How the NHL Ratings Work
Methods, validation and limits behind the Pascal Rating for the NHL, which ranks all 32 teams in the Pascal Power Rankings.
Published by Pascal Institute. Edited by Christopher Merry, Chief Decision Scientist.
- What the rating measures
- Goals and shots
- Opponent adjustment
- Last season as a prior
- Scaling
- Weekly expectations
- Validation
- Interpretive limits
- Sources and data
What the rating measures
The rating estimates how many more goals per game a team scores than it allows against an average opponent, adjusted for the opponents it has faced. It is a forecast of strength, not a judgment of who has earned the most standings points.
The rating uses only regular-season results and shots on goal. It ignores betting markets, starting goaltenders, injuries, trades, rest, travel and standings points.
Goals and shots
Goals are the result that matters, but in a single game they are mostly luck. Shots on goal are more stable and say more about the next game early in a season. The rating uses both, half each. Shots are converted to goals at the previous season's league shooting percentage, so both halves are measured in goals per game.
A shootout counts as a tie in goals, because the extra goal a shootout winner is credited with says little about team strength. Overtime goals count. Only games completed before a rating's date are used, and all games on the same day are rated from the same frozen state.
Opponent adjustment
A ridge regression (penalty 2) fits each team's goals in every game as its attacking strength minus the opposing team's defensive strength, plus a home-ice term. The same fit is run on shots. A team's net strength is attack plus defense. No separate strength-of-schedule term is added, because the joint fit already accounts for opponents.
Last season as a prior
Each team starts the season at half of its final rating from last season, and that starting point is worth 30 games of evidence. The prior has all the weight before the first game, half after 30 games and about a quarter by the end of an 82-game season. Halving last season reflects how much NHL team strength regresses from one season to the next. The Arizona Coyotes, Utah Hockey Club and Utah Mammoth are treated as one franchise.
strength = n / (n + 30) × this season + 30 / (n + 30) × 0.5 × last season, where n is games played.
Scaling
| Step | Definition |
|---|---|
| Net strength | Half goals, half shots in goals, each opponent-adjusted, blended with the prior, in goals per game |
| Public rating | rating = 100 + 10 × (strength + 0.0017) / 0.340 |
| Reading the scale | 100 is an average team. Ten rating points equal one standard deviation of pregame team strength in 2021–22 to 2023–24. |
| Timing | Ratings use only games completed before the cutoff. On days without games, ratings do not change. |
Weekly expectations
To turn ratings into goals of margin, we fitted final margin against the pregame rating gap and a home-ice term on 3,936 games from 2021–22 to 2023–24, froze the result and tested it once on 2,624 games from 2024–25 and 2025–26.
| Quantity | Definition |
|---|---|
| Expected margin | (home rating − away rating) × 0.0548 + 0.22, in goals, from the home team's view |
| Margin per rating point | 0.0548 goals (95% bootstrap interval 0.049 to 0.060). A 10-point rating gap is worth about half a goal. |
| Home ice | 0.22 goals. Neutral-site games, including international and most outdoor games, get none. |
| Win probability | The final margin treated as normal around the expected margin, with a standard deviation of 2.54 goals, the holdout error. Evenly rated teams give the home team about 53%, a half-goal favorite wins about 58% of the time, and a one-goal favorite about 65%. |
| Scorecard | Each edition locks its expectations when it is published. The next edition reports how many favorites won against the number the probabilities implied, and the average distance between final and expected margins. |
Validation
| Check | Result |
|---|---|
| Candidates | 210 combinations of goal and shot weight, ridge penalty, prior length and prior shrinkage, each replayed game day by game day using only earlier games |
| Selection | Leave-one-season-out log loss on 2021–22 to 2023–24. The half-goals, half-shots blend beat goals alone and shots alone, and a regressed prior beat no prior. |
| Holdout, 2024–25 and 2025–26 | Log loss 0.676 for the rating, against 0.690 for always picking the home team and 0.693 for a coin flip. Favorites won 56.8% of games. |
| Simpler ratings, holdout | Goals alone without a prior scored 0.684, shots alone 0.678 |
| Margin fit, holdout | Typical error 2.54 goals, against 2.60 for predicting a 0 margin |
| Early season | Games in which either team had played fewer than 10 games scored 0.670, better than later games, because the prior carries the early weeks |
Interpretive limits
Hockey is close to a coin flip. The rating gap explains about 4% of the variation in final margins, and even well-separated teams lose often. Teams within several rating points of each other are effectively even.
No goaltender adjustment. The starting goaltender changes a game's odds, and the rating does not know who starts.
Shots, not shot quality. The rating counts shots on goal, not shot attempts, shot locations or expected goals. Empty-net goals count, which slightly inflates late-game margins.
Lopsided games are slightly overconfident. In the holdout, home teams given less than a 35% chance won about 39% of the time.
Outdoor games lose home ice. The source data marks outdoor games as neutral site even when one team is the host.
The source can change. Results come from public NHL.com data, which is occasionally corrected after the fact.
Live ratings
Every team's current rating, updated as games finish, is on the NHL Team Ratings page.
Track record
See how the ratings have been scored in public, season by season and in retrospective studies, in the Track Record.
Sources and data
- AnalytIQ, NHL Rating v1 research (methodology nhl-rating-v1, frozen): candidate selection, the definition, the 0.0548-goal slope, the 0.22-goal home effect and the 2.54-goal holdout error.
- AnalytIQ NHL results dataset: regular-season schedules, final scores and shots on goal since 2020–21, collected from NHL.com and checked against ESPN final scores.
- Weekly ratings, records and schedules: computed from that dataset as of each edition's cutoff. Each edition's data file is kept with the published pages.