Error of the Day · Thinking

Regression to the mean: why the scolding seemed to work

A terrible week is usually followed by a better one, whatever you did in between. Credit your intervention for that and you've learned the wrong lesson, as a famous psychologist discovered at an air force flight school.

WrongI yelled at the team after their worst quarter and the next quarter was better, so yelling works.

RightTheir worst quarter was partly bad luck, so the next one was likely to be better anyway. I'd need a comparison group to know whether yelling did anything.

Memory trick After an extreme, expect ordinary. Don't hand the credit to whatever happened in between.

What people get wrong#

Something goes unusually badly. You step in: a pep talk, a new diet, a new manager, a stern email. Things improve. Obviously, your intervention worked.

Maybe. But results that are unusually high or low are usually followed by results closer to average, even when nobody does anything. That is regression to the mean, and the error is treating it as proof of cause. Statisticians sometimes call that mistake the regression fallacy.

The flight-school story#

The best-known illustration comes from Daniel Kahneman. In his Nobel Prize autobiography, he describes trying to convince Israeli Air Force flight instructors that praise beats punishment for teaching skills. A veteran instructor pushed back: when he praised cadets for a clean maneuver, "in general when they try it again, they do worse," and when he screamed at them for a bad one, "in general they do better the next time" (Nobel Prize: Kahneman).

The instructor's observations were accurate. His conclusion was not. Kahneman calls it the most satisfying "Eureka" moment of his career: because we reward people when they do well and punish them when they do badly, and because performance regresses to the mean, "it is part of the human condition that we are statistically punished for rewarding others and rewarded for punishing them."

He then ran a demonstration: each participant tossed two coins at a target behind his back, with no feedback. Those who did best on the first throw mostly did worse on the second, and vice versa. Nobody had praised or yelled at anyone.

Why it happens#

Any measured result is part skill and part luck. An unusually great day usually means skill plus good luck; an unusually awful one means skill plus bad luck. Next time, the skill is still there, but the luck is redrawn, and it is more likely to be ordinary. So the next result drifts back toward the person's typical level.

The idea is old. In 1886, Francis Galton reported that the offspring of seeds "did not tend to resemble their parent seeds in size, but to be always more mediocre than they," smaller than unusually large parents and larger than very small ones. He found the same pattern in human height and called it "regression towards mediocrity" (Galton, 1886). Galton's work is where the statistical term regression comes from.

Regression is not a force pulling things back, and it doesn't mean a streak is "due" to end. It is simply what happens when you pick a case because it was extreme and then measure it again.

Where it fools people#

  • Medicine. People see a doctor when symptoms peak, and peaks tend to subside. Epidemiologists Adrian Barnett, Jolieke van der Pols and Annette Dobson warn that the effect can make natural variation look like real change, that it grows with measurement error, and that it is strongest when you follow up only the people selected for extreme baseline values. Their conclusion: it "should always be considered as a possible cause of an observed change" (Int J Epidemiol, 2005).
  • Management. The worst-performing branch gets a turnaround consultant and improves. The star salesperson gets an award and has an average next quarter.
  • Sports. A rookie's record-breaking season is followed by a "sophomore slump."
  • Policy. A safety measure installed right after a spike in accidents looks effective when the next year is quieter.

In each case something real might also be happening. The point is that the before-and-after comparison can't tell you.

The fix#

Our worst store got the new training program and sales jumped 20%, so the program works.

Our worst store got the new training program and sales jumped 20%; let's see how similar low-performing stores without it did.

Stores chosen for a terrible quarter would tend to bounce back anyway.

  • Ask how the case was chosen. If it was picked for being extreme, expect some return toward average no matter what.
  • Find a comparison group that was equally extreme but didn't get the intervention. Randomized controls exist largely for this reason.
  • Measure more than once before acting. A baseline built from several readings is less likely to be a fluke.
  • Judge feedback by long-run results, not by what happened on the very next attempt.

For more ways intuition misreads data, see our guide to cognitive biases.

Sources

Every factual claim on this page links to one of these. Checked November 22, 2026. See how we source.

  1. Nobel Prize: Daniel Kahneman, Biographical nobelprize.org
  2. Galton.org: Francis Galton, Regression towards Mediocrity in Hereditary Stature (1886, PDF) galton.org
  3. PubMed: Barnett, van der Pols & Dobson, Regression to the mean: what it is and how to deal with it (Int J Epidemiol, 2005) pubmed.ncbi.nlm.nih.gov