NHL Team Power Ratings and Predictions (SRS Algorithm)

[Doctor No]

After games of May 5th...

Outcome | Prob CHI in 4 games | 48.1%
CHI in 5 games|30.0%
CHI in 6 games|10.6%
CHI in 7 games|6.4%
MIN in 7 games|4.9%
CHI wins | 95.1%
MIN wins|4.9%

Outcome | Prob ANA in 5 games | 25.5%
ANA in 6 games|23.0%
ANA in 7 games|21.3%
CAL in 7 games|17.0%
CAL in 6 games|13.2%
ANA wins | 69.8%
CAL wins|30.2%

 

[Doctor No]

After games of May 6...

Outcome | Prob
NYR in 7 games|16.9%
WAS in 7 games|11.6%
WAS in 6 games|30.2%
WAS in 5 games | 41.3%
WAS in 4 games|0.0%
NYR wins|16.9%
WAS wins | 83.1%

Outcome | Prob
MTL in 7 games|3.7%
TBL in 7 games|3.8%
TBL in 6 games|11.6%
TBL in 5 games|20.5%
TBL in 4 games | 60.4%
MTL wins|3.7%
TBL wins | 96.3%

 

[Doctor No]

After games of May 7:

Outcome | Prob
MTL in 7 games|10.4%
TBL in 7 games|10.4%
TBL in 6 games|29.6%
TBL in 5 games | 49.5%
TBL in 4 games|0.0%
MTL wins|10.4%
TBL wins | 89.6%

Chicago wins their series, putting the algorithm at 7-2 so far (although picking the higher seeds would put you at 6-3).

 

[Doctor No]

Quiz time - throughout the season (so far), which NHL team has been the most consistent this year, and which NHL team has been the most inconsistent?

"Consistent" is defined here as how close they come to doing what the SRS algorithm expects them to do.

 

[hatterson]

Quiz time - throughout the season (so far), which NHL team has been the most consistent this year, and which NHL team has been the most inconsistent?

"Consistent" is defined here as how close they come to doing what the SRS algorithm expects them to do.

Are you talking game to game? As in, which team matched the SRS predictions most and which team matched them the least? Or do you mean it in a different way?

 

[billcook]

Tampa Bay and Anaheim.

 

[Doctor No]

Are you talking game to game? As in, which team matched the SRS predictions most and which team matched them the least? Or do you mean it in a different way?

The SRS algorithm produces a best estimate of a team's strength, and with that, one can generate best estimates of each game retrospectively.

However, the algorithm produces a single estimate for each team, and throughout the year a team will out-perform in some games, and under-perform in others. For instance, if Team A has an SRS of +1.2, Team B has an SRS of -0.3, and home-ice advantage is measured at +0.5, then we would predict Team A to beat Team B by 2.0 goals on home ice. When they don't win by precisely 2.0 goals, that's a process deviation.

This is in part because a team's composition changes over the course of the season, or because teams play better against certain opponents than others, but more largely because teams' abilities are not constant from game to game.

(Stated differently, which team has the lowest algorithm standard deviation, and which team has the highest?)

(Not Tampa, and not Anaheim. )

 

[Doctor No]

Through games of May 8...

Outcome | Prob
NYR in 7 games|29.0%
WAS in 7 games|20.3%
WAS in 6 games | 50.8%
NYR wins|29.0%
WAS wins | 71.0%

Outcome | Prob ANA in 5 games | 56.2%
ANA in 6 games|20.2%
ANA in 7 games|13.3%
CAL in 7 games|10.3%
ANA wins | 89.7%
CAL wins|10.3%

 

[hatterson]

The SRS algorithm produces a best estimate of a team's strength, and with that, one can generate best estimates of each game retrospectively.

However, the algorithm produces a single estimate for each team, and throughout the year a team will out-perform in some games, and under-perform in others. For instance, if Team A has an SRS of +1.2, Team B has an SRS of -0.3, and home-ice advantage is measured at +0.5, then we would predict Team A to beat Team B by 2.0 goals on home ice. When they don't win by precisely 2.0 goals, that's a process deviation.

This is in part because a team's composition changes over the course of the season, or because teams play better against certain opponents than others, but more largely because teams' abilities are not constant from game to game.

(Stated differently, which team has the lowest algorithm standard deviation, and which team has the highest?)

(Not Tampa, and not Anaheim. )

I wanted to confirm that it was based on game to game predictions versus a season long combined number. ie, it would seem possible for a team to deviate from their prediction in every single game but yet be perfectly on the mark season long simply by overscoring by 1 in half the games and underscoring by 1 in the rest.

As far as most vs least consistent, I kinda want to say the Sabres were the most consistent because every single one of their numbers said they were terrible all year and they played terrible all year save for that 10 or 15 game stretch about 1/4 of the way into the season.

The Leafs seem like they may be the least consistent. They played 1/3 of the season at a pace that would have finished them in the top 5 in the league and the last 2/3 at a pace comparable to the Sabres, but their season long numbers wouldn't be near as terrible as Buffalo's so would not only dramatically underestimate their first third, but also over-estimate their last two thirds.

Another potential inconsistent team may be Minnesota because of the massive deviation from one half of the season to the next.

Perhaps St. Louis as a consistent team.

 

[Doctor No]

Sabres is absolutely right! It sounds odd, although when you lose all of the time, it makes it easier to be consistent.

Vancouver's least consistent as far as the algorithm is concerned (that is, Vancouver's the biggest problem when trying to assign a single "strength" to them for the season.

Here's the full list, from most consistent to least consistent. For sense of scale, these are the standard deviations of the team strengths (and you can see that these dwarf the strength values, suggesting either that there's a lot of luck in hockey, or that the algorithm isn't super sophisticated - or both).

Team | St. Dev
Buffalo|

1.54​

|
Carolina|

1.88​

|
New Jersey|

1.88​

|
Florida|

1.95​

|
Edmonton|

1.95​

|
Washington|

1.95​

|
Nashville|

1.96​

|
Arizona|

1.96​

|
Ottawa|

1.97​

|
Boston|

1.98​

|
NY Rangers|

2.04​

|
Calgary|

2.06​

|
Philadelphia|

2.08​

|
NY Islanders|

2.11​

|
Detroit|

2.15​

|
Los Angeles|

2.16​

|
Winnipeg|

2.16​

|
Tampa Bay|

2.24​

|
Anaheim|

2.25​

|
Chicago|

2.25​

|
Colorado|

2.25​

|
Minnesota|

2.28​

|
Pittsburgh|

2.33​

|
Montreal|

2.34​

|
St. Louis|

2.35​

|
Dallas|

2.37​

|
San Jose|

2.41​

|
Columbus|

2.45​

|
Toronto|

2.54​

|
Vancouver|

2.60​

|

 

[Doctor No]

After last night's (May 9) win by the Habs...

Outcome | Prob
MTL in 7 games|21.1%
TBL in 7 games|20.4%
TBL in 6 games | 58.5%
MTL wins|21.1%
TBL wins | 78.9%

 

[hatterson]

Missed going 2/2 by .06

Also, my money is on the "both" option, with a little more weight on "luck"

 

[Doctor No]

Missed going 2/2 by .06

Also, my money is on the "both" option, with a little more weight on "luck"

True. Luck's so pervasive in hockey, and people really seem to be bothered by the term. Including people who've played hockey all of their life, or should I say *especially* people who've played hockey all of their life, even though those are exactly the people who should know better (I'm sure that we've all seen some bizarre stuff while playing).

Anyhow, the method dropped one last night (or this morning) - and is now at 7-3 (the same as you'd get picking the higher seeds). Sad trombone.

Here's the update for Rangers/Caps:

Outcome | Prob NYR in 7 games | 59.5%
WAS in 7 games|40.5%

 

[Doctor No]

Algorithm liked Tampa Bay over Montreal (an upset per the seeds).

Algorithm is 8-3 now, whereas choosing the higher seeds gets you 7-4. Not a lot of lift, admittedly.

As a reminder, the algorithm likes the Rangers (top team overall per the method) over Washington.

 

[supsens]

Algorithm liked Tampa Bay over Montreal (an upset per the seeds).

Algorithm is 8-3 now, whereas choosing the higher seeds gets you 7-4. Not a lot of lift, admittedly.

As a reminder, the algorithm likes the Rangers (top team overall per the method) over Washington.

Tampa 47 ROW Mont 43 that seed wasn't earned right, lol i know it won't matter but wasn't an upset on any level

 

[Doctor No]

Tampa 47 ROW Mont 43 that seed wasn't earned right, lol i know it won't matter but wasn't an upset on any level

It's an upset on the seeding level.

Saying it wasn't earned suggests that either Tampa or Montreal didn't understand the rules by which playoff seeds would be awarded.

 

[supsens]

TB got more points playing hockey and Mont got more in the shootout,the shootout is not in the playoffs, and I am not sure anyone had TB as an underdog.

 

[Doctor No]

Nevertheless.

I suppose that you went on the record, publicly, and picked Tampa Bay prior to the start of the series?

Not that this would be a highly-controversial pick, but by any standard definition, the Canadiens were the favorites in the series.

 

[Doctor No]

Algorithm takes the Rangers...after the division finals, it's at 9-3 now (picking the higher seeds each series gets you 8-4).

If you picked each series randomly, you'd have about a 7% chance of going 9-3 or better.

 

[Doctor No]

SRS algorithm power ratings after the division finals...

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

0.681​

|

0.009​

|

0.690​

|
Tampa Bay |

0.600​

|

-0.014​

|

0.586​

|
St. Louis|

0.500​

|

0.048​

|

0.548​

|
Chicago |

0.478​

|

0.062​

|

0.540​

|
Washington|

0.406​

|

0.047​

|

0.454​

|
Minnesota|

0.283​

|

0.089​

|

0.371​

|
Nashville|

0.295​

|

0.068​

|

0.364​

|
Montreal|

0.298​

|

0.032​

|

0.330​

|
Anaheim |

0.297​

|

-0.008​

|

0.289​

|
NY Islanders|

0.236​

|

0.036​

|

0.272​

|
Ottawa|

0.239​

|

-0.005​

|

0.234​

|
Winnipeg|

0.151​

|

0.060​

|

0.211​

|
Calgary|

0.204​

|

-0.006​

|

0.198​

|
Los Angeles|

0.183​

|

-0.019​

|

0.164​

|
Vancouver|

0.182​

|

-0.020​

|

0.162​

|
Detroit|

0.135​

|

0.024​

|

0.159​

|
Pittsburgh|

0.092​

|

0.036​

|

0.128​

|
Dallas|

0.012​

|

0.056​

|

0.068​

|
Boston|

0.049​

|

-0.008​

|

0.040​

|
Colorado|

-0.085​

|

0.064​

|

-0.021​

|
San Jose|

-0.049​

|

-0.019​

|

-0.068​

|
Columbus|

-0.171​

|

0.013​

|

-0.158​

|
Florida|

-0.207​

|

-0.013​

|

-0.220​

|
Philadelphia|

-0.232​

|

0.006​

|

-0.226​

|
New Jersey|

-0.427​

|

0.029​

|

-0.398​

|
Carolina|

-0.463​

|

0.029​

|

-0.434​

|
Toronto|

-0.622​

|

0.023​

|

-0.599​

|
Edmonton|

-1.037​

|

0.034​

|

-1.003​

|
Arizona|

-1.244​

|

0.047​

|

-1.197​

|
Buffalo|

-1.378​

|

0.058​

|

-1.320​

|

Teams remaining in BOLD.
Home-ice advantage currently is worth +0.253 goals/game.

 

[Doctor No]

And of course, conference final prediction distributions:

Outcome | Prob
NYR in 4 games|7.2%
NYR in 5 games|15.4%
NYR in 6 games|15.6%
NYR in 7 games | 18.0%
TBL in 7 games|13.3%
TBL in 6 games|15.4%
TBL in 5 games|9.9%
TBL in 4 games|5.1%
NYR wins | 56.2%
TBL wins|43.8%

Outcome | Prob
ANA in 4 games|4.2%
ANA in 5 games|10.2%
ANA in 6 games|11.7%
ANA in 7 games|15.2%
CHI in 7 games|15.2%
CHI in 6 games | 20.0%
CHI in 5 games|14.7%
CHI in 4 games|8.8%
ANA wins|41.3%
CHI wins | 58.7%

One favorite, and one underdog. (In advance of anyone who responds to that effect, "underdog" refers to the lower seed. Feel free to use your own definition.)

 

[supsens]

Nevertheless.

I suppose that you went on the record, publicly, and picked Tampa Bay prior to the start of the series?

Not that this would be a highly-controversial pick, but by any standard definition, the Canadiens were the favorites in the series.

I am not sure what you mean by any standard definition but ya I made some money, tampa was 5-0 against the habs this year and finished the season a +51.
They were 3-0 against the rangers but I am not sure who to bet on, it's close. I think the rangers play more of a "playoff" style but I am on the fence. The Caps punished the rangers too so that may slow them down. It's going to be a good final in the east and west this year.
tampa also outscored chi 6-3 and the ducks 9-4 in the two meetings with each so I might just bet on them to take it all, the odds seem decent.
But that D core on the rangers is top notch... fun times

I wasn't trying to fight or pick on anything, I was just saying it was weird that people would pick the habs in that series

 

[Doctor No]

I agree - Tampa was one of the algorithm's stronger picks. Montreal had the extra home game, which probably led to a lot of picks.

 

[Doctor No]

I'm loving whatever thing they're showing over and over again on the NHL Network (on the hour right now).

Where they take five random metrics, imply that THESE ARE THE FIVE THINGS and that Chicago is better in three of them, and therefore they will win the series

 

[Doctor No]

After today's Rangers/Lightning Game One...

Outcome | Prob
NYR in 4 games|12.6%
NYR in 5 games | 21.8%
NYR in 6 games|17.9%
NYR in 7 games|17.6%
TBL in 7 games|13.1%
TBL in 6 games|12.0%
TBL in 5 games|5.0%
NYR wins | 69.9%
TBL wins|30.1%