Elections under Biased Candidate Endorsements — An Experimental Study

We study the impact of media bias on voters' behavior and election outcomes in a laboratory experiment. We model this interaction and derive the Bayesian Nash Equilibria. These predict for a single media that, ex-ante, an increased media bias affects candidates' winning probabilities non-monotonically and increases voter turnout. Introducing a second media outlet can affect the election outcome and voter turnout in either direction. We test these predictions in a laboratory experiment and find that both observed election outcomes and vote shares are well predicted. Voter turnout, however, is much less responsive to media bias than predicted. We show that subjects' observed behavior can be rationalized, to a substantial extent, by a quantal response equilibrium model combined with (a) distinct noise parameters for candidate choice and turnout decisions, (b) non-Bayesian belief updating, and (c) “partial competition neglect” (i.e., voters’ imperfect ability to infer closeness of elections from messages announced by media outlets).