APBRmetrics

Posted: **Mon Nov 05, 2012 9:35 pm**

After 1 to 3 games per team, the fast starters in the league:

e82     per36 rates    tm   G   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
17.7   Harden,James   Hou   3   42   .627   34.5   5.3   5.4   1.4   4.0   .6   2.50
16.5   Lowry,Kyle     Tor   3   36   .724   27.8   8.1   7.0   3.7   3.0   .3   2.71
15.8  Anthony,Carmelo NYK   2   39   .528   32.0   7.6   1.8   1.4   1.4   .9   2.39
15.6   Durant,Kevin   Okl   3   42   .553   21.4  12.9   6.1   1.4   4.6  1.1   2.19
15.3   Paul,Chris     LAC   3   34   .582   21.2   4.8  15.1   2.9   2.9   .0   2.68

14.9   Davis,Glen     Orl   2   33   .495   30.5   9.6   3.2   1.1    .6  1.1   2.67
14.9 Jennings,Brandon Mil   2   33   .523   20.2   3.6  14.8   4.4   3.3   .6   2.66
14.4   James,Lebron   Mia   3   35   .611   22.7  10.0   6.7    .7   1.7  1.7   2.42
13.7   Dunleavy,Mike  Mil   2   29   .878   26.7  12.1   4.8   2.4   1.8   .6   2.79
13.5   Duncan,Tim     SAS   3   33   .558   25.4  12.2   1.9    .7   2.2  3.3   2.39

e82     per36 rates    tm   G   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
12.6   Howard,Dwight  LAL   4   36   .628   25.2  11.2   2.3    .8   3.6  2.6   2.10
12.5   Redick,J.J.    Orl   2   35   .763   28.2   2.5   6.8   1.0   1.5   .0   2.14
11.7 Varejao,Anderson Cle   3   34   .672   15.2  17.1   3.6    .4    .7   .7   2.01
11.6   Parker,Tony    SAS   3   34   .517   23.6   3.1   9.8    .7   2.1   .0   2.00
11.6   Gasol,Marc     Mem   2   38   .641   21.2   6.9   5.4    .5   1.4  1.0   1.80

11.5   Williams,Mo    Uta   3   33   .623   26.0   2.3   7.0   1.8   2.2   .7   2.04
11.2 Aldridge,Lamarcu Por   3   39   .450   21.3   8.9   3.1    .6   1.9  1.2   1.70
11.2   Lopez,Brook    Brk   1   32   .551   31.5   6.5    .0   1.1   1.1  1.1   2.07
11.1   Kidd,Jason     NYK   2   24   .774   24.8   5.9   8.7   3.8    .8   .0   2.74
11.0   Bosh,Chris     Mia   3   34   .612   23.7  11.0   1.5    .7   2.1  1.4   1.89

e82     per36 rates    tm   G   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
10.8   Lillard,Damian Por   3   38   .566   21.8   3.8   8.8    .6   3.5   .0   1.67
10.7   Scola,Luis     Phx   3   31   .553   21.2  10.2   2.0   2.0    .4  1.6   2.04
10.3   Holiday,Jrue   Phi   2   41   .562   21.6   3.5   7.6   1.3   4.0   .0   1.49
10.2   Bryant,Kobe    LAL   4   37   .699   28.5   5.5   3.0   1.0   3.9   .0   1.63
10.2 Westbrook,Russel Okl   3   36   .444   22.4   6.3   7.5    .0   3.4   .0   1.68

10.1   Crawford,Jamal LAC   3   31   .688   31.1   3.4   1.3   1.5    .8   .0   1.93
10.0  Collison,Darren Dal   3   32   .675   20.8   3.3   8.5   1.9   1.9   .0   1.83
10.0 Kirilenko,Andrei Min   2   30   .750   19.7   8.2   5.3    .6   2.4  2.4   1.96
9.9    Rondo,Rajon    Bos   3   42   .575   13.5   5.1   9.7   1.4   2.3   .0   1.41
9.9    West,David     Ind   3   36   .471   21.6   8.3   1.4   1.0    .7   .7   1.61

e82 is the player's eWins rate over the whole season, if they would continue at this production and minutes.
e484 is their eWins per 484 minutes; NBA average rate is 1.00
Per minute, Jason Kidd looks like the best player in the league, other than Jose Barea.
Per-36-minute rates are also per 100 points and 44 rebounds per team, along with other adjustments.

Posted: **Tue Nov 06, 2012 6:53 pm**

Who bites on trying to trade for Jennings?

Posted: **Tue Nov 06, 2012 9:08 pm**

Crow wrote:Who bites on trying to trade for Jennings?

I doubt he gets traded, at least mid season. Bucks have his RFA, seems more like any deal will occur over the offseason.

Posted: **Mon Nov 12, 2012 8:01 pm**

After .076 of season's games having been played, it's mostly the usual suspects.

Code: Select all

e82     per36 rates    tm   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
14.8   James,Lebron   Mia   34   .595   25.1  11.7   7.3    .9   2.1   .9   2.58
13.7   Durant,Kevin   Okl   38   .590   25.1  10.9   4.4   1.5   4.2  1.1   2.11
13.5   Paul,Chris     LAC   33   .597   21.6   4.4  12.2   2.2   2.2   .0   2.44
13.1  Anthony,Carmelo NYK   36   .534   32.5   7.4   2.0    .8   2.0  1.3   2.18
12.8   Howard,Dwight  LAL   35   .590   24.0  11.5   2.7   1.0   3.4  2.5   2.18

12.7   Duncan,Tim     SAS   30   .574   25.0  12.5   2.9   1.7   1.7  2.9   2.48
12.6   Bryant,Kobe    LAL   36   .649   30.8   6.0   4.6   1.4   4.0   .0   2.06
11.7   Rondo,Rajon    Bos   41   .530   13.6   4.4  12.1   1.6   2.9   .0   1.67
11.6   Harden,James   Hou   40   .554   28.0   4.7   4.1   1.5   4.6   .6   1.71
10.9   Noah,Joakim    Chi   39   .563   16.6   9.1   3.3   1.6   2.5  2.3   1.66

e82     per36 rates    tm   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
10.6 Westbrook,Russel Okl   35   .484   22.1   5.0   9.7   1.0   3.6   .1   1.79
10.5   Gasol,Marc     Mem   36   .552   16.2   7.7   6.0    .3    .8  1.2   1.73
10.1   Holiday,Jrue   Phi   39   .548   20.4   3.7   8.8   1.6   5.3   .6   1.55
10.1   Smith,J.R.     NYK   34   .579   21.6   5.2   4.5   2.3   1.0   .5   1.76
10.0   Bosh,Chris     Mia   32   .611   24.6   8.7   1.6   1.3   1.6  1.3   1.85

10.0  Randolph,Zach   Mem   37   .492   17.1  14.7   1.3    .8   2.6   .8   1.58
9.9  Kirilenko,Andrei Min   33   .615   14.7   9.0   4.7   1.8   2.5  2.2   1.75
9.8   Conley,Mike     Mem   33   .614   19.3   3.8   8.5   2.2   3.5   .2   1.75
9.6   Pierce,Paul     Bos   36   .534   22.0   7.7   3.2   1.5   2.0   .2   1.57
9.2   Griffin,Blake   LAC   32   .521   19.4  10.5   4.5   1.6   3.5   .8   1.70

e82     per36 rates    tm   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
9.2  Aldridge,Lamarcu Por   39   .461   19.2   7.3   3.6    .8   1.6  1.6   1.41
9.1   Batum,Nicolas   Por   40   .566   17.7   6.0   2.7   2.5   1.5   .9   1.37
9.0   Irving,Kyrie    Cle   35   .561   24.3   4.7   6.1    .9   4.3   .3   1.53
8.9  Varejao,Anderson Cle   35   .567   13.1  14.8   3.5   1.4   1.6   .5   1.77
8.7   Walker,Kemba    Cha   34   .497   19.0   4.1   5.7   3.0   1.7   .2   1.51

8.6   Teague,Jeff     Atl   28   .640   22.1   3.9   8.3   2.1   3.9   .5   1.84
8.6   Kidd,Jason      NYK   20   .805   19.2   5.2   7.9   3.9    .4   .0   2.50
8.6   Lowry,Kyle      Tor   31   .689   23.9   7.2   7.0   3.5   2.9   .6   2.42
8.5   Ellis,Monta     Mil   36   .473   22.1   3.5   5.4   1.4   2.8   .6   1.40
8.5   Garnett,Kevin   Bos   31   .542   19.7  11.4   2.2    .8   2.2  1.4   1.63

e82     per36 rates    tm   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
8.4   Faried,Kenneth  Den   30   .548   17.9  12.5    .7   1.0    .9  1.4   1.69
8.4   Felton,Raymond  NYK   31   .476   16.4   3.8  10.4   2.0   3.2   .0   1.60
8.3   Gay,Rudy        Mem   35   .455   21.5   7.2   2.2   1.7   2.6  1.0   1.41
8.3   Deng,Luol       Chi   40   .535   18.7   6.7   2.8    .6   2.0   .6   1.25
8.2   Mayo,O.J.       Dal   35   .657   25.1   3.9   3.2    .7   2.8   .1   1.40

8.1   Millsap,Paul    Uta   30   .526   17.0  11.5   2.9   1.2   2.6  1.7   1.62
8.1   Gortat,Marcin   Phx   34   .561   12.9  11.4    .6    .9   1.5  4.1   1.41
8.1   Martin,Kevin    Okl   30   .697   25.3   3.8   3.0   1.9   2.5   .3   1.61
8.1   Horford,Al      Atl   35   .574   16.5   9.4   2.2    .4   1.0  1.3   1.37
8.1   Gasol,Pau       LAL   37   .459   14.1  10.0   3.3    .8   1.8  1.4   1.27

#41-50: BLopez, Monroe, Al Jefferson, Aminu, Lin, Sanders, Jennings, DWest, Dragic, Wade

Posted: **Sat Nov 17, 2012 3:44 pm**

At 10.5% of the season

Code: Select all

e82     per36 rates   tm   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
15.2   James,Lebron   Mia   36  .585   27.4  10.4   7.6    .8   2.3  1.1   2.50
14.6   Durant,Kevin   Okl   39  .608   27.3  10.5   4.5   1.5   3.8  1.4   2.22
14.0   Paul,Chris     LAC   33  .602   22.6   4.2  12.7   2.6   2.4   .0   2.55
13.3   Bryant,Kobe    LAL   36  .635   32.6   5.4   5.4   1.4   3.5   .0   2.15
12.5   Duncan,Tim     SAS   31  .558   24.4  12.9   3.0   1.3   1.5  3.2   2.36

11.9   Howard,Dwight  LAL   35  .575   22.7  12.4   2.2   1.1   3.9  2.8   1.98
11.4 Varejao,Anderson Cle   35  .620   17.5  16.1   3.5   1.5   1.5   .5   2.19
11.1 Kirilenko,Andrei Min   36  .655   18.2   9.0   3.5   1.5   3.0  2.3   1.82
11.0   Noah,Joakim    Chi   40  .559   17.1   9.7   3.4   1.3   2.4  2.1   1.63
10.8   Harden,James   Hou   40  .557   26.6   4.5   4.3   1.4   3.8   .6   1.59

e82     per36 rates   tm   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
10.7 Westbrook,Russel Okl   36  .470   21.1   5.5  10.3   1.3   3.1   .1   1.75
10.6   Rondo,Rajon    Bos   39  .528   14.3   5.1  12.5   2.0   3.0   .0   1.82
10.6   Gasol,Marc     Mem   37  .562   17.3   7.7   4.7    .6    .9  1.2   1.69
10.6   Randolph,Zach  Mem   37  .531   18.7  15.3   1.0    .9   2.4   .8   1.70
10.4 Jennings,Brandon Mil   35  .520   19.5   3.5   8.3   3.4   2.8   .5   1.78

10.0   Batum,Nicolas  Por   40  .597   20.3   6.3   2.9   2.2   2.4  1.2   1.49
10.0   Griffin,Blake  LAC   32  .522   20.6  11.9   5.2   1.5   3.7  1.0   1.86
9.9   Millsap,Paul    Uta   32  .570   21.3  11.4   3.0   1.2   2.6  1.6   1.83
9.7   Conley,Mike     Mem   34  .577   18.2   4.0   8.8   1.9   3.2   .1   1.70
9.5   Irving,Kyrie    Cle   35  .576   26.1   4.8   6.2   1.2   4.4   .3   1.60

e82     per36 rates   tm   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
9.4   Lopez,Brook     Brk   30  .571   26.3   9.1   1.2    .5   2.2  3.2   1.86
9.4   Holiday,Jrue    Phi   38  .538   21.1   4.0   7.9   1.5   5.1   .5   1.45
9.3   Pierce,Paul     Bos   35  .541   22.6   7.3   3.5   1.8   2.0   .4   1.58
9.3   Anthony,Carmelo NYK   37  .522   25.7   7.9   2.0    .7   2.6   .7   1.50
9.2   Bosh,Chris      Mia   32  .598   23.5   9.0   1.6    .9   1.7  1.5   1.68

9.1   Williams,Deron  Brk   35  .560   22.5   2.6   8.6    .8   4.2   .5   1.52
8.8   Walker,Kemba    Cha   36  .506   20.0   3.8   5.1   3.0   1.9   .3   1.46
8.8   Faried,Kenneth  Den   30  .536   18.2  14.5    .5   1.1   1.6  1.4   1.72
8.7   Monroe,Greg     Det   34  .506   17.9  10.9   3.5   1.7   3.0  1.2   1.51
8.6   Deng,Luol       Chi   41  .558   19.6   6.8   2.6    .6   2.0   .6   1.24

e82     per36 rates   tm   Min   Eff%   Sco   Reb   Ast   Stl   TO   Blk   e484
8.6   Jefferson,Al    Uta   33  .476   16.7  13.4   2.3   1.4   1.8  1.0   1.54
8.6  Aldridge,Lamarcu Por   39  .478   19.6   6.9   3.8    .8   1.9  1.5   1.31
8.4   Kidd,Jason      NYK   25  .822   17.7   3.6   5.4   2.8    .9   .6   1.98
8.4   Gay,Rudy        Mem   36  .478   21.5   6.6   2.8   1.5   2.3  1.3   1.37
8.4   Dragic,Goran    Phx   34  .576   18.7   3.0   6.8   2.0   2.4   .2   1.46

8.3   Felton,Raymond  NYK   33  .496   19.7   3.7   8.3   1.6   2.9   .0   1.47
8.2   Gasol,Pau       LAL   37  .474   14.1  10.3   3.7    .7   1.8  1.5   1.30
8.2   Teague,Jeff     Atl   27  .588   22.1   4.1   8.4   2.1   4.3   .7   1.77
8.1   Smith,J.R.      NYK   34  .574   19.4   5.5   3.5   1.7   1.2   .6   1.40
7.9   Mayo,O.J.       Dal   35  .630   25.9   3.9   3.1    .5   2.8   .2   1.34

Posted: **Sun Nov 18, 2012 10:46 am**

The correlation this season between eWins/minute and minutes per game is .58, among the top 300 players in total minutes. These players all have at least 64 minutes and average > 10 mpg.
For PER, the correlation is .46, and for WinShares/48 its .30

Separated into teams, MPG has an average correlation of .61 with eWins rate; still .46 with PER, and just .286 with WS/Min.
I can't imagine why WS correlation is even lower on teams than across the league, nor why PER is not better. Maybe over a full season, all would be higher?

The most typical team, for these correlations, is Washington. Ranked by MPG:

Code: Select all

mpg      Wizards        e484    PER    WS/48
31.4    Price,A.J.       .81    10.9    .017
28.7    Beal,Bradley     .62    10.4    .027
24.9    Ariza,Trevor    1.01    12.3    .046
24.6    Booker,Trevor    .86    14.0    .059
24.1    Okafor,Emeka    1.15    16.0    .092

21.8    Seraphin,Kevin   .77    13.4    .023
21.7   Crawford,Jordan  1.06    16.5    .064
21.4    Webster,Martell  .65    12.0    .099
14.6    Pargo,Jannero   -.36      .3   -.138
13.7    Vesely,Jan       .06     4.9    .009
12.4    Singleton,Chris  .56    11.9    .075

239.4    correlations   .625    .473    .272

NBA averages are 1.00 for eW/484, 15.0 for PER, and about .100 for WS/48
This study does not include last night's games. eWins suggests Crawford should perhaps start over Beal, and I guess he did.
PER seems to suggest 2 or 3 starters should be replaced; and WS looks almost random, relative to minutes.

Coincidentally the Wiz' opponents last night, the Jazz have the best correlations across the board:

Code: Select all

mpg      Jazz           e484    PER    WS/48
35.7    Williams,Mo     1.13    16.4    .088
33.0    Jefferson,Al    1.54    18.5    .121
32.1    Millsap,Paul    1.83    22.5    .188
30.9    Hayward,Gordon   .90    15.3    .113
27.9    Williams,Marvin  .68    12.6    .083

25.7    Foye,Randy       .51    13.0    .110
24.1    Favors,Derrick  1.18    19.1    .135
17.6    Tinsley,Jamaal   .23     4.4   -.078
13.9    Kanter,Enes      .29     8.7   -.015

240.8    correlations   .763    .750    .752

The sum of minutes is included to fill space, but when it's close to 240, you may suppose players getting fewer mpg are only getting minutes when someone else is unavailable.
Utah is one of 4 teams where WS has a better MPG correlation than PER has.
There are 3 teams for which PER has a better correlation than eWins.
This is the closest that Win Shares comes to eWins; and for a more severe discrepancy, look at Philly:

Code: Select all

mpg      Sixers           e484    PER    WS/48
38.2    Holiday,Jrue      1.45    18.0    .090    
34.1    Young,Thaddeus    1.11    16.1    .136    
32.2    Turner,Evan       1.01    11.6    .062    
27.2    Richardson,Jason  1.28    18.9    .185    
25.8    Wright,Dorell     1.00    15.0    .094 
   
24.0    Young,Nick         .28     9.4    .039    
22.0    Allen,Lavoy        .49    10.6    .068    
21.1    Hawes,Spencer      .98    14.3    .088    
15.0    Ivey,Royal         .51    12.8    .136
    
239.6    correlations     .724    .481   -.019

If WS should rank Royal Ivey as an average player (.100), the WS/mpg correlation rises to .15, and if he's .050, it improves to .36

The Kings get the lowest correlation by all 3 measures.

Code: Select all

mpg      Sixers           e484    PER    WS/48
32.2    Evans,Tyreke       .73    12.0    .013    
31.1    Cousins,Demarcus  1.48    18.9    .048    
30.7    Thompson,Jason    1.14    17.2    .150    
30.1    Thornton,Marcus    .64    15.3    .066    
24.7    Johnson,James      .10     4.9   -.101
    
23.4    Thomas,Isaiah      .62    12.6    .032    
22.4    Hayes,Chuck        .63    10.9    .061    
19.4    Brooks,Aaron       .22     8.4    .006    
18.8    Salmons,John       .60    15.3    .076    
16.4    Robinson,Thomas    .51    12.1    .055    
10.0    Fredette,Jimmer   1.26    26.5    .233
    
259.3    correlations     .149   -.233   -.363

Jimmer is lighting it up but only getting 10 minutes. PER and WS think he's a superstar, but eWins discounts his numbers largely because he's only playing vs about 40% starters.
If his PER and WS/48 are scaled down to (for example) 20 and .150, those correlations improve to -.02 and -.21, respectively.

Posted: **Mon Nov 19, 2012 1:55 pm**

I assume here that you are explaining minutes by PER etc rather than predicting via that method, which would explain the low correlation for PER to some extent. In other words you are using the actual performance of players to describe the coaching decisions, but coaches don't have a crystal ball, and I suggest using a prior estimate of expected values for each player, updated throughout the season, would provide a better prediction for minutes played.

Posted: **Mon Nov 19, 2012 3:01 pm**

The correlations are just what they are, independent of what I may theorize. It does seem as though coaches could not all be suffering the same delusions regarding player value -- volume scoring or any efficiency measure, if overvalued, would not be expected to prevail too heavily across the whole league.

For the whole 2011-12 season, but without players traded midseason, the correlations are .743 for eWins/484, .663 for PER, and .544 for Win Shares per 48.
All these are improvements over the short season at hand; but I wanted to see how it went for now, and to check it again later in the season.
Which types of metric might coaches use for midseason adjustment?

The numbers above are averages for teams; for the league at large, they're .73, .56, and .51
PER is most improved by ranking players on their respective teams, if minutes are a valid indicator of player worth.

The 2012 survey includes the top 400 players in minutes, so it's everyone averaging 6 mpg or more.
I'm curious how other metrics correlate to minutes allotment.

Posted: **Mon Nov 19, 2012 3:18 pm**

Based on these correlations for last season, it seems 28 of 30 NBA head coaches would prefer PER over WS/48 for allotment of minutes. Exceptions: Min and Atl
It also seems that only 4 of 30 would take PER over eWins: Det, Tor, LAC, and GS.

Between eWins and Win Shares, the 2012 correlation advantage ranges from .01 (Clips) to .63 (Wiz). Avg is .20

Posted: **Mon Nov 19, 2012 9:44 pm**

I did some work a while ago working merely with exponentially smoothed values and a genetic algorithm to find parameter estimates and found that when using the box score to estimate coaching decisions that minutes per game were best described in the linear case by:

Minutes Per Game = 1.5*PTS48 - 0.2*FGA48 - 0.9*FTA48 - 1.0*TOV48 + 0.4*DRB48 - 0.2*ORB48 + 0.5*BLK48 - 0.2*PF48 + 0.5*AST48 + 0.1*STL48

This was built across the league so obviously some players will get more/less minutes than this suggests by necessity, but I believe it is a very good equation to describe coaching decisions on average, and I personally use it to create a depth chart for each team.

I will say again that this was built to predict and not explain, but I do believe it shows that most coaches take little interest in the stats side of things and favour the eye test. For instance I would pose that the ORB coefficient is negative because coaches see missed shots prior to those rebounds and decide that there is an issue with this five man unit. Likewise coaches appear quick to forgive missed shots if eventually points appear, even if they are low-efficiency - however if you get to the line then a low free throw percentage will hurt your minutes. Steals are low-value, perhaps because coaches believe that the players they see taking risks to make steals are taking too much risk?

In any case this equation to me says that most NBA coaches aren't very able to recognise what is causing them to win, but perhaps they are and it is the advanced stats crowd that have it bent out of shape. I know there is a sentiment amongst some here that using coaching decisions as a prior might be a good idea, but from this at least I believe it really isn't.

Posted: **Mon Nov 19, 2012 10:43 pm**

v-zero, before jumping to conclusions about the coaching abilities, did you consider that your regression lacked important informations? How about taking consistency into account? Or fatigue? How about fit and team chemistry? These are all things which might very well influence the coaching decisions. A high coefficient of determination, does not mean you found causality here. When you find a result, which is letting you conclude that most coaches aren't able to "recognise what is causing them to win", you better check your method.

Posted: **Mon Nov 19, 2012 10:55 pm**

... this was built to predict and not explain, but I do believe it shows that most coaches take little interest in the stats side of things and favour the eye test.

Well, you're looking 'after the fact' at how these aggregate productions correlate with minutes.
I'd think it says nothing about whether the coaches are looking for these particular numbers and playing who they do because of such a stat.

You've also got nothing much representing defense in your stat. That's half the game, so right there you have some 30-40% of a coach's judgment which is not contained in the box score.
Of course, the one stat that really matters is point differential, and that's why we have +/- .

Last year, PER picks better teams' minutes a tad worse than it picks weaker teams'. eWins was just slightly better for stronger teams. And WS was much better with stronger teams; or you might say it's worse for weaker teams.

Posted: **Mon Nov 19, 2012 11:01 pm**

v-zero wrote:I did some work a while ago working merely with exponentially smoothed values and a genetic algorithm to find parameter estimates and found that when using the box score to estimate coaching decisions that minutes per game were best described in the linear case by:

Minutes Per Game = 1.5*PTS48 - 0.2*FGA48 - 0.9*FTA48 - 1.0*TOV48 + 0.4*DRB48 - 0.2*ORB48 + 0.5*BLK48 - 0.2*PF48 + 0.5*AST48 + 0.1*STL48

This was built across the league so obviously some players will get more/less minutes than this suggests by necessity, but I believe it is a very good equation to describe coaching decisions on average, and I personally use it to create a depth chart for each team.

I will say again that this was built to predict and not explain, but I do believe it shows that most coaches take little interest in the stats side of things and favour the eye test. For instance I would pose that the ORB coefficient is negative because coaches see missed shots prior to those rebounds and decide that there is an issue with this five man unit. Likewise coaches appear quick to forgive missed shots if eventually points appear, even if they are low-efficiency - however if you get to the line then a low free throw percentage will hurt your minutes. Steals are low-value, perhaps because coaches believe that the players they see taking risks to make steals are taking too much risk?

In any case this equation to me says that most NBA coaches aren't very able to recognise what is causing them to win, but perhaps they are and it is the advanced stats crowd that have it bent out of shape. I know there is a sentiment amongst some here that using coaching decisions as a prior might be a good idea, but from this at least I believe it really isn't.

That's pretty cool. Are the weights that far off from what you would've expected? Maybe it would help if you removed teams that are tanking or developing rookies from your datapool? To see how coaches that care more about winning select MPG for their players. Also, fatigue is hard to quantify, but +/- and consisteny could prove helpful

Mike, are you (again/still?) adjusting for home/road assists? What teams are especially bad at padding their players' totals? I just saw this

and now I think it might be a good idead to only use away assists in a player metric

Posted: **Mon Nov 19, 2012 11:20 pm**

With just a game or two at home or away, for each team, I haven't "turned on" the home/away assists and blocks adjustments yet. But I will do so directly.

Last year, players around the league blocked only .914 as many FGA on the road as they did at home.
The Rockets "led" the league with just .68 as many on the road as at home. Followed by Mil and Mem (.72), Chi (.74), Den (.75).

The Clippers assisted just .846 as many of their FG on the road as they did at home. I think I've pointed out Blake Griffin's astonishing home/away assist differences.
Next were the Lakers at .87, then Den (.87), Cle (.876), Okl (.884)
For the whole league, the average was .968, which is an improvement over previous years.

I don't get every players H/A rates, but rather assign a single multiplier to every player on a team.
Rockets got credit for just .81 of their blocks -- all of their away blocks and .68 of their home blocks.

Posted: **Tue Nov 20, 2012 12:03 am**

mystic wrote:v-zero, before jumping to conclusions about the coaching abilities, did you consider that your regression lacked important informations? How about taking consistency into account? Or fatigue? How about fit and team chemistry? These are all things which might very well influence the coaching decisions. A high coefficient of determination, does not mean you found causality here. When you find a result, which is letting you conclude that most coaches aren't able to "recognise what is causing them to win", you better check your method.

There's nothing wrong with the method, it was never intended to take everything into account, only to look at the league as a whole, using nothing other than data available from previous box scores prior to each game that is the simple linear combination of box score per 48 stats which best predicts the average minute allocations. I really don't care that you think coaches aren't that naive, I believe that the game has a few great coaches and a large number of mediocre coaches, but I didn't create this stat with any bias in mind, I just wanted a useful, simple equation. If you use it throughout the season to predict minutes for each game (with knowledge of player availability) it has around a 0.7 coefficient of determination, and that is in predicting, not retrodicting. I'm not looking for you to agree, but you can't accuse the numbers of lying because there is a place in your heart for the eye test.

Mike G wrote:Well, you're looking 'after the fact' at how these aggregate productions correlate with minutes.
I'd think it says nothing about whether the coaches are looking for these particular numbers and playing who they do because of such a stat.

You've also got nothing much representing defense in your stat. That's half the game, so right there you have some 30-40% of a coach's judgment which is not contained in the box score.
Of course, the one stat that really matters is point differential, and that's why we have +/- .

I may be looking after the fact but as I said above it is the case that no information from after the fact was used to predict the fact, only information available up to that point. And I'm not suggesting at all that coaches use a stat such as this, rather that they mostly trust their gut, not stats, because I don't believe anybody looking at the stats would produce this equation on purpose. Most come from an era before advanced stats and probably baulk at the idea that a computer could have told them that Kenneth Faried would be one of the best three picks of the draft last season despite ending up 22nd to the Nuggets, but it could have. I am not a plus minus person. It's useless in anything other than RAPM form, but even then I am not a big fan - obviously that is my choice, and I have my reasons. And yes, defence is under represented in the box score, that isn't news, as it turns out though it seems we can do a bloody good job (0.7 coefficient of determination as stated above, without even taking into account teammate ability) without it.

J.E. wrote:That's pretty cool. Are the weights that far off from what you would've expected? Maybe it would help if you removed teams that are tanking or developing rookies from your datapool? To see how coaches that care more about winning select MPG for their players. Also, fatigue is hard to quantify, but +/- and consisteny could prove helpful

I will run it using marginal production relative to teammates at some points so that players are rated relative to available replacements, but it takes about six hours to find coefficients to a good degree of precision so it may have to wait a bit. I was really hoping it would show that coaches value efficiency more than PER would suggest, but it doesn't look like it.

APBRmetrics

2012-13 eWins

2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins

Re: 2012-13 eWins