November 26th, 2019
"The Social Exchange of Biased Beliefs"
We directly elicit experimental subjects’ beliefs about their relative performance on a cognitive ability task and track the evolution of these beliefs over time. Part of the way through the task, we pair subjects of similar ability with one another and allow them to observe each other's beliefs in real time. Compared to control settings without social exchange of beliefs or without bias (beliefs over objective lotteries), we find an echo-chamber effect where systematic bias on the individual level (overconfidence) is reinforced and amplified through social-exchange. We find that updating behavior with social-exchange is not consistent with Bayes' rule even if subjects are assumed to discount the overconfidence of others or have uncertainty about the informativeness of others' beliefs. Credible public information is effective at reducing the echo-chamber effect, but does not eliminate it.