Non-regressive prediction
Also known as: Prediction by representativeness
Forecasting an extreme result from an extreme cue, ignoring the noise.
What it means
The tendency to make predictions that are as extreme as the evidence they are based on, failing to regress the forecast toward the average to account for the unreliability of the evidence. When an indicator is an imperfect predictor — as nearly all are — the optimal forecast should be a compromise between the indicator and the base rate, more moderate than the cue itself; intuition instead matches the prediction to the impression. The error follows from representativeness: people ask how well the case resembles the outcome, not how diagnostic the resemblance is. It guarantees systematic overprediction of extremes and is a chief reason people are blindsided by regression to the mean, mistaking an ordinary statistical rebound for the effect of an intervention. It matters in hiring, admissions, sports scouting, and investing, where bold forecasts from striking-but-noisy signals are routinely too extreme.
Examples
Told a student's score on one practice test, people predict a final grade just as high or low — instead of a more moderate estimate that allows for the test's imperfect reliability.
A candidate dazzles in one interview, so the panel predicts a star performer — rather than a good-but-ordinary hire, allowing that a single hour is a noisy read on a career.
A sales rep posts a record quarter and management sets next quarter's target just as high, treating one exceptional stretch as the new baseline rather than skill mixed with luck.
First described in Kahneman & Tversky (1973).