Skip to main content
Category: Effects
Type: Statistical Regularities & Interpretation Trap
Origin: Francis Galton’s family-height studies (1880s); generalized across domains
Also known as: Reversion to mediocrity (historical phrase)
Quick Answer — Regression to the mean says that an unusually high or low score is often followed by a value closer to the average—not necessarily because anything “fixed” you, but because extreme draws partly reflect noise. It explains illusory causation after praise, punishment, medical interventions, and sports “slumps.” Confusing it with treatment effects is one of the most expensive thinking mistakes in policy and management.

​
What is Regression to the Mean?

Regression to the mean is a pattern in repeated measurement: extreme outcomes tend to be followed by more moderate ones, even when nothing meaningful changed. The phenomenon arises whenever an observed extreme mixes true underlying level with random fluctuation—so part of the extremeness is “luck,” which is unlikely to repeat.
Extremes are often partly accident; the next draw is rarely as accidental in the same direction.
The idea appears everywhere: test scores, athletic performance, corporate earnings, and clinical symptoms. It interacts with survivorship bias (you see winners, not the full draw) and contrasts with the gambler’s fallacy, which wrongly expects balance rather than independent reversion toward an average. It complements the law of large numbers: averages stabilize as samples grow; single extreme points do not.

​
Regression to the Mean in 3 Depths

  • Beginner: After a very good or very bad day, the next day is often closer to ordinary—partly because the first day was unusual.
  • Practitioner: Before crediting a coach, drug, or policy for improvement after an extreme low, ask what naive prediction would be without any intervention.
  • Advanced: Build selection models that shrink extreme estimates toward the prior mean—this is the statistical soul of “regression” thinking.

​
Origin

Francis Galton studied the heights of parents and children, documenting that very tall parents tend to have tall—but less exceptionally tall—children, relative to the population (“regression towards mediocrity”). The finding was not moralistic; it reflected correlation less than one between measured traits across generations. Later, the insight generalized: whenever two imperfectly correlated variables are involved, predicting one from an extreme value of the other implies a less extreme prediction. Psychologists and behavioral scientists emphasized how this creates faux causality—people interpret natural reversion after selection as proof that their action worked.

​
Key Points

Regression to the mean is a lens for separating signal from post-selection noise.
1

Extremes mix signal and noise

A record month, fever spike, or test peak usually over- or under-estimates the steady underlying level.
2

Selection creates illusions

Studying only “worst cases” or “best performers” guarantees some reversion even with no treatment effect.
3

Repeat measurement reveals partial bounce-back

The second measurement is not independent in people’s stories—it is often pulled toward the mean by mathematics.
4

Controls and baselines matter

Randomized trials and historical baselines help distinguish real impact from statistical rebound.

​
Applications

Use the concept to audit praise, blame, and “what worked.”

Education & Testing

Students who scored extremely low often improve on retest partly by regression—tutor effects must beat that baseline expectation.

Sports & Performance

Rookie standouts and cold streaks frequently move toward league averages; narratives still invent “slumps” and “clutch.”

Healthcare & Wellness

People seek care at symptom peaks; improvement afterward partly reflects natural symptom fluctuation—trials need controls.

Management & KPIs

Punishing teams after the worst quarter or rewarding the best can misread luck; look at longer runs and distributions.

​
Case Study

Galton’s family-height analysis is the classic measurable illustration: when mid-parent height is very high, children’s heights remain correlated but less extreme relative to the population than their parents’—reflecting correlation below 1.0. That mathematical fact is why “tall parents, somewhat less extremely tall children” is the normal pattern, not a mysterious biological “drive to average.” Modern textbooks use the same structure to warn that any single extreme measurement—blood pressure, sales, defect rate—should be expected to move toward the long-run mean on retest unless the measurement is perfectly reliable and the world perfectly stable.

​
Boundaries and Failure Modes

Regression to the mean explains part of change, not all of it. Boundary 1: Real treatments exist
Therapy, training, and process fixes can shift the true mean—not only statistical bounce. Boundary 2: Sometimes extremes signal structural change
A sustained regime shift breaks the simple “revert to old mean” story. Common misuse: Dismissing all improvement after a crisis as “just regression”—without comparing to a credible counterfactual.

​
Common Misconceptions

Humility about luck is a professional skill.
Reality: It predicts movement toward the mean from extremes, not identical outcomes for everyone.
Reality: The gambler’s fallacy expects balance in independent sequences; regression arises from measurement error and imperfect correlation.
Reality: Extremes on “before” almost guarantee some movement without any intervention—design comparisons accordingly.
Pair these when judging evidence and selection.

Survivorship Bias

Why visible winners hide the full pool of tries.

Law of Large Numbers

Why averages stabilize as sample size grows.

Gambler’s Fallacy

A different error about sequences and independence.

​
One-Line Takeaway

If you start from an extreme, expect the next datapoint to look more ordinary—even if you did nothing.