Regression to the Mean: The Statistical Force Behind Every Slump
Francis Galton discovered it in pea plants. It explains why the Sports Illustrated cover curse is real, why bad students improve after punishment, and why no hot streak lasts.
In 1886, Francis Galton was studying the relationship between the heights of parents and their children. He noticed something unexpected. Tall parents tended to have children taller than average — but shorter than themselves. Short parents tended to have children shorter than average — but taller than themselves. Extreme values in one generation moderated toward the population mean in the next.
He called this "regression toward mediocrity." We call it regression to the mean. It is one of the most important and most frequently misunderstood forces in statistics.
The mechanism
Any measured outcome is the sum of a stable underlying component (skill, true effect, structural cause) and a random transient component (luck, noise, measurement error). Extreme outcomes require extreme values of both components simultaneously. On subsequent measurement, the stable component persists — but the random component regresses toward zero.
Observed score = True ability + Random noise If a student scores 3σ above mean on a test: This requires both high ability AND positive luck. On a second test, luck reverts toward zero. Expected score on retest: mean + r × (first score − mean) Where r = correlation between two test scores (0 < r < 1) If r = 0.7: expected retest = mean + 0.7 × 3σ = mean + 2.1σ Apparent "regression" of 0.9σ is entirely statistical.
Real-world manifestations
The Sports Illustrated Cover Curse: Athletes featured on the cover are almost always at a peak performance moment. Peak performances contain extraordinary luck components. Subsequent performance regresses toward the athlete's true mean. The curse is regression to the mean, not a jinx.
Punishment and reward in education: If a student performs exceptionally poorly on a test, a teacher punishes them. The student performs better next time. The teacher concludes punishment works. In fact, the poor performance was partly bad luck — the improvement would have happened anyway. This statistical artifact led to systematic overestimation of punishment's effectiveness across 20th-century educational psychology.
| Phenomenon | Naïve interpretation | Correct interpretation |
|---|---|---|
| Fund manager outperforms market 3 years in a row, then underperforms | Lost their touch | 3-year streak was partially luck; regression inevitable |
| Drug reduces symptoms after taking at peak severity | Drug works | Patients seek treatment at peak — would improve anyway |
| Cities with high crime rates show improvement after policing interventions | Policy worked | High-crime years contain noise; partial regression expected |
How to detect it
Regression to the mean is present whenever: (a) you select subjects based on an extreme score, and (b) the score contains any random component. The stronger the random component (lower test-retest correlation), the more dramatic the apparent regression. A randomized controlled trial with a proper control group is the standard method for isolating true effects from regression artifacts — because the control group regresses at the same rate as the treatment group.