Political actors face a trade-off when they try to influence the beliefs of voters about the effects of policy proposals. They want to sway voters maximally, yet voters may discount predictions that are inconsistent with what they already hold to be true. Should political actors moderate or exaggerate their predictions to maximize persuasion? I extend the Bayesian learning model to account for confirmation bias and show that only under strong confirmation bias are predictions far from the priors of voters self-defeating. I use a preregistered survey experiment to determine whether and how voters discount predictions conditional on the distance between their prior beliefs and the predictions. I find that voters assess predictions far from their prior beliefs as less credible and, consequently, update less. The paper has important implications for strategic communication by showing theoretically and empirically that the prior beliefs of voters constrain political actors.

Rational voters care about outcomes, while parties campaign on policy proposals, the outcomes of which are never perfectly known. Can parties exploit this uncertainty to shape public opinion? This article presents a spatial preference model for policy proposals with uncertain outcomes. It reports the results of a large pre-registered survey experiment that involved presenting respondents with predictions about the effects of three policy proposals. The findings show that respondents update their attitudes to the proposals as their beliefs about outcomes change, and that parties are no less able to influence beliefs than non-partisan experts. Contrary to previous research, respondents discount outcome uncertainty by giving equal weight to conflicting optimistic and pessimistic predictions. The study shows that parties can shape public opinion by influencing voter beliefs, and that voters are not repelled by the uncertainty inherent in conflicting information.