Author: Aristotelis Boukouras (University of Leicester), Will Jennings (University of Southampton), Lunzheng Li (University of Southampton) and Zacharias Maniadis (University of Southampton)
Paper number: 1902
Biased exposure of voters to the outcome of polls constitutes a risk to the principle of balanced and impartial elections. We first show empirically how modern communication (through social media) may naturally result in such biased exposure. Then, in a series of experiments with a total of 375 participants, we investigate the impact of such biased exposure on election outcomes in an environment where only a strict subset of voters has information on the quality of the two candidates. Thus, polls serve to communicate information to uninformed voters. In our treatment conditions, participants have access to a biased sample of polls’ results, favouring systematically one candidate. Participants in the biased treatment conditions consistently elect the candidate favoured by polls more often than in the unbiased control conditions. Remarkably, this holds even when voters are a priori informed about the bias. Accordingly, our results indicate that – in an experimental setting at least – biased polls distort election results via two channels: (i) by distorting the information set of voters, and (ii) by providing an anchor for subjects’ expectations regarding the election outcome. Overall, biased exposure distorts elections in a very robust manner.
Keywords: biased polls, candidate valence, information aggregation