Expect extreme results to regress toward average

Unusually good or bad performance predicts more average performance next time — account for this before giving praise or blame.

Why it works

Regression to the mean is a mathematical inevitability whenever performance involves both skill and chance. The representativeness heuristic causes people to ignore it: they explain extreme outcomes as if they fully reflect the underlying reality ("they’ve turned a corner" / "this person is hopeless"), rather than partly reflecting chance fluctuation. This produces the famous intervention illusion: harsh feedback after a bad performance "works" and praise after a good one "hurts" — simply because extreme outcomes regress, regardless of the intervention.

How to do it

1. After observing an unusually high or low result, ask: "How much of this could be due to chance fluctuation around the true average?"
2. Predict that the next result will likely be closer to the person’s or process’s long-run average, not as extreme.
3. Before attributing a change to an intervention, verify it with more than one or two follow-up observations.
4. Separate feedback on what you can observe from conclusions about underlying talent or state.

Evidence

Regression to the mean is a mathematical property of correlated variables (Francis Galton identified it in 1886). Kahneman’s intervention illusion example (flight instructors praising vs. criticizing) illustrates how ignoring regression causes false causal attributions; this is a teaching example, not a formal study. (mechanistic)

How much any result regresses depends on how strongly skill vs. luck determines it; regression matters more in high-variance, luck-influenced domains than in highly skill-determined ones.

Sources

• Galton (1886), Regression towards mediocrity in hereditary stature, Journal of the Anthropological Institute of Great Britain and Ireland

Common mistake

Treating regression as a reason to ignore feedback — real skill improvements happen, but they should be confirmed over multiple observations, not inferred from a single extreme outcome.