k-Majority Digraphs and the Hardness of Voting with a Constant Number of Voters

TL;DR

This paper investigates the structure of majority digraphs with a fixed small number of voters, providing characterizations, efficient computation methods, and complexity results showing voting problems remain hard even with few voters.

Contribution

It characterizes majority digraphs for 2 and 3 voters, develops SAT-based methods for minimal voter count, and proves certain voting rules are computationally hard with a small constant number of voters.

Findings

Characterized digraphs induced by 2 and 3 voters.

Developed SAT-based approach for minimal voter count.

Proved Kemeny's rule is hard to evaluate with 7 voters.

Abstract

Many hardness results in computational social choice make use of the fact that every directed graph may be induced as the pairwise majority relation of some preference profile. However, this fact requires a number of voters that is almost linear in the number of alternatives. It is therefore unclear whether these results remain intact when the number of voters is bounded, as is, for example, typically the case in search engine aggregation settings. In this paper, we provide a systematic study of majority digraphs for a constant number of voters resulting in analytical, experimental, and complexity-theoretic insights. First, we characterize the set of digraphs that can be induced by two and three voters, respectively, and give sufficient conditions for larger numbers of voters. Second, we present a surprisingly efficient implementation via SAT solving for computing the minimal number of voters that is required to induce a given digraph and experimentally evaluate how many voters are required to induce the majority digraphs of real-world and generated preference profiles. Finally, we leverage our sufficient conditions to show that the winner determination problem of various well-known voting rules remains hard even when there is only a small constant number of voters. In particular, we show that Kemeny's rule is hard to evaluate for 7 voters, while previous methods could only establish such a result for constant even numbers of voters.