All scores →
Loading…
Loading…
Loading…

Pythagorean wins: using points to predict records

By 4Basketball Staff · · 11 min read

Pythagorean wins turn the points a team scores and allows into an expected record. Learn the formula, work through an example, and see why luck, injuries and trades push teams away from it.

A team's win-loss record tells you what happened. It does not always tell you what should have happened. A team that wins eight games in the final seconds and loses a few blowouts can finish with a handsome record that its scoring does not support, while a team that loses a string of one-possession games can end up under .500 despite outscoring its opponents all year. For decades, analysts have used a compact formula to separate the two stories. It is called the Pythagorean expectation, and it is one of the most useful and least intimidating tools in basketball analysis.

The idea is this: the share of games a team should win depends on the share of the total points it accounts for. Teams that score a lot more than they allow tend to win a lot of games. Teams that score and allow about the same amount tend to win about half. The formula puts numbers on that tendency, and it gives you an expected winning percentage you can compare against the actual one.

In this article we explain where the name comes from, how the formula works, what exponent to use in basketball, and how to put it to work, including a worked example. We also look at why real teams drift from their expected records, which is where the most interesting questions live. If you have read our piece on net rating, you will recognize the family resemblance, and you can see current records on the standings page.

Where the idea comes from

The method was invented for baseball by Bill James, the writer who did more than anyone to bring statistical rigor into the sport. He noticed that a baseball team's winning percentage was very close to the square of its runs scored divided by the sum of the squares of its runs scored and runs allowed. Because squaring and summing squares looks like the Pythagorean theorem, the name stuck, even though no triangle is involved.

The basic template is:

winning percentage is approximately points scored raised to a power, divided by the sum of points scored raised to that power and points allowed raised to that power.

In baseball, the power, or exponent, is about two. In basketball it is much larger. The reason is that basketball scoring is more deterministic: with so many possessions and such a large share of them going to the better team's advantage, a small edge in points translates into a large edge in winning chances. A baseball team that outscores opponents by ten percent might win roughly 55 percent of its games; a basketball team with a ten percent edge in points would win close to 80 percent. To capture that steepness, the exponent for the NBA has been estimated at around 14 by many analysts, with various studies suggesting values in the range of roughly 13 to 17 depending on the period and the method of fitting. College versions of the formula, which are fit to a different level of competition and a different number of possessions, tend to use a smaller exponent, closer to 10.

You do not need to memorize any of these numbers. The point is that the exponent is large, that basketball is steep, and that the formula is not very sensitive to the exact value. Using 13.9 or 14.3 or 14 gives expected wins within a fraction of a game of each other for most teams.

How to compute it

The calculation takes three steps.

  1. Take the team's total points scored and total points allowed for the season (or the per-game averages; the result is the same).
  2. Raise each to the chosen power, say 14.
  3. Divide the scored figure by the sum of the two. Multiply by the number of games to get expected wins.

Because raising big numbers to the 14th power is awkward by hand, it is easier to work with the ratio. If you divide points scored by points allowed, call it R, then the expected winning percentage equals R to the power 14 divided by one plus R to the power 14. A spreadsheet or calculator makes this trivial.

A worked example

Consider a hypothetical team that scores 8,900 points over an 82-game season and allows 8,500. That is an average score of about 108.5 to 103.7, or a margin of roughly plus 4.9 per game.

The ratio R is 8,900 divided by 8,500, or about 1.047. Raised to the 14th power, that comes out to about 1.90. The expected winning percentage is therefore 1.90 divided by 2.90, or about .656. Multiply by 82 games and you get roughly 54 expected wins.

Now compare that with the rule of thumb linking margin of victory to wins: each point of average margin is worth somewhere around two and a half wins over a full season. A margin of nearly 5 points suggests about 12 wins above .500, which is 41 plus 12, or 53 or so. The two methods agree within a game, which is just what you would expect, since they are two ways of expressing the same relationship.

Another feature of the formula shows up at the symmetric point: a team that scores exactly as many points as it allows has a ratio of one, and one raised to any power is one, so the expected winning percentage is exactly one half. That sounds trivial, but it is a helpful check that the formula behaves sensibly.

What the formula is good for

Judging a record

The most direct use is to ask whether a team's record is flattering or harsh. Put the actual wins next to the expected wins. A team that has won 10 more games than the formula suggests is a candidate to have been fortunate. A team that has won 6 fewer than expected is a candidate to have been unlucky. Neither verdict is certain, but the pattern is informative, because the sources of the gap have a characteristic profile, which we examine below.

Forecasting what comes next

Pythagorean wins predict future winning percentage better than past winning percentage does. That finding has been repeated in many studies of many sports. The reason is the one you would suspect: points scored and allowed are, in a sense, a larger sample of the team's true ability than wins and losses are. A single game gives one bit of information, win or loss, but it also gives a margin, which says how decisively. Using the margin throws away less information.

This does not mean the formula is a crystal ball. It means that, if you have to guess how a team will perform in the second half of a season, its points-based expected record is a better starting point than its actual record. The same applies from one season to the next: teams with large gaps between actual and expected wins tend, on average, to see the gap shrink the following year. This is the same behavior described in regression to the mean.

Building other tools

Pythagorean wins are a building block for other ideas. A season simulation, for instance, needs to translate team strength into the probability of winning each game, and the formula's logic is one way to do that. We discuss the construction of those tools in simulating a season: how win probabilities are built. Rating systems like Elo use a different mechanism but arrive at a similar conclusion: margin matters, and the best guide to a team's quality is how much it outscores opponents, not just whether it wins.

Why real teams drift from the formula

If the formula were perfect, every team's record would match it. They do not, and the deviations teach you something. Here are the main sources.

Luck in close games

This is the largest and most common source. A team that goes 15-5 in games decided by five points or fewer will beat its expected record, and a team that goes 5-15 will fall short. Outcomes in close games have a strong random component. Analysts who have looked at how well teams repeat their close-game records from one year to the next generally find that the repeatable part is small. A few coaches and rosters may have a real edge in late-game execution, but most of the variation in close-game records appears to be chance. Our article on clutch statistics covers why so little of this is persistent.

To appreciate how large the luck can be, consider a perfectly average team that wins each game with a probability of one half. Over 82 games, the standard deviation of its win total is about 4.5 wins. That means a team that is truly average can finish anywhere from the high 30s to the mid 40s with ordinary good or bad luck. Every team's record has that kind of noise built in, so gaps of three or four wins between actual and expected totals are entirely ordinary.

The shape of the margins

The formula treats a 30-point win and a 3-point win in terms of total points. Two teams with the same total points scored and allowed can have very different distributions of results. One may win 50 games by narrow margins and lose 32 in blowouts. Another may win and lose evenly. The first team will tend to outperform its Pythagorean total, and the second will track it more closely. In general, the teams that win lots of close games and lose a few by big margins finish above the formula.

That pattern is not a mistake in the formula, it is a feature of the data, and it often has a basketball explanation: a team that clinches a game early and rests its starters will see its margin shrink in the fourth quarter, while a team that is losing badly may give up on defense. The raw margin is distorted in both directions. Some analysts therefore trim blowouts, or reduce the weight of margins beyond a certain size, to improve the fit.

Injuries and changing rosters

The formula uses the whole season's points. If a star is lost for half of the schedule, the points total reflects a mixture of two different teams. The record, which counts games won rather than points, may be a better representation of one or the other. When a team trades for several rotation players at the deadline, the season totals tell you about the team that existed before and the team that exists after, blended together. In these cases the gap between actual and expected wins may be less about luck than about the fact that the formula is averaging a changing object.

This is one reason the formula is most trustworthy for teams that stayed healthy and intact. It is also a reason to compute the expected record for sub-periods, say the games before and after a major change, when the question is how good the current team is.

Schedule and rest

A team that plays a weaker schedule scores and allows points against weaker opponents. Its Pythagorean record is therefore inflated, and so is its actual record, so the two can agree while both overstate the team. The remedy is to adjust margins for opponent quality first, the approach outlined in strength of schedule and how to adjust for it, and then apply the formula to the adjusted numbers.

Using it in practice

A compact checklist makes the formula easy to apply.

  • Get the totals. Points scored and allowed across the same set of games.
  • Pick the exponent. Use roughly 14 for the NBA, and note that other leagues and levels call for adjustments. The result is not sensitive to small changes.
  • Compute the expected win percentage and convert to wins. Compare with the actual wins.
  • Read the gap with care. A gap of two or three wins is noise. A gap of eight or more is a signal that luck, a lopsided pattern of margins or a roster change is at work.
  • Look at the split. If the team's record in close games is far out of line with its overall strength, that points to luck as the explanation.
  • Do not forget the sample. In a shorter season, or in a league such as the WNBA with fewer games, the same win-loss randomness is a larger share of the picture, and the expected-record gap needs more careful reading.

One more point deserves emphasis. Pythagorean wins describe what a team's scoring record implies, not what it will do. The formula does not know about next week's injuries, the trade that has not yet happened or the opponent coming in on a rested night. It is a diagnostic, not a prophecy. If you want a view of the whole league through the lens of results and ratings, the rankings page puts several of these ideas side by side.

Why a points-based view is so persuasive

There is a deeper reason that the points-based record predicts better than the actual record, and it is worth spelling out because it gives you a way of seeing many other statistics. Wins and losses are a coarse summary of a game. They compress a rich story into a single bit. If two teams meet and one wins by 1, the standings treat that result exactly as they would a win by 30, though one tells you the teams are close and the other tells you they are not.

Points are a finer measuring stick. They carry information about how good each team was on that night, and over many games, the average margin converges on the underlying gap in quality with far less noise than the win percentage does. A statistician would say that margin is a more efficient estimator. A basketball fan can simply think of it as listening to the entire conversation instead of just the last sentence.

This idea reaches beyond Pythagorean wins. It is why net rating, adjusted efficiency and nearly every serious team rating start with points rather than outcomes. It is also why a fan should be cautious about narratives built on a single record. The question to ask is how a team won, and by how much, and against whom, and not only how many times.

What to take away

Pythagorean wins give you a quick, honest check on a record. The formula turns points scored and allowed into an expected winning percentage, using a large exponent of about 14 for the NBA, and it usually lands within a few wins of reality. When a team is far above or far below its expected total, the cause is usually luck in close games, a skewed pattern of blowouts, a roster that changed during the season or a lopsided schedule. Use the formula to question a record, not to replace it, and combine it with other evidence, such as schedule-adjusted ratings and healthy-roster sub-samples, whenever the stakes of the question are high.

Search more on this

Keep reading

The four factors: what wins basketball games

Shooting, turnovers, rebounding and free throws decide who scores more per possession. Learn how the four factors work, how they are weighted, and how to use them to see why a team is winning or losing.

4Basketball Staff · May 27, 2026 · 12 min read

Elo ratings in basketball: the idea and its uses

Elo began as a chess rating and now powers many basketball forecasts. This article explains the formula with a worked example, the adjustments basketball needs, and what an Elo rating can and cannot tell you.

4Basketball Staff · Jun 4, 2026 · 12 min read

Small samples in basketball: when a hot streak is noise

A hot week or a cold month often says less than it seems. Learn how small samples fool us, how big a sample you need for common stats, and a simple way to shrink extreme numbers toward reality.

4Basketball Staff · Jun 15, 2026 · 11 min read