wasn’t even consistent in his athletic prime in NO. Good but not elite defense, and offense with almost no creating for teammates.

He probably has two or three honest elite seasons with NO and 2019 AD is one of them. His last two seasons in LA were like his 2016 year. Maybe worse.

I don’t find Middleton or Brown to be better than prime Bev anyway. Same kind of high volume role players. It’s been years since Khris was consistently a good defender, and I don’t think Brown has ever been above average.

Booker had a hell of a 2021 playoffs on both ends bit 2022 was his first complete regular season with good D and O.

Lowry’s old.

Draymond is offensively neutered, and defensively only half as good in the RS.

Bam is not elite at anything and can’t shoot as a big man. Massive cap on his game.

Sabonis is similar.

FVV has put together a couple really good years though.

[ceiling raiser]

General thought on RAPM -

Do we need to better distinguish between floor raising and ceiling raising efforts?

i.e. take Player A and Player B

Player A has a 5.0 RAPM (+5.25/-0.25 O/D)
Player B also has a 5.0 RAPM (+3.00/+2.00 O/D)

However say Player A has an on-court of +11.0, Player B + 2.50.

How would you interpret this breakdown?

[ceiling raiser]

General thought on RAPM -

Do we need to better distinguish between floor raising and ceiling raising efforts?

i.e. take Player A and Player B

Player A has a 5.0 RAPM (+5.25/-0.25 O/D)
Player B also has a 5.0 RAPM (+3.00/+2.00 O/D)

However say Player A has an on-court NetRtg of +11.0, Player B + 2.50.

How would you interpret this breakdown?

the way rapm works i think it would be unlikely a player with a 5 RAPM would have a netrtg so low wouldnt it?

edit: by net rating you mean team rating om court or on/off?

Meant just on-court not on/off. Edited.

[GeorgeMarcus]

Reposting what I posted in the other thread:

.

Source of this RAPM ranking?

What are the actual numbers/values, and associated confidence intervals or standard errors, supporting these rankings? I have seem that the RAPM confidence intervals can be quiet large: https://squared2020.com/2019/10/03/exercising-error-quantifying-statistical-tests-under-rapm-part-iv/amp/

Not that the ranking is worthless without these, but a lot of context is missing.

I think you are misunderstanding how I incorporate the numbers into my analysis. That's why I tried to emphasize how no stat will ever translate to some kind of all-encompassing player value. I'll reuse the analogy I used last night... Think of it like profit margins in a business. A lot of factors go into those margins- marketing schemes, administrative changes, the world economy, and so on. In that sense it doesn't make sense to say "our Q3 margins were best so X practice we engaged at the time was the reason for that". Evaluating the why of the profit margins requires a lot more nuance. And yet, if businesses were to ignore profit margins entirely for being "too noisy" it would be incredibly dumb. The singular goal of a business is to make money. The singular goal of basketball is to win. Those are the fundamental lenses through which decision making should be considered.

[giberish]

General thought on RAPM -

Do we need to better distinguish between floor raising and ceiling raising efforts?

i.e. take Player A and Player B

Player A has a 5.0 RAPM (+5.25/-0.25 O/D)
Player B also has a 5.0 RAPM (+3.00/+2.00 O/D)

However say Player A has an on-court NetRtg of +11.0, Player B + 2.50.

How would you interpret this breakdown?

the way rapm works i think it would be unlikely a player with a 5 RAPM would have a netrtg so low wouldnt it?

edit: by net rating you mean team rating om court or on/off?

Meant just on-court not on/off. Edited.

Both are overall good to very good players (or at least very effective in the roles that they're asked to play) (assuming both are playing significant minutes).

Player A has essentially all his value on offense, but he's not a defensive liability; while Player B is more balanced in terms of offensive and defensive value.

Player A gets to play with much better teammates, likely on a team that would be solid even if he were out injured. Player B plays on a weaker team that's probably under .500 without him.

[Doctor MJ]

General thought on RAPM -

Do we need to better distinguish between floor raising and ceiling raising efforts?

i.e. take Player A and Player B

Player A has a 5.0 RAPM (+5.25/-0.25 O/D)
Player B also has a 5.0 RAPM (+3.00/+2.00 O/D)

However say Player A has an on-court NetRtg of +11.0, Player B + 2.50.

How would you interpret this breakdown?

Short answer to the first is: RAPM isn't going to be very effective on its own to distinguish between floor raising and ceiling raising.

On the second, we'd certainly guess that Player A is more likely to be ceiling-raising than Player B.

[WestGOAT]

I think you are misunderstanding how I incorporate the numbers into my analysis.

1) Can you give an example of how you are incorporating which specific numbers?

2) I understand your analogy, but profit margins are precise values if all the information is complete, it's the difference between revenue and expenses.... there is no standard error or confidence intervals that come into play, whereas the single RAPM values do have large variance terms, did you look in the example I provided in: https://squared2020.com/2019/10/03/exercising-error-quantifying-statistical-tests-under-rapm-part-iv/amp/

If not here is a screenshot of the variance associated with 2018-2019 Danny Green's +/-:


This ranges indicates that, with 95% certainty, Danny Green's +/- falls between these lower and higher bound values.

As you can see in the table below these intervals are huge:


So, by simply making a ranking by averaging out yearly +/-, you are missing a lot of context. How certain can you be about this? Well that's where standard error can shed some light.

[LukaTheGOAT]

Also, will share this thread, as it potentially shines some good light on things to consider when view RAPM
https://apbr.org/metrics/viewtopic.php?f=2&t=8566&hilit=16+20+RAPM

[GeorgeMarcus]

I think you are misunderstanding how I incorporate the numbers into my analysis.

1) Can you give an example of how you are incorporating which specific numbers?

2) I understand your analogy, but profit margins are precise values if all the information is complete, it's the difference between revenue and expenses.... there is no standard error or confidence intervals that come into play, whereas the single RAPM values do have large variance terms, did you look in the example I provided in: https://squared2020.com/2019/10/03/exercising-error-quantifying-statistical-tests-under-rapm-part-iv/amp/

If not here is a screenshot of the variance associated with 2018-2019 Danny Green's +/-:


This ranges indicates that, with 95% certainty, Danny Green's +/- falls between these lower and higher bound values.

As you can see in the table below these intervals are huge:


So, by simply making a ranking by averaging out yearly +/-, you are missing a lot of context. How certain can you be about this? Well that's where standard error can shed some light.

My bad, I meant to respond to this before and got distracted.

I understand your viewpoint better and believe I can identify our disconnect. You are talking about RAPM as a predictive value- which is often the case in RAPM discussions. I'm in the minority who looks at RAPM as more of a descriptive value. My approach can be hard to articulate because RAPM is not as digestible as raw +/-, Net Rtg, etc. However I believe all derivations of +/- data should be viewed as descriptive first and foremost, and in that sense RAPM as a sort of weighted +/- metric.

Honestly I think this taps into my broader approach to basketball analytics, not just impact data. IMO predictive analysis should require a more layered approach with distinct subsets of information as a means of forming the 'big picture'. Naturally this sort of approach hinges on one's subjective outlook like which subsets to consider, how to fuse them together into a single valuation, etc.

Am I making sense?