Diverging Opinions

We commonly observe cases where two people who see the same evidence draw opposite conclusions, and that additional information seems only to polarize them further. One view of this is that it demonstrates failure of Bayesian reasoning, or the need for non-common priors. By contrast, we explain the phenomenon using a Bayesian model with common priors. We assume a model with two-dimensional uncertainty in which there is no prior disagreement about an true state. Nevertheless, a disagreement among agents may arise and continually increase if additional information becomes available on only one of the states. We provide support for our model by testing it experimentally. Furthermore, in the lab, the disagreement tends to increase more for the subjects whose behavior is more consistent with Bayesian behavior.