Commitment and Strategic Communication in Bayesian Persuasion: Theory and Experiment
In many scenarios, a persuader (she) has conflicting interests from a decision-maker (DM, or he) and attempts to influence his action. Kamenica and Gentzkow (2011) pioneered the literature on Bayesian persuasion, wherein persuasion is achieved by strategically designing DM’s information environment. We consider a variant scenario wherein the persuader has superior information over the decision-maker and communicates with the DM. We further explore theoretically and experimentally how the persuader’s private information and the communication protocols affect the effectiveness of persuasion and the information transmission. In our setting, the persuader commits to a designed information structure but privately observes the signal realization, and she can exchange her private information of the signal with the DM via cheap talk or partially verifiable messages. The DM receives the message and chooses a state-contingent bet that determines the payoffs for both players. We design a laboratory persuasion task in which the persuader can use a simple graphic colouring task to influence a DM’s conditional probability inference, and further the DM’s bet decision. We find that (1) the persuader's choices of information structure are in line with theoretical predictions, showing a trade-off between effectiveness and informativeness of the information design; (2) the DMs can predict the persuader’s information design in the correct direction, but quantitatively not all of them respond with optimal actions; (3) the observed treatment effects are consistent with the theory, in that a combination of private signal observation with low credibility in communication reduces the effectiveness of persuasion, but increases the information transmission.