NHL Team Power Ratings and Predictions (SRS Algorithm)

[Doctor No]

Here are the updated SRS rankings, using NHL games through March 15, 2015:

Team | 2014-15 Goal Diff | 2014-15 Schedule | 2014-15 Strength
NY Rangers|

0.721​

|

-0.052​

|

0.669​

|
St. Louis|

0.638​

|

-0.022​

|

0.616​

|
Chicago|

0.603​

|

0.010​

|

0.613​

|
Tampa Bay|

0.629​

|

-0.063​

|

0.566​

|
Nashville|

0.423​

|

0.020​

|

0.443​

|
Washington|

0.471​

|

-0.048​

|

0.423​

|
Montreal|

0.420​

|

-0.034​

|

0.386​

|
Pittsburgh|

0.333​

|

0.043​

|

0.376​

|
NY Islanders|

0.310​

|

0.019​

|

0.328​

|
Minnesota|

0.338​

|

-0.028​

|

0.310​

|
Calgary|

0.348​

|

-0.040​

|

0.308​

|
Detroit|

0.250​

|

-0.043​

|

0.207​

|
Anaheim|

0.183​

|

-0.004​

|

0.180​

|
Los Angeles|

0.191​

|

-0.020​

|

0.171​

|
Boston|

0.159​

|

-0.013​

|

0.147​

|
Ottawa|

0.176​

|

-0.032​

|

0.145​

|
Vancouver|

0.132​

|

-0.011​

|

0.121​

|
Winnipeg|

0.072​

|

0.024​

|

0.096​

|
Colorado|

-0.116​

|

0.102​

|

-0.013​

|
San Jose|

-0.015​

|

-0.004​

|

-0.019​

|
Dallas|

-0.157​

|

0.060​

|

-0.097​

|
Philadelphia|

-0.225​

|

0.001​

|

-0.224​

|
New Jersey|

-0.304​

|

0.015​

|

-0.289​

|
Florida|

-0.319​

|

-0.029​

|

-0.348​

|
Carolina|

-0.382​

|

0.026​

|

-0.356​

|
Toronto|

-0.500​

|

0.045​

|

-0.455​

|
Columbus|

-0.551​

|

0.032​

|

-0.519​

|
Edmonton|

-1.130​

|

0.057​

|

-1.074​

|
Arizona|

-1.232​

|

0.063​

|

-1.169​

|
Buffalo|

-1.485​

|

0.099​

|

-1.386​

|

Home-ice advantage is currently measured at +0.238 goals/game (down a fair amount thanks to recent results).

 

[Doctor No]

Updated SRS rankings, through games of March 22, 2015:

Team | 2014-15 Goal Diff | 2014-15 Schedule | 2014-15 Strength
NY Rangers|

0.761​

|

-0.052​

|

0.708​

|
Tampa Bay|

0.685​

|

-0.059​

|

0.626​

|
Chicago|

0.577​

|

0.031​

|

0.609​

|
St. Louis|

0.589​

|

0.004​

|

0.593​

|
Nashville|

0.438​

|

0.006​

|

0.444​

|
Montreal|

0.466​

|

-0.038​

|

0.428​

|
Washington|

0.438​

|

-0.054​

|

0.385​

|
Minnesota|

0.366​

|

-0.001​

|

0.366​

|
Pittsburgh|

0.306​

|

0.026​

|

0.332​

|
NY Islanders|

0.301​

|

0.024​

|

0.325​

|
Calgary|

0.306​

|

-0.048​

|

0.257​

|
Winnipeg|

0.167​

|

0.029​

|

0.196​

|
Ottawa|

0.239​

|

-0.045​

|

0.195​

|
Detroit|

0.197​

|

-0.028​

|

0.169​

|
Vancouver|

0.181​

|

-0.041​

|

0.140​

|
Anaheim|

0.135​

|

0.004​

|

0.139​

|
Los Angeles|

0.141​

|

-0.033​

|

0.108​

|
Boston|

0.068​

|

-0.018​

|

0.051​

|
Colorado|

-0.085​

|

0.095​

|

0.010​

|
Dallas|

-0.083​

|

0.071​

|

-0.013​

|
San Jose|

-0.042​

|

0.001​

|

-0.041​

|
New Jersey|

-0.278​

|

-0.002​

|

-0.280​

|
Florida|

-0.278​

|

-0.024​

|

-0.302​

|
Philadelphia|

-0.311​

|

0.003​

|

-0.308​

|
Carolina|

-0.451​

|

0.045​

|

-0.405​

|
Columbus|

-0.444​

|

0.020​

|

-0.425​

|
Toronto|

-0.589​

|

0.032​

|

-0.557​

|
Edmonton|

-1.042​

|

0.027​

|

-1.015​

|
Arizona|

-1.274​

|

0.068​

|

-1.206​

|
Buffalo|

-1.472​

|

0.104​

|

-1.368​

|

Home-ice advantage is currently measured at +0.251 goals/game.

SRS algorithm really doesn't like Anaheim.

 

[Moops]

Home-ice advantage is currently measured at +0.251 goals/game.

What is it over a sample of multiple seasons? Say, last five years, just for instance.

 

[Doctor No]

Here are all of the seasons for which I currently have game log data:

SEASON|HIA
1978-79|0.669
1979-80|0.650
1980-81|0.728
1981-82|0.831
1982-83|0.856
1983-84|0.599
1984-85|0.571
1985-86|0.523
1986-87|0.521
1987-88|0.516
1988-89|0.453
1989-90|0.541
1990-91|0.589
1991-92|0.694
1992-93|0.434
1993-94|0.337
1994-95|0.394
1995-96|0.359
1996-97|0.271
1997-98|0.179
1998-99|0.265
1999-00|0.221
2000-01|0.244
2001-02|0.280
2002-03|0.272
2003-04|0.255
2005-06|0.301
2006-07|0.317
2007-08|0.265
2008-09|0.339
2009-10|0.327
2010-11|0.195
2011-12|0.286
2012-13|0.331
2013-14|0.273
2014-15|0.247

2014-15 is through last night (so slightly different from the above number). It bounces around a bit from season to season, but it's roughly proportional to the season's goal-scoring environment.

 

[Moops]

Wow. So the Hawks are just the wire to wire runaway best team in this model?

9 standings points off the President's trophy and only 2 goals ahead in GD.

I just wonder what the model loves so much about the Hawks to be 140 points clear of the league.

This is only some back-of-the-envelope, tangential BS on my part, but .144*82 games comes out to 11.8 goals, which is an improvement of about 2 wins or 4 points over an 82 game schedule. It's worth about 1 goal in a best-of-seven. It's not nothing, but it really goes to show how tightly bunched the teams are, even at the top.

 

[Moops]

How does SRS translate to win probability?

 

[Doctor No]

How does SRS translate to win probability?

Terrific question - most directly, SRS translates to goal relationships (not to win relationships). With that said, it's natural to want to go from one to the other.

When I do this, I typically (*) look at the variance of each team's performance. If each team has an expectation and a variance, and we assume a performance distribution (I prefer normal), then the difference between the two teams' performances will also have a distribution (the difference between normal distributions is normal), and then you can ask questions like "when is this difference greater than zero?" (so, the probability that one team wins).

These vary from team to team and it's related to the goal scoring environment. To give extreme examples on both ends, if two teams were playing and both had zero variance, then the team with the higher SRS would win 100% of the time. If two teams were playing and both had infinite variance, then each team would win 50% of the time no matter the SRS values. (Obviously, the correct answer is somewhere in between).

Complicating matters is the existence of three-point games in the NHL, which I haven't studied in enough detail to get something that predicts very well. As a first order approximation, I typically look at the proportion of games where each team has a third point. More directly, I don't care as much since this stuff is more interesting to me (1) to evaluate aspects of goaltenders - such as their schedules, and (2) in the playoffs - where three-point games don't matter.

(*) "Typically" means that I haven't done this here yet, but will when I get a chance (at least before the playoffs).

 

[Doctor No]

SRS rankings, NHL games through March 29, 2015:

Team | 2014-15 Goal Diff | 2014-15 Schedule | 2014-15 Strength
NY Rangers|

0.680​

|

-0.039​

|

0.641​

|
St. Louis|

0.560​

|

0.006​

|

0.566​

|
Tampa Bay|

0.605​

|

-0.053​

|

0.552​

|
Chicago|

0.507​

|

0.026​

|

0.533​

|
Nashville|

0.416​

|

0.030​

|

0.445​

|
Minnesota|

0.440​

|

0.000​

|

0.440​

|
Washington|

0.461​

|

-0.043​

|

0.418​

|
Montreal|

0.408​

|

-0.031​

|

0.377​

|
NY Islanders|

0.260​

|

0.022​

|

0.282​

|
Calgary|

0.303​

|

-0.031​

|

0.272​

|
Pittsburgh|

0.263​

|

0.007​

|

0.270​

|
Winnipeg|

0.184​

|

0.031​

|

0.215​

|
Los Angeles|

0.160​

|

-0.007​

|

0.153​

|
Anaheim|

0.141​

|

0.010​

|

0.151​

|
Detroit|

0.187​

|

-0.039​

|

0.148​

|
Ottawa|

0.173​

|

-0.047​

|

0.127​

|
Vancouver|

0.160​

|

-0.035​

|

0.125​

|
Boston|

0.092​

|

-0.020​

|

0.072​

|
Colorado|

-0.040​

|

0.066​

|

0.026​

|
San Jose|

-0.053​

|

0.015​

|

-0.038​

|
Dallas|

-0.092​

|

0.047​

|

-0.045​

|
Florida|

-0.224​

|

-0.015​

|

-0.239​

|
Philadelphia|

-0.276​

|

0.007​

|

-0.270​

|
Columbus|

-0.333​

|

0.035​

|

-0.299​

|
New Jersey|

-0.342​

|

0.003​

|

-0.339​

|
Carolina|

-0.400​

|

0.040​

|

-0.360​

|
Toronto|

-0.605​

|

0.026​

|

-0.580​

|
Edmonton|

-0.973​

|

0.024​

|

-0.949​

|
Arizona|

-1.211​

|

0.062​

|

-1.149​

|
Buffalo|

-1.467​

|

0.081​

|

-1.386​

|

Home ice advantage is currently estimated at +0.237 goals/game.

 

[Sharppi]

4/6 in top-6 from Central.

 

[Doctor No]

For quite awhile (posts #23 to #35 inclusive), the top three were all from the Central.

It makes an Avalanche fan weep.

 

[Doctor No]

"Typically" means that I haven't done this here yet, but will when I get a chance (at least before the playoffs).

Just doing some farting around while watching Avs/Oilers, and here are some predictions based on these data. Admittedly, these are for tonight's games already in progress, but I'll do these daily in the mornings when I have time.

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
EDM|COL|1.21|72%|28%
LAK|CHI|0.62|61%|23%
CAL|DAL|-0.08|48%|24%
BUF|ARI|0.47|59%|23%
TBL|MTL|0.06|51%|19%
VAN|STL|0.68|62%|22%

(Total home wins predicted: 3.5, total third points predicted: 1.4)

E(Home) = Number of goals by which the home team is favored

Pr(H>0) = Probability of the home team winning the game

Pr(3PG) = Probability that the game involves OT or a SO (with the losing team receiving one point).

The last factor is the fudgiest at the moment.

 

[Doctor No]

And for tomorrow (March 31st):

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
FLO|BOS|0.54|61%|32%
NJD|CBS|0.27|55%|21%
CAR|WAS|1.01|69%|23%
TBL|TOR|-0.93|34%|16%
OTT|DET|0.25|55%|27%
NYR|WIN|-0.20|46%|25%
VAN|NAS|0.50|59%|24%

 

[Doctor No]

For games of April 1st:

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
TOR|BUF|-0.57|40%|19%
PHI|PIT|0.77|64%|29%
EDM|ANA|1.28|73%|28%
COL|SJS|0.22|54%|25%

 

[Doctor No]

For NHL games of April 2nd:

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
NYI|CBS|-0.34|44%|22%
BOS|DET|0.29|56%|31%
WAS|MTL|0.15|53%|24%
TBL|OTT|-0.17|47%|20%
CAR|FLO|0.36|57%|25%
NYR|MIN|0.02|50%|19%
CAL|STL|0.45|59%|25%
VAN|CHI|0.60|61%|22%
EDM|LAK|1.29|74%|24%

I made a small tweak to the variance calculation, since I realized that I could explain some of the deviation. The result is a (slightly) lower standard deviation calculation, which increases the predictability.

 

[seventieslord]

Look at Anaheim in 1st overall, yet very mediocre by goal differential.

Is this unprecedented?

 

[Doctor No]

For NHL games of April 2nd:

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
CHI|BUF|-1.65|20%|22%
MTL|NJD|-0.45|41%|23%
STL|DAL|-0.38|43%|23%
COL|ANA|0.51|60%|29%
ARI|SJS|1.45|76%|22%

 

[Doctor No]

Look at Anaheim in 1st overall, yet very mediocre by goal differential.

Is this unprecedented?

It certainly seems unprecedented - at least in recent history. It's not apples to apples, because my database has "full" SRS numbers for each season - so they're through the playoffs, and don't stop at the end of the regular season.

With that said, here are the final SRS numbers for each of the Presidents' Trophy winners in recent history (basically, back through what I have in my goalie database):

Season | Team | SRS
2013-14|Boston|+0.847
2012-13|Chicago|+0.981
2011-12|Vancouver|+0.508
2010-11|Vancouver|+0.647
2009-10|Washington|+0.823
2008-09|San Jose|+0.471
2007-08|Detroit|+1.083
2006-07|Buffalo|+0.598
2005-06|Detroit|+0.923
2003-04|Detroit|+0.670
2002-03|Ottawa|+0.844
2001-02|Detroit|+0.954
2000-01|Colorado|+1.099
1999-00|St. Louis|+1.001
1998-99|Dallas|+0.820
1997-98|Dallas|+0.793
1996-97|Colorado|+0.880
1995-96|Detroit|+1.465
1994-95|Detroit|+1.219
1993-94|NY Rangers|+0.975
1992-93|Pittsburgh|+1.218
1991-92|NY Rangers|+0.923
1990-91|Chicago|+0.696
1989-90|Boston|+0.580
1988-89|Calgary|+1.542
1987-88|Calgary|+1.052
1986-87|Edmonton|+1.147
1985-86|Edmonton|+1.221
1984-85|Philadelphia|+1.129
1983-84|Edmonton|+1.594
1982-83|Boston|+1.078
1981-82|NY Islanders|+1.651
1980-81|NY Islanders|+1.393
1979-80|Philadelphia|+0.992
1978-79|NY Islanders|+1.746

Right now, Anaheim's at +0.187, so if they win the Presidents' Trophy, it would far and away be the lowest schedule-adjusted goal differential (noting that they could roll off a few blowout wins the rest of the regular season, and also that since these are apples-to-pears, if they have a strong playoff run, their SRS will increase).

 

[Doctor No]

Some good games today, at least according to the algorithm:

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
PIT|PHI|-0.18|47%|29%
MTL|FLO|-0.30|44%|27%
WAS|DET|-0.04|49%|29%
OTT|TOR|-0.49|41%|22%
STL|CHI|0.25|55%|24%

 

[Doctor No]

NHL games through April 5th, 2015:

Team | 2014-15 Goal Diff | 2014-15 Schedule | 2014-15 Strength
NY Rangers|

0.744​

|

-0.035​

|

0.709​

|
St. Louis|

0.570​

|

0.011​

|

0.581​

|
Tampa Bay|

0.613​

|

-0.046​

|

0.567​

|
Chicago|

0.544​

|

0.010​

|

0.554​

|
Washington|

0.475​

|

-0.034​

|

0.441​

|
Minnesota|

0.403​

|

0.004​

|

0.406​

|
Nashville|

0.380​

|

0.026​

|

0.405​

|
Montreal|

0.375​

|

-0.021​

|

0.354​

|
Calgary|

0.329​

|

-0.033​

|

0.296​

|
NY Islanders|

0.278​

|

0.001​

|

0.280​

|
Winnipeg|

0.179​

|

0.033​

|

0.212​

|
Los Angeles|

0.218​

|

-0.017​

|

0.201​

|
Anaheim|

0.163​

|

-0.005​

|

0.158​

|
Vancouver|

0.165​

|

-0.007​

|

0.157​

|
Pittsburgh|

0.152​

|

0.003​

|

0.155​

|
Ottawa|

0.190​

|

-0.036​

|

0.154​

|
Detroit|

0.152​

|

-0.020​

|

0.132​

|
Boston|

0.127​

|

-0.028​

|

0.099​

|
San Jose|

0.000​

|

-0.015​

|

-0.015​

|
Colorado|

-0.127​

|

0.057​

|

-0.069​

|
Dallas|

-0.127​

|

0.056​

|

-0.070​

|
Philadelphia|

-0.203​

|

0.009​

|

-0.194​

|
Columbus|

-0.269​

|

0.030​

|

-0.240​

|
Florida|

-0.250​

|

-0.005​

|

-0.255​

|
New Jersey|

-0.392​

|

0.015​

|

-0.377​

|
Carolina|

-0.462​

|

0.044​

|

-0.418​

|
Toronto|

-0.563​

|

0.017​

|

-0.545​

|
Edmonton|

-1.063​

|

0.037​

|

-1.026​

|
Arizona|

-1.203​

|

0.039​

|

-1.164​

|
Buffalo|

-1.392​

|

0.067​

|

-1.326​

|

Home-ice advantage: +0.243 goals/game.

Correlations:

Opp Strength | GAA | Sv% | S/60 | Min
-1.0|2.06|0.928|28.5|7,235
-0.5|2.29|0.918|28.0|16,797
+0.0|2.56|0.913|29.6|45,056
+0.5|2.81|0.909|30.8|36,112
+1.0|2.93|0.908|31.7|2,354
R^2|97%|85%|85%|

And today's games:

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
CAR|BUF|-0.66|36%|20%
CBS|NYR|1.19|71%|21%
WIN|MIN|0.44|58%|25%
LAK|VAN|0.20|54%|20%
DAL|SJS|0.30|55%|21%

 

[Doctor No]

Today's games (April 7th):

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
NYI|PHI|-0.23|45%|28%
NYR|NJD|-0.85|34%|22%
CAR|DET|0.81|66%|25%
PIT|OTT|0.24|55%|27%
WIN|STL|0.58|61%|27%
MIN|CHI|0.42|58%|21%
ARI|CAL|1.70|80%|23%
NAS|COL|-0.23|45%|29%
LAK|EDM|-0.98|31%|24%

 

[Doctor No]

Right now, Anaheim's at +0.187, so if they win the Presidents' Trophy, it would far and away be the lowest schedule-adjusted goal differential

Well, I guess that's that. Rangers win the Presidents' Trophy, and will likely end the regular season with the highest power rating as well.

 

[Doctor No]

Today:

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
TOR|CBS|0.55|60%|21%
BOS|WAS|0.58|61%|30%
DAL|ANA|0.42|58%|26%

 

[Doctor No]

Games of April 9:

Road | Home | E(Home) | Pr(H>0) | Pr(3PG)
WIN|COL|-0.07|49%|28%
OTT|NYR|0.80|66%|24%
CAR|PHI|0.51|60%|26%
BOS|FLO|-0.08|48%|31%
DET|MTL|0.46|59%|28%
NJD|TBL|1.20|72%|18%
CHI|STL|0.27|55%|23%
MIN|NAS|0.24|55%|25%
LAK|CAL|0.38|57%|23%
SJS|EDM|-0.68|37%|22%
ARI|VAN|1.57|77%|21%

 

[Doctor No]

And here are what the algorithm currently predicts for the first round (I will finalize these once matchups are set and all regular season games are in the book):

Outcome | Prob
MTL in 4 games|8.3%
MTL in 5 games|16.7%
MTL in 6 games|16.8%
MTL in 7 games | 17.9%
PIT in 7 games|12.9%
PIT in 6 games|14.0%
PIT in 5 games|9.0%
PIT in 4 games|4.5%
MTL wins | 59.6%
PIT wins|40.4%

Outcome | Prob
TBL in 4 games|11.6%
TBL in 5 games | 20.8%
TBL in 6 games|18.4%
TBL in 7 games|18.1%
DET in 7 games|10.7%
DET in 6 games|11.1%
DET in 5 games|6.4%
DET in 4 games|3.0%
TBL wins | 68.9%
DET wins|31.1%

Outcome | Prob
NYR in 4 games|15.2%
NYR in 5 games | 24.7%
NYR in 6 games|19.9%
NYR in 7 games|17.2%
BOS in 7 games|8.6%
BOS in 6 games|8.2%
BOS in 5 games|4.3%
BOS in 4 games|1.9%
NYR wins | 77.0%
BOS wins|23.0%

Outcome | Prob
WAS in 4 games|8.4%
WAS in 5 games|16.9%
WAS in 6 games|16.8%
WAS in 7 games | 18.0%
NYI in 7 games|12.8%
NYI in 6 games|14.2%
NYI in 5 games|8.6%
NYI in 4 games|4.3%
WAS wins | 60.1%
NYI wins|39.9%

Outcome | Prob
ANA in 4 games|4.7%
ANA in 5 games|11.4%
ANA in 6 games|12.9%
ANA in 7 games|16.1%
WIN in 7 games|15.0%
WIN in 6 games | 18.8%
WIN in 5 games|13.4%
WIN in 4 games|7.7%
ANA wins|45.1%
WIN wins | 54.9%

Outcome | Prob
VAN in 4 games|5.2%
VAN in 5 games|11.8%
VAN in 6 games|13.2%
VAN in 7 games|16.1%
CAL in 7 games|14.9%
CAL in 6 games | 18.3%
CAL in 5 games|13.1%
CAL in 4 games|7.4%
VAN wins|46.3%
CAL wins | 53.7%

Outcome | Prob
STL in 4 games|7.9%
STL in 5 games|16.3%
STL in 6 games|16.3%
STL in 7 games | 17.8%
MIN in 7 games|13.1%
MIN in 6 games|14.6%
MIN in 5 games|9.2%
MIN in 4 games|4.8%
STL wins | 58.3%
MIN wins|41.7%

Outcome | Prob
NAS in 4 games|4.8%
NAS in 5 games|11.3%
NAS in 6 games|12.8%
NAS in 7 games|16.3%
CHI in 7 games|14.8%
CHI in 6 games | 18.8%
CHI in 5 games|13.6%
CHI in 4 games|7.7%
NAS wins|45.1%
CHI wins | 54.9%

 

[charlie1]

Terrific question - most directly, SRS translates to goal relationships (not to win relationships). With that said, it's natural to want to go from one to the other.

When I do this, I typically (*) look at the variance of each team's performance. If each team has an expectation and a variance, and we assume a performance distribution (I prefer normal), then the difference between the two teams' performances will also have a distribution (the difference between normal distributions is normal), and then you can ask questions like "when is this difference greater than zero?" (so, the probability that one team wins).

These vary from team to team and it's related to the goal scoring environment. To give extreme examples on both ends, if two teams were playing and both had zero variance, then the team with the higher SRS would win 100% of the time. If two teams were playing and both had infinite variance, then each team would win 50% of the time no matter the SRS values. (Obviously, the correct answer is somewhere in between).

Complicating matters is the existence of three-point games in the NHL, which I haven't studied in enough detail to get something that predicts very well. As a first order approximation, I typically look at the proportion of games where each team has a third point. More directly, I don't care as much since this stuff is more interesting to me (1) to evaluate aspects of goaltenders - such as their schedules, and (2) in the playoffs - where three-point games don't matter.

(*) "Typically" means that I haven't done this here yet, but will when I get a chance (at least before the playoffs).

Really interesting work, Doc. Is the assumption of normality merely a convenience thing, or have you done some tests?