A data-driven approach to point spread analysis
Successful spread betting isn't about picking winners - it's about finding spreads that don't accurately reflect the true difference between teams.
"Is this spread too high or too low based on my analysis?"
Expected Points Added (EPA) is the foundation of modern spread analysis. Here's how to use it:
Team Strength = Offensive EPA/play - Defensive EPA/play allowed
NFL teams average about 60-65 plays per game. Use this to estimate point differential:
Expected Margin ≈ (EPA Diff) × 60 plays
Home field advantage in the NFL has declined but still exists:
Final estimate: If Team A is home: 15.6 + 2.5 = 18.1 points
If the spread is Team A -14, and your model says -18, you have a potential edge on Team A.
If the spread is Team A -21, you might have value on Team B +21.
Certain margins occur more frequently in NFL games. Understanding key numbers is critical for spread betting.
Shares of all 7,276 completed games, 1999–2025. Data: bundled nflverse games.csv, computed by the author.
Only buy points through key numbers (3 and 7). Buying from -5 to -4.5 rarely provides value - the cost outweighs the benefit.
| Situation | Measured Record | Why It Works |
|---|---|---|
| Road dogs +3 to +10 | 52.2% ATS (3,370 games) | Real lean; still under the 52.4% break-even |
| Teams off a 17+ point loss | 50.8% ATS (1,768 games) | No measurable bounce-back edge |
| Unders below 32°F outdoors | Unders hit 46.5% (338 games) | Folklore reversed: cold games beat the market total more often |
| Home dogs | 50.2% ATS (2,418 games) | A coin flip over 27 seasons |
Data: nflverse games.csv bundled with this site, computed by the author. Pushes excluded from cover rates.
Create point values for each team based on EPA, DVOA, or your own metrics
Team A rating - Team B rating + home field adjustment
Rest, travel, weather, injuries, motivation
If your line differs by 2+ points, investigate further
Keep records, calculate CLV, adjust weights over time
Use our tools to find value and size your bets optimally.
Before building a model to out-predict the market, look at what you are up against. Grouping all 7,245 non-pick'em games in the bundled nflverse log (1999–2025) by the size of the closing spread and comparing the market's number to what actually happened:
| Closing spread | Games | Avg line | Avg favorite margin | Favorite won SU | Dog won outright |
|---|---|---|---|---|---|
| 0.5–2.5 | 1,450 | 1.7 | +0.6 | 52.8% | 47.0% |
| 3–3.5 | 1,835 | 3.2 | +3.3 | 60.1% | 39.8% |
| 4–6.5 | 1,758 | 5.3 | +5.6 | 67.8% | 32.0% |
| 7–9.5 | 1,299 | 7.9 | +8.8 | 75.8% | 23.9% |
| 10–13.5 | 688 | 11.2 | +11.4 | 82.8% | 17.0% |
| 14+ | 215 | 15.4 | +15.7 | 91.2% | 8.8% |
Data: nflverse games.csv bundled with this site (7,276 completed games, 1999–2025), computed by the author. At a closing spread of exactly 3, the favorite won 58.0% straight up.
Read the third and fourth columns together: at every spread size, the average actual margin lands within about a point of the average line. There is no bucket where favorites or dogs are systematically mispriced by enough to matter. Your model is not trying to find a broken market; it is trying to be half a point sharper than a well-calibrated one, game by game. That is achievable in spots — it is not achievable everywhere, which is why bet volume discipline shows up in every serious bettor's writeup.
Keep reading: do favorites cover, by spread size, how accurate the spread is, and key numbers before you pay for any half-point.
Nothing here is betting advice, and no number on this page predicts any single game. Sports betting is legal only in some jurisdictions and only for adults (21+ in most U.S. states). If betting stops being entertainment, call or text 1-800-GAMBLER. Read our full disclaimer.