Free Trial Ad
Why Subscribe?
  • Player/Prospect News
  • Exclusive Insider Info
  • Members-Only Forums
  • Exclusive Videos
  • Subscribe Now!
Inbox

Comparison of Win Shares and Player Efficiency Rating (PER)

Posted: 05/08/2012 10:27 PM

Comparison of Win Shares and Player Efficiency Rating (PER) 


I have been a big believer in Win Shares as the best measurement of a players overall contribution.  I have always felt that PER, the stat developed by John Hollinger of ESPN, over valued some players, and it was my gut instinct that it overvalued players with that dominated the ball on the offensive end.  I decided to do a study to determine the efficacy of each relative to the other.

What I did was first sort out the players who played minimal minutes, and players who were hurt and missed a substantial number of games, and also to separate out players traded to new teams mid-season.  For 2011-2012, I considered only players playing 33 games (1/2 the season), and a minimum of 10 minutes played per game.  I used the stats available from basketball-reference.com.

For the players that met both criteria (319 players) I then sorted players from best to worst in Win Shares per 48 minutes, and PER, with 1 being best 319 being the worst.  Best player using both PER and WS was LeBron, and the worst player using WS was Corey Higgins of Charlotte, and  using PER Jamal Magloire of Toronto.

My goal was to determine how each player was ranked using each methodology, and find the players where PER and WS differed in a meaningful way.

For instance here is how some various players compared in each methodology.
Player Per Rank WS Rank Differential
LeBron James 1 1 0
Chris Paul 2 2 0
Manu Ginobili 7 3 -4
Kevin Durant 4 4 0
James Harden 29 5 -24
Dwyane Wade 3 6 3
Kevin Love 5 7 2
Ryan Anderson 25 10 -15
Brandan Wright 22 11 -11
Kenneth Faried 18 12 -6
Derrick Rose 9 13 4
Tiago Splitter 32 14 -18
DeShawn Stevenson 317 307 -10
Jamaal Magloire 319 309 -10
Rasual Butler 315 310 -5
Tyrus Thomas 286 313 27
Nolan Smith 307 314 7
Lance Stephenson 310 315 5
Toney Douglas 308 316 8
Austin Daye 309 317 8
Matt Carroll 314 318 4
Cory Higgins 318 319 1

These types of players are where both tools measure the players in a similar way.  I arbitrarily chose to include 1/2 of the 319 players to fit within this category.  I then found the 25% of players where Win Shares rated a players contribution as being substantially greater than Per, and the 25% of where PER rated a players contribution as being substantially greater than WS.  Here is the top 10-12 for those two player sets.  The negative numbers are those players where WS ranked a player much higher than per, and those with a positive number is where PER ranked a player far higher than WS.

Player Per Rank WS Rank Differential
Joel Anthony 262 97 -165
James Jones 272 107 -165
Kyle Korver 172 29 -143
Greg Stiemsma 174 33 -141
Shane Battier 271 138 -133
Nick Collison 215 84 -131
Matt Bonner 188 64 -124
Jodie Meeks 229 106 -123
Jared Jeffries 250 129 -121
Mike Miller 236 120 -116
Udonis Haslem 249 133 -116
Thabo Sefolosha 270 154 -116
Jose Barea 126 237 111
Deron Williams 34 146 112
DeMarcus Cousins 21 136 115
Gerald Henderson 160 280 120
Klay Thompson 127 252 125
D.J. Augustin 158 291 133
John Wall 65 200 135
Tyreke Evans 89 225 136
Jordan Crawford 139 275 136
Chris Kaman 111 283 172
Kemba Walker 124 299 175
Monta Ellis 51 240 189

So to recap, I found 160 players where WS and PER tend to agree on a players contribution, both good players and bad players.  LeBron is great in WS and PER, and Jamal Magloire is bad using WS and PER.  Ignore those players, lets focus on the 160 players where WS and PER disagree.  For instance PER say's Joel Anthony is one of the 60 worst players in the league, but Win Shares says he is a top 100 player.  By the same token, PER says Monta Ellis in Golden State was a top 50 player, and Win Shares says he is 79th worst player in the league in 2011-12.

Now let me explain my logic in why I am doing what I am doing.  The point of either stat is to measure a players contribution, that is what the various proponent's of each claim.  I am assuming that better contribution should help your team win games.  That is the objective of playing the games after all, TO WIN GAMES.  So a measurement of a players contribution should help predict and measure team wins.

I sorted all players based upon their teams actual wins.  This is what I found;

Measurement Category Calculated Wins Winning %
Winning % for Players where WS is Good and Per is Bad 38.96 59.0%
Winning % for Players where WS and Per are consistent 33.72 51.1%
Winning % for Players where WS is Bad and Per is Good 24.72 37.5%

For players rated much higher in Win Shares than in PER, their actual team winning % is 59%, which over the 66 games season equals 39 wins.  In short for those players that WS said made a bigger contribution than PER, those players played on teams winning 60% of the time.

For the players where Win Shares and PER agree, on both good players and bad players the average team winning % is 51.1%, or 33.7 wins.  This makes perfect sense to me.  In the cases where both tools agree, good and bad, the winning % was essentially 50%.

For players rated much higher in PER than in Win Shares, their actual team winning % is 37.5%, which equals 24.7 wins.  For those players where PER said a player had a significant contribution, but WS said they didn't the winning % of those players was an abysmal 37.5%.

The difference between WS and PER winning % is an astounding 21.5% or 17.63 wins over an 82 games season.

I did the same thing for 2010-11 and here are the results based upon an 82 games season not a 66 games season.

Measurement Category Calculated Wins Winning %
Winning % for Players where WS is Good and Per is Bad 49.6 60.5%
Winning % for Players where WS and Per are consistent 39.8 48.5%
Winning % for Players where WS is Bad and Per is Good 31.7 38.7%

The results are almost identical compared to the 2011-12 results

I did it also for 2009-10, again on an 82 game basis.
Measurement Category Calculated Wins Winning %
Winning % for Players where WS is Good and Per is Bad 50.1 61.1%
Winning % for Players where WS and Per are consistent 40.6 49.5%
Winning % for Players where WS is Bad and Per is Good 30.7 37.4%
Once again the results are almost identical.


It is quite clear that Win Shares is a far better predictor of a players contribution to a teams wins than PER.

What statistical categories seem to define the difference

Measurement Category Tm Wins PER eFG% ORB% DRB% TRB% AST% STL% BLK% TOV% USG%
Winning % for Players where WS and Per are consistent 41 14.855 0.503 5.573 15.09 10.32 13.86 1.587 1.673 13.3 19.55
Winning % for Players where WS is Bad and Per is Good 31 15.219 0.491 4.703 13.41 8.988 17.44 1.628 1.438 13.67 22.72
Winning % for Players where WS is Good and Per is Bad 49 12.144 0.519 5.765 14.7 10.29 8.953 1.476 1.613 13.32 14.68
Differential between the 2 tools 36.73% 5.39% 18.42% 8.80% 12.64% -94.82% -10.30% 10.85% -2.62% -54.74%

Players who are High WS / Low PER
1.)  Have a higher eFG%.  This measurement iseFG% -- Effective Field Goal Percentage; this statistic adjusts for the fact that a 3-point field goal is worth one more point than a 2-point field goal, which basically is (points scored - free throws made) / shots taken.  These players FG% is 5.4% higher than High Per / Low WS players.
2.)  These players are better rebounders, particularly better offensive rebounders.  These players get 18.4% more offensive rebounds, and 12.6% more total rebounds.  The stats used are
ORB% -- Offensive Rebound Percentage; an estimate of the percentage of available offensive rebounds a player grabbed while he was on the floor.
DRB% -- Defensive Rebound Percentage; an estimate of the percentage of available defensive rebounds a player grabbed while he was on the floor.
TRB% -- Total Rebound Percentage; an estimate of the percentage of available rebounds a player grabbed while he was on the floor.
3.)  These players tend to get more blocks, at a rate of 10.85% higher.  This stat is BLK% -- Block Percentage; an estimate of the percentage of opponent two-point field goal attempts blocked by the player while he was on the floor.


Players who are High PER / Low WS
1.)  ...get assists at almost twice the rate of High WS / Low PER.  The measurement I am using is AST% -- Assist Percentage; an estimate of the percentage of teammate field goals a player assisted while he was on the floor".
2.)  ... dominate a team offense, based upon Usage % which is USG% -- Usage Percentage; an estimate of the percentage of team plays used by a player while he was on the floor.
3.)  ...  get steals at a rate 10.3% higher than their counterpart, based upon STL% -- Steal Percentage; an estimate of the percentage of opponent possessions that end with a steal by the player while he was on the floor.

PER gives far more credence to players who dominate the ball, and tend to shoot the ball a lot, or set up other players, while shooting a low %.  The most obvious examples would be Monta Ellis, Tyreke Evans, Kemba Walker, John Wall, or Jamal Crawford, and/or poor rebounding big men who don't shoot particularly well, such as Chris Kamen and Demarcus Cousins.

WS gives far more credit to players who are very efficient shooters, ie take shots they have a high probability of making, and as such do not shoot as often, are very good 3 point shooters, and players who rebound very well.  Some examples would be
James Jones, Kyle Korver, Shane Battier, and Mike Miller, and/or Joel Anthony, Nick Collison, Udonis Haslem, Zaza Pachulia.

Here is Portland's Player stats for 2011-12, sorted by the players with the greatest to worst WS to PER differential.

Player Tm MP MPG PER TS% eFG% ORB% DRB% TRB% AST% STL% BLK% TOV% USG% ORtg DRtg OWS DWS WS WS/48 Per Rank WS Rank Differential
415 Kurt Thomas POR 803 15.2 9.5 0.483 0.465 5.5 22.1 13.7 8.7 1.6 3 15.8 11.2 99 104 0.2 0.9 1.1 0.067 275 211 -64
295 Wesley Matthews POR 2228 33.8 14.1 0.539 0.496 2.9 9 5.9 8.3 2.3 0.5 8 18.7 110 108 3.4 1.4 4.8 0.104 156 135 -21
25 Luke Babbitt POR 537 13.4 11.5 0.558 0.535 3.4 17.5 10.3 4.8 1 0.6 12.9 17.9 104 108 0.4 0.3 0.8 0.067 233 212 -21
77 Marcus Camby POR 894 22.4 14.5 0.428 0.416 13.4 33.2 23.1 12.4 1.9 4.8 20.9 11.5 100 98 0.3 1.6 1.9 0.1 141 145 4
399 Nolan Smith POR 541 12.3 7.8 0.434 0.408 3.6 8.6 6 18.2 1.8 0.4 17.9 19.8 88 109 -0.6 0.3 -0.3 -0.029 307 314 7
35 Nicolas Batum POR 1791 30.4 17.3 0.575 0.534 5.2 12.6 8.8 7.9 1.7 2.6 11.2 20.5 112 107 3.3 1.4 4.8 0.127 75 88 13
446 Gerald Wallace POR 1503 35.8 15.5 0.556 0.502 4.8 17 10.7 12.4 2.2 1.3 14.4 17.8 108 105 2 1.4 3.5 0.111 104 124 20
6 LaMarcus Aldridge POR 1994 36.3 22.7 0.56 0.513 8.6 17.5 12.9 13.2 1.3 1.7 9.5 27 113 106 5.4 1.6 7 0.169 13 34 21
141 Raymond Felton POR 1906 31.8 13.4 0.491 0.455 1.6 7.6 4.6 33.3 2.2 0.4 19.6 20.8 99 109 0.7 1 1.7 0.042 179 261 82
97 Jamal Crawford POR 1613 26.9 15.7 0.506 0.438 1.1 7.5 4.3 20.5 1.8 0.7 11.8 26.6 102 109 1.6 0.7 2.3 0.069 101 206 105


I hope you find this interesting and helpful.

Last edited 05/08/2012 10:34 PM by ziggythebeagle

Reply | Quote