Fooling Yourself: how narratives shape beliefs

Decision-makers often receive information through narratives combining diagnostic evidence, which favors one state over another, with nondiagnostic details, which provide no information about the true state. We study whether such uninformative details embedded in a narrative affect belief updating. In a laboratory experiment, participants report incentivized beliefs in a Bayesian updating task. We implement three experimental conditions with the same statistical structure: an urn problem with diagnostic colored balls and nondiagnostic white balls; an investigation narrative in which these signals appear as incriminating and non-incriminating clues; and the same narrative with non-incriminating clues replaced by no-information messages. Although nondiagnostic signals should leave beliefs unchanged, participants revise beliefs toward equal odds, attenuating the impact of diagnostic evidence. This response occurs in the urn condition, is stronger in the investigation narrative, and largely disappears when narrative clues are replaced by no-information messages. Simulations show such revision increases evidence required for threshold decisions.