Trembling toward Duverger: Convergence and Cycles in Synchronous Plurality Dynamics

Abstract

Duverger's law states that plurality elections tend to concentrate support on two candidates. Pivotal-voter models typically derive this from aggregate uncertainty about votes, so that two-candidate competition emerges in the limit of large electorates. We instead propose a model for a fixed, finite population. A poll reveals the current support counts, and each voter best responds under trembling-hand beliefs, assuming that every other voter may deviate with a small probability. Starting from truthful votes, we show that a single update typically suffices for the dynamics to converge to a stable profile in which every voter supports her preferred candidate among the two poll leaders. For any number of candidates and any full-support tremble belief, a four-vote lead of the runner-up over all lower candidates suffices; for three candidates, a lead of two suffices. As a consequence, under random preferences, Duverger's law emerges after one update with probability $1-\mathcal O(N^{-1/2})$. While these results may suggest that the dynamics always settle, convergence can fail. Mild period-two cycles already arise, with a few voters swapping ballots around an otherwise stable profile. We show that substantially richer behavior is possible, even with three candidates. Period-three cycles arise with six voters changing their ballots, and no such cycle can involve fewer than four. Cycles can also alternate the plurality winner. Moreover, larger electorates support ever longer cycles: for every odd $\ell\geq3$, truthful voting can lead to a cycle whose vote counts repeat every $\ell$ rounds and whose full ballot profile repeats every $2\ell$ rounds.