How the NFL Ratings Work
Methods, validation and limits behind the Pascal Rating for the NFL, which ranks all 32 teams in the Pascal Power Rankings and drives our published NFL forecasts.
Published by Pascal Institute. Edited by Christopher Merry, Chief Decision Scientist.
- What the rating measures
- Scoring every play
- Opponent adjustment
- Last season as a prior
- Combining and scaling
- Weekly expectations
- Validation
- Interpretive limits
- Sources and data
What the rating measures
The rating estimates how efficiently a team gains and prevents expected points on each play, adjusted for the opponents it has faced. It is a forecast of strength, not a judgment of who has earned the best record.
The rating uses play-by-play efficiency only. Unlike our college football rating, it does not use scoring margin. It also ignores betting markets, injuries, quarterback changes, weather, home field, roster continuity and win-loss record.
Scoring every play
Each pass or rush play receives its expected points added (EPA) from nflverse/nflfastR. No-plays and quarterback kneels are excluded. Only games completed before a rating's cutoff are used, and games that kick off at the same time are rated from the same frozen state.
Opponent adjustment
A single ridge regression (penalty 5.0) fits every play as the offense's strength minus the opposing defense's strength. A team's offense is credited for the defenses it faced, and its defense for the offenses it faced. Both components are centered on zero, and a positive defensive value means a team prevents expected points better than average. No separate strength-of-schedule term is added, because the joint fit already accounts for opponents.
Last season as a prior
Offense and defense are each blended with the team's final values from the previous season, worth two games of evidence. The prior has full weight before Week 1, two thirds after one game, half after two and one third after four, and keeps shrinking as games are played.
component = n / (n + 2) × this season + 2 / (n + 2) × last season, where n is games played.
Combining and scaling
| Step | Definition |
|---|---|
| Overall strength | overall = offense + defense, in EPA per play |
| Public rating | rating = 100 + 10 × (overall − 0.00377) / 0.0740 |
| Reading the scale | 100 is the 2020–2022 league average. Ten rating points equal one standard deviation of team strength in that period, or 0.074 EPA per play. |
| Timing | Ratings use only games completed before the cutoff. On days without games, ratings do not change. |
Weekly expectations
Because the rating is measured in EPA per play, turning it into points of margin needs a separate fit. We fitted final margin against the pregame rating gap and a home-field term on 1,107 games from 2020–2023, froze the result and tested it once on 570 games from 2024–2025.
| Quantity | Definition |
|---|---|
| Expected margin | (home rating − away rating) × 0.361 + 1.7, from the home team's view |
| Margin per rating point | 0.361 points (95% bootstrap interval 0.303 to 0.418). A 10-point rating gap is worth about 3.6 points. |
| Home field | 1.68 points, rounded to 1.7. Neutral-site games get none. |
| Win probability | The final margin treated as normal around the expected margin, with a standard deviation of 13.0 points, the 2024–2025 holdout error. A 3-point favorite wins about 59% of the time, and a 7-point favorite about 70%. |
| 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 |
|---|---|
| Rating holdout, 2024–2025 | Pregame rating gaps correlated 0.40 with final point differential across 570 games |
| Opponent adjustment | Improved prediction error in every season from 2020 through 2025, though the 2023 gain was small |
| Prior | Last season's information materially improved predictions in Weeks 1–4 |
| Rejected variants | Stronger recency weighting, an extra strength-of-schedule term and heavier passing weights did not help |
| Margin fit, holdout | Typical error 13.0 points with the frozen 2020–2023 fit, against 14.45 for predicting a 0 margin; the slope fitted on the holdout alone was 0.387, inside the development interval |
| Margin fit, by season | 0.32 to 0.41 points per rating point in every season from 2020 to 2025, with no trend |
Interpretive limits
Small gaps are ties. A rating point is worth about a third of a point of margin, and the typical error is 13 points. Teams within a few rating points of each other are effectively even.
It ignores wins and points. A team that wins close games with an inefficient offense will rank below its record, and a team that loses close games efficiently will rank above it. Close results are treated mostly as noise, which is usually right but not always.
Defense is the weaker signal. Validation found defensive components less stable and less predictive than offense, so defense-first teams carry more uncertainty than their ratings suggest.
No quarterback or injury adjustment. A starter's injury shows up only after the team plays without him.
Last season counts at full strength early. The prior is not regressed toward average, so teams that were extreme last season move the most in the first weeks.
The provider's EPA can change. The play values come from nflverse, which occasionally corrects past data.
Accuracy is modest. With home field included, the rating gap explains about 19% of the variation in final margins. It is a disciplined baseline, not an oracle.
Live ratings
Every team's current rating, updated as games finish, is on the NFL 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, NFL Rating v1 specification (methodology nfl-rating-v1, frozen): the definition, opponent adjustment, prior and public scale.
- AnalytIQ, NFL Sprint 2 validation report: development, validation and holdout results for the rating.
- AnalytIQ, NFL rating-to-margin calibration: the 0.361-point slope, the 1.68-point home effect and the 13.0-point holdout error.
- Weekly ratings, records and schedules: stored AnalytIQ rating builds and game catalog, exported with read-only queries. Each edition's data file and the queries are kept with the published pages.