Why is belief–action consistency so low? The role of belief uncertainty

Experimental research typically shows that best-response rates are below what plausible error rates would suggest. We experimentally test the conjecture that observed action–belief inconsistencies are related to belief uncertainty. We rely on a belief-sampling model that has been highly successful in explaining behavior in multi-armed bandit problems and aggregate outcomes in games, markets, and surveys. Our data shows that inducing higher belief uncertainty leads more frequently to choices that are inconsistent with stated beliefs and – in an experiment directly testing the mechanism – to stochastic belief reports. The uncertainty–inconsistency relationship continues to hold when we control for error costs econometrically in several ways.