National Hockey League Computer Ratings


Introduction

This is my rough attempt at a computer rating that can, among other things, come somewhat close to predicting final game scores. The actual computer rating, which describes team performance based on games played to date, is found under the "Rating" column. To determine a hypothetical margin of victory, use the "Points" column (just subtract the teams in question) and add the home "field" advantage, which is on the right below the predictions.

All of the predictions on the right already have the home "field" added in, and also show a predicted total. Shootouts and overtimes make this tricky, and are a lot of why the average error is 1.5 goals or so.

Early-season predictions are based almost entirely on last year's games and the preseason, so use those with extreme caution.


Ratings last updated Friday 06/11/21, 12:40 PM ET Rank Team W L Rating Points BCS 1 Tampa Bay 44 23 3.25 51.04 1 2 Carolina 41 26 2.75 50.93 4 3 Florida 39 23 2.57 50.71 2 4 NY Islanders 40 28 2.09 50.56 5 5 Pittsburgh 39 23 1.95 50.41 3 6 Colorado 46 21 1.90 50.93 12 7 Las Vegas 48 22 1.71 50.79 11 8 Boston 39 28 1.67 50.36 7 9 Toronto 38 25 1.46 50.55 9 10 Washington 37 24 1.38 50.17 6 Rank Team W L Rating Points BCS 11 Nashville 33 29 1.07 50.24 8 12 Edmonton 35 25 1.05 50.33 10 13 Dallas 23 33 0.82 50.43 16 14 NY Rangers 27 29 0.58 50.15 13 15 Winnipeg 34 30 0.33 50.11 15 16 Montreal 32 35 0.13 50.06 18 17 Calgary 26 30 0.01 50.09 19 18 Chicago 24 32 -0.03 49.96 17 19 Minnesota 38 25 -0.22 49.98 20 20 Philadelphia 25 31 -0.60 49.50 14 Rank Team W L Rating Points BCS 21 Columbus 18 38 -0.95 49.72 23 22 Detroit 19 37 -1.08 49.65 24 23 Vancouver 23 33 -1.10 49.61 22 24 Ottawa 23 33 -1.19 49.53 21 25 St Louis 27 33 -1.93 49.56 27 26 New Jersey 19 37 -2.07 49.14 25 27 Arizona 24 32 -2.24 49.37 28 28 Buffalo 15 41 -2.80 48.95 26 29 Los Angeles 21 35 -2.95 49.30 29 30 San Jose 21 35 -3.50 48.98 30 31 Anaheim 17 39 -4.04 48.90 31
(The "BCS style" ranking is one based entirely on wins and losses, similar to what is used in college football.)

When I get them working, divisional rankings will go here:

Rank Conference Mean Rating ---- ------------------------- ----------- 1 Central 1.05 2 East 0.27 3 North 0.10 4 West -1.41

PREDICTIONS FOR UPCOMING GAMES Date Away Team Home Team Total Prediction ----------- -------------------- -------------------- ----- --------------

Current home field advantage is: 0.28 MAE for games to date: 1.97 These ratings fit to produce 0.61 of the correct winners. Pct when predicted MOV is above 0.56: 0.68 A favored away team rarely loses when favored by more than -0.82. A favored home team rarely loses when favored by more than 0.80.

Above are some statistics about the ratings model. Each team has its own home "field" advantage, but the average all of them is shown here. The MAE is the mean absolute error of the ratings fit to all the games played to date. This number is usually larger than you think it should be, but it's a good measure of how variable (or maybe "predictable") game outcomes can be.

Immediately below that, you can see how this best fit does in retro-predicting (there's a better word I'm sure) just the game winners. My favorite stats are the last two--when the home or away team is favored by the given margin, they only lose 30 percent of the time. This is the kind of information that people in the sports wagering world might find useful.


About the author

I have a Ph.D. in Atmospheric Science from the University of Alabama in Huntsville. I am now a faculty member at Indiana University in Bloomington, teaching courses in the broad areas of weather and climate. Don't hesitate to contact me using this email form if you have questions or non-hateful comments. shopify stats