Regression to the Mean

An unusually good or unusually bad result tends to be followed by a more average one, for purely statistical reasons, which means whatever happened right before that shift often gets credited or blamed for something it didn’t actually cause.

4 min read

What Is It?

Francis Galton first documented regression to the mean in the 1880s, studying the heights of parents and children: unusually tall parents tended to have somewhat shorter children, and unusually short parents tended to have somewhat taller ones, in both cases drifting back toward the population average. The mechanism has nothing to do with intervention. Any result that reflects both a stable underlying level and some temporary variation, luck, circumstance, measurement noise, will, on average, be followed by a result closer to that underlying level, simply because an extreme outcome is unusually likely to have contained an unusually large temporary component, good or bad, and there’s no reason to expect that component to be just as extreme next time. The pattern is easy to mistake for cause and effect because it always shows up right where an intervention would also show up: after the extreme result. A coach benches a slumping player and the player’s next game is more average, a manager gives a harsh warning after an unusually bad quarter and performance improves, a team gets extra scrutiny after a rare failure and the next period looks better. In every case, some of that improvement, sometimes all of it, would have happened without the intervention, purely because extreme results tend to be followed by less extreme ones.

Why Does It Matter?

Organizations routinely act on the assumption that whatever they did right after an extreme result is what caused the return to normal, and that belief then gets reinforced every time it happens again, because it will happen again, regardless of what the intervention actually was. This creates a specific, durable trap: interventions that do nothing get credit for working, because the natural statistical drift back toward average looks exactly like a successful correction. It also works in the other direction, praise given after an unusually good result can look like it caused a subsequent decline, when the decline is often just the same statistical pattern running the other way. The risk isn’t that interventions never work. It’s that regression to the mean makes it hard to tell whether a given intervention actually worked, because the same pattern that a real fix would produce is also what would have happened on its own. The problem is especially acute in management because interventions aren’t randomly timed, they’re disproportionately triggered by unusually bad results, exactly the conditions under which some improvement should be expected anyway, intervention or not.

What Changes Once You See It?

You start asking whether an improvement after an extreme result would have happened anyway, before crediting whatever intervention came right before it. You start looking for a comparison group, a repeated observation, or some other evidence that separates the intervention’s actual effect from the return toward normal you’d expect anyway, before concluding that a change in performance proves the intervention worked. You also get more cautious about punishing an unusually bad result or over-rewarding an unusually good one, since both are more likely than a typical result to be partly a product of chance that won’t repeat the same way next time.

Common Misunderstandings

It isn’t a claim that interventions never work, or that skill and effort don’t matter. It’s a claim about what to expect from chance alone, which needs to be separated from any change actually caused by an intervention.
It isn’t the same as Twyman’s Law, which is about scrutinizing a surprising number before trusting it happened at all. Regression to the mean assumes the number is real and asks what naturally happens next, not whether the number itself should be doubted.
It doesn’t only apply to individual performance. It applies to any measured outcome with a chance component: sales figures, safety incidents, customer satisfaction scores, error rates.
It isn’t evidence that extreme results are meaningless. An unusually good or bad outcome can still reflect something real, the point is that some of it is also chance, and chance alone predicts a less extreme result next time.
It doesn’t mean everyone’s performance eventually drifts toward the whole organization’s average. “The mean” is whatever underlying level is relevant to what’s being repeatedly measured, a consistently exceptional performer regresses toward their own unusually high average, not toward the company-wide one.

Diagnostic Question

If we’d done nothing at all after that result, how much movement back toward normal should we have expected anyway?

Explore Further

Field Notes

None yet.

Related Field Guide

Origin

Francis Galton, “Regression Towards Mediocrity in Hereditary Stature” (1886), Journal of the Anthropological Institute; the statistical principle has since become foundational to interpreting any measured outcome with both a skill component and a chance component.