Some thoughts and questions on BPM for Dsmok1:
1. Have you looked into adding Games Played (% of teams games played to account for lockouts and the 2015 season) into the formula? How about Games started? Games started will tell you who coaches think are the best players plus its a good stat to use an opponent adjustment.
2. Guys like Steve Nash don't get much credit from BPM for their playmaking. Maybe you should add a variable that uses assist% with one or more of the following: Teammate efficiency, Teammate 3point rate, Teammate lack of turnovers, FTR to account for penetrators. Steve Nash increased his teammates efficiency but BPM actually punishes him for that. If his teammates shot lower %, Nash gets more credit from the TS%-team TS% variable even though he played worse. Obviously the TS%-team TS% variable is good for recognizing low TS% guys like McGrady and Iverson as good offensive players but it punishes PG's.
The debut and popularization of BPM
Re: The popularization of BPM
Following logic of first suggestion, could also consider % of last 5 minutes of game.
Concern BPM doesn't give enough credit for assists? On surface I had opposite concern but haven't looked deeply at it.
Concern BPM doesn't give enough credit for assists? On surface I had opposite concern but haven't looked deeply at it.
-
permaximum
- Posts: 416
- Joined: Tue Nov 27, 2012 7:04 pm
Re: The popularization of BPM
Low TS% guys like McGrady and Iverson? According to what? I calculated TS% in Iverson years including it's last two strongly declining years and it's 52.6%. Iverson's TS% is 51.8%. League average USG is around 18%. Iverson's USG is a lot more. League average TS% is a little bit higher in McGrady years but he can't be classified as low TS% guy either. His USG is a lot higher than league average too.colts18 wrote: Obviously the TS%-team TS% variable is good for recognizing low TS% guys like McGrady and Iverson as good offensive players but it punishes PG's.
Re: The popularization of BPM
This is an area that we hoped plus-minus analysis would give insight. Did his teammates' shooting improve when he joined them? Does his backup have the same effect upon the same teammates? Such things aren't revealed by any boxscore analysis.colts18 wrote:2. Guys like Steve Nash don't get much credit from BPM for their playmaking. Maybe you should add a variable that uses assist% with one or more of the following: Teammate efficiency, Teammate 3point rate, Teammate lack of turnovers, FTR to account for penetrators. Steve Nash increased his teammates efficiency but BPM actually punishes him for that. ..
Ast% by itself really isn't that great a tool. Assisting on 50% of teammate scoring in a great offense, and doing the same in a very weak offense, aren't equal accomplishments.
Of course, Ast% * TmEfficiency is pretty much equivalent to Ast/100 poss.
http://bkref.com/tiny/D60gk
That is Steve Nash's first MVP season, 2005. That year, he ranked 15th in WS, 9th in WS/48, 18th in PER, 48th in VORP, 57th in BPM -- below luminaries like Bobby Simmons, Brian Cardinal, Greg Buckner, Damon Jones ...
In 2006, he was 16th in PER, 10th in WS/48 and WS, 22nd in VORP and BPM -- just behind Battier.
Apparently the BPM term (Usg*Ast) doesn't help much, and neither does (Reb*Ast).
-
permaximum
- Posts: 416
- Joined: Tue Nov 27, 2012 7:04 pm
Re: The popularization of BPM
To support Mike on Steve Nash case;
He's ranked 4th in the single-year NPI-RAPM (vanilla RAPM) of 2005. The following season he's ranked 14th in JE's single-year NPI-RAPM and 41st in colts18's single-year NPI-RAPM. There's a strong decline but box-score metrics don't really reflect that. BTW, I don't know why J.E's and colts18's NPI-RAPM rankings are this different.
He's ranked 4th in the single-year NPI-RAPM (vanilla RAPM) of 2005. The following season he's ranked 14th in JE's single-year NPI-RAPM and 41st in colts18's single-year NPI-RAPM. There's a strong decline but box-score metrics don't really reflect that. BTW, I don't know why J.E's and colts18's NPI-RAPM rankings are this different.
Re: The popularization of BPM
Yes, we've talked about the Nash issue before with this model, over at Tom Tango's blog: http://tangotiger.com/index.php/site/co ... nus-in-nba
Basically: he is another extreme outlier. Generally, players with very few rebounds or steals or blocks are not very good, as those are the prime indicators of physical ability to play in the NBA. Assists and true shooting percentage indicate skills and awareness. The BPM regression showed quite strongly one must have both the physical ability and the skills in order to be valuable. There seem to be a few outlying cases, particularly Nash, where pure skill on the offensive end can outweigh the other factors.
Basically: he is another extreme outlier. Generally, players with very few rebounds or steals or blocks are not very good, as those are the prime indicators of physical ability to play in the NBA. Assists and true shooting percentage indicate skills and awareness. The BPM regression showed quite strongly one must have both the physical ability and the skills in order to be valuable. There seem to be a few outlying cases, particularly Nash, where pure skill on the offensive end can outweigh the other factors.
Re: The popularization of BPM
DSMok1, obviously it's water under the bridge now -- but does it make sense, in future perhaps, to discuss basketball stats at the premier basketball stats analysis website, rather than at a baseball blog?
I still wonder if the notion of minimum wage players winning 20% of their games is perhaps tied to a baseball environment, with the equivalent of `27 possessions per team per game; and that you might find that to be the case after about 15 minutes (or 27 poss) of a basketball game.
I still wonder if the notion of minimum wage players winning 20% of their games is perhaps tied to a baseball environment, with the equivalent of `27 possessions per team per game; and that you might find that to be the case after about 15 minutes (or 27 poss) of a basketball game.
Re: The popularization of BPM
Mike G wrote:DSMok1, obviously it's water under the bridge now -- but does it make sense, in future perhaps, to discuss basketball stats at the premier basketball stats analysis website, rather than at a baseball blog?
I still wonder if the notion of minimum wage players winning 20% of their games is perhaps tied to a baseball environment, with the equivalent of `27 possessions per team per game; and that you might find that to be the case after about 15 minutes (or 27 poss) of a basketball game.
Tom Tango's blog is where many of the best sports statistics minds post, across many disciplines/sports. Many discussions that are not restricted to baseball, focused on the larger statistics and modeling questions. It's where I learned sports statistics, largely...
-
AcrossTheCourt
- Posts: 237
- Joined: Sat Feb 16, 2013 11:56 am
Re: The popularization of BPM
That's odd because by "pure" RAPM Stockton rated really well defensively in the late 90's/early 00's, I believe, and of course he has a great reputation for defense.colts18 wrote:Am I reading the stat correctly?
AST% is a massive negative according to the formula. Thats probably why it seems like the stat is not high on John Stockton. Its interesting that Stockton never had a positive in defensive BPM in his career despite having a ton of steals (which the stat loves) and playing on some really good defensive teams.
You would think AST% would be a positive because PG's rate pretty well in Offensive RAPM.
Also Personal fouls aren't in the stat even though they are a box score stat.
I think the versatility term (the assist/rebound thing) really hurts some point guards -- the ones who don't rebound. A lot of point guards derive their value from distributing, and if that term is attached to rebounding it hurts them more than others.
Re: The popularization of BPM
Neil, I'm sorry to say this but you're doing this community a giant disservice by acting like your way of testing metrics - computing error on the same data that was actually used to fit these models - is a valid test.
Everyone, for the love of god, please get a better sense of what constitues in-sample and out-of-sample data and why it's important to not mix these up. Maybe take part in a kaggle.com competition or two to get a sense of what out-of-sample really means, and how these kind of competitions are set up in a correct way
Everyone, for the love of god, please get a better sense of what constitues in-sample and out-of-sample data and why it's important to not mix these up. Maybe take part in a kaggle.com competition or two to get a sense of what out-of-sample really means, and how these kind of competitions are set up in a correct way
Re: The popularization of BPM
Isn't that only true for BPM values going from 2001 to 2013? Meaning, any BPM values derived from 2000 and before and then subsequentially tested for the performance in the +1, +2, etc. season don't fit your critique, right?J.E. wrote:Neil, I'm sorry to say this but you're doing this community a giant disservice by acting like your way of testing metrics - computing error on the same data that was actually used to fit these models - is a valid test.
In that sense, it is not a question of the method used by Neil, but rather that this method is not applicable for all metrics in all years. Other than that I agree, that testing the 2001 to 2013 BPM values in that fashion is not a out-of-sample test. So, not including such years in the test would be the appropiate way.
Re: The popularization of BPM
I think I found a somewhat good solution for that issue, just that the method itself needs to be tested out of sample, before really declaring that as a "solution". Right now the numbers I get for players with a certain numbers of assists looks like this (an improved version of my previously posted numbers in this thread):AcrossTheCourt wrote:I think the versatility term (the assist/rebound thing) really hurts some point guards -- the ones who don't rebound. A lot of point guards derive their value from distributing, and if that term is attached to rebounding it hurts them more than others.
Code: Select all
Player OV DV OV+DV O/U
LeBron James 7.21 2.09 9.30 0.46
Manu Ginobili 4.93 1.25 6.18 0.11
Chris Paul 5.66 0.40 6.06 -0.12
Jason Kidd 2.96 2.73 5.69 0.94
Dwyane Wade 5.21 0.48 5.69 0.37
John Stockton 4.12 0.90 5.02 0.78
Stephen Curry 4.56 0.37 4.93 0.95
Russell Westbrook 3.00 1.19 4.19 0.26
Deron Williams 5.03 -1.06 3.97 0.95
Baron Davis 3.39 0.25 3.64 -0.43
Steve Nash 5.98 -2.52 3.46 -0.68
Ricky Rubio 0.96 2.17 3.13 -0.18
Chauncey Billups 4.11 -1.20 2.90 0.30
Mike Conley 1.98 0.80 2.79 -0.28
Andre Miller 2.04 0.75 2.78 0.47
Devin Harris 2.86 -0.24 2.62 0.54
Kyle Lowry 2.06 0.42 2.48 -0.12
Isaiah Thomas 3.75 -1.42 2.33 0.28
Sam Cassell 2.18 0.14 2.32 0.20
Goran Dragic 2.90 -0.73 2.16 0.32
Ty Lawson 3.80 -1.80 2.00 -0.42
Derrick Rose 2.57 -0.57 2.00 0.50
Damian Lillard 3.20 -1.23 1.97 -0.04
Pablo Prigioni 1.57 0.39 1.96 0.17
Gilbert Arenas 2.70 -1.06 1.64 0.07
Steve Francis 2.00 -0.45 1.56 0.18
Allen Iverson 2.30 -1.25 1.05 -0.06
Jrue Holiday 1.15 -0.17 0.98 -0.27
Jeremy Lin 0.71 0.24 0.95 0.61
Jameer Nelson 1.26 -0.32 0.94 -0.07
Antonio Daniels 2.58 -1.66 0.92 0.35
Brandon Jennings 1.64 -0.74 0.90 1.09
Eric Bledsoe -1.67 2.43 0.77 -0.36
Stephon Marbury 2.55 -1.89 0.66 0.88
Eric Snow -1.23 1.58 0.36 -0.10
John Wall 1.10 -0.79 0.31 0.03
T.J. Ford 0.22 0.07 0.28 -0.05
Darrell Armstrong -0.33 0.59 0.26 0.25
Scottie Pippen 0.38 -0.19 0.19 0.48
Tony Parker 1.60 -1.49 0.11 -0.55
Speedy Claxton -1.07 1.16 0.09 0.54
Raymond Felton 0.82 -0.85 -0.03 -0.25
Terry Porter 0.17 -0.24 -0.07 -1.06
Mike James 1.30 -1.40 -0.10 -0.14
Jose Calderon 2.37 -2.55 -0.18 0.25
Mo Williams 0.92 -1.12 -0.21 0.42
Kemba Walker -0.09 -0.22 -0.31 -0.30
Mike Bibby 0.95 -1.29 -0.34 -0.49
Terrell Brandon -0.48 -0.01 -0.50 0.62
Damon Stoudamire 0.18 -0.71 -0.53 0.22
Beno Udrih 1.32 -1.87 -0.55 -0.36
Matthew Dellavedova 0.62 -1.28 -0.66 -0.62
Jeff Teague 0.60 -1.26 -0.66 0.09
Tim Hardaway 0.52 -1.23 -0.70 -0.39
Gary Payton 0.56 -1.37 -0.82 -0.58
Earl Watson -0.56 -0.34 -0.90 -0.40
Bob Sura -0.45 -0.49 -0.93 -0.16
Greivis Vasquez 1.02 -1.97 -0.95 -0.74
Shaun Livingston -0.81 -0.25 -1.06 -0.14
Charlie Ward -1.64 0.52 -1.11 -0.20
Sergio Rodriguez -0.08 -1.04 -1.12 0.83
Jason Williams 1.06 -2.25 -1.19 0.27
Kyrie Irving 1.10 -2.30 -1.20 0.19
Travis Diener -0.06 -1.29 -1.35 0.45
Trey Burke 0.51 -1.86 -1.35 0.27
Mookie Blaylock -2.50 1.11 -1.39 0.47
Rajon Rondo -0.56 -0.86 -1.42 -0.11
Darrick Martin -0.72 -0.76 -1.47 -0.49
Jarrett Jack 0.36 -1.91 -1.55 0.62
Greg Anthony -0.83 -0.76 -1.58 0.93
Kirk Hinrich -1.22 -0.38 -1.60 -0.75
Anthony Carter -1.65 0.04 -1.61 0.17
Rafer Alston -1.33 -0.39 -1.72 -0.48
Nick Calathes -1.49 -0.35 -1.83 -0.63
Chris Duhon -0.48 -1.36 -1.83 -0.63
Darren Collison 0.36 -2.20 -1.84 -0.21
Mark Jackson -0.65 -1.21 -1.86 0.89
Brian Roberts -0.04 -1.84 -1.88 0.63
Chris Whitney -0.65 -1.25 -1.90 0.41
Jordan Farmar -0.89 -1.07 -1.97 -0.55
Steve Blake -0.49 -1.52 -2.01 -0.42
Jerryd Bayless -0.21 -1.82 -2.04 0.17
Anthony Johnson -1.16 -0.96 -2.12 -0.57
Sundiata Gaines -1.14 -1.06 -2.20 0.07
Shelvin Mack -0.22 -1.99 -2.21 -0.32
Jamaal Tinsley -1.15 -1.09 -2.24 -0.73
Luke Ridnour -0.19 -2.11 -2.30 0.03
Brevin Knight -3.36 1.04 -2.31 -0.05
D.J. Augustin 0.86 -3.21 -2.34 0.15
Moochie Norris -1.31 -1.11 -2.41 -0.05
Jose Barea -0.27 -2.16 -2.43 -0.20
Phil Pressey -1.49 -0.94 -2.43 -0.42
Sarunas Jasikevicius -1.80 -0.65 -2.45 -0.07
Rick Brunson -1.37 -1.17 -2.53 -0.67
Dan Dickau -0.99 -1.59 -2.58 0.38
Carlos Arroyo -1.04 -1.59 -2.63 -0.64
Travis Best -1.33 -1.31 -2.64 -0.36
Howard Eisley -1.87 -0.79 -2.66 0.38
Nick Van Exel -0.38 -2.30 -2.68 0.02
Alvin Williams -0.92 -1.77 -2.70 -0.10
Khalid El-Amin -2.06 -0.69 -2.74 0.28
Ishmael Smith -2.78 0.01 -2.77 -0.48
John Crotty -1.61 -1.20 -2.81 -0.57
Ramon Sessions 0.17 -3.02 -2.85 -0.23
Kenny Anderson -2.68 -0.18 -2.86 -0.95
Eric Maynor -1.76 -1.20 -2.96 -0.63
Rod Strickland -1.78 -1.19 -2.97 -0.21
Sebastian Telfair -1.87 -1.12 -2.99 0.10
Earl Boykins -0.13 -2.90 -3.03 0.23
Roko Ukic -2.37 -0.80 -3.16 -0.47
Charles Jenkins -2.27 -0.90 -3.17 0.28
Raul Lopez -2.08 -1.13 -3.21 -1.16
Chris Childs -2.76 -0.46 -3.22 -0.17
Jeff McInnis 0.89 -4.16 -3.26 -0.04
Jason Hart -2.32 -0.98 -3.30 -0.27
Jacque Vaughn -2.64 -0.74 -3.37 0.27
Jay Williams -2.01 -1.48 -3.49 0.33
Avery Johnson -1.78 -1.71 -3.49 -0.14
Will Bynum -1.62 -1.88 -3.50 0.39
Kendall Marshall -1.75 -1.81 -3.56 0.30
Eugene Jeter -1.58 -2.08 -3.65 -0.49
Robert Pack -2.36 -1.33 -3.69 0.32
Marcus Williams -1.01 -2.80 -3.81 -0.23
Donald Sloan -2.20 -1.80 -4.01 1.47
Frank Williams -3.11 -1.38 -4.48 -0.79
Tierre Brown -2.69 -1.80 -4.49 0.10
Michael Carter-Williams -2.10 -2.51 -4.61 -0.34
Keith McLeod -2.84 -1.81 -4.65 -0.22
Troy Hudson -0.44 -4.23 -4.67 -0.83
Kevin Ollie -2.56 -2.21 -4.77 -0.70
Kenny Satterfield -4.31 -0.90 -5.21 -0.02
Mateen Cleaves -5.37 -0.85 -6.22 -0.96
Jonny Flynn -3.46 -3.99 -7.45 -0.44
Nash still underrated, but mostly due to the defensive aspect.
-
Neil Paine
- Posts: 73
- Joined: Mon Apr 18, 2011 1:18 am
- Location: Philadelphia
- Contact:
Re: The popularization of BPM
Hey Jerry, you know the great thing about historical metrics? We can cut them off at 2000 (or whenever you want) and still have a large sample to test on. No need for the rudeness.J.E. wrote:Neil, I'm sorry to say this but you're doing this community a giant disservice by acting like your way of testing metrics - computing error on the same data that was actually used to fit these models - is a valid test.
Everyone, for the love of god, please get a better sense of what constitues in-sample and out-of-sample data and why it's important to not mix these up. Maybe take part in a kaggle.com competition or two to get a sense of what out-of-sample really means, and how these kind of competitions are set up in a correct way
-
permaximum
- Posts: 416
- Joined: Tue Nov 27, 2012 7:04 pm
Re: The popularization of BPM
I know that someone would say this. So I did not use any 2014-15 data to fit my model. If "someone"'s not satisfied with the results there'll always be 2014-15 season to test. Besides, I tested BPM out-of-sample 2014-15 + 2000-1974. Prediction doesn't really suffer.J.E. wrote:Neil, I'm sorry to say this but you're doing this community a giant disservice by acting like your way of testing metrics - computing error on the same data that was actually used to fit these models - is a valid test.
Everyone, for the love of god, please get a better sense of what constitues in-sample and out-of-sample data and why it's important to not mix these up. Maybe take part in a kaggle.com competition or two to get a sense of what out-of-sample really means, and how these kind of competitions are set up in a correct way
BTW my metric (two metrics in fact, MAX_1 and MAX_5) has been ready since a week ago but I still couldn't find the time to fix b-r linkfy errors for player codenames. Hopefully, it'll be done tomorrow.
Re: The popularization of BPM
We talking about Dave Berri & Wins Produced or BPM?AcrossTheCourt wrote: I think the versatility term (the assist/rebound thing) really hurts some point guards -- the ones who don't rebound. A lot of point guards derive their value from distributing, and if that term is attached to rebounding it hurts them more than others.
Joking btw, I just remember the WP days in which practically all the best rebounders at every position ranked at the top of the overall rankings.