NHL Team Power Ratings and Predictions (SRS Algorithm)

[Doctor No]

Final 2015-16 NHL regular season SRS power ratings:

Team | 2015-16 Goal Diff | 2015-16 Schedule | 2015-16 Strength
Washington|

0.720​

|

-0.005​

|

0.714​

|
Pittsburgh|

0.512​

|

0.000​

|

0.513​

|
Dallas|

0.451​

|

0.004​

|

0.455​

|
Florida|

0.439​

|

-0.007​

|

0.432​

|
San Jose|

0.378​

|

-0.037​

|

0.341​

|
Los Angeles|

0.366​

|

-0.032​

|

0.334​

|
Chicago|

0.317​

|

0.011​

|

0.328​

|
Tampa Bay|

0.317​

|

-0.009​

|

0.308​

|
St. Louis|

0.280​

|

0.013​

|

0.294​

|
Anaheim|

0.317​

|

-0.037​

|

0.280​

|
NY Rangers|

0.232​

|

0.017​

|

0.248​

|
NY Islanders|

0.195​

|

0.006​

|

0.201​

|
Nashville|

0.145​

|

0.020​

|

0.165​

|
Minnesota|

0.122​

|

0.018​

|

0.140​

|
Boston|

0.122​

|

-0.007​

|

0.115​

|
Philadelphia|

-0.049​

|

0.027​

|

-0.022​

|
Ottawa|

-0.134​

|

0.016​

|

-0.118​

|
Detroit|

-0.159​

|

0.010​

|

-0.149​

|
Montreal|

-0.183​

|

0.014​

|

-0.169​

|
Buffalo|

-0.256​

|

0.015​

|

-0.241​

|
Winnipeg|

-0.284​

|

0.036​

|

-0.248​

|
Colorado|

-0.293​

|

0.038​

|

-0.254​

|
New Jersey|

-0.293​

|

0.025​

|

-0.267​

|
Carolina|

-0.341​

|

0.038​

|

-0.303​

|
Calgary|

-0.354​

|

0.002​

|

-0.352​

|
Columbus|

-0.402​

|

0.037​

|

-0.365​

|
Arizona|

-0.439​

|

0.017​

|

-0.422​

|
Edmonton|

-0.512​

|

0.011​

|

-0.502​

|
Toronto|

-0.585​

|

0.039​

|

-0.546​

|
Vancouver|

-0.634​

|

0.026​

|

-0.608​

|

Home-ice advantage is measured at +0.207 goals/game.

Some of the closest top-to-bottom rankings I've ever seen (noting that these will move during the playoffs).

I'll start a new thread for the playoffs.

 

[TheCLAM]

Final 2015-16 NHL regular season SRS power ratings:

Team | 2015-16 Goal Diff | 2015-16 Schedule | 2015-16 Strength
Washington|

0.720​

|

-0.005​

|

0.714​

|
Pittsburgh|

0.512​

|

0.000​

|

0.513​

|
Dallas|

0.451​

|

0.004​

|

0.455​

|
Florida|

0.439​

|

-0.007​

|

0.432​

|
San Jose|

0.378​

|

-0.037​

|

0.341​

|
Los Angeles|

0.366​

|

-0.032​

|

0.334​

|
Chicago|

0.317​

|

0.011​

|

0.328​

|
Tampa Bay|

0.317​

|

-0.009​

|

0.308​

|
St. Louis|

0.280​

|

0.013​

|

0.294​

|
Anaheim|

0.317​

|

-0.037​

|

0.280​

|
NY Rangers|

0.232​

|

0.017​

|

0.248​

|
NY Islanders|

0.195​

|

0.006​

|

0.201​

|
Nashville|

0.145​

|

0.020​

|

0.165​

|
Minnesota|

0.122​

|

0.018​

|

0.140​

|
Boston|

0.122​

|

-0.007​

|

0.115​

|
Philadelphia|

-0.049​

|

0.027​

|

-0.022​

|
Ottawa|

-0.134​

|

0.016​

|

-0.118​

|
Detroit|

-0.159​

|

0.010​

|

-0.149​

|
Montreal|

-0.183​

|

0.014​

|

-0.169​

|
Buffalo|

-0.256​

|

0.015​

|

-0.241​

|
Winnipeg|

-0.284​

|

0.036​

|

-0.248​

|
Colorado|

-0.293​

|

0.038​

|

-0.254​

|
New Jersey|

-0.293​

|

0.025​

|

-0.267​

|
Carolina|

-0.341​

|

0.038​

|

-0.303​

|
Calgary|

-0.354​

|

0.002​

|

-0.352​

|
Columbus|

-0.402​

|

0.037​

|

-0.365​

|
Arizona|

-0.439​

|

0.017​

|

-0.422​

|
Edmonton|

-0.512​

|

0.011​

|

-0.502​

|
Toronto|

-0.585​

|

0.039​

|

-0.546​

|
Vancouver|

-0.634​

|

0.026​

|

-0.608​

|

Home-ice advantage is measured at +0.207 goals/game.

Some of the closest top-to-bottom rankings I've ever seen (noting that these will move during the playoffs).

I'll start a new thread for the playoffs.

You should think about ranking the system from 1-100 after your results have been processed.

 

[Doctor No]

You should think about ranking the system from 1-100 after your results have been processed.

Can you describe more what you're thinking here?

 

[wgknestrick]

Can you describe more what you're thinking here?

I think he means that you use a "10 must" scale. Washington gets 100 and every team under is scored relatively.

 

[Doctor No]

Ah, I see - so Washington would get a 100, Vancouver would get a 0, and everyone else would go in between (essentially stretching the current scale to a 0-100)?

I do something like that when measuring goalies - I count the percentage of games played against the "top quartile" and against the "bottom quartile", which I think shows schedule strength a bit differently than my standard algorithm (which is a minutes played weighting of the Simple Rating System values).

 

[wgknestrick]

The other thing that is nice is to have the scale be in a common unit, like goals or goal differential. This way you can easily calculate the expected goal differential when 2 teams play each other. IE WSH plays VAN and should expect to win by an average of 1.2goals or whatever that comes out to.

 

[Doctor No]

Exactly - that's (I think) a main advantage of the metric here; there's a common unit (goals/game).

 

[Doctor No]

Final 2015-16 SRS rankings (through the Stanley Cup playoffs):

Team | 2015-16 Goal Diff | 2015-16 Schedule | 2015-16 Strength
Washington|

0.702​

|

0.038​

|

0.740​

|
Pittsburgh|

0.566​

|

0.100​

|

0.666​

|
San Jose|

0.472​

|

0.044​

|

0.516​

|
Florida|

0.398​

|

0.011​

|

0.409​

|
Tampa Bay|

0.343​

|

0.040​

|

0.383​

|
Dallas|

0.316​

|

0.037​

|

0.353​

|
Chicago|

0.303​

|

0.035​

|

0.339​

|
St. Louis|

0.235​

|

0.086​

|

0.322​

|
Anaheim|

0.337​

|

-0.025​

|

0.312​

|
Los Angeles|

0.287​

|

0.008​

|

0.295​

|
NY Islanders|

0.108​

|

0.058​

|

0.166​

|
NY Rangers|

0.092​

|

0.064​

|

0.156​

|
Boston|

0.122​

|

-0.002​

|

0.120​

|
Minnesota|

0.068​

|

0.042​

|

0.110​

|
Nashville|

0.000​

|

0.083​

|

0.083​

|
Philadelphia|

-0.136​

|

0.080​

|

-0.057​

|
Ottawa|

-0.134​

|

0.021​

|

-0.113​

|
Detroit|

-0.195​

|

0.038​

|

-0.158​

|
Montreal|

-0.183​

|

0.019​

|

-0.164​

|
Buffalo|

-0.256​

|

0.020​

|

-0.236​

|
Winnipeg|

-0.284​

|

0.035​

|

-0.249​

|
Colorado|

-0.293​

|

0.040​

|

-0.252​

|
New Jersey|

-0.293​

|

0.030​

|

-0.263​

|
Carolina|

-0.341​

|

0.044​

|

-0.297​

|
Calgary|

-0.354​

|

0.009​

|

-0.344​

|
Columbus|

-0.402​

|

0.041​

|

-0.361​

|
Arizona|

-0.439​

|

0.025​

|

-0.414​

|
Edmonton|

-0.512​

|

0.016​

|

-0.496​

|
Toronto|

-0.585​

|

0.045​

|

-0.541​

|
Vancouver|

-0.634​

|

0.034​

|

-0.600​

|

These are the rankings that I use on my goaltender site to determine a goalie's SoS (average Strength of Schedule).

 

[triggrman]

Wow, I would have thought Nashville would move up after the playoffs but they dropped a spot.

 

[Doctor No]

Wow, I would have thought Nashville would move up after the playoffs but they dropped a spot.

Their schedule strength got better (as happens with most playoff teams) although their goal differential took a hit.

http://hockeygoalies.org/bio/nhl/logs/NSH2015.html

Looking at their individual game outcomes, their rating would have benefitted from deflating the "blowout" games. Under the standard SRS algorithm, a four-goal win is twice as "good" as a two-goal win. In reality, it's better, but probably not twice as good.

Many SRS improvements calibrate some level of deflator, which does improve the predictive accuracy to some degree. I'll play with some of these tricks later in the summer.

 

[wgknestrick]

Their schedule strength got better (as happens with most playoff teams) although their goal differential took a hit.

http://hockeygoalies.org/bio/nhl/logs/NSH2015.html

Looking at their individual game outcomes, their rating would have benefitted from deflating the "blowout" games. Under the standard SRS algorithm, a four-goal win is twice as "good" as a two-goal win. In reality, it's better, but probably not twice as good.

Many SRS improvements calibrate some level of deflator, which does improve the predictive accuracy to some degree. I'll play with some of these tricks later in the summer.

Do you count ENG the same as 5v5? If you do, there is your area of improvement to the system. Actually weighting the situations each goal was scored would help tremendously. A team that wins 4-2 via 5v5 scoring is generally better than one that wins scoring a couple of PPG (but even at 5v5) in their 4-2 victory.

 

[Doctor No]

Both of those also give some improvement over the base SRS algorithm (although not as much as you might expect - especially the ENG counting - from a prospective standpoint).

 

[wgknestrick]

Both of those also give some improvement over the base SRS algorithm (although not as much as you might expect - especially the ENG counting - from a prospective standpoint).

I can agree with that.

They probably don't matter much on large samples, but probably improve the accuracy with small samples. They also should help the head-to-head accuracy and prefer the better 5v5 GF% team.

It is not that uncommon to have 1-2 ENG in a single playoff game.