Regression to the Mean: Why Extremes Don't Last

You already learned to think in probabilities — to trade the yes/no switch for a dial, anchor on base rates, and weigh evidence with Bayes. This course installs one of the most useful and most misread consequences of that worldview: regression to the mean. Whenever an outcome mixes stable skill with fluky luck, an extreme result was almost certainly extreme partly by luck — and luck, by definition, does not repeat. So the next measurement from the same source drifts back toward the average, with no cause, no intervention, and no story required.

That single fact is a wrecking ball for everyday reasoning. It means the business that posted a blowout quarter probably posts a milder one next time — not because it got complacent, but because record quarters ride a wave of good luck that recedes. It means the athlete on the magazine cover “slumps” the next season, the patient who saw the doctor at their sickest “recovers” thanks to the treatment, and the risky intersection where a speed camera was just installed “gets safer” — all of it, at least partly, mathematics wearing the costume of cause and effect. The regression fallacy — crediting an intervention for an improvement that would have happened anyway — is one of the most expensive thinking errors there is, and it hides in plain sight everywhere results are measured.

This course builds the model from the ground up. First the core mechanic: any measured result = a stable signal + transient noise, so selecting on an extreme means selecting for extreme noise, which won’t hold. Then Galton’s discovery — tall parents having (on average) shorter children — the observation that literally named the entire field of regression analysis. You’ll work a numeric skill+luck model and watch a top group’s average fall on a retest while the population mean sits still. Then the payoff: the regression fallacy in the wild (the Sports Illustrated “curse”, speed cameras and worst-sites-first, feeling better after any visit to the doctor), and Kahneman’s brutal flight-instructor anecdote — why instructors “learn” that punishment works and praise fails when it is pure reversion, and how this same illusion breeds superstition, quack cures, and managerial over-reaction. Finally, how to defend: control groups, larger samples, and telling regression apart from a real trend and from the gambler’s fallacy. You’ll drive an interactive scatter that crowns the luckiest performers and watches them slide back down. By the end you’ll never again mistake the tide going out for something you did.