Retrospective Voting: An Experimental Study

This essay reports on some experiments designed to study two candidate electoral competition when voters are 'retrospective' voters. The experiments consist of a sequence of elections in which subjects play the part of both voters and candidates. In each election the incumbent adopts a policy position in a one-dimensional policy space, and voters are paid (on the basis of single peaked utility function over that space) for the position adopted by the incumbent. Neither voters nor candidates are informed of the voter utility functions, and the only information received by the voter is the payoff he has received from the present and previous incumbent administrations. Despite the severely limited information of candidates and voters, we find that, generally, candidates converge toward the median voter ideal point.