Friday, 25 July 2014

Puckerings archive: Neutral Winning Percentage (27 Jul 2001)

What follows is a post from my old hockey analysis site (later It is reproduced here for posterity; bear in mind this writing is over a decade old and I may not even agree with it myself anymore. This post was originally published on July 27, 2001 and was updated on April 9, 2002.

Neutral Winning Percentage
Copyright Iain Fyffe, 2002

This essay describes a method for evaluating goaltenders which is, in theory, free from bias created by the team the goalie plays for and the teams he faces. It developed from an idea proposed by Marc Foster, president of the Hockey Research Association. This idea was to adapt Michael Wolverton's support-neutral pitcher records for used in hockey.

Support-neutral records remove the effect of team offence from a pitcher’s won-lost record, thereby producing a result that more fairly evaluates a pitcher’s performance. Foster proposed to adapt this method to calculate offence-neutral records for goaltenders, as follows.

Each goaltender’s season is broken down by the number of times which he allows zero goals in a game, one goal in a game, two goals, and so on. The records for all teams in the league are then broken down based on the number of goals they allowed in each game. We then use these league numbers to compute an expected record for the goaltender, based on the number of times he allowed each number of goals against.

For instance, the breakdown by goals against for the 2000/01 NHL season is as follows:

 GA  #  W  W%  L  L%  T  T%
 0  186  172  .92  0  .00  14  .08
 1  406  316  .78  28  .07  62  .15
 2  595  340  .57  141  .24  114  .19
 3  520  171  .33  275  .53  74  .14
 4  365  58  .16  275  .75  32  .09
 5  228  16  .07  206  .90  6  .03
 6  111  5  .05  104  .94  2  .01
 7+  49  0  .00  49  .00  0  .00
 2460  1078  1078  304

So, if a goalie allows 2 goals in a game, we expect him to win 57% of the time, lose 24% of the time, and tie 19%. We therefore give him credit for 0.57 wins, 0.24 losses, and 0.19 ties for each game in which he allows 2 goals. We multiply the number of games in which he allows each number of goals by the appropriate factors for wins, losses and ties. Adding the results gives us a won-lost-tied record. We then convert the record into a winning percentage, so we can compare goaltenders directly.

This winning percentage is offence-neutral. The number of goals the goaltender’s team scores has no effect upon the percentage. The bias resulting from playing for a high- or low-scoring team is eliminated.

Foster’s idea, as presented above, is a good first step. But we can go further and remove even more team-related bias from a goaltender’s record. This is, in fact, the point of this exercise: to remove all team effects from a goalie’s record, so we can evaluate the goalie based solely upon his efforts.

We have already eliminated the distortion caused by team offence. Two types of distortion remain: distortion from team defence, and distortion from opponent’s offence. Fortunately, both of these can be compensated for.

Distortion from team defence

In the essay entitled “Goaltender Perseverance: a Useless Stat”, I demonstrate that the number of shots a goalie faces is a function of his team. Thus, a goaltender who faces a large number of shots is being unfairly penalized for playing for his particular team. This distortion must be removed to effectively compare goaltenders.

To control for this distortion, we should not evaluate a goalie based on the actual number of goals he allows, but on the number of goals he would allow when facing an average number of shots. In this was, we remove the bias resulting from facing a high or low number of shots.

Unfortunately, some distortion will remain. Team defence affects not only the number of shots faced, but the quality of shots faced. However, this distortion cannot be removed because we cannot determine the effect it has on a goalie’s save percentage. Still, this distortion is present in all methods currently used for evaluating goaltenders (including save percentage), so even if it is present in Neutral Winning Percentage, this method is still an improvement.

Distortion from opponent's offence

The idea here is basically the same as Keith Woolner’s Pitcher’s Quality of Opposition. If a goalie faces teams who are more better shooters (that is, have higher scoring percentages), he will give more goals, even when his shots against are normalized.

A goaltender’s adjusted goals against in a game (based on average shots) should therefore be further adjusted, depending on the shooting percentage of the team faced. When facing a team with an above-average shooting percentage, the goals against would be adjusted downward, reflecting the greater challenge faced. Conversely, a low-shooting-percentage team produces an upward adjustment.

Application of the method

For each game, the goaltender’s actual save percentage for that game is applied to an adjusted number of shots to produce an adjusted number of goals against. This goals against figure is rounded to the nearest whole number, since it is impossible to allow 2.3 goals in a game (for example).

The adjusted shots against is calculated thusly:

LgShotsPerGame x (LgShootPct/OppShootPct)


LgShotsPerGame is the total shots in the league divided by the total games played in the league
LgShootPct is the total goals in the league divided by the total shots in the league
OppShootPct is the opponent’s goals divided by the opponent’s shots

Each level of adjusted goals against (0,1,2...) is compiled for each goaltender, and the record is then computed as described earlier.

One further thing needs discussion: what to do when a goalie does not play a full game. In such a case, the same process is used, but instead of receiving credit for a full game, the goalie receives credit for whatever portion of the opponent's shots he faced. For instance, if a goaltender faces half the shots in a game, the resulting wins, losses and ties for that game are each multiplied by one-half.

2000/01 NHL results

I computed the Neutral Winning Percentage (NWP) for all goalies for the 2000/01 NHL season. Complete results follow. For discussion purposes, here are the leaders and trailers in NWP (minimum 30 ‘decisions’):

 Leaders  Trailers
 1.  Manny Fernandez  .611  1.  Dan Cloutier  .430
 2.  Mike Dunham  .610  2.  J.S.Aubin  .437
 3.  Dominik Hasek  .604  3.  Mike Vernon  .449
 4.  Yevgeny Nabokov  .599  4.  Trevor Kidd  .451
 5.  Sean Burke  .594  5.  Guy Hebert  .452
 6.  Roman Cechmanek  .593  Mike Richter  .452
 Roberto Luongo  .593  7.  Bob Essensa  .456
 8.  Manny Legace  .566  8.  Marc Denis  .456
 9.  Ron Tugnutt  .556  9.  Jocelyn Thibault  .472
 10.  Patrick Lalime  .554  10.  Damian Rhodes  .482
 Patrick Roy  .554

So, even though Fernandez and Dunham were marginally better, Hasek was a good choice as Vezina winner. It remains to be seen whether the young goalies, such as Fernandez, Dunham, Nabokov, Luongo and Legace, will be able to maintain their high levels of performance. Players like Vernon and Richter seem to be surviving on reputation alone.

One of the great advantages of this method is that numbers from one year to the next are directly comparable. Since NWP is based on the averages for that season, we need make no further adjustments to compare different seasons. A .600 NWP is just as impressive in 2000/01 as it is in 1990/91 or 1980/81.

2000/01 NHL Neutral Winning Percentages
 Goalie  Team(s)  Dec  NWP  Goalie  Team(s)  Dec  NWP
 Aebischer  Col  23.0  .489  Kochan  TB  5.1  .422
 Aubin  Pgh  34.2  .437  Kolzig  Wsh  70.8  .545
 Belfour  Dal  60.9  .541  LaBarbera  NYR  0.1  1.000
 Bierk  Min  1.0  .100  Lalime  Ott  59.7  .554
 Billington  Wsh  10.9  .583  Larocque  Chi  2.5  .240
 Biron  Buf  15.3  .529  Legace  Det  35.0  .566
 Boucher  Phi  24.8  .355  Luongo  Fla  43.4  .593
 Brathwaite  Cgy  44.5  .528  Maracle  Atl  12.5  .448
 Brodeur  NJ  69.8  .527  Mason  Nsh  0.9  .389
 Burke  Phx  59.8  .594  McLean  NYR  20.3  .451
 Cechmanek  Phi  56.4  .593  McLennan  Min  36.4  .533
 Cloutier  TB-Van  31.5  .430  Moss  Car  9.2  .250
 Dafoe  Bos  41.4  .545  Nabokov  SJ  60.9  .599
 Denis  Clb  30.2  .465  Naumenko  Ana  1.2  .083
 Dipietro  NYI  18.2  .390  Noronen  Buf  1.9  .395
 Dunham  Nsh  46.5  .610  Osgood  Det  46.9  .504
 Esche  Phx  22.2  .471  Ouellet  Phi  1.3  .538
 Essensa  Van  34.0  .456  Parent  Pgh  5.3  .509
 Fankhouser  Atl  4.3  .535  Passmore  LA-Chi  17.6  .497
 Fernandez  Min  40.5  .611  Potvin  Van-LA  57.3  .496
 Fichaud  Mtl  1.0  .550  Raycroft  Bos  10.8  .454
 Fiset  LA  5.3  .377  Rhodes  Atl  34.0  .482
 Flaherty  NYI-TB  18.5  .389  Richter  NYR  43.8  .452
 Fountain  Ott  0.9  .389  Roussel  Ana-Edm  16.6  .434
 Gage  Edm  4.4  .307  Roy  Col  58.9  .554
 Garon  Mtl  9.2  .505  Salo  Edm  71.7  .507
 Giguere  Ana  33.5  .534  Scott  LA  0.4  .000
 Grahame  Bos  7.9  .418  Shields  SJ  18.6  .548
 Gustafson  Min  4.0  .463  Skudra  Bos  18.5  .384
 Hackett  Mtl  16.3  .387  Snow  Pgh  33.5  .525
 Hasek  Buf  64.2  .604  Tallas  Chi  10.5  .405
 Healy  Tor  14.3  .339  Terreri  NJ-NYI  14.7  .422
 Hebert  Ana-NYR  49.8  .452  Theodore  Mtl  54.9  .530
 Hedberg  Pgh  9.1  .560  Thibault  Chi  63.4  .472
 Hirsch  Wsh  0.3  1.000  Tugnutt  Clb  51.8  .556
 Hnilicka  Atl  31.0  .495  Turco  Dal  21.1  .609
 Holmqvist  NYR  2.0  .300  Turek  StL  53.1  .505
 Hurme  Ott  21.3  .491  Vanbiesbrouck  NYI-NJ  43.4  .486
 Irbe  Car  72.7  .545  Vernon  Cgy  37.1  .449
 Johnson  StL  28.0  .546  Vokoun  Nsh  34.6  .542
 Joseph  Tor  67.6  .538  Weekes  TB  55.5  .509
 Khabibulin  TB  2.0  .525  Whitmore  Bos  3.6  .139
 Kidd  Fla  38.7  .451  Yeremeyev  NYR  3.6  .208
 Kiprusoff  SJ  2.6  .558

No comments:

Post a Comment

Hostgator promo codes