Pythagorean Expectation: Estimating a Team's Record from Runs Scored and Allowed
Bill James's Pythagorean formula turns runs scored and allowed into an expected win total. Learn the math, the 1.83 exponent, worked examples from real seasons, and how to use it on a standings table.
A baseball team's record tells you how many games it won. Its runs scored and runs allowed tell you how well it played. The two usually agree, and when they do not, the gap says something useful. Pythagorean expectation is the formula that measures the gap: it takes a team's runs scored (R) and runs allowed (RA) and returns the winning percentage a team with that run profile should have had. Multiply by the games played and you have an expected win total.
The formula is simple enough to do with a pocket calculator, and it works better than most things that look this simple. Across more than a century of data its average miss is about three wins per team-season. This article explains the formula, where it came from, why an exponent of 1.83 beats the original 2, how to apply it to a standings table, and what it cannot tell you.
The formula
The original version, devised by Bill James around 1980, is:
Expected winning percentage = R squared divided by (R squared plus RA squared)
The name comes from its resemblance to the Pythagorean theorem, in which the squares of the two shorter sides of a right triangle add up to the square of the longest. A team that scores and allows the same number of runs gets an expected percentage of exactly .500, whatever the number. A team that scores more than it allows gets above .500, and the share rises faster than the raw ratio of runs, which is the point of squaring.
Take a team that scores 800 runs and allows 700 in a 162-game season. Squaring gives 640,000 and 490,000. The expected winning percentage is 640,000 divided by 1,130,000, which is .566, and multiplying by 162 gives about 91.8 wins. An actual record of 92-70 would match the expectation almost perfectly.
The generalized form uses an exponent other than 2: the expected percentage is R to the power x divided by (R to the power x plus RA to the power x). With x set to 1.83, the same team comes out at .561, or about 90.8 wins. The 1.83 is not a mystical constant; it is the number that fits the historical data a little better than 2. The next section shows why.
Why 1.83 and not 2
If the formula were exact, every team would land on its expectation, and none does. The practical question is which exponent produces the smallest errors. A least-squares fit of the formula to the historical team seasons with at least 100 games played gives an exponent of 1.83 across all seasons from 1901 to 2025, and 1.83 again for the 1,558 team seasons since 1969. The fit is that stable, which is why 1.83 became the standard improvement on James's original 2.
The improvement in accuracy is small, which is also worth knowing. Using the 1969-and-later seasons, the average miss is 3.18 wins with an exponent of 1.83 and 3.26 wins with an exponent of 2. In either case, about 54 percent of team seasons land within three wins of the estimate, and about 80 percent land within five. The extra decimal places buy a little precision, not a different answer.
The exponent differs from 2 for a reason that is worth understanding. The formula's exponent depends on how many runs there are to go around. In a low-scoring environment a single run matters more, so the exponent should be a little lower; in a high-scoring environment it should be a little higher. One refinement, often called Pythagenpat, sets the exponent from the total runs per game in the teams' games, using a power of roughly 0.285. For most seasons it lands between 1.8 and 1.9, and it matters mainly in seasons with unusually low or high scoring.
Statisticians have also shown that the formula is not an accident. If the runs a team scores and allows in single games follow a certain kind of statistical distribution, a formula of this shape emerges, with an exponent close to 1.8. That explains why the same approach works in other sports with different exponents: basketball, with its far larger scoring totals, uses exponents well above ten.
Worked examples from real seasons
Here are the 2001 American League clubs that most illustrate how the formula works. Runs and wins come from the historical record, and the expected wins use the 1.83 exponent.
| Team (2001) | Record | R | RA | Run difference | Expected wins | Actual minus expected |
|---|---|---|---|---|---|---|
| Seattle Mariners | 116-46 | 927 | 627 | +300 | 108.8 | +7.2 |
| Oakland Athletics | 102-60 | 884 | 645 | +239 | 103.7 | -1.7 |
| New York Yankees | 95-65 | 804 | 713 | +91 | 89.3 | +5.7 |
| Cleveland Indians | 91-71 | 897 | 821 | +76 | 87.5 | +3.5 |
| Minnesota Twins | 85-77 | 771 | 766 | +5 | 81.5 | +3.5 |
| Detroit Tigers | 66-96 | 724 | 876 | -152 | 67.0 | -1.0 |
The Mariners are the headline. Their run difference of +300 is the highest in the table, and the formula says it was worth about 109 wins, which would have been the best record in the league on its own. Their actual total was 116. That is an enormous season by any measure, and it also contained several wins' worth of good fortune, most likely in close games. Oakland, with a +239 difference, was almost exactly where the formula put it.
The Yankees are the second lesson. They won 95 games with a +91 run difference, which the formula says should have produced about 89, so they outperformed it by nearly six wins. The formula is not saying the Yankees were fortunate, only that their record was ahead of their run difference, and the reasons for such gaps are discussed below. The Twins at 85-77 with a run difference of +5 are a smaller version of the same pattern. The Tigers (66-96, -152) landed within a win of their expectation, a result that is typical of poor teams.
The same exercise for 2019 gives a second look at how well the formula holds at the extremes:
| Team (2019) | Record | R | RA | Expected wins | Actual minus expected |
|---|---|---|---|---|---|
| Houston Astros | 107-55 | 920 | 640 | 106.9 | +0.1 |
| New York Yankees | 103-59 | 943 | 739 | 98.8 | +4.2 |
| Minnesota Twins | 101-61 | 939 | 754 | 97.0 | +4.0 |
| Detroit Tigers | 47-114 | 582 | 915 | 49.0 | -2.0 |
| Baltimore Orioles | 54-108 | 729 | 981 | 59.5 | -5.5 |
| Kansas City Royals | 59-103 | 691 | 869 | 64.3 | -5.3 |
The Astros hit their expectation almost to the decimal. The Yankees and Twins finished about four wins above theirs, the Tigers two wins below, and the Orioles and Royals about five wins below. All of that is normal. A team with a run difference of -333, like the 2019 Tigers, is about as bad as a team can be, and the formula puts them at 49 wins, close to their 47.
Why teams beat or miss their Pythagorean record
The gap between actual and expected wins is where the information is. Four influences matter most.
Distribution of runs. Wins come from outscoring an opponent in a single game, but the formula works on season totals. A team that wins many games by a run or two and loses a few by ten runs will have a record that beats its run difference. The 2001 Yankees fit that pattern: a run difference of +91 produced 95 wins. The reverse, a team that is blown out often and wins narrowly less often, will fall short.
Bullpen and close games. A strong bullpen, especially at the end of games, helps a club turn narrow leads into wins. Clubs with good relief pitching tend to beat their Pythagorean record, though the effect is modest and hard to carry from one year to the next.
Luck. A large part of any gap in a single season is chance, particularly in games decided by one run. Researchers have found that a team's one-run record has little predictive value from year to year, which is another way of saying it behaves like luck. See Sample Size and Regression to the Mean in Baseball Statistics for why a few games can mislead.
Park and context. The formula does not adjust for ballpark. A team playing in a high-scoring park accumulates a large number of runs scored and allowed, which does not change its expectation much, because both sides inflate. The extra detail is in Park Factors Explained.
Among the 1,558 team seasons since 1969 with at least 100 games played, the largest overperformance by the 1.83 formula belongs to the 2021 Seattle Mariners, who went 90-72 with 697 runs scored and 748 allowed. The formula expected about 76 wins (75.8), so the Mariners finished 14.2 wins ahead of it, the most in that sample. They missed the postseason. The largest underperformance belongs to the 1993 New York Mets, who went 59-103 with a run difference of -72 and were expected to win about 73 and a half. Both teams are in the tails; most seasons fall much closer to the line.
What the formula predicts
Pythagorean expectation has a second use that matters more than the first. A team's expected winning percentage is a better predictor of its future winning percentage than its actual winning percentage is. If a team is 60-40 on the strength of a +20 run difference, the formula suggests a lower rest-of-season pace than its record. The reason is the one above: the part of a record that exceeds the run difference is mostly luck, and luck does not repeat.
This is useful in three ways. For a fan trying to judge whether a surprising team is real, the run difference is the first thing to check. For the front office deciding whether to buy or sell at the trade deadline, expected wins are a more reliable reading of strength than the standings. And for a bettor or a fantasy player evaluating a team's outlook, the run difference is a cleaner starting point than the record. Wagering carries risk, is for adults of legal age where it is legal, and nothing here is advice to bet; if gambling is a problem, the US helpline is 1-800-GAMBLER.
A rule of thumb comes out of the formula. At about 700 runs scored and allowed, a change of one run in the run difference moves the expected total by about 0.12 wins with an exponent of 2, or 0.11 with 1.83. That means roughly 8.6 to 9.4 runs per win. The familiar rule of thumb of about 10 runs per win is a round number for this range, and the exact figure depends on how many runs are scored in the league.
Using it on a standings table
You can run the calculation on any standings page that lists runs scored and runs allowed. Start from the standings page and write down R, RA and games played for the team. Then:
- Raise R to the power 1.83, and do the same with RA.
- Divide the first by the sum of the two. That is the expected winning percentage.
- Multiply by the games played to get expected wins, and subtract the actual wins.
If you have no tool for exponents, use the stat calculator or a spreadsheet, where the formula is a single line. For a quick estimate without the exponent, subtract RA from R and divide by 10 to find the wins above or below 81 in a 162-game season. A +100 run difference points to about 91 wins; it will not be exact, but it will be close.
The first month of a season is where this is least reliable. With 30 games played, a team's run difference is dominated by a few blowouts, and both its record and its expectation will move a great deal. The formula works best when a season is at least a third done, and it works best of all on the full 162-game year.
What it cannot tell you
The formula is descriptive, not a theory of why a team won. It says nothing about which players were responsible, and it assumes the runs are scored and allowed in a typical pattern. It cannot see injuries that happen after a team's best stretch, trades that change the roster, or the effect of a one-game playoff. It also ignores sequencing: the same season total can come as runs bunched into a few games or spread across many, and the spread-out pattern wins more games. Run expectancy shows how the base-out state shapes how many runs an inning is likely to produce.
It is also a team formula; applying it to a single player is a loose analogy, not a calculation. Finally, it describes the regular season. In the postseason, series are short and the formula's tidy averages give way to luck. For game-level decisions, look at Win Probability, Leverage Index and Win Probability Added, which describe how a game's outcome shifts pitch by pitch.
Common questions
Who invented Pythagorean expectation? Bill James developed it around 1980 and wrote about it in his annual Baseball Abstract. The 1.83 refinement came later from fits to historical data.
Is an exponent of 2 wrong? No. It is a slightly less accurate version of the same idea. On team seasons since 1969 it misses by 3.26 wins on average against 3.18 for 1.83.
Does it work for single seasons or only for averages? It works for single seasons, with typical errors of about three wins. Individual teams can still miss by ten or more.
Why does it matter? It separates a team's quality from its luck. Over a full season, run difference is the better guide to how good a team really is.
Where to go next
Open the standings and try the formula on your favorite team; the head-to-head tools show how two clubs compare. For a deeper explanation of the number a club needs to clinch, read Magic Numbers Explained, and for more reading on measures like this one visit the stats and analytics topic page.