The Numbers
What actually happens in NFL games, measured on 7,276 games since 1999 with a closing line and a final score. Sample size and a confidence range travel with every number.
Why 3 is not just another number
Three points decides 15.1% of all NFL games — 1,100 of them. Seven decides another 9.0%. Everything from 15 up, combined, decides less than three does alone.
The market agrees. Of the 81 NFL games currently priced, 22 sit on 3 or 3.5 and 15 sit exactly on one of the key numbers. The most common line on the board right now is 1.5 (15 games). Books build around the same margins the archive says decide the most games.
Does laying the points work?
Cover rate from the favorite's side, grouped by how big the number is. Every band is a coin flip, and that includes the double-digit favorites people are convinced never cover.
Bar = 95% confidence range, tick = measured rate. Faint line is 50%; the green line is 52.38%, the break-even rate at −110 once the hold is included.
Overs, unders and blowouts
Grouped by spread size, because the theory worth testing is that lopsided games run the clock. They don't, measurably.
Bar = 95% confidence range, tick = measured rate. Faint line is 50%; the green line is 52.38%, the break-even rate at −110 once the hold is included.
Situations people bet on
Home side against the spread in each situation. Division games and home favorites both land slightly under 50%, which means the other side of those was marginally better — and still short of the 52.38% you need.
Bar = 95% confidence range, tick = measured rate. Faint line is 50%; the green line is 52.38%, the break-even rate at −110 once the hold is included.
What to take from this
The NFL spread market is priced about as well as a market can be — 27 seasons and nothing here clears the vig. What survives is the shape of the distribution: because 3 and 7 carry so much of the probability, the price you pay around those numbers matters more than any angle on this page. Player props are a different market with a different answer — that is where our model actually publishes picks.