Letting the Lamppost Illuminate

[ElGee]

I can't think of a basketball statistic that is perfect, in a vacuum, and doesn't need interpretation. (Off the top of my head, I'd argue free throw shooting is pretty darn close. After that? Gets really tricky.) So the notion of "throwing out a stat," which I see a lot from people based on outliers or something not adhering to their current picture of the elite, seems bizarre to me.

In another thread, I listed the eFG% -- a measure of shooting accuracy -- for the Lakers. Kobe Bryant was 9th. Does that mean we should throw out eFG% because Bryant is clearly a better shooter than Derrick Caracter?
--
As for Kobe, it's quite easy to understand, in theory, why his APM is so low. Doc MJ alluded to it: APM models will "look" at stuff like Gasol+no Kobe and Kobe+no Gasol to decide teammate quality when they play together. If Gasol+no Kobe comes out way ahead, APM "thinks" Gasol is the superstar and Kobe is benefitting from playing with him.

Then the same logic extends to the quality of the opponent. It wouldn't take much for Kobe to do well against "weaker" opponent lineups and the Lakers to play well when he's on the bench against comparable (or better) lineups. As you can see from lineup data, there are a bunch of small-sampled variables (thus the big error).

Then again, it's *likely* that Bryant isn't one of the best players in the league this year or else he wouldn't be generating *such* a low APM. Which is to say the more obvious basketball observation: lineup data suggests that the LA Lakers are really good without Kobe Bryant, even against good competition.

Couple that with what we saw last year when he was injured, the play of Bynum and Odom, his reduction in minutes, and that his 6-year APM from Ilardi was 18th in the league anyway (+5.6) and it's not a very radical stance to say Kobe isn't as valuable (a) as in year's past and (b) relative to the top MVP candidates.

[mysticbb]

Adding Ridge Regression can cut out some of the tremendous collinearity, but at the expense of losing some of the validity of the stat.

I agree with you, which is also the reason I prefer those values: http://stats-for-the-nba.appspot.com/4-year-ranking

3.x year ranking From 07/08 to now. No coaches included. Lambda: 3000. 3.x year analysis is better at predicting 10/11-out-of-sample-data than any other combination of years.

Checking the results of the lineups gives us the most accurate rating here. We can now add boxscore metrics either in a similar fashion as Rosenbaum did it, or we can use some kind of a formula to do that afterwards.

1yr data with PER and WS/48 incorparated after:

Code: Select all

Nowitzki            Dirk          7.6
James               LeBron        7.6
Howard              Dwight        7.2
Paul                Chris         7.2
Gasol               Pau           6.3
Ginobili            Manu          6.1
Wade                Dwyane        6.0
Hilario             Nene          5.5
Garnett             Kevin         5.4
Pierce              Paul          5.4
Rose                Derrick       5.3
Durant              Kevin         5.1
Chandler            Tyson         5.0
Bosh                Chris         4.9
Bynum               Andrew        4.9
Nash                Steve         4.7
Bryant              Kobe          4.6
Aldridge            LaMarcus      4.6
Love                Kevin         4.4
Duncan              Tim           4.3
Anderson            Ryan          4.3
Odom                Lamar         4.1
Horford             Al            3.9
Randolph            Zach          3.9
Millsap             Paul          3.6
Martin              Kevin         3.3
Westbrook           Russell       3.2
Noah                Joakim        3.1
Griffin             Blake         3.1
O'Neal              Shaquille     3.0
Parker              Tony          3.0
West                David         2.9
Johnson             Amir          2.9
Okafor              Emeka         2.8
Miller              Andre         2.7
Lowry               Kyle          2.7
Young               Thaddeus      2.7
Boozer              Carlos        2.7
Deng                Luol          2.6
Evans               Jeremy        2.6
Allen               Ray           2.5
Stoudemire          Amare         2.5
Gordon              Eric          2.5
Williams            Deron         2.4
Ibaka               Serge         2.4
Andersen            Chris         2.4
Anthony             Carmelo       2.3
Bass                Brandon       2.3
Brand               Elton         2.2
Curry               Stephen       2.1
Iguodala            Andre         2.1
Wallace             Gerald        2.1
Smith               Josh          2.0
Gortat              Marcin        2.0
Billups             Chauncey      2.0
Conley              Mike          1.9
Lawson              Ty            1.9
Mahinmi             Ian           1.9
Allen               Tony          1.9
Scola               Luis          1.9
Rondo               Rajon         1.8
Foster              Jeff          1.8
Hill                George        1.8
Smith               J.R.          1.8
Harden              James         1.8
Hayes               Chuck         1.8
Terry               Jason         1.7
Williams            Louis         1.6
Gay                 Rudy          1.6
Asik                Omer          1.5
McRoberts           Josh          1.5
Bogut               Andrew        1.5
Dunleavy            Mike          1.4
Korver              Kyle          1.4
Bonner              Matt          1.4
Nelson              Jameer        1.3
Stojakovic          Peja          1.3
Granger             Danny         1.2
Richardson          Jason         1.2
Thomas              Tyrus         1.2
Brewer              Ronnie        1.2
Blair               DeJuan        1.2
Wilcox              Chris         1.2
Miller              Brad          1.2
Gallinari           Danilo        1.1
Jefferson           Al            1.1
Gasol               Marc          1.1
Collison            Nick          1.1
Turkoglu            Hedo          1.1
Meeks               Jodie         1.0
McGee               JaVale        1.0
Humphries           Kris          1.0
Boykins             Earl          1.0
Kirilenko           Andrei        1.0
Chandler            Wilson        0.9
Jones               James         0.9
Powe                Leon          0.9
Barnes              Matt          0.9
Holiday             Jrue          0.9
Gibson              Taj           0.9
Patterson           Patrick       0.9
Lopez               Brook         0.9
Arthur              Darrell       0.9
White               D.J.          0.8
Johnson             Joe           0.8
Monroe              Greg          0.8
Speights            Marreese      0.8
Fields              Landry        0.8
Stuckey             Rodney        0.8
Smith               Craig         0.7
Hansbrough          Tyler         0.7
Tolliver            Anthony       0.6
Carter              Vince         0.6
Fernandez           Rudy          0.6
Davis               Baron         0.6
Felton              Raymond       0.6
Martin              Kenyon        0.6
Ilyasova            Ersan         0.6
Jordan              DeAndre       0.5
Afflalo             Arron         0.5
Thornton            Marcus        0.5
Harris              Devin         0.5
Booker              Trevor        0.5
Gray                Aaron         0.4
Miles               C.J.          0.4
Ilgauskas           Zydrunas      0.4
Budinger            Chase         0.4
Radmanovic          Vladimir      0.4
McDyess             Antonio       0.3
Dudley              Jared         0.3
Williams            Reggie        0.3
Matthews            Wes           0.3
Wright              Dorell        0.3
Brockman            Jon           0.3
Kidd                Jason         0.2
Lee                 David         0.2
Haslem              Udonis        0.2
Turiaf              Ronny         0.2
Battier             Shane         0.2
Roy                 Brandon       0.1
Camby               Marcus        0.1
Splitter            Tiago         0.1
Dooling             Keyon         0.1
Williams            Marvin        0.1
Mohammed            Nazr          0.0
Warrick             Hakim         0.0
Butler              Caron         0.0
Ridnour             Luke          0.0
Calderon            Jose          0.0
Garcia              Francisco     0.0
Batum               Nicolas       0.0
Udrih               Beno         -0.1
Hibbert             Roy          -0.1
Redick              J.J.         -0.1
Mbah a Moute        Luc          -0.1
Watson              C.J.         -0.1
Brown               Shannon      -0.1
Jefferson           Richard      -0.1
Varejao             Anderson     -0.2
Artest              Ron          -0.2
Douglas             Toney        -0.2
Krstic              Nenad        -0.2
Frye                Channing     -0.2
Jones               Dahntay      -0.3
Hill                Grant        -0.3
Cook                Daequan      -0.3
Dorsey              Joey         -0.3
Augustin            D.J.         -0.4
Marion              Shawn        -0.4
Uzoh                Ben          -0.4
George              Paul         -0.4
Martin              Cartier      -0.4
McGrady             Tracy        -0.4
Barea               Jose         -0.4
Ellis               Monta        -0.4
Maggette            Corey        -0.5
Davis               Glen         -0.5
Gooden              Drew         -0.5
Prince              Tayshaun     -0.5
Teague              Jeff         -0.5
Neal                Gary         -0.5
Bibby               Mike         -0.5
Wright              Julian       -0.5
Dalembert           Samuel       -0.5
Landry              Carl         -0.5
Crawford            Jamal        -0.6
Morrow              Anthony      -0.6
Villanueva          Charlie      -0.6
Thomas              Kurt         -0.6
Delfino             Carlos       -0.7
Fisher              Derek        -0.7
Maynor              Eric         -0.7
Livingston          Shaun        -0.7
Diogu               Ike          -0.7
Beaubois            Rodrigue     -0.7
Daniels             Marquis      -0.7
Chalmers            Mario        -0.8
Jamison             Antawn       -0.8
Lee                 Courtney     -0.8
House               Eddie        -0.8
Thompson            Jason        -0.8
Collison            Darren       -0.9
Ajinca              Alexis       -0.9
Lewis               Rashard      -0.9
Douglas-Roberts     Chris        -0.9
Sefolosha           Thabo        -0.9
Harrington          Al           -0.9
Jennings            Brandon      -0.9
Williams            Shawne       -1.0
Brown               Derrick      -1.0
Green               Jeff         -1.0
Wafer               Von          -1.0
Evans               Reggie       -1.0
Favors              Derrick      -1.0
Davis               Ed           -1.0
Wright              Brandan      -1.0
Young               Sam          -1.0
Diaw                Boris        -1.1
Walker              Bill         -1.1
Bynum               Will         -1.1
Bogans              Keith        -1.1
Lopez               Robin        -1.1
Kaman               Chris        -1.1
Belinelli           Marco        -1.1
Sessions            Ramon        -1.1
Childress           Josh         -1.1
Erden               Semih        -1.2
Barbosa             Leandro      -1.2
Salmons             John         -1.2
Casspi              Omri         -1.2
Mayo                O.J.         -1.2
Dampier             Erick        -1.2
Wilkins             Damien       -1.2
Gibson              Daniel       -1.2
Ely                 Melvin       -1.3
Perkins             Kendrick     -1.3
Johnson             James        -1.3
Young               Nick         -1.3
Hawes               Spencer      -1.4
Bargnani            Andrea       -1.4
Henderson           Gerald       -1.4
Udoh                Ekpe         -1.4
Pachulia            Zaza         -1.4
Blatche             Andray       -1.5
Thornton            Al           -1.5
Cardinal            Brian        -1.5
Gordon              Ben          -1.5
Smith               Jason        -1.6
Daye                Austin       -1.6
Evans               Tyreke       -1.7
Head                Luther       -1.7
Anderson            James        -1.7
Watson              Earl         -1.8
Stevenson           DeShawn      -1.8
Brown               Kwame        -1.8
Pittman             Dexter       -1.8
Farmar              Jordan       -1.8
Andersen            David        -1.9
Haywood             Brendan      -1.9
Carroll             Matt         -1.9
Anthony             Joel         -1.9
Hamilton            Richard      -2.0
Derozan             DeMar        -2.0
Williams            Shelden      -2.0
Battie              Tony         -2.0
O'Neal              Jermaine     -2.0
Wallace             Ben          -2.1
Moon                Jamario      -2.1
Jackson             Darnell      -2.2
Gomes               Ryan         -2.2
Jackson             Stephen      -2.2
Arroyo              Carlos       -2.2
Pietrus             Mickael      -2.2
Johnson             Wesley       -2.2
Robinson            Nate         -2.2
Biedrins            Andris       -2.3
Greene              Donte        -2.3
Green               Willie       -2.3
Forbes              Gary         -2.3
Randolph            Anthony      -2.3
Hill                Jordan       -2.3
Cook                Brian        -2.3
Ariza               Trevor       -2.3
Collins             Jason        -2.4
Jeter               Eugene       -2.4
Vujacic             Sasha        -2.4
West                Delonte      -2.4
Miller              Mike         -2.4
Richardson          Quentin      -2.5
Jack                Jarrett      -2.5
Parker              Anthony      -2.5
Foye                Randy        -2.5
Williams            Jason        -2.6
Ager                Maurice      -2.6
Caracter            Derrick      -2.6
Blake               Steve        -2.6
Taylor              Jermaine     -2.7
Okur                Mehmet       -2.7
Brooks              Aaron        -2.7
Pondexter           Quincy       -2.7
Armstrong           Hilton       -2.7
Hinrich             Kirk         -2.7
Ford                T.J.         -2.7
Clark               Earl         -2.7
Dowdell             Zabian       -2.7
Arenas              Gilbert      -2.7
Posey               James        -2.7
James               Damion       -2.7
Turner              Evan         -2.7
McGuire             Dominic      -2.7
Howard              Juwan        -2.8
Jeffries            Jared        -2.8
Mozgov              Timofey      -2.8
Brewer              Corey        -2.8
Jones               Solomon      -2.8
Cunningham          Dante        -2.8
Cousins             DeMarcus     -2.9
Rush                Brandon      -3.0
Hollins             Ryan         -3.0
Webster             Martell      -3.0
Wall                John         -3.0
Nocioni             Andres       -3.0
Telfair             Sebastian    -3.0
Price               A.J.         -3.1
Gaines              Sundiata     -3.1
Elson               Francisco    -3.1
Williams            Mo           -3.1
Samuels             Samardo      -3.1
Carney              Rodney       -3.1
Amundson            Louis        -3.1
Harangody           Luke         -3.1
Bayless             Jerryd       -3.1
Najera              Eduardo      -3.2
Maxiell             Jason        -3.2
Mills               Patrick      -3.3
Koufos              Kosta        -3.3
Graham              Joey         -3.3
Kleiza              Linas        -3.3
Milicic             Darko        -3.3
Hickson             J.J.         -3.3
Petro               Johan        -3.4
Sanders             Larry        -3.4
Vasquez             Greivis      -3.4
Hayward             Lazar        -3.4
Harris              Manny        -3.4
Mason               Roger        -3.5
Ross                Quinton      -3.5
Pekovic             Nikola       -3.5
Aminu               Al-Farouq    -3.6
Quinn               Chris        -3.8
Ellington           Wayne        -3.8
Walton              Luke         -3.9
Przybilla           Joel         -3.9
Johnson             Armon        -3.9
Jianlian            Yi           -4.0
Outlaw              Travis       -4.0
Gadzuric            Dan          -4.0
Gee                 Alonzo       -4.0
Law                 Acie         -4.1
Howard              Josh         -4.2
Murphy              Troy         -4.2
Henry               Xavier       -4.2
Bledsoe             Eric         -4.2
Pavlovic            Sasha        -4.3
Bell                Raja         -4.3
Hayward             Gordon       -4.3
Duhon               Chris        -4.3
Fesenko             Kyrylo       -4.4
Dragic              Goran        -4.4
Carter              Anthony      -4.5
Eyenga              Christian    -4.6
Seraphin            Kevin        -4.7
Barron              Earl         -4.8
Smith               Ishmael      -4.9
Powell              Josh         -5.0
Price               Ronnie       -5.0
Thabeet             Hasheem      -5.0
Marks               Sean         -5.0
Weems               Sonny        -5.0
Crawford            Jordan       -5.3
Butler              Rasual       -5.3
Flynn               Jonny        -5.6
Williams            Jawad        -6.3
Williams            Terrence     -6.3
Graham              Stephen      -6.5
Harris              Mike          8.9
Greene              Orien         8.3
Novak               Steve         2.8
Haddadi             Hamed         1.7
Ming                Yao           1.7
Ebanks              Devin         1.3
Green               Danny         0.7
Owens               Larry         0.4
Thomas              Etan          0.2
Carroll             DeMarre       0.2
Magloire            Jamaal       -0.6
Johnson             Chris        -0.8
Adrien              Jeff         -1.1
Balkman             Renaldo      -1.1
Warren              Willie       -1.2
Lin                 Jeremy       -1.8
Siler               Garret       -2.4
Aldrich             Cole         -2.7
Scalabrine          Brian        -2.8
Evans               Maurice      -2.9
Jones               Dominique    -3.2
N'Diaye             Hamady       -3.2
Ivey                Royal        -3.2
Janning             Matt         -3.3
Banks               Marcus       -3.6
Cousin              Marcus       -3.8
Stackhouse          Jerry        -4.2
Ratliff             Theo         -4.2
Udoka               Ime          -4.3
Peterson            Morris       -4.8
Johnson             Trey         -5.0
Bell                Charlie      -5.0
Diop                DeSagana     -5.1
Allen               Malik        -5.1
Jeffers             Othyus       -5.1
Skinner             Brian        -5.3
Mensah-Bonsu        Pops         -5.4
Collins             Jarron       -5.4
Shakur              Mustafa      -5.7
Oberto              Fabricio     -5.8
Temple              Garrett      -5.9
Songaila            Darius       -6.0
Hudson              Lester       -6.0
Stephenson          Lance        -6.0
Kapono              Jason        -6.4
Summers             DaJuan       -6.5
Beasley             Michael      -6.8
Collins             Sherron      -7.2
Dupree              Ronald       -7.2
Lawal               Gani         -7.8
Lucas               John         -7.9
Mullens             Byron        -8.3
Smith               Joe          -8.6
Babbitt             Luke         -8.7
Brackins            Craig        -8.7
Alabi               Solomon      -8.8
Simmons             Bobby        -9.5
Whiteside           Hassan       -9.6
Wright              Antoine     -10.4
Bradley             Avery       -10.6
Rautins             Andy        -12.2

In both cases (the 3x yr data and the mix with the boxscore metrics) will eliminate the "Bryant-problem" of this season. Obviously there is a problem with collinearity due to the large amount of minutes Bryant plays with Gasol together on the court. The APM algorithm by Barzilai is giving all the credit to Gasol while Bryant ends up with a negative value. When we look at the raw +/- numbers, we can see it maybe more clearly. Bryant and Gasol played 2159.5 minutes together so far with +410, Gasol has 567 minutes without Bryant and +85, while Bryant has 317 minutes without Gasol and +13.

Despite the fact that Bryant isn't playing that well on the defensive end in this season (he is rather badly in terms of help and team defense, still better as 1on1 defender), he gets heavily underrated in Barzilai's algorithm due to collinearity. But he is still worse than his usual average value for the last couple of seasons.

[Doctor MJ]

Adding Ridge Regression can cut out some of the tremendous collinearity, but at the expense of losing some of the validity of the stat.

I agree with you, which is also the reason I prefer those values: http://stats-for-the-nba.appspot.com/4-year-ranking

I'd actually appreciate an explanation of ridge regression and RAPM. It looks promising, but I don't have my around it.

The appspot site, while I look at it, could really use more information such as the standard error.

(I also wish they'd drop the coach APM, or at least keep it totally separate from the players. It makes zero sense to use APM on coaches and take the results seriously.)

[laika]

Here is the most likely explanation for why Kobe's numbers are so strange this year. Adjusted +/- has gotten his number wildly wrong. I would say a mistake this severe means that adj +/- is useless(at least on basketballvalue), but it seems that most of you disagree.

-------------min------% with Kobe-----ppg with---ppg w/o
Gasol-------2727-------79--------------9.1-------7.2
Fisher------2017-------97-------------10.2-------5.4
Artest------2123-------89-------------10.4-------(-7.0)
Odom--------2360-------69--------------7.7-------6.3
Bynum-------1245-------76--------------9.9-------2.1
Brown-------1408-------26--------------4.5-------2.9
Blake-------1451-------34--------------3.0-------2.5


So the only strange result is with Gasol and Odom. The other players are helped by playing with Kobe. But this result has a logical explanation.

---------------------------pts--min----ppg
Bryant/Odom/Gasol--- 238--- 1330--- 8.6
Bryant/Bynum/Gasol-- 153--- 696 --- 10.6
Bryant/Bynum/Odom---- 44--- 245--- 8.6


Gasol is not carrying Kobe at all. There is virtually no difference when Odom/Bynum are in.
Almost the entire gap in performance between Gasol and Kobe is explained by the Laker team dynamics. When Kobe is out, Gasol plays with Blake, Barnes or Brown. While they are worse than the starters they are still reasonably decent. When Gasol is out though, Kobe does not have close to the same situation. In those minutes without Bynum/Odom he is forced to play with Caracter, Ratliff or a weird experimental lineup. Those two have been truly horrible this year.

We already have an unusually low number of comparable minutes due to Jackson's strict rotations. It seems that the APM models don't calculate the values for low minute players, instead assigning them an average value. Doing this for Caracter and Ratliff is making a massive mistake.
When you combine these two problems the error is really large. The difference in rating between Kobe and Gasol/Odom/Bynum can be explained almost entirely by this error, rather than any actual performance difference. Kobe's true performance is probably reasonably close to his unadjusted 7.9 Net, which would place him near the top of the league.

Even this might understate Kobe's value though. +/- assumes that Kobe/Fisher/Artest are basically the same player since they nearly always play together. Based on production, Kobe is massively better than those 2. It's quite possible that Kobe is doing the majority of the work among the 3 to help the Lakers, but only getting 33% of the credit in a +/- system.

[Doctor MJ]

In those minutes without Bynum/Odom he is forced to play with Caracter, Ratliff or a weird experimental lineup

Okay laika you're frustrating me now. I'm frustrated with myself to be honest because I have a tendency to want to argue and to teach, and this is causing me to waste my time. I've interacted quite a bit with you and exposed some huge misconceptions you have without any acknowledgement from you that you're absorbing any of it.

So now you write this post, and I do research. I'm just going to respond to the part above because it illustrates the issue succinctly.

You're saying Kobe's numbers are hopeless crippled because of the time he's spending with Caracter and Ratliff, okay. Let's make a time measurement called "Kobe minutes". It's the amount of time the rest of Kobe's teammates get to play with him. There are 5 Lakers on the floor, so the number of Kobe minutes is 4 times the amount of minutes Kobe played.

How many "Kobe minutes" did the Lakers have during this regular season? 11,116.
How many of those minutes were allocated to either Caracter or Ratliff? 76

And you're writing a post here effectively stating with confidence that these 76 minutes render Kobe's adjusted +/- useless? Gah! (For the record, taking these minutes out of the equation does close the gap in terms of raw +/-, but Gasol's raw +/- still easily surpasses Kobe's)

Seriously dude, it's seems that every bit of statistical analysis you do is working backwards. You're never using the stats to inform your opinion, you using them to support your opinions, and when they don't, the research you do seemingly is only for the purpose of trying to show why your opinion is still right, and what's worst for me is that you never do your research all the way you simply go far enough to develop a theory that helps your case, and then you assert it as truth.

I'm sorry I'm being so harsh but man, I'm getting nothing out of this interaction with you but a headache.

[mysticbb]

Very well said Doc. It is not like we already pointed out that normal APM has collinearity issues, we know that and we have to keep that in mind. Writing a post only to support a specific player and make it look like we have to throw the stats completely out of the window due to that specific player being "hated" by that stats is not only giving you a headache. Especially when the guy mixing up technical terms like APM, NET and so on as if there wouldn't be a difference.

[floppymoose]

<begin hand waving>
I've found APM to be unreliable and worse than raw +-. I don't have any way to prove this, other than to seay that I've found the outliers worse when engaging the sniff test. This is from several years of looking at APM. I don't believe the mathematics works. And I say this as a former math person. I've got nothing against the idea of APM. I just don't believe the state of the art there is good enough yet.
<end hand waving>

[laika]

1. I've interacted quite a bit with you and exposed some huge misconceptions you have without any acknowledgement from you that you're absorbing any of it.

2.How many "Kobe minutes" did the Lakers have during this regular season? 11,116.
How many of those minutes were allocated to either Caracter or Ratliff? 76

And you're writing a post here effectively stating with confidence that these 76 minutes render Kobe's adjusted +/- useless? Gah! (For the record, taking these minutes out of the equation does close the gap in terms of raw +/-, but Gasol's raw +/- still easily surpasses Kobe's)

3. you never do your research all the way you simply go far enough to develop a theory that helps your case, and then you assert it as truth.

1.What you call "exposing misconceptions" I would call disagreements. I could just as easily say that you have some huge misconceptions from my perspective. List the specific facts I have gotten wrong and not corrected if you think I have.

2.Your example is wrong and you should know it. APM is calculated using the independent minutes. The original numbers were from a while ago that I hadn't bothered to post- I'll recalculate with updated numbers so there is no doubt about where they are coming from.

Gasol--- 534 pts- in -3037 minutes
Bryant--- 423- in- 2779

Bryant/Gasol------ 418 -in- 2411 --- 8.3 ppg
Bryant w/o Gasol-- 5 -in - 368---- .7 ppg
Gasol w/o Bryant-- 116 -in- 626--- 8.9 ppg

Bryant/Odom/Gasol-- 258- in -1438 --- 8.6 ppg
Bryant/Bynum/Gasol- 146 - in -829 --- 8.5 ppg
Bryant/Odom/Bynum-- 45 - in - 292 --- 7.4 ppg
Bryant/scrubs------ (-40)- in - 76 --- (-25.3 ppg)

There are not 11000 minutes. Bryant played only 368 independent minutes from Gasol this year. Those 76 excluded minutes were more than enough to wildly distort his numbers. Playing with reasonable backups Gasol w/o Bryant won by 8.9 ppg while Bryant w/o Gasol won by 7.4 ppg. Once you adjust for the fact that Gasol was 4pts better than Bynum/Odom that difference is what you would expect if Gasol and Bryant were equal players this year.

If you account for all the other matchups then maybe it's possible that Gasol was a point or two better at best. Although keep in mind that Gasol played against worse competition than Kobe since he played more minutes and coaches avoid playing their backups with Kobe in.
But basketballvalue claims that Gasol was more than 20 points better than Kobe. This number is so absurd I'm amazed I even have to bother to prove it was wrong.

3.I don't know how far you think I have to go. I think I've gone far enough to prove beyond a reasonable doubt that the basketballvalue adj +/- numbers are wrong.

If collinearity were the only problem then you might be able to say that APM gave correct values subject to very high standard deviations. But as I've shown the exclusion of data is also a big problem and actually invalidates the APM model they are using. You really do have to throw the basketballvalue APM stats completely out the window- they are that far wrong.

[DSMok1]

You guys might want to read my feature on Adjusted Plus/Minus:

http://godismyjudgeok.com/DStats/2011/n ... ilization/

laika's right in that APM is really unstable. I, like him, would not use BBV's 2 year APM for much of anything. On the other hand, APM is the only unbiased statistic that actually completely captures a player's value that we have. And there are ways to manipulate it (Such as Jeremias Engelmann's RAPM at http://stats-for-the-nba.appspot.com/ ) so as to reduce the collinearity/instability problem.

And here's what his RAPM looks like over the past few years:

[Nivek]

So, the RAPM numbers suggest that Kobe this season helped the Lakers on offense (though not as much as in previous seasons) and hurt them defensively. Overall, he helped some, but his "helping" dropped significantly from previous seasons. Am I reading the graph correctly, DSMok?

[mysticbb]

Nivek, that is correct. That is also in agreement with Synergy Stats data.

DSMok1, we discussed the problems already and pointed out that RAPM is the way to go. I read your post about the review before and just want to give you props for that. It is a great read and overview.
The problem with a guy like laika isn't that he points out flaws, he is doing it for a specific reason. if Bryant would do great in that stat, he would use it to justify Bryant's greatness. That a biased discussion which is hardly constructive at all.

[Doctor MJ]

1.What you call "exposing misconceptions" I would call disagreements. I could just as easily say that you have some huge misconceptions from my perspective. List the specific facts I have gotten wrong and not corrected if you think I have.

laika, you came on here shouting about how people were doing adjusted +/- "wrong" using buzzwords like collinearity as part of your statements. After a few exchanges with you I responded to this statement of yours:

I also have a question about whether they use oncourt or net +/- numbers in their calculations. I think the oncourt number is much better than the offcourt number. At least with the on number the player is being directly measured along with his teammates. The offcourt number primarily measures the substitute quality and seems to have little to do with how good a player is. I personally would prefer an adjusted +/- that is calculated using oncourt rather than net.

By saying:

Hmm, see I don't think you have your head around this. Here's roughly what a data input looks like:

Brown,Bryant,Fisher,Turiaf,Walton,Alston,James,McGrady,Yao,Wells,2

I'm simplifying here to make it more intuitive. Literally, the program takes as input what happened when each 10 player combo was on the floor. Thousands upon thousands of these lines, and treats each player as a variable. The goal is to assign values to the players so that adding their values together you get the scoreboard result at the end.

See what I'm saying? Technically, it is using on court +/-, but not in the way I think you mean. Adjusted +/- doesn't take a player's season-long +/- and apply adjustments to it, it literally breaks everything down to granules of data and solves for thousands of linear equations.

Basically, it made clear that after several back & forths with you expressing such confidence that you knew best how to calculate adjusted +/-, you hadn't the foggiest idea of how adjusted +/- even works.

That is not a disagreement. That is you having misconceptions.

And how did you respond to me taking so much of my sweet time to expose misconceptions? You never responded to what I wrote at all. Your next post in the thread was:

After two rounds of the playoffs Anthony's net +/- is 36.5. Lebron's is negative 10.0. I guess it's a thankless job carrying Lebron but you can hardly expect anyone in the stats community to point this out.

Can't afford to blemish the first great pillar of truth, "Lebron is awesome and Kobe sucks" that their religion is built upon.

As if that entire exchange we had meant nothing...and yet still, when you've posted more, I've responded. Clearly I'm insane to try so hard to teach someone who doesn't want to learn.

[Doctor MJ]

<begin hand waving>
I've found APM to be unreliable and worse than raw +-. I don't have any way to prove this, other than to seay that I've found the outliers worse when engaging the sniff test. This is from several years of looking at APM. I don't believe the mathematics works. And I say this as a former math person. I've got nothing against the idea of APM. I just don't believe the state of the art there is good enough yet.
<end hand waving>

floppy,

I don't know that we have any disagreement, although a statement like "I don't believe the mathematics works" makes it hard to know. If you want to understand where I'm coming from with laika read this thread and the Joel Anthony thread.

I don't have any illusions about APM being a perfect stat. What I am steadfast about is that every analyst should be using +/- data in some form along with box score data when evaluating players over long period of time.

[Doctor MJ]

You guys might want to read my feature on Adjusted Plus/Minus:

http://godismyjudgeok.com/DStats/2011/n ... ilization/

laika's right in that APM is really unstable. I, like him, would not use BBV's 2 year APM for much of anything. On the other hand, APM is the only unbiased statistic that actually completely captures a player's value that we have. And there are ways to manipulate it (Such as Jeremias Engelmann's RAPM at http://stats-for-the-nba.appspot.com/ ) so as to reduce the collinearity/instability problem.

Sweet, will read it today.

[laika]

1.laika, you came on here shouting about how people were doing adjusted +/- "wrong" using buzzwords like collinearity as part of your statements. After a few exchanges with you I responded to this statement of yours:

I also have a question

By saying:

See what I'm saying? Technically, it is using on court +/-, but not in the way I think you mean. Adjusted +/- doesn't take a player's season-long +/- and apply adjustments to it, it literally breaks everything down to granules of data and solves for thousands of linear equations.

Basically, it made clear that after several back & forths with you expressing such confidence that you knew best how to calculate adjusted +/-, you hadn't the foggiest idea of how adjusted +/- even works.

That is not a disagreement. That is you having misconceptions.

And how did you respond to me taking so much of my sweet time to expose misconceptions? You never responded to what I wrote at all. Your next post in the thread was:

After two rounds of the playoffs Anthony's net +/- is 36.5. Lebron's is negative 10.0. I guess it's a thankless job carrying Lebron but you can hardly expect anyone in the stats community to point this out.

Can't afford to blemish the first great pillar of truth, "Lebron is awesome and Kobe sucks" that their religion is built upon.

As if that entire exchange we had meant nothing...and yet still, when you've posted more, I've responded. Clearly I'm insane to try so hard to teach someone who doesn't want to learn

I don't remember shouting, but most people seem to agree with me that basketballvalue is calculating adj +/- incorrectly.

How is asking a question showing "huge misconceptions"?
I asked a question and you answered it. I didn't feel the need to keep responding, especially after mystic attacked me with a bunch of lame insults. Also, it didn't help that we were starting to talk in circles.

I didn't respond again until you made basically a troll thread about Kobe using extremely poor analysis.

Usually I respond to posts that I disagree with. I guess in the future I could make a note of posts that I agree with.

[Doctor MJ]

I don't remember shouting, but most people seem to agree with me that basketballvalue is calculating adj +/- incorrectly.

How is asking a question showing "huge misconceptions"?

I don't think you'll find anyone here who would agree to the statement "basketballvalue is calculating APM incorrectly."

On the other hand, every single one of us would agree that APM is an imperfect stat.

Previously, you chafed about the distinction between these two statements saying they essentially mean the same thing, but then you made clear that you literally thought that basketballvalue was taking a player's raw season-wide +/- and doing a series of adjustments to it to come up with their adjusted +/-, and you were sure that those adjustments were wrong. Thus it was made clear that the language difference between how described things and how we described things was actually at the very root behind you misconception, which I tried to explain to you once I understood where you went off track.

You now ask "How is asking a question showing a huge misconception?".

When someone asks a question where literally answering the question won't yield any useful information, it typically means that the person asking the question has a misconception. You asked:

I also have a question about whether they use oncourt or net +/- numbers in their calculations.

I recognized that someone would ask this question if they didn't understand how adjusted +/- was calculated, and thus I spend time explaining to you how they calculated the stat form the ground up.

I asked a question and you answered it. I didn't feel the need to keep responding, especially after mystic attacked me with a bunch of lame insults. Also, it didn't help that we were starting to talk in circles.

I didn't respond again until you made basically a troll thread about Kobe using extremely poor analysis.

Usually I respond to posts that I disagree with. I guess in the future I could make a note of posts that I agree with.

Well there's a couple social things here:

-You being sick of "lame insults" is a perfectly reasonable reason for you to stop participating.

-However if you're going to end up continuing the same conversation with the same people on a board where there weren't a large amount of topics in between, then you come off looking pretty bad when don't acknowledge your own mistakes while you continue on your attack.

Obviously you don't see the situation as others do, but from my perspective I'm doing you a favor by helping you to better understand something you're interested in. If you don't see value to my insights, and insist on pushing forward with you mind unchanged, then there's no point in me conversing with you.

[DSMok1]

So, the RAPM numbers suggest that Kobe this season helped the Lakers on offense (though not as much as in previous seasons) and hurt them defensively. Overall, he helped some, but his "helping" dropped significantly from previous seasons. Am I reading the graph correctly, DSMok?

Yes, you are. The scale is points per 100 possessions played, with positive good for everything.

[floppymoose]

You guys might want to read my feature on Adjusted Plus/Minus:

http://godismyjudgeok.com/DStats/2011/n ... ilization/

Nice article. My instincts, worth what you paid for them, are that the SPM approach is total garbage, and Rosenbaum didn't accomplish anything when he tried to use that to "fix" his APM results.

I like basketballvalues approach of simply lumping all the low minute players together in one bucket to reduce the noise. Their APM values are the best I've seen so far.

I don't have a good feel for regularized APM yet, but I agree with the criticism that assuming the low minute players are average is pretty crippling. That is going to depress the value of guys who get rested against the scrubs, and inflate the value of guys who have a mix of playing with the starters but also stay in against scrub opponents. The results at stats-for-the-nba.appspot.com are.... interesting.

[Doctor MJ]

Nice article. My instincts, worth what you paid for them, are that the SPM approach is total garbage, and Rosenbaum didn't accomplish anything when he tried to use that to "fix" his APM results.

Worse than accomplishing nothing, it lowered the credibility. I've got a lot of respect for Rosenbaum, and I'm forever grateful to Roland Beech for his work with 82games, but when people come up with metrics that arbitrarily assigning weights combining completely different metrics I always shake my head. It just convolutes matters.

That he combined things really is evidence that he didn't think much of the work he was doing. It was just a lark for him. But for me it's never been an issue that +/- stats are flawed because all stats are flawed. I'm not trying to find the end all be all, just stuff I can use.

[floppymoose]

But for me it's never been an issue that +/- stats are flawed because all stats are flawed. I'm not trying to find the end all be all, just stuff I can use.

I think we are coming at this from similar vantage points, but my angle is: I want to use the best +- stat I can, and I feel that there is progress still to be made on how to generate decent APM numbers.