Toronto Raptors using SportsVU

[Mike G]

Is that a problem with WS/48 or a truism--when a player plays better, his team will be better?

You can swap 2 identical players between a good team and a bad team.
The teams will still be just as good/bad, and the players' WS/48 will flip.

Here's an example, not of identical players, but PG's Jason Williams and Mike Bibby, who were swapped in 2001/2002.

Code: Select all

year  JWill    tmW%    PER    WS/48
2001  Kings    .70    12.8    .079
2002  Grizz    .24    15.1    .020
            
year  Bibby    tmW%    PER    WS/48
2001  Grizz    .30    16.8    .081
2002  Kings    .74    14.9    .113

PER, for corroboration purposes, says the players were better (more productive?) with their poorer teams.
WS/48 says the opposite.
I've used pythagorean W% for the teams.

[deepak]

A given player will almost always have higher WS/48 with a better team.

Is that a problem with WS/48 or a truism--when a player plays better, his team will be better?

I think, in general, the issue here is that WS, being based on DRTG, is dependent on team defense. If a player goes to a significantly better or worse defensive team, his DRTG will correspondingly change.

Though for Rudy Gay's case, his drop in WS/48 can just as well be explained by his drop in offensive efficiency.

[xkonk]

I mentioned 'quick and dirty'. Any non-WS measures that could better inform the issue? Also, the team quality issue doesn't apply to Calderon, who appears to be playing better on a worse team.

[Mike G]

Win Shares will better predict players' and teams' win rates after this conversion:

WS2 = WS/48 + (.500 - pyth%)*.17
... where pyth% is the team's pythagorean W%, based on point differential.

This adjustment works to remove the 'exaggeration' of players' WS rate due to team excellence/suckiness.
For players on a .600 pyth% team, the formula suggests subtracting .017 from their WS/48.
For teams at the extremes, the adjustment is more profound: Players from last year's historically bad Bobcats get .063 added to their WS/48 rates.

DJ Augustin therefore goes from .018 to .081 -- from abysmal to respectable.
This year, he's listed at .113 in Indiana. But they're a .600 team, so he's "really" just a .096 player on a .500 team.

Has he improved by .095 WS/48, or by merely .015 ? The latter seems a lot more likely.

With this new WS/48 applied, players' WS will not add up to team wins (or pyth wins). You have to un-convert: Multiply by 1.17*(Pyth% - .500) to get "expected" team wins.

Of 223 players with at least 500 minutes both last year and this year, 60% are closer to last year's WS/48 after conversion.
The average change in WS/48 is .033, and after adjustment it's .027

[Mike G]

I did this a few weeks ago, so the 2013 team and player rates have changed. Still, for what it's worth, the 30 players with biggest changes in WS/48 rates, last year to this year. Minimum of 500 minutes each season.
Shown are team pythagorean W% , WS/48 rates, and "adjusted" WS/48 for both years. The final column is "improvement" in the change: Less measured change is generally "better"; and the vast majority are less changed after adjustment. Minus sign indicates a bigger change.

Code: Select all

.Win Shares        2011-12     2012-13     Win Shares/48   Adjusted WS/48 
.most changed     Tm   pyth%  Tm   pyth%     2012   2013    2012   2013    diff
Andray Blatche    WAS  .342   BRK  .515     -.011   .160    .016   .157    .029
Kevin Love        MIN  .426   MIN  .392      .223   .096    .236   .114    .006
Greg Stiemsma     BOS  .585   MIN  .392      .170   .044    .156   .062    .033
Toney Douglas     NYK  .602   HOU  .585     -.046   .075   -.063   .061   -.003
Kevin Seraphin    WAS  .342   WAS  .363      .094  -.014    .121   .009   -.004

Brendan Haywood   DAL  .529   CHA  .231      .114   .011    .109   .057    .051
Marvin Williams   ATL  .611   UTA  .487      .156   .053    .137   .055    .021
Will Bynum        DET  .338   DET  .440     -.023   .078    .005   .088    .017
D.J. Augustin     CHA  .128   IND  .600      .018   .113    .081   .096    .080
Andrea Bargnani   TOR  .384   TOR  .452      .104   .009    .124   .017   -.012

.Win Shares        2011-12     2012-13     Win Shares/48   Adjusted WS/48 
DeShawn Stevenson NJN  .307   ATL  .524     -.014   .080    .019   .076    .037
James Johnson     TOR  .384   SAC  .284      .079  -.014    .099   .023    .017
Larry Sanders     MIL  .506   MIL  .462      .072   .158    .071   .164   -.007
Joakim Noah       CHI  .754   CHI  .552      .223   .139    .180   .130    .034
Luc Mbah a Moute  MIL  .506   MIL  .462      .112   .031    .111   .037    .007

Kemba Walker      CHA  .128   CHA  .231      .009   .087    .072   .133    .018
Andre Iguodala    PHI  .641   DEN  .609      .158   .081    .134   .062    .005
Marcin Gortat     PHO  .489   PHO  .326      .172   .095    .174   .125    .028
Dirk Nowitzki     DAL  .529   DAL  .452      .175   .098    .170   .106    .013
Martell Webster   MIN  .426   WAS  .363      .064   .140    .077   .163   -.011

.Win Shares        2011-12     2012-13     Win Shares/48   Adjusted WS/48 
Ryan Anderson     ORL  .524   NOH  .396      .219   .145    .215   .163    .022
Gustavo Ayon      NOH  .368   ORL  .324      .142   .068    .164   .098    .007
Earl Clark        ORL  .524   LAL  .520      .027   .100    .023   .097   -.001
Spencer Hawes     PHI  .641   PHI  .391      .152   .081    .128   .100    .043
Kevin Durant      OKC  .684   OKC  .753      .230   .300    .199   .257    .012

Michael Beasley   MIN  .426   PHO  .326      .022  -.047    .035  -.017    .017
Avery Bradley     BOS  .585   BOS  .498      .085   .016    .071   .016    .015
Arron Afflalo     DEN  .586   ORL  .324      .121   .052    .106   .082    .045
Carlos Boozer     CHI  .754   CHI  .552      .187   .118    .144   .109    .034
Rodney Stuckey    DET  .338   DET  .440      .131   .063    .159   .073   -.017

On average, 2012 to 2013 WS/48 differences are lessened by .018 in this group.
Beasley is the only player of 223 who carries a negative Adj WS/48 this year. Only 8 are below .020

[xkonk]

Does it make sense to include players who haven't changed teams in an analysis like this? Kevin Love, for example, gets his rating bumped up because the Wolves are playing worse this year than last year. But certainly that's due in part to the fact that Love has missed time and played injured. Should he get credit for a decrease in team performance that is at least partly his 'fault'?

[Mike G]

Good question. After adjustment, the formula asserts that last year his .426 Wolves dragged his personal WS/48 down from .236 to .223. And this year, the .392 mark has depressed it from .114 to .096 .

Code: Select all

.Win Shares        2011-12     2012-13     Win Shares/48   Adjusted WS/48
.most changed     Tm   pyth%  Tm   pyth%     2012   2013    2012   2013    diff
Andray Blatche    WAS  .342   BRK  .515     -.011   .160    .016   .157    .029
Kevin Love        MIN  .426   MIN  .392      .223   .096    .236   .114    .006

The depressive effect of his weak team has increased from .013 to .018, last year to this year.
Or conversely, the adjustment is slightly bigger this year, to gauge what he'd be with a .500 team.

The .018 bump-up is a small rebate on the .127 dropoff WS/48 has measured.

The Nets, meanwhile have improved from .307 to .515 . Their average player, therefore, would be .061 last year and .103 this year. These are just 1/5 of the team pyth%

Deron Williams' WS/48 has gone from .099 to .140 . After adjustment, it's from .132 to .137
Kris Humphries' WS/48 has gone from .118 to .112. Adjusted, it's .151 to .109

Is DWill greatly improved and Humphries slightly diminished?
Or is DWill slightly up and Humphries way off his game?