Bayesian persuasion with fact-checking: An experimental investigation

Abstract

Despite the ever-increasing accessibility of fact-checking, there is little empirical evidence on how it influences a person’s ability to persuade another. This paper experimentally investigates the impact of a fact-checking device that probabilistically flags false messages in a Bayesian persuasion framework. In theory, if fact-checking occurs with a sufficiently low probability, the sender can and should offset its effects by lying more frequently, rendering the device ineffective. However, our experiment contradicts this prediction. We find that senders do not lie any more frequently in the presence of fact-checking than in its absence, a behavior consistent with lying aversion. We also find that receivers’ actions are monotonic in their induced posterior, a behavior that resembles Bayesian rationality. Finally, we discuss how these results apply to a variety of real-world persuasive contexts such as litigation, lobbying, and disinformation dissemination.