RealGM

The list of this stat on ESPN's NBA page has Butler as the best Bulls player at 41st which should make this a total B.S. stat.

Mirotic is the second listed Bull at 64th and Rose clocks in at 145th.

WTF?

You don't agree that Butler is the best Bulls player this year or the fact that he's only 41st?

What is real plus-minus?

RPM is inspired by the same underlying +/- logic: If a team outscores its opponents when a player is on the court, he's probably doing something to help his team, whether or not he's putting up big numbers.

But the familiar +/- stat has a serious flaw: Each player's rating is heavily influenced by the play of his on-court teammates.

For example, in the basic +/- numbers, Thunder backup point guard Reggie Jackson is ranked 27th in the league. But he's also spent the majority of his minutes playing alongside Kevin Durant, the league's likely MVP. What we really want to know is how much of Jackson's elite rating is attributable to his own play, and basic +/- simply can't tell us.

But real plus-minus can.

Where does real plus-minus come from?

WHY "REAL" PLUS-MINUS?

"Real" comes from economics and such terminology as real income, real interest rate and real value. Instead of "nominal" value, for instance, economists prefer to talk about real value, which is adjusted for inflation.

Likewise, real plus-minus includes important adjustments to raw stats.

Drawing on advanced statistical modeling techniques (and the analytical wizardry of RPM developer Jeremias Engelmann, formerly of the Phoenix Suns), the metric isolates the unique plus-minus impact of each NBA player by adjusting for the effects of each teammate and opposing player.

The RPM model sifts through more than 230,000 possessions each NBA season to tease apart the "real" plus-minus effects attributable to each player, employing techniques similar to those used by scientific researchers when they need to model the effects of numerous variables at the same time.

RPM estimates how many points each player adds or subtracts, on average, to his team's net scoring margin for each 100 possessions played. The RPM model also yields separate ratings for the player's impact on both ends of the court: offensive RPM (ORPM) and defensive RPM (DRPM).

ORIGINS OF RPM

Real plus-minus (RPM) was developed by Jeremias Engelmann, formerly of the Phoenix Suns, in consultation with Steve Ilardi, University of Kansas psychology professor and former NBA consultant.

It follows the development of adjusted plus-minus (APM) by several analysts and regularized adjusted plus-minus (RAPM) by Joe Sill.

RPM reflects enhancements to RAPM by Engelmann, among them the use of Bayesian priors, aging curves, score of the game and extensive out-of-sample testing to improve RPM's predictive accuracy.

As the name suggests, real plus-minus shares a family resemblance with the +/- stat in the box score, which merely registers the net change in score (plus or minus) while each player is on the court.

http://espn.go.com/nba/story/_/id/10740 ... plus-minus

Is it meaningful? Well better than raw +/- given enough time and data. Usually best used to capture the value of defensive players, that traditional counting stats don't promote, or 'intangible' or glue guys, who don't put up a lot of box stats, but are doing a lot of little things to help your team.

Butler has a negative DRPM this year which hurts his overall RPM (last year it was positive).

Most guys with a top RPM have both positive ORPM & DRPM, like Steph Curry or Lebron or AD.

Whether any of this is meaningful...who knows.

I think the DRPM listing is surprisingly good though, more so than the overall RPM.

What I have not been able to figure out is it xRAPM or another xRAPM flavor for ESPN? It's all Jeremias Engelmann's work apparently.

I applaud ESPN for venturing into APM stats--if you are picking one an xRAPM based one probably correlates best to actual wins over time. Strange I did not see team wins estimated by using it this year by ESPN-still using some other method it seems.

I think they are pushing it too much as a one number to explain everything (essentially just replacing PER on that front) which is always a risk. I'd like to see an article describing how these are measures of outcomes based on past and current performance--and not an intrinsic valuation of a player (they have a value in that but not end all be all). They could point to Hinrich (not this year) when he first came to CHI from ATL. There are a few other players who switched teams and improved Rudy Gay to SAC--typically by finding a proper role/fit. Both had significant RPM growth. It should be in the basket of stats to evaluate a guy but not be thrown about as a guys intrinsic value...which some times without qualifiers seemingly lean this way.

But then you read articles and it's obvious Doolittle, etc are clearly looking at a multitude of other stats as well. So I like it but --caution---"Here be dragons".

micromonkey wrote:What I have not been able to figure out is it xRAPM or another xRAPM flavor for ESPN? It's all Jeremias Engelmann's work apparently.

very similar if not identical. here is the ESPN version for the entirety of last season:

http://espn.go.com/nba/statistics/rpm/_/year/2014

compare to englemann's site. note that the last update was only at the end of march:

http://stats-for-the-nba.appspot.com/ratings/xRAPM.html

xRAPM uses box score weights to "stabilize" the the plus-minus data. i interpret that as meaning that it kind of bastardizes the whole idea of +/- in order to make the results more reasonable (less outliers, more stable from season-to-season). i'd prefer to just see some form of advanced +/- without box score seasoning - warts and all. don't know if that stat is readily available

dice wrote:

micromonkey wrote:What I have not been able to figure out is it xRAPM or another xRAPM flavor for ESPN? It's all Jeremias Engelmann's work apparently.

very similar if not identical. here is the ESPN version for the entirety of last season:

http://espn.go.com/nba/statistics/rpm/_/year/2014

compare to englemann's site. note that the last update was only at the end of march:

http://stats-for-the-nba.appspot.com/ratings/xRAPM.html

xRAPM uses box score weights to "stabilize" the the plus-minus data. i interpret that as meaning that it kind of bastardizes the whole idea of +/- in order to make the results more reasonable (less outliers, more stable from season-to-season). i'd prefer to just see some form of advanced +/- without box score seasoning - warts and all. don't know if that stat is readily available

Here is 2015 prior informed RAPM (RPM without the box score): http://stats-for-the-nba.appspot.com/ratings/RAPM_2015.html

2008-2014: http://www.gotbuckets.com/statistics/rapm/2014-rapm/

Data going as far back as 1994 these days, you just have to know where to find it. Also, work is being done to develop a statistical plus minus based in RAPM and player tracking Data, which is currently outperforming every other single metric in predicting wins in 2015 by a landslide.

Dr Spaceman wrote:Here is 2015 prior informed RAPM (RPM without the box score): http://stats-for-the-nba.appspot.com/ratings/RAPM_2015.html

prior informed takes the past couple of years into account, correct?

dice wrote:

Dr Spaceman wrote:Here is 2015 prior informed RAPM (RPM without the box score): http://stats-for-the-nba.appspot.com/ratings/RAPM_2015.html

prior informed takes the past couple of years into account, correct?

It's "multi-year" which is kind of muddy. The idea is 2015 is informed by 2014, which is informed by 2013... and so on. The 2008-2014 link I posted above I know for sure uses the previous two year APM (no ridge regression), but for J.E.'s datasets I'm not exactly clear on what he's using. Since he's been working for ESPN, RAPM has kind of become a lost art. http://www.apbr.org/metrics/viewtopic.php?f=2&t=8827

If you're interested, here is non prior informed RAPM covering 2014 and 1st half 2015: https://docs.google.com/spreadsheets/d/17EHZ5HwmBRv3Y6UC1dKrtcfSKUNTqGWarfRUpnhGTno/edit#gid=1655629432 This version is far less "skeptical" than prior informed, so it has no problem with players like Korver and Lowry who took massive leaps in a single season. But it's also a far noisier approach, so take these rankings with a gigantic mountain of salt and remember to weight for minutes played, role, etc.

JerrySloan wrote:The list of this stat on ESPN's NBA page has Butler as the best Bulls player at 41st which should make this a total B.S. stat.

Mirotic is the second listed Bull at 64th and Rose clocks in at 145th.

WTF?

Rose has been an average NBA player this year. Jimmy has clearly been the best. Whether Niko has truly been better than Pau is questionable, but it makes for a good discussion. We know that Pau's defense has given the Bulls a lot of problems, and we also know that Niko is very disruptive to opposition defenses. Maybe he really has been that effective for the Bulls. Or maybe not. But I'm glad that the stat raised the issue.

As for Jimmy at #41, I am shaking my head at that one, as well as his low defensive rating. I don't think that's quite right. But no stat is perfect.

Dr Spaceman wrote:If you're interested, here is non prior informed RAPM covering 2014 and 1st half 2015: https://docs.google.com/spreadsheets/d/17EHZ5HwmBRv3Y6UC1dKrtcfSKUNTqGWarfRUpnhGTno/edit#gid=1655629432 This version is far less "skeptical" than prior informed, so it has no problem with players like Korver and Lowry who took massive leaps in a single season. But it's also a far noisier approach, so take these rankings with a gigantic mountain of salt and remember to weight for minutes played, role, etc.

this seems to be the purest form of advanced +/-, however noisy. is it available going farther back?

JerrySloan wrote:The list of this stat on ESPN's NBA page has Butler as the best Bulls player at 41st which should make this a total B.S. stat.

Mirotic is the second listed Bull at 64th and Rose clocks in at 145th.

WTF?

So I'll say this:

Stats in the (+/-) family are not meant to be taken as "player x is better than player y". At the heart, these statistics are meant to measure "when player x steps on the court, his team improves by x amount". This is an absolutely crucial distinction to make, because so often people look at stats like this and say "No way are Manu Ginobili and Amir Johnson better than Kevin Durant, this is B.S." So this isn't a stat you can just look at and stop thinking, in fact these stats are designed to inform your thinking. People who get really into plus minus stats (like myself) do so because the numbers that are spit out match pretty much exactly what we see on film.

So what is this measuring? Basically, the idea is that every player in the NBA plays in a multitude of lineups every season and that using all of this lineup data, you can measure the precise impact of each individual player on the scoring margin.. Think of it like an algebra problem, where

scoring margin= Bulls1+Bulls2+Bulls3+Bulls4+Bulls5-Jazz1-Jazz2-Jazz3-Jazz4-Jazz5

If you do this over thousands of lineups and thousands of opponent lineups, you can start to get a good picture of each of the individual variables in this equation. (Note here that Bulls 1,2,3 would be Jordan, Pippen, Rodman, etc.)

RPM is the latest version of this type of stat, which uses box score data to make the results look more "intuitive". I wholeheartedly disagree with this approach, but since this is the most easily available plus minus stat complaining doesn't do a whole lot of good.

The important thing to remember is that you should take these number for exactly what they are: a given player's impact on the scoring margin. Nothing more, nothing less.So for example, I still think Durant is a better player than Ginobili, because his skill set is superior, he plays the #1 role on offense, and he plays far more minutes than Ginobili. I sort of "filter" RAPM data through my own observations to come up with a clear picture of a given player. So the data are telling me that Ginobili, in his role, is one of the most impactful players in the league. But when you consider how limited his role actually is, I'm not likely to rank him very highly overall.

Regarding your questions about specific Bulls players, all I'm going to say is that years and years of data agree that players who play the way Rose has this season do not help their teams win. And players like Mirotic who shoot well and space the floor are far more impactful than you'd think (look at nba.com's lineup data, Mirotic is a part of basically all of the Bulls' best offensive lineups).

dice wrote:

Dr Spaceman wrote:If you're interested, here is non prior informed RAPM covering 2014 and 1st half 2015: https://docs.google.com/spreadsheets/d/17EHZ5HwmBRv3Y6UC1dKrtcfSKUNTqGWarfRUpnhGTno/edit#gid=1655629432 This version is far less "skeptical" than prior informed, so it has no problem with players like Korver and Lowry who took massive leaps in a single season. But it's also a far noisier approach, so take these rankings with a gigantic mountain of salt and remember to weight for minutes played, role, etc.

this seems to be the purest form of advanced +/-, however noisy. is it available going farther back?

https://sites.google.com/site/rapmstats/2006-npi
colts18's site has several years of NPI data

You can generally find NPI data by googling (acrossthecourt, shutupandjam usually post it on their blogs), but for the most part people prefer prior informed. If you're looking for the "purest", you might do well looking at simple APM (adjusted plus minus) which does not use any ridge regression, meaning that outliers are generally given free reign. The idea of ridge regression being that you can smooth out datapoint that don't seem to make sense.

https://docs.google.com/spreadsheets/d/1h20JYcZJu2tGNIyOwVbNfez0-zXXy5ItLyXC4qTE5D8/edit#gid=0
This is DoctorMJ's scaled PI spreadsheet, with data going back to 98, excepting 2001.

In general you should no that PI RAPM is the preferred way of doing things for the people who know these stats best,and there's good reason for that. The methodology behind it is peer reviewed and it's pretty accepted that this is the "best" way to use RAPM. Looking at NPI can be helpful, but it's important to remember that the outputs are far less reliable than PI.

Dr Spaceman wrote:

dice wrote:

Dr Spaceman wrote:If you're interested, here is non prior informed RAPM covering 2014 and 1st half 2015: https://docs.google.com/spreadsheets/d/17EHZ5HwmBRv3Y6UC1dKrtcfSKUNTqGWarfRUpnhGTno/edit#gid=1655629432 This version is far less "skeptical" than prior informed, so it has no problem with players like Korver and Lowry who took massive leaps in a single season. But it's also a far noisier approach, so take these rankings with a gigantic mountain of salt and remember to weight for minutes played, role, etc.

this seems to be the purest form of advanced +/-, however noisy. is it available going farther back?

If you're looking for the "purest", you might do well looking at simple APM (adjusted plus minus) which does not use any ridge regression, meaning that outliers are generally given free reign. The idea of ridge regression being that you can smooth out datapoint that don't seem to make sense.

i'm fine with the regression because that's just pure math. there is no distorting of the general philosophy of +/- by using box scores nor muddying the waters with prior season data

In general you should no that PI RAPM is the preferred way of doing things for the people who know these stats best,and there's good reason for that. The methodology behind it is peer reviewed and it's pretty accepted that this is the "best" way to use RAPM. Looking at NPI can be helpful, but it's important to remember that the outputs are far less reliable than PI.

makes sense. thanks

I cannot say that I read everything in all the above posts - I freely admit to my aged eyes glazing over after a while - but I appreciate the responses.

However, with all due respect to the mathletes who came up with these calculations - although I don't know if their acumen equals that of The Simpsons crew who discovered the Higgs Boson - I think the idea that Draymond Green(12), Kyle Korver(14), Hassan Whitside(18), Markieff Morris(25) and Nene(32) have had a more positive impact on their teams this year than Jimmy Butler is plain and simple rubbish.

JerrySloan wrote:However, with all due respect to the mathletes who came up with these calculations - although I don't know if their acumen equals that of The Simpsons crew who discovered the Higgs Boson - I think the idea that Draymond Green(12), Kyle Korver(14), Hassan Whitside(18), Markieff Morris(25) and Nene(32) have had a more positive impact on their teams this year than Jimmy Butler is plain and simple rubbish.

Fair enough. I won't argue that. What I take away from that stat is those guys might be more useful than most of us think. Hell, Green beat our ass once this year, Whiteside did in another game, and Nene did during last year's playoffs. Plus we know how Korver can be. So yes while I am not saying they are better than Butler, I can see how they affect games. Morris ... hmmm ... I gotta think more about that.

Ice Man wrote:

JerrySloan wrote:However, with all due respect to the mathletes who came up with these calculations - although I don't know if their acumen equals that of The Simpsons crew who discovered the Higgs Boson - I think the idea that Draymond Green(12), Kyle Korver(14), Hassan Whitside(18), Markieff Morris(25) and Nene(32) have had a more positive impact on their teams this year than Jimmy Butler is plain and simple rubbish.

Fair enough. I won't argue that. What I take away from that stat is those guys might be more useful than most of us think. Hell, Green beat our ass once this year, Whiteside did in another game, and Nene did during last year's playoffs. Plus we know how Korver can be. So yes while I am not saying they are better than Butler, I can see how they affect games. Morris ... hmmm ... I gotta think more about that.

It tells you why Morris plays a lot for Phoenix.

iceman wrote

Fair enough. I won't argue that. What I take away from that stat is those guys might be more useful than most of us think. Hell, Green beat our ass once this year, Whiteside did in another game, and Nene did during last year's playoffs. Plus we know how Korver can be. So yes while I am not saying they are better than Butler, I can see how they affect games. Morris ... hmmm ... I gotta think more about that

.

It's one thing to say that you can see how they affect games which I can't argue with.

But do you actually accept that their respective rankings - 12th, 14th, 18th, 25th and 32nd in the entire NBA - are valid?

Dr Spaceman wrote:

JerrySloan wrote:The list of this stat on ESPN's NBA page has Butler as the best Bulls player at 41st which should make this a total B.S. stat.

Mirotic is the second listed Bull at 64th and Rose clocks in at 145th.

WTF?

So I'll say this:

Stats in the (+/-) family are not meant to be taken as "player x is better than player y". At the heart, these statistics are meant to measure "when player x steps on the court, his team improves by x amount". This is an absolutely crucial distinction to make, because so often people look at stats like this and say "No way are Manu Ginobili and Amir Johnson better than Kevin Durant, this is B.S." So this isn't a stat you can just look at and stop thinking, in fact these stats are designed to inform your thinking. People who get really into plus minus stats (like myself) do so because the numbers that are spit out match pretty much exactly what we see on film.

So what is this measuring? Basically, the idea is that every player in the NBA plays in a multitude of lineups every season and that using all of this lineup data, you can measure the precise impact of each individual player on the scoring margin.. Think of it like an algebra problem, where

scoring margin= Bulls1+Bulls2+Bulls3+Bulls4+Bulls5-Jazz1-Jazz2-Jazz3-Jazz4-Jazz5

If you do this over thousands of lineups and thousands of opponent lineups, you can start to get a good picture of each of the individual variables in this equation. (Note here that Bulls 1,2,3 would be Jordan, Pippen, Rodman, etc.)

RPM is the latest version of this type of stat, which uses box score data to make the results look more "intuitive". I wholeheartedly disagree with this approach, but since this is the most easily available plus minus stat complaining doesn't do a whole lot of good.

The important thing to remember is that you should take these number for exactly what they are: a given player's impact on the scoring margin. Nothing more, nothing less.So for example, I still think Durant is a better player than Ginobili, because his skill set is superior, he plays the #1 role on offense, and he plays far more minutes than Ginobili. I sort of "filter" RAPM data through my own observations to come up with a clear picture of a given player. So the data are telling me that Ginobili, in his role, is one of the most impactful players in the league. But when you consider how limited his role actually is, I'm not likely to rank him very highly overall.

Regarding your questions about specific Bulls players, all I'm going to say is that years and years of data agree that players who play the way Rose has this season do not help their teams win. And players like Mirotic who shoot well and space the floor are far more impactful than you'd think (look at nba.com's lineup data, Mirotic is a part of basically all of the Bulls' best offensive lineups).

This is an amazing description of this stat/stats in general. I love the way you describe why Ginobili rats out higher than Durant.

Good work.

Are you studying/do you have a job doing something statistics related?

JerrySloan wrote:But do you actually accept that their respective rankings - 12th, 14th, 18th, 25th and 32nd in the entire NBA - are valid?

Depends what you mean by valid. When these guys have been on the court, their teams have done unusually well. That is a valid statement. However, nobody would pick Kyle Korver 12th in a league-wide draft, in forming actual teams trying to win actual game. Nor any of those other guys according to their placements.

What that tells me is that some of their high placement is situational, and would not be repeated on other teams at other times. Some is luck. Some is the guy playing over his head. And some, perhaps, is that the player is underrated, and the RAPM statistic captures the underrating effect.

Invalid? I dunno. It's a stat. Steve Kerr has the highest winning percentage of any active NBA coach. That is a valid statistic. If I try to make too much it, though, then I start talking nonsense. Which is what I would be doing if I took RAPM as God's answer and pounded the table that Kyle Korver is the 12th best player in the NBA.