False Consensus Voting and Welfare Reducing Polls

We consider a process of costly majority voting where people anticipate that others have similar preferences. This perceived consensus of opinion is the outcome of a fully rational Bayesian updating process where individuals consider their own tastes as draws from a population. We show that the correlation in preferences lowers expected turnout. The intuition is that votes have a positive externality on those who don’t participate, which reduces incentives to participate. We study the effects of the public release of information (“polls†) on participation levels. We find that polls raise expected turnout but reduce expected welfare because they stimulate the “wrong†group to participate. As a result, polls frequently predict the wrong outcome. While this lack of prediction power is usually attributed to an imperfect polling technology, we show it may result from the reaction of rational voters to the poll’s accurate information.