Persuasion of a confirmation-biased agent

Abstract

In a basic game of Bayesian persuasion, I let the Receiver display confirmation bias as in Rabin and Schrag (1999). Hence, I assume that the Receiver can misinterpret the signal generated by the Sender's experiment, if this does not support the most likely state under her prior beliefs. I find that confirmation bias can raise the Receiver's welfare, as it can lead the Sender to disclose more information. In this case, the Sender is also better-off and confirmation bias generates a Pareto improvement. I show this with a simple example where an online trading platform designs its information environment to induce an investor and user of the platform to invest with higher frequency and in more risky assets. Then, I generalize the analysis, derive sufficient conditions for more information transmission and provide a graphical illustration thereof. The direction of the bias is important, as the Sender must prefer a Bayesian Receiver updating in the direction of her prior, to reveal more information. Finally, I show that, with confirmation bias, even uninformative experiments can be effective at persuasion, and that my results easily extend to the case of imperfect information about the strength of the Receiver's bias (i.e., her type).