Social Exchange of Motivated Beliefs

Abstract

We use laboratory experiments to study whether biases in beliefs grow more severe when people socially exchange these beliefs with one another. We elicit subjects’ (naturally biased) beliefs about their relative performance in an intelligence quotient (IQ) test and allow them to update these beliefs in real time. Part of the way through the task we give each subject access to the beliefs of a counterpart who performed similarly on the test and allow them both to observe the evolution of one another’s beliefs. We find that subjects respond to one another’s beliefs in a highly asymmetric way, causing a severe amplification of subjects’ initial bias. We find no such patterns in response to objective public signals or in control treatments without social exchange or scope for motivated beliefs. We also provide evidence that the pattern is difficult to reconcile with Bayesianism and standard versions of confirmation bias. Overall, our results suggest that bias amplification is likely driven by “motivated assignment of accuracy” to others’ beliefs: subjects selectively attribute higher informational value to social signals that reinforce their motivation.