Robust Voting under Uncertainty

Abstract

This paper proposes normative criteria for voting rules under uncertainty about individual preferences. The criteria emphasize the importance of responsiveness, i.e., the probability that the social outcome coincides with the realized individual preferences. Given a convex set of probability distributions of preferences, denoted by $P$, a voting rule is said to be $P$-robust if, for each probability distribution in $P$, at least one individual's responsiveness exceeds one-half. Our main result establishes that a voting rule is $P$-robust if and only if there exists a nonnegative weight vector such that the weighted average of individual responsiveness is strictly greater than one-half under every extreme point of $P$. In particular, if the set $P$ includes all degenerate distributions, a $P$-robust rule is a weighted majority rule without ties.