Empirical Evaluation of Voting Rules with Strictly Ordered Preference Data

Abstract

The study of voting systems often takes place in the theoretical domain due to a lack of large samples of sincere, strictly ordered voting data.We derive several million elections (more than all the existing studies combined) from a publicly available data, the Netflix Prize dataset. The Netflix data is derived from millions of Netflix users, who have an incentive to report sincere preferences, unlike random survey takers. We evaluate each of these elections under the Plurality, Borda, k-Approval, and Repeated Alternative Vote (RAV) voting rules.We examine the Condorcet Efficiency of each of the rules and the probability of occurrence of Condorcet's Paradox. We compare our votes to existing theories of domain restriction (e.g., single-peakedness) and statistical models used to generate election data for testing (e.g., Impartial Culture). We find a high consensus among the different voting rules; almost no instances of Condorcet's Paradox; almost no support for restricted preference profiles, and very little support for many of the statistical models currently used to generate election data for testing.