Disambiguating Information Interventions

Information experiments read reported probabilities as beliefs. When the odds are unknown, a reported probability also reflects two ambiguity attitudes: aversion pushes it in the pessimistic direction; coarse perception pulls it toward one half. Information changes these attitudes as well as the belief, so a change in a reported probability is not a change in belief. We characterize the designs that identify the belief and both attitudes. A single elicited probability does not identify even the sign of the belief change; we give the threshold beyond which the sign is trustworthy. A known six-bet design identifies all three, run once in each information condition. We use it in a randomized experiment in a Dutch population panel, betting on a matched panel member’s doctor visits. At face value, the reported probabilities show no detectable effect, which is not evidence about beliefs. Recovered beliefs show improved predictive accuracy and lower ambiguity aversion.