Probability matching
Guessing each option as often as it actually occurs — when always picking the likeliest wins.
What it means
A choice pattern in which people allocate their responses in proportion to the observed probabilities of outcomes, rather than always selecting the most probable one as expected-value maximization prescribes. If a light flashes red 70% of the time and green 30%, the optimal strategy is to predict red every trial, yet people tend to predict red about 70% of the time and green 30%, lowering their hit rate. The behavior is robust across species and tasks, and its causes are debated: it may reflect a search for nonexistent patterns, an attempt to gather information or hedge, or sensitivity to sequential structure that would make matching adaptive in a changing world. It matters as a clean case where intuition systematically underperforms a trivially simple optimal rule, and as a window into how the mind treats randomness — often refusing to accept that the best response to noise is monotonous repetition.
Examples
In repeated prediction games, participants keep occasionally betting on the rarer outcome to 'catch' it, instead of always choosing the more frequent one and maximizing correct guesses.
In a town that is dry four days in five, a forecaster who says 'rain' one day in five to match the climate is right less often than one who flatly says 'dry' daily.
A checklist that beats clinical judgement most of the time is best followed every time, yet many clinicians override it now and then to catch the rare exception — and get more calls wrong.
First described in Studied since Humphreys (1939); Estes and colleagues.