ESPN adds new overall stat...REAL PLUS/MINUS..Finally!

[ceiling raiser]

im familiar (im an actuary).

i still dont see anything as to how it negates the other 9 players on the floor. It moves the needle a bit but not nearly enough.

Then you simply don't understand the stat. And let's be honest, you are just saying that to back up your original assertion, you don't want to hurt your pride of admitting you're wrong. Be better than that. No one who understands the stat would agree with that assessment.

Or else you can write a detailed paper explaining how the math is wrong and that all NBA teams who spend hundreds of thousands of dollars a year these past 10 years hiring mathematicians to calculate this stat for them are wrong and you are right. Even though you have at most spent 2 minutes looking at it.

Good Luck.

clearly i dont, and it has nothing to do with pride. i understand the math, i dont understand how it removes the impact of the other 9 players on the floor. im not saying its a useless stat that doesnt paint a good picture of a players value... i just dont see how its a defensive stat that removes the impact of your 4 teammates and the 5 opponents.

here is what i find flawed:

It is in this regression where the effects of the other players on the floor are accounted for. The bs in equation (1) measure the point differential difference (measured per 100 possessions) of the given player relative to the reference players, holding constant all of the players that shared the floor with that player (and with the reference players), i.e. holding the other players constant. What does this “holding other players constant” mean? Strictly speaking, it means that we can take a player and surround him with four teammates and five opponents and compare how that player’s team would do versus how it would do if he was replaced by a replacement player keeping all of the other players the same. This is what is meant by “holding the other players constant,” since we can repeat this exercise with any other combination of other players.

you are taking something that is inherently variable, and making it constant. you are also ignoring situation circumstances, which is fine when evaluating team defense, but it becomes immensely important for evaluating individual defense. it is operating under the assumption that all things are constant outside of the player in question, at that the level all 9 other players play at is the same regardless of the player who is swapped out -- and that any change would be as a result of the player who was replaced -- which simply can not be assumed.

again, im not saying its a garbage stat that cant help indentify a player worth. but i dont see anything introduced that takes plus/minus down to an accurate individual metric

That is exactly what RAPM does NOT do. It indeed takes context into account, and finds the best fit based on all of the moving parts involved (which in this case, are all of the players in the dataset). There's a ton of literature out there on the calculation of RAPM and the math in general, read at least some of it before attempting to construct a faulty argument.

[JohnsHopkins]

If your new stat says that Channing Frye is better than Blake Griffin, it might not be a very good stat. And Iggy > Durant. And Nick Collison > Tim Duncan. This thing is

Brandan Wright is better than Griffin on PER. PER is trash, too? Every advanced stat is flawed, but having them helps us understanding the game much more.

I used RAPM this years to see which role players are hugely undervalued. Like with PER, it's flawed on some players, but it's still a good stat.

And Iguodala is not having a bad year, he's been outrageously good defensively.

F*** John Hollinger. He has ruined a once very good Memphis Grizzlies team. Outside of the trade of Rudy Gay, all his moves have caused the grizz to take a huge step back, esp the firing of lionell hollins.

[Prokorov]

Then you simply don't understand the stat. And let's be honest, you are just saying that to back up your original assertion, you don't want to hurt your pride of admitting you're wrong. Be better than that. No one who understands the stat would agree with that assessment.

Or else you can write a detailed paper explaining how the math is wrong and that all NBA teams who spend hundreds of thousands of dollars a year these past 10 years hiring mathematicians to calculate this stat for them are wrong and you are right. Even though you have at most spent 2 minutes looking at it.

Good Luck.

clearly i dont, and it has nothing to do with pride. i understand the math, i dont understand how it removes the impact of the other 9 players on the floor. im not saying its a useless stat that doesnt paint a good picture of a players value... i just dont see how its a defensive stat that removes the impact of your 4 teammates and the 5 opponents.

here is what i find flawed:

It is in this regression where the effects of the other players on the floor are accounted for. The bs in equation (1) measure the point differential difference (measured per 100 possessions) of the given player relative to the reference players, holding constant all of the players that shared the floor with that player (and with the reference players), i.e. holding the other players constant. What does this “holding other players constant” mean? Strictly speaking, it means that we can take a player and surround him with four teammates and five opponents and compare how that player’s team would do versus how it would do if he was replaced by a replacement player keeping all of the other players the same. This is what is meant by “holding the other players constant,” since we can repeat this exercise with any other combination of other players.

you are taking something that is inherently variable, and making it constant. you are also ignoring situation circumstances, which is fine when evaluating team defense, but it becomes immensely important for evaluating individual defense. it is operating under the assumption that all things are constant outside of the player in question, at that the level all 9 other players play at is the same regardless of the player who is swapped out -- and that any change would be as a result of the player who was replaced -- which simply can not be assumed.

again, im not saying its a garbage stat that cant help indentify a player worth. but i dont see anything introduced that takes plus/minus down to an accurate individual metric

That is exactly what RAPM does NOT do. It indeed takes context into account, and finds the best fit based on all of the moving parts involved (which in this case, are all of the players in the dataset). There's a ton of literature out there on the calculation of RAPM and the math in general, read at least some of it before attempting to construct a faulty argument.

i read the information you gave me. if there is another explanation or if i got the wrong thing, then please give me another source

[ceiling raiser]

clearly i dont, and it has nothing to do with pride. i understand the math, i dont understand how it removes the impact of the other 9 players on the floor. im not saying its a useless stat that doesnt paint a good picture of a players value... i just dont see how its a defensive stat that removes the impact of your 4 teammates and the 5 opponents.

here is what i find flawed:



you are taking something that is inherently variable, and making it constant. you are also ignoring situation circumstances, which is fine when evaluating team defense, but it becomes immensely important for evaluating individual defense. it is operating under the assumption that all things are constant outside of the player in question, at that the level all 9 other players play at is the same regardless of the player who is swapped out -- and that any change would be as a result of the player who was replaced -- which simply can not be assumed.

again, im not saying its a garbage stat that cant help indentify a player worth. but i dont see anything introduced that takes plus/minus down to an accurate individual metric

That is exactly what RAPM does NOT do. It indeed takes context into account, and finds the best fit based on all of the moving parts involved (which in this case, are all of the players in the dataset). There's a ton of literature out there on the calculation of RAPM and the math in general, read at least some of it before attempting to construct a faulty argument.

i read the information you gave me. if there is another explanation or if i got the wrong thing, then please give me another source

I didn't give you information, I just jumped in after being shocked at your response. The link you were supplied indicates that this is a regression of the coefficients that give the best fit. You're not solving for one player in said regression, but finding the coefficients for ALL players to get the best fit.

If you want help in breaking down how this type of regression works (including what is being solved for and how exactly it's being done), you just have to ask, and somebody here will be able to point you in the right direction.

Once you understand that, check out the APBRmetrics board:

http://www.apbr.org/metrics/viewforum.php?f=2

there are tons of posts on APM, RAPM, xRAPM (stat modified to produce RPM), and their development.

If you want to calculate it on your own, here are some links that mystic sent me (to the data and instructions you'll need) that I've used to play around with the calculations:

http://basketballvalue.com/downloads.php
http://www.countthebasket.com/blog/2008 ... lus-minus/
http://www.countthebasket.com/blog/2008 ... lus-minus/
http://cran.r-project.org/web/packages/ ... index.html

You'll just need excel/R to calculate said metrics (it's a bit involved, you said you were an actuary which uses a bit of math, but I'm not sure what your mathematical programming background is; regardless, it's somewhat straightforward).

[Johnlac1]

I see Ricky Rubio is no. 12. That's pretty damned high. I wonder what all the Rubio haters will say now.

[Prokorov]

That is exactly what RAPM does NOT do. It indeed takes context into account, and finds the best fit based on all of the moving parts involved (which in this case, are all of the players in the dataset). There's a ton of literature out there on the calculation of RAPM and the math in general, read at least some of it before attempting to construct a faulty argument.

i read the information you gave me. if there is another explanation or if i got the wrong thing, then please give me another source

I didn't give you information, I just jumped in after being shocked at your response. The link you were supplied indicates that this is a regression of the coefficients that give the best fit. You're not solving for one player in said regression, but finding the coefficients for ALL players to get the best fit.

If you want help in breaking down how this type of regression works (including what is being solved for and how exactly it's being done), you just have to ask, and somebody here will be able to point you in the right direction.

Once you understand that, check out the APBRmetrics board:

http://www.apbr.org/metrics/viewforum.php?f=2

there are tons of posts on APM, RAPM, xRAPM (stat modified to produce RPM), and their development.

If you want to calculate it on your own, here are some links that mystic sent me (to the data and instructions you'll need) that I've used to play around with the calculations:

http://basketballvalue.com/downloads.php
http://www.countthebasket.com/blog/2008 ... lus-minus/
http://www.countthebasket.com/blog/2008 ... lus-minus/
http://cran.r-project.org/web/packages/ ... index.html

You'll just need excel/R to calculate said metrics (it's a bit involved, you said you were an actuary which uses a bit of math, but I'm not sure what your mathematical programming background is; regardless, it's somewhat straightforward).

thanks, thats helpful... im not great with excel or macros but there are plenty of people here who are so i think ill be ok

[dlts20]

there is a very good 2 man debate going on in this thread and I am honored to be apart of your listening audience........lol

[mysticbb]

ESPN has finally gotten on board with the statistical revolution and added their version of adjusted plus/minus.

It is actually J.E.'s version, and at that xRAPM. You can find the numbers for the other seasons here: https://stats-for-the-nba.appspot.com/ (only 2001 to 2014 are based on real play-by-play numbers, the values before are based on simulated play-by-play data and shouldn't be taken that seriously)

So, it is a RAPM using a boxscore prior and some of the other stuff presented on the website (e.g. aging curve, coaching adjustment, also some adjustment based on the respective score at the time, etc. pp.). It is an improvement in terms of anything else we see in public available stats, but we should keep in mind that a different prior can mean a lot here. To give an impression about that: http://talkingpracticeblog.com/
Those guys are also using a ridge-regression and a boxscore prior with several adjustments and are coming up with a different result (although they present only a few players here). My own version gives actually results which are located somewhat between xRAPM and IPV, but I'm using a linear regression approach to merge RAPM and SPM values. (All three versions have also in common that they assume a baysian distribution, which is really important, but probably only worth noting for guys having a clue what that means anyway ...).

Really important note: A better RPM value (or xRAPM) does not mean a player is per se better than another player with a lower value. Such values are also dependent on the role a player has to fulfill within a team, where the coaches may choose to play him more often in situations where he can succeed. Even an coaching adjustment may only cover a part of that. Also, a player being able to play more minutes against any kind of opponents provides additional value. Thus, a player playing 36 mpg with a lower RPM may very well be considered overall better than a player only giving a team 20 mpg.
So, I hope people are not just trying to dismiss the results based on perceived outliers in the ranking. The metric itself proved already to be capable of predicting out of sample data better than any other public avaiable stats like PER or Win Shares or whatever.

[ceiling raiser]

i read the information you gave me. if there is another explanation or if i got the wrong thing, then please give me another source

I didn't give you information, I just jumped in after being shocked at your response. The link you were supplied indicates that this is a regression of the coefficients that give the best fit. You're not solving for one player in said regression, but finding the coefficients for ALL players to get the best fit.

If you want help in breaking down how this type of regression works (including what is being solved for and how exactly it's being done), you just have to ask, and somebody here will be able to point you in the right direction.

Once you understand that, check out the APBRmetrics board:

http://www.apbr.org/metrics/viewforum.php?f=2

there are tons of posts on APM, RAPM, xRAPM (stat modified to produce RPM), and their development.

If you want to calculate it on your own, here are some links that mystic sent me (to the data and instructions you'll need) that I've used to play around with the calculations:

http://basketballvalue.com/downloads.php
http://www.countthebasket.com/blog/2008 ... lus-minus/
http://www.countthebasket.com/blog/2008 ... lus-minus/
http://cran.r-project.org/web/packages/ ... index.html

You'll just need excel/R to calculate said metrics (it's a bit involved, you said you were an actuary which uses a bit of math, but I'm not sure what your mathematical programming background is; regardless, it's somewhat straightforward).

thanks, thats helpful... im not great with excel or macros but there are plenty of people here who are so i think ill be ok

No problem. I see mystic's chimed in below (er, above), so if you have any questions he might be able to help you better than I can.

[Foliohattu]

RK NAME TEAM GP MPG ORPM

435 Kendrick Perkins, C OKC 56 19.5 -6.19

Seems to be working.

[NyCeEvO]

Really important note: A better RPM value (or xRAPM) does not mean a player is per se better than another player with a lower value. Such values are also dependent on the role a player has to fulfill within a team, where the coaches may choose to play him more often in situations where he can succeed. Even an coaching adjustment may only cover a part of that. Also, a player being able to play more minutes against any kind of opponents provides additional value. Thus, a player playing 36 mpg with a lower RPM may very well be considered overall better than a player only giving a team 20 mpg.
So, I hope people are not just trying to dismiss the results based on perceived outliers in the ranking. The metric itself proved already to be capable of predicting out of sample data better than any other public avaiable stats like PER or Win Shares or whatever.

Thank you for writing this.

While there are many people who blindly use advanced stats to "prove" their case, there are just as many if not more who will look at an advanced stat and throw it away, because they see a ranking where Andre Iguodala is higher than KD and they think "Oh this stat suggests that AI is better than KD...what a useless stat."

That's not what the stat is saying at all.

The stat is attempting to give a numerical representation of the impact that a player has on offense/defense given the player's context (i.e. minutes played, position, etc.) as you said.

If you change the context for a player who thrives playing 20mpg as a #1 or #2 option off the bench to a #1 option and star for the entire team, there's a very high probability their RAPM numbers would drop drastically.

[Winsome Gerbil]

Nick Collison = biggest All Star snub this year.

Seriously, if I invented a stat that told me that Matt Barnes was better than Melo, that James Harden was the 2nd most important guard his own team, that Marcin Gortat was worth more than Brow and Cousins combined, and that Channing Frye was amongst the league's elite...I'd be too embarrassed to put it up.

[dice]

milwaukee bucks individual WAR sums up to around 1 or 2 - plausible only if you think a D league all-star team could win double digit games in the NBA

chicago bulls individual WAR sums to around 31 whereas they have won 45 games

d.j. augustin WAR rank 425 out of 435, indicating he has been one of the most damaging players in the league to his team, this despite the reality that he has been a savior for the chicago bulls

i don't question the math here. i question the application of the math in measuring a basketball player's net impact on his team. at the very least there are a number of substantial outliers

[WhateverBro]

Nick Collison = biggest All Star snub this year.

Seriously, if I invented a stat that told me that Matt Barnes was better than Melo, that James Harden was the 2nd most important guard his own team, that Marcin Gortat was worth more than Brow and Cousins combined, and that Channing Frye was amongst the league's elite...I'd be too embarrassed to put it up.

That's not what the stat is saying. Read MysticBBs post in this thread, that will probably make things clearer.

I'm very happy that ESPN are pushing this stat, I hope they will use it when they're doing games too so that people will be more familiar with advanced stats.

[Effigy]

2 TWolves in the top 12? Only one Spur (at 11)? Sure, this stat seems really accurate.

[crazy_me_87]

could someone sum it up real simple..??? i am a little confused how it works

anthony Davis only 83 best offensive rating?... Nene a DPOY canidate?

i´m totaly new to this stat could someone enlighten me,thanks!

[Pan Jia Yuan]

Nick Collison = biggest All Star snub this year.

Seriously, if I invented a stat that told me that Matt Barnes was better than Melo, that James Harden was the 2nd most important guard his own team, that Marcin Gortat was worth more than Brow and Cousins combined, and that Channing Frye was amongst the league's elite...I'd be too embarrassed to put it up.

It has nothing to do with inventing stats. Anthony Davis is the perfect example. Everyone knows that he's having a great individual season, PPG, RPG, BPG, efficiency, PER, ORtg, Win Shares, everything's good. But it's also true that for the team's success it doesn't really matter whether he's on the court or on the couch. You don't need a formula for that, it has nothing to do with "inventing a stat", it's simply counting the points. Offensively, they're slightly better with him, defensively there's basically no difference with or without him.
It's a phenomenon, isn't it? I for one think it'd be much more interesting to understand it than just dismissing it.

[Winsome Gerbil]

Nick Collison = biggest All Star snub this year.

Seriously, if I invented a stat that told me that Matt Barnes was better than Melo, that James Harden was the 2nd most important guard his own team, that Marcin Gortat was worth more than Brow and Cousins combined, and that Channing Frye was amongst the league's elite...I'd be too embarrassed to put it up.

That's not what the stat is saying. Read MysticBBs post in this thread, that will probably make things clearer.

I'm very happy that ESPN are pushing this stat, I hope they will use it when they're doing games too so that people will be more familiar with advanced stats.

I'm hoping they don't, because a lot more would be learned if those people spent less time punching a calculator and more time watching the actual game.

[dice]

Only one Spur (at 11)? Sure, this stat seems really accurate.

WAR for the spurs adds up to 51. 49 for OKC, 43 for miami. pretty reasonable

the spurs are a balanced TEAM

[WhateverBro]

I'm hoping they don't, because a lot more would be learned if those people spent less time punching a calculator and more time watching the actual game.

I'm fairly certain that these people actually watches ALOT of basketball.

There's no need to dismiss the stat just because you don't agree with a couple of cherry picked examples.