Pythagorean Expectation, Explained

After reading this you will be able to turn any team's points scored and allowed into an expected winning percentage, compare it to their actual record, and read the gap as luck rather than skill.

What Pythagorean expectation is

Bill James noticed something about baseball teams in the late 1970s. If you know how many runs a team scored and how many it allowed, you can predict its winning percentage almost as well as the win column itself does. He wrote the relationship as a ratio of squared run totals, which looked like the Pythagorean theorem, so the name stuck. It has nothing to do with right triangles.

The idea now runs across every scoring sport. Feed in points for and points against, pick an exponent tuned to the sport, and you get the record a team "deserved" given how it outscored opponents. A team that won 95 games but only deserved 88 got lucky in close games. That gap is the useful part.

Here is the hook. The 2016 Texas Rangers went 95-67 in real life but scored only 765 runs while allowing 757. Their run differential of +8 points to a team that should have been roughly 82-80. They beat their expected record by about 13 wins, one of the largest gaps in modern baseball, and the next season they fell to 78-84. Run differential saw it coming.

The formula and why it works

The expected winning percentage is:

W\% = \frac{S^n}{S^n + A^n}

Here S is points, runs or goals scored, A is the same allowed, and n is the exponent for the sport. Expected wins are just W\% \times G, where G is games played.

The intuition: if you score exactly as many as you allow, then S = A and the fraction is 0.5, no matter what n is. That is correct. A break-even team should win half its games. As S pulls ahead of A, the ratio climbs toward 1. The exponent controls how steeply. A larger n means a small scoring edge translates into a large winning edge.

Why does the exponent differ by sport? It reflects how much scoring variance separates good teams from bad ones per game. In basketball, teams score around 110 points a night, so a 5 percent scoring edge is huge and reliable, which forces a high exponent near 14. In baseball, single runs swing games and randomness is large, so the exponent sits near 1.83. These values were fit to historical data, not derived from theory.

Common exponents by sport
Sport	Exponent n	Typical season total
Baseball (MLB)	1.83	700 to 850 runs
Basketball (NBA)	~14	8,500 to 9,500 points
American football (NFL)	2.37	300 to 500 points
Ice hockey (NHL)	2.15	200 to 300 goals
Soccer (goals)	~1.35	40 to 90 goals

The exponents are averages fit across many seasons. Any single team can miss its Pythagorean projection by several wins in a year without anything being wrong. The formula describes a tendency, not a law.

A worked example with the demo data

An 82-run differential over 162 games

The demo loads a baseball team that scored 745 runs, allowed 678, and played 162 games. Use the baseball exponent 1.83.

Raise scored to the exponent: 745^{1.83} \approx 172{,}600.
Raise allowed to the exponent: 678^{1.83} \approx 144{,}900.
Add them: 172{,}600 + 144{,}900 = 317{,}500.
Divide: 172{,}600 / 317{,}500 \approx 0.5436.
Multiply by games: 0.5436 \times 162 \approx 88.1 expected wins.

So this team should finish about 88-74. If you also enter an actual win total, say 92, the tool reports a luck figure of 92 - 88.1 = +3.9 wins. That is a modestly lucky season, the kind that often shrinks the following year.

You can check the powers on any calculator. The exponent step is the only part that needs a machine: 745^1.83 means 745 raised to the 1.83 power, which is e^(1.83 × ln 745) if you want to verify by hand.

How the exponent bends the curve

The single most instructive move is to hold the score fixed and slide the exponent. Watch how the same run differential produces a different expected winning percentage in each sport.

The same 745-678 differential yields 0.544 in baseball but would imply 0.765 at a basketball-style exponent of 14.

Reading the luck gap

The difference between actual and expected wins is the number worth watching. A positive gap means a team won more than its scoring suggests, usually by winning a disproportionate share of one-run or one-goal games. A negative gap means the opposite.

Pythagorean wins: Expected wins from the formula, rounded to the game.
Luck (or cluster luck): Actual wins minus Pythagorean wins. Over a full season, gaps beyond about \pm 6 in baseball are rare.
Run (or point) differential: Scored minus allowed. This is the raw input, and on its own it predicts next season better than this year's win total.

Rough guide for a full MLB season: a gap under \pm 3 wins is ordinary noise, \pm 3 to \pm 6 is notable, and beyond \pm 6 is extreme and almost always regresses. In the NFL the season is only 17 games, so the same formula produces gaps that look large in percentage terms but tiny in raw wins. Read every gap against the length of the season.

If you want a single number to bet on for next season, use the expected record, not the actual one. Teams that overshoot their Pythagorean projection tend to fall back toward it, and teams that undershoot tend to rise.

The Rangers won 13 more games than their +8 run differential justified, then regressed the next year.

Common mistakes

The most frequent error is mixing units. If you enter total season points for one team and per-game points for the other, the ratio is meaningless. Both inputs must cover the same span, usually a full season.

The second error is using the wrong exponent. Plugging basketball scores into the baseball exponent of 1.83 flattens the curve badly. An NBA team that outscores opponents by 5 points a game is a title contender, but at exponent 1.83 that edge looks like a coin flip. Match the exponent to the sport.

Soccer breaks the formula because draws are common and the model has no slot for them. It treats every match as a win or a loss, so a soccer Pythagorean figure is a rough strength index only, not a projected points total. Do not convert it directly into league points.

The third mistake is overreading a small sample. Ten games is not enough to trust the gap. A single blowout can swing a team's differential by 30 points in football and distort the estimate. The formula settles down as games accumulate.

Related tools on this site

Pythagorean expectation sits alongside other quick estimators here. For the raw baseball inputs that feed it, the Batting Average and OPS Calculator and the ERA and WHIP Calculator summarize scoring and run prevention. On the basketball side, the True Shooting % Calculator measures the efficiency behind those points. For football, the Passer Rating Calculator scores the passing that drives point totals. In cricket the closest cousins are the Cricket Run Rate Calculator and the Net Run Rate Calculator, which rank teams by scoring rate rather than by wins.

Frequently asked questions

Why is the baseball exponent 1.83 and not 2?

James first used 2, which was close. Later fits across thousands of team seasons found that 1.83 minimizes the average error, so it became the standard. The difference is small: a 745-678 team gets 0.5436 at 1.83 versus 0.5477 at 2, about two-thirds of a win over 162 games.

What is a "good" luck gap?

There is no good or bad, only large or small. A gap of +2 wins in baseball is noise. A gap of +10 is extreme and signals a record built on close-game results that rarely repeat.

Can I use it mid-season?

Yes, but scale your expectation to games played. After 40 baseball games the estimate is noisy, and gaps of several wins mean little. By game 100 the formula is fairly stable.

Does it work for soccer?

Only loosely. Because draws are so common and the formula has no draw category, treat a soccer result as a comparative strength number, not a projected points haul.

Which exponent should I pick for a custom sport?

Start near 2 and adjust. If small scoring edges reliably decide games (high-scoring, low-variance sports), raise the exponent. If single scores swing outcomes and upsets are common, lower it toward 1.5.

Try it yourself in the tool