Pythagorean Wins: Why Point Differential Predicts the NFL

The Idea: Records Lie, Point Differential Doesn't (As Much)

A team's win-loss record is the thing everyone looks at, but it is a surprisingly noisy summary of how good a team actually was. Win a fistful of one-score games and your record outruns your performance; lose those same coin-flips and it undersells you. Pythagorean expectation cuts through that noise by asking a different question: given how many points a team scored and allowed, how many games should it have won?

The method was borrowed from baseball - Bill James named it after the Pythagorean theorem because the original formula resembled one - and adapted to football. The insight that makes it powerful is not just descriptive. A team whose actual record sits far above or below its Pythagorean expectation tends to regress toward that expectation the following season, which makes Pythagorean wins a better predictor of next year than this year's actual wins.

The takeaway up front: point differential is more stable than record. Much of the gap between a team's record and its scoring profile is close-game variance - and variance does not repeat reliably.

The Formula

Pythagorean win percentage is a ratio of points scored, raised to an exponent, against total scoring:

$$ \text{Win\%} = \frac{\text{PF}^{x}}{\text{PF}^{x} + \text{PA}^{x}} $$

where PF is points for, PA is points against, and x is an exponent fit to the sport. For the NFL, research from Football Outsiders and Pro-Football-Reference has long put the exponent at roughly:

x ≈ 2.37 (the well-established NFL Pythagorean exponent)

To turn the win percentage into expected wins, multiply by the number of games in the season:

$$ \text{Expected Wins} = \text{Win\%} \times \text{Games} $$

Why an exponent at all? The exponent controls how strongly point differential maps to wins. A larger exponent means scoring margin separates teams more sharply. The value near 2.37 is the one that historically best fits actual NFL outcomes; you will occasionally see slightly different figures from different datasets, but 2.37 is the standard reference.

A Worked Example

Illustrative: a team that scores 400, allows 340

Round, hypothetical numbers chosen to demonstrate the arithmetic.

Plug PF = 400 and PA = 340 into the formula with x = 2.37:

Win% = 400^2.37 / (400^2.37 + 340^2.37)

Raising each to the 2.37 power and taking the ratio gives a win percentage of roughly 0.59. Over a 17-game season:

~0.59

Pythagorean win %

~10

Expected wins (17 games)

+60

Point differential

So this scoring profile "deserves" about 10 wins. If the team actually finished 12-5, it likely got lucky in close games and is a regression candidate. If it finished 8-9, it probably underachieved its scoring and may be undervalued heading into next year.

Why It Predicts Next Season

Here is the genuinely useful part. Two teams can both finish 11-6, but one outscored opponents by a wide margin while the other squeaked through a string of three-point wins. Their records are identical; their underlying quality is not. The wide-margin team's record matches its play, so it is on solid ground. The close-win team's record is propped up by a skill that is mostly luck - close games are close to coin flips, and coin flips do not stay hot.

The Regression Principle

A team that wins far more games than its Pythagorean expectation tends to decline next season; a team that wins far fewer tends to improve - even if nothing about the roster changes. Point differential carries forward; close-game record largely does not.

This is why analysts and bettors look at the gap between actual wins and Pythagorean wins as a regression signal. It is one of the simplest, most durable tools for finding teams the market may have over- or under-rated based on a shiny or ugly record. For how to act on signals like this, see our guide to betting on regression.

Caveats and Limits

Not all close-game record is luck

A great quarterback, a reliable kicker, or a strong situational coaching staff can earn a few extra one-score wins repeatedly. Pythagorean expectation assumes most of it is variance, which is usually but not always true.

Rosters change

A quarterback injury, a major trade, or coaching turnover can break the link between last year's scoring and next year's team. Regression toward Pythagorean assumes rough continuity.

Blowouts distort margin

A couple of garbage-time-inflated routs can pad point differential without reflecting true strength. Margin is more stable than record, but it is not immune to noise.

It ignores schedule

Raw PF and PA do not account for opponent quality. Pair Pythagorean wins with a strength-of-schedule adjustment for a fairer read.

The bottom line

Pythagorean expectation estimates how many games a team's scoring deserved using Win% = PF^x / (PF^x + PA^x), with the NFL exponent around 2.37, then multiplies by games played. Because point differential is far more stable than close-game record, the gap between a team's actual wins and its Pythagorean wins is a reliable regression signal - over-performers tend to fall back and under-performers tend to climb. Just remember it is a baseline, not a verdict: quarterback play, injuries, and schedule can all bend the relationship.