Why Do the Numbers 3 and 7 Matter When Pricing an NFL Spread?
15.0% of NFL games end on a 3-point margin and a bell curve puts 5.3% there. Here is why that gap decides what a half point on a spread costs.
Updated Sep 2026 · Part of the models series
Football scores in threes and sevens, so final margins pile up on a few numbers instead of spreading smoothly. On the pooled 1999 to 2025 nflverse games file, 15.0% of games end on a margin of exactly 3 points, and a smooth bell curve fitted to the same games puts only 5.3% there.
Which margins are the key numbers?
Five of them, and each one is a scoring combination. A field goal makes 3, a touchdown with the extra point makes 7, two field goals or a touchdown with a missed extra point make 6, a touchdown plus a field goal makes 10, and two touchdowns make 14.
Their measured shares of final margin sit at 15.0% for 3, 8.9% for 7, 6.3% for 6, 5.3% for 10 and 5.0% for 14. Those five numbers between them account for four games in ten. Nothing about a normal distribution produces that pattern, because a normal distribution knows nothing about how points arrive.
Why does a smooth curve get it so wrong?
A bell curve spreads probability evenly across neighboring margins, which is exactly what football does not do. At the fitted spread of about 13.1 points, the curve puts 5.3% of games on a 3-point margin against a measured 15.0%. It understates the single most common outcome in the sport by nearly three times.
That error is invisible on a spread of 8 and expensive on a spread of 3. So the NFL model separates projecting from pricing. A regression produces two numbers, a projected margin and a projected total. A second layer turns each into a probability for every whole-number outcome, with fitted weight added back at 3, 6, 7, 10 and 14. Every market price is a partial sum of one of those two distributions.
What does that change about a 3-point spread?
It changes what the half point costs. A bettor laying 3 pushes on every game his team wins by exactly 3, and a bettor laying 3.5 loses those same games instead.
The fit is checked against a tolerance at every key number, and those tolerances were pinned on 2026-08-09 before any candidate was scored. The tolerance at 3 is 0.030. The smooth-normal baseline misses by about 0.097 there. It fails the check at four of the five numbers.
Do totals cluster the same way?
Far more weakly. The most common final total, 37, holds about 3.9% of games against a smooth curve’s 3.0%, a gap nothing like the one at a 3-point margin.
The same discretizer runs on totals, tracking 37, 41, 44, 47 and 51, but no preregistered tolerance check guards it. Margins need the added weight at key numbers. Totals mostly do not.
What this does not tell you
Key numbers say nothing about which side to bet. They describe the shape of the margin distribution, not its direction. Fairline flags no value on any NFL moneyline, spread or total. All six market-and-side combinations failed a preregistered holdout before launch, and the NFL model page publishes that table. The shares here are fitted from pooled seasons with post-2015 games weighted double for the extra-point rule change, so a future rule change moves them again.