Sticking with What (Barely) Worked

Outcome bias occurs when an evaluator considers ex-post outcomes when judging whether a choice was correct, ex-ante. We formalize this cognitive bias in a simple model of distorted Bayesian updating. We then examine strategy changes made by professional football coaches. We find they are more likely to revise their strategy after a loss than a win -- even for narrow losses, which are uninformative about future success. This increased revision following a loss occurs even when a loss was expected, and the offensive strategy is revised even when failure is attributable to the defense. These results are consistent with our model's predictions.