NFL Spread Dial

How it works

For each game we compute seven numbers about the home team relative to the away team, multiply each by a fitted coefficient and the weight you set, and add them up. That sum is our expected home margin. The edge is that margin minus the line.

Our margin (home − away) =
  w₁ × +0.0256 × Elo difference
+ w₂ × +0.1027 × form difference   (avg margin, last 4 games)
+ w₃ × +0.2528 × rest difference   (days)
+ w₄ × −1.4355 × cover-rate difference (last 6 games)
+ w₅ × +1.8010 × home game         (1, or 0 at a neutral site)
+ w₆ × +0.0745 × efficiency difference (net EPA per 100 plays, last 16 games)
+ w₇ × −2.6633 × QB-change difference (1 if the starter isn't the usual one)

Edge         = our margin − the line (written as a home margin)
Cover chance = how often past games with an edge that size covered

The cover chance is the record, not a bell-curve reading of the edge. Read off a normal curve with the 12.7-point spread of games around the line, a 4-point edge would say 62%. In the back-test, games with an edge that big covered about half the time. So each game shows the back-test's cover rate for edges its size, at the settings you have chosen, and a “strong lean” means the formula and the line are far apart, not that the formula is likely right.

The coefficients are an ordinary least-squares fit of actual final margins on those seven factors over every regular-season and playoff game from 2015 through 2026 (3,150 games), refit each time the page regenerates. Each slider (w) scales one coefficient: 1.0× is the fit, 0× removes the factor.

The seven factors

• Power rating. Our own Elo rating for each team, updated after every game since 1999 using FiveThirtyEight's published NFL method (K = 20, margin-of-victory multiplier, one-third reversion to the mean each new season).
• Recent form. Each team's average points margin over its last four games, whichever season they fell in.
• Rest edge. Days since each team last played, straight from the schedule. A team off a bye has 14; a team off a Thursday game has 10.
• Cover trend. The share of its last six games each team covered the closing spread (a push counts one half).
• Home field. 1 when the home team is at its own stadium, 0 at a neutral site such as London or Melbourne.
• Play efficiency. Expected points added per play on offense, minus the same figure allowed on defense, over each team's last sixteen games, from nflverse's weekly team stats. EPA scores every snap by how much it changed the offense's expected points, so it credits a team for moving the ball even in games the scoreboard doesn't reflect. Shown per 100 plays.
• QB change. 1 when a team's starting quarterback in nflverse's schedule file is not the one who has started most of its games this season (in week 1, most of its last eight), home minus away. That catches a backup in for an injury or a benching, and a new starter in week 1. The line moves for these; without this factor the formula would rate the team as if its usual starter played.

Play efficiency joined in September 2026 as the sixth factor. With the other five already in the fit, it still carries weight (a t-statistic of about 2.4, and a slightly better fit on 2022–26 games when it is left out of the fitting years), but it does not lift the back-test off a coin flip. It is in because it measures something the scoreboard misses, not because it beats the line.

QB change joined the same month as the seventh. nflverse's schedule file names each game's starting quarterback, including the coming week's once a starter is known, so the formula can see when a team starts someone other than the quarterback who has started most of its games that season. Fitted with the other six it is worth −2.7 points (a t-statistic of about −5.9 when it joined), and it lowers the typical miss on 2022–26 games when they are left out of the fit. Without it the dial rated a team with its backup in as if the usual starter played, and leaned hardest exactly where the line had already moved. It sharpens the margin; it does not lift the record off a coin flip.

What the back-test is, and isn't

Two records, and the first is the one to trust. The out-of-sample record scores each season from 2016 on with coefficients fitted only on the seasons before it, at the fitted weights, so no graded game was seen by the fit that graded it. It is refit every time the page regenerates, like everything else here.

The back-test below the games replays the fitted games with the factors exactly as they stood before each kickoff (no look-ahead) and counts how often the side the formula leaned to covered nflverse's closing line. It is in-sample: the coefficients were fitted on those same games, so it flatters the default settings slightly, and moving a dial to push the rate up is curve-fitting. It stays because it is the only way to see what your own weights would have done. Both records are graded against the closing line, not the line you see during the week.

What it leaves out

• Injuries other than a change at quarterback, and how good a replacement quarterback is: the factor only knows the starter is not the usual one. Weather and coaching changes too. None of these are in the schedule file, so none are in the score.
• Line movement. The line here is a snapshot at the last refresh; a book's number when you look may differ.
• Anything non-linear. Seven factors, added up, is the whole model. That is deliberate: every point in the result can be traced to one row of the table under each game.

Sources

• nflverse schedules and results (games.csv), CC-BY 4.0 — scores, closing lines, rest days and sites for every game since 1999.
• nflverse weekly team stats (stats_team_week), CC-BY 4.0 — offensive EPA and play counts for every team-game since 2012, the input to play efficiency.
• FiveThirtyEight, “How our NFL predictions work” (archived copy, January 2024; the site closed in 2025) — the Elo update rule the power rating follows (K = 20, margin-of-victory multiplier, one-third reversion between seasons, 48-point home edge inside the rating).
• The Odds API — current US sportsbook spreads, when the live connection is on.

Last reviewed: September 2026. Not affiliated with or endorsed by the NFL or any team.