Paying is believing: The effect of costly information on Bayesian updating

Abstract

The difficulties posed by Bayesian updating are recognized across many domains. In this paper we explore whether individual belief updating is affected by the cost of information. Our conjecture is that this effect should be observed if individuals are prone to the sunk cost fallacy. We design an experimental environment where subjects perform a belief updating task after receiving useful and identical information on the state of the world. Our treatments vary the way in which information is made available to subjects. We find a systematic effect of the cost of information on belief updating. Subjects overweigh costly information relative to free information, which results in a ‘push’ of beliefs towards the extremes. The cost-driven shift can lead to posterior beliefs more attuned with Bayesian updating. We argue that an intensification of the representativeness bias is the most likely explanation of our results.