Confidence Sharing Improves Social Learning

Abstract

How does communicating confidence affect social influence? We study this question in a large-scale experiment in which ``senders” answer quiz questions and transmit information to ``receivers,” who then make their own choices. In one treatment, senders can share only their answer; in another, they can additionally report their confidence. While answer sharing improves receiver accuracy, allowing senders to report confidence roughly doubles this effect. Answer-only sharing improves accuracy only for neutral questions, whereas confidence reporting improves accuracy for both political and neutral questions. Confidence sharing also reduces misinformation in misleading questions, whereas answer-only sharing does not. The effects are highly heterogeneous: respondents with lower education and lower income benefit exclusively from confidence reports, whereas higher educated respondents and higher income benefit similarly across treatments. We show that a simple framework in which individuals differ in how they model others’ competence can rationalize these patterns. Our findings suggest that confidence reporting is a low-cost, scalable intervention for improving information in online environments.