Does Goal Difference Predict League Points? Pythagorean Expectation in Football Explained

Pythagorean expectation estimates how many points a football team should have earned from the goals it scored and conceded. Borrowed from baseball, it treats the ratio of goals for to goals against as a measure of strength. It explains most of a league table, though not all of it, and the goal totals it needs come straight from results on RubiScore.

This data-study asks one question: how well do goals scored and conceded predict a team's league points? It covers the idea's origin, how the published football versions differ, a worked calculation and the limits. Final scores, standings and match pages at https://rubiscore.com carry the season totals of goals for and against that every version starts from.

Where Does Pythagorean Expectation Come From?

Baseball writer Bill James introduced the formula around 1980. It estimates a team's winning percentage from runs scored and runs allowed: runs scored squared, divided by the sum of runs scored squared and runs allowed squared. The name comes from its resemblance to the Pythagorean theorem.

It works well in baseball. The 2002 New York Yankees scored 897 runs and allowed 697, which the formula turns into a winning percentage of .624, or about 101 wins in 162 games. They won 103.

Later work found that a single-number exponent of about 1.83, rather than 2, fits baseball better, and Baseball-Reference uses it. Other sports need very different values. Basketball, where chance plays a smaller role, uses about 13.9, an adaptation credited to analyst Daryl Morey. The exponent is a fitted number, not a law.

Why Is Football Harder to Fit Than Baseball?

Goals are scarce and draws are common. Baseball has no ties, so equal runs scored and allowed implies a .500 record. In football the same balance of goals can produce many mixes of wins, draws and defeats, and a win is worth three points against one for a draw. In the Premier League between 2000/01 and 2016/17, 3,304 of 12,920 matches, or 25.6 percent, ended level, according to a study by Katherine Reinmuth and Paul Sommers.

Researchers have handled the problem in different ways:

• Hamilton (2011). Howard Hamilton's paper in the Journal of Quantitative Analysis in Sports estimates points per game instead of win percentage. It fits goal distributions to a Weibull distribution and finds that a league's fitted exponent stays stable across leagues in the same year and within a league across seasons.
• Bertin (2016). Writing for StatsBomb, Michael Bertin predicted the share of available points using the older two-points-for-a-win convention and found an exponent of about 1.2. He converted the result to a three-point scale and derived a linear shortcut: in a 38-match league, expected points are roughly 52 plus two-thirds of goal difference.
• Reinmuth and Sommers (2018). In the Journal of Student Research, they found that with draws included, no simple single-exponent formula explains Premier League points. With draws excluded, the ratio of wins to losses was close to the ratio of goals for to goals against raised to the power 1.70.

Bertin notes that Hamilton himself argued a conventional single-exponent version may not be possible for football. The exponents 1.2 and 1.70 do not conflict, because they predict different quantities, a share of points on one scale and a win-loss ratio on the other.

What Is the Formula in Words?

A team's expected share equals its goals scored raised to a power, divided by the sum of its goals scored and its goals conceded, each raised to that same power. Equal goals for and against give exactly one half, and a higher exponent makes a given goal ratio count as more decisive. The football versions change what is predicted, either points per game, a share of points or a win-loss ratio, and fit the exponent to match.

How Do You Calculate It? A Hypothetical Walk-Through

Every number in this section is hypothetical, chosen only to show the arithmetic. It is not a real club or season. Take a team that scores 60 goals and concedes 40 in a 38-match league with three points for a win.

1. Divide goals for by goals against: 60 ÷ 40 = 1.5.
2. Raise the ratio to the fitted exponent, here the 1.70 from the Premier League study. 1.5 to the power 1.70 is about 1.99, so wins should outnumber losses roughly two to one.
3. Assume the 25.6 percent draw share from that study. That is about 9.7 draws in 38 matches, leaving about 28.3 decisive matches.
4. Split the decisive matches in the ratio of about 1.99 to 1. That gives roughly 18.8 wins and 9.5 losses.
5. Convert to points: 3 × 18.8 + 9.7 draws is about 66 points.

Now run the shortcut as a check. Goal difference is +20, so 52 + two-thirds of 20 is about 65 points. The two routes land within about a point here. That agreement is a feature of this hypothetical team, not a guarantee, because the methods were fitted on different data. The shortcut also gives a rule of thumb for 38-match leagues with three points for a win: each extra goal of goal difference is worth about two-thirds of a point.

How Well Does It Predict Points?

The evidence is strong but not tight.

• Win-loss ratios. In the Premier League sample from 2000/01 to 2016/17, Reinmuth and Sommers reported that more than 90 percent of the variation in teams' win-loss ratios could be explained by goals scored and allowed.
• Points totals. Bertin tested his versions on 300 team-seasons from La Liga, Serie A and the Premier League, covering 2010/11 to 2014/15. After converting to three-point scoring, the two-step formula had a root-mean-square error of 4.35 points. The linear shortcut had 4.7, a mean absolute error of 3.57, and 92 of 300 predictions within one point.
• Other versions. In Bertin's account, Hamilton's model sat at 4 to 5 points of error, and Martin Eastwood's version at 4.08.

So a typical miss is about four to five points over a 38-match season, roughly one and a half wins. That is small next to the spread of a full table, but large enough to matter in a tight finish.

What Does Goal Difference Capture?

Goal difference carries more information than the points column, in two ways.

• Margin. A 3-0 win and a 1-0 win both earn three points. Goal difference records that the first was more convincing.
• Close-game luck. A side that wins many 1-0 matches and loses a few heavily will tend to collect more points than its goal difference implies. The reverse pattern tends to collect fewer.

Where Does the Method Break Down?

No fitted formula removes these confounders.

• Margins beyond one goal. A team that keeps attacking at 9-1 down, rather than minimising the damage, produces points that goal difference cannot explain. Bertin found that the biggest under-predictions in his sample came largely from one unusually open attacking side.
• Game state. Teams that lead often protect the lead and teams that trail push forward, so final goal totals partly reflect the scoreline along the way.
• Penalties and red cards. A converted penalty is a single event worth a full goal, and a dismissal reshapes a match. Both move goal difference without necessarily showing underlying strength.
• Schedule strength. Goal totals are raw. They do not adjust for opponent quality or the balance of home and away matches, which matters most mid-season.
• Small samples. Bertin's shortcut is almost meaningless early in a season. After one 1-0 win it predicts about 53 points, against three actually earned, unless it is scaled by matches played.
• Table adjustments. Points deductions and awarded results change points without changing goals.

How Is It Different From Expected Points?

Pythagorean expectation starts from goals that were scored. Expected points built on xG start from the chances that were created. The first asks what the final scorelines imply, and the second asks what the shot quality implies.

That makes them layers rather than rivals. Actual points, goals-based expected points and chance-based expected points form a ladder, and each step removes a different source of noise. The first gap points to close-game results and the distribution of margins. The second points to finishing and goalkeeping, which Pythagorean expectation inherits because goals already contain them.

How Should You Read Points vs Expected Points Gaps?

A gap is a question, not a verdict. A short routine keeps it honest.

1. Compare the gap with the typical error. With misses of four to five points common, a gap of two or three points is not a story. A hypothetical team on 58 points whose goals imply 65 is seven points adrift, which is outside the usual band and worth checking.
2. Check the sample. Wait until most of the season has been played, or scale the formula by matches played.
3. Count the close results. Final scores and standings on RubiScore make it quick to tally one-goal wins and one-goal defeats, which is where most gaps are made.
4. Compare with a chance-based table. If xG-based expected points tell the same story, the case is stronger. If they disagree, finishing or goalkeeping is likely involved.
5. Check the context. Look for red cards, penalties, a lopsided early schedule and any points deductions.

None of the studies cited here shows that gaps reliably reverse next season, so treat a gap as a prompt to look closer, not a forecast.

Verdict

Goals do most of the work. Across the samples cited here, a goal-based estimate typically misses a team's actual total by about four to five points, and goal ratios explain more than 90 percent of the variation in Premier League win-loss ratios.

The caveats are real. Draws force every football version to depart from the baseball original, the published exponents are not interchangeable, and margins, game state, penalties and schedule all blur the picture. Treated as a sanity check on a league position rather than a forecast, Pythagorean expectation is a sound tool.