Regression to the mean is a purely statistical phenomenon: an unusually extreme result (very good or very bad) tends to be followed by a result closer to the average, simply because extreme results are partly due to chance factors that aren't likely to repeat in the same direction twice. The trap is that this natural statistical drift gets mistaken for a causal effect of whatever happened in between — punishment, praise, a new policy, a new coach.
First documented in the nineteenth century, this remains one of the most consistently misunderstood ideas in everyday reasoning, precisely because the mistaken causal story (punishment works, or praise backfires) is so much more emotionally satisfying than the correct, duller explanation (extreme results regress toward average on their own).
While teaching behavioral psychology to Israeli Air Force flight instructors, Kahneman was told by a veteran instructor that praising a cadet after a good maneuver was reliably followed by worse performance next time, while criticizing a bad maneuver was reliably followed by improvement — leading the instructor to conclude punishment simply worked better than praise. Kahneman recognized this as pure regression to the mean: an unusually good or unusually bad performance is naturally followed by a more average one, regardless of what feedback was given in between. The underlying statistical phenomenon itself traces back over a century to Francis Galton, who found that unusually tall or short parents produced children whose heights were, on average, closer to the population mean than the parents' own heights — the founding empirical demonstration of the effect.