Mathematical structure of voting paradoxes. I: Pairwise votes (Q1969020)

This important paper, together with its companion [Mathematical structure of voting paradoxes. II: Positional voting. Econ. Theory 15, No. 1, 55--102 (2000; Zbl 1081.91007)], represent the culmination of the author's long and deep analysis of voting systems and the various paradoxes (counter-intuitive outcomes) that they generate. The author is well-known for his introduction of geometric methods into the field and it is that geometric and linear algebraic viewpoint that he here deploys to obtain a large number of significant results. Voting methods can conveniently be divided into positional and pairwise. A positional system assigns points to candidates or alternatives based on how individual voters rank them and aggregate the points; a pairwise system matches candidates in pairs and aggregates based on majority votes. A key concept for pairwise systems is that of a Condorcet winner, an alternative preferred by a majority of voters over any other alternative. If a Condorcet winner exists, and a given method chooses a different alternative, then regardless of which alternative is chosen a majority of the electorate would have preferred the Condorcet winner. May's Theorem shows that majority rule is (in a certain sense) the only fair method of choosing between two alternatives. The ideal of majority choice is deeply embedded in voting theory and underlies the desire of pairwise systems to aggregate majority decisions. It is this central majority ideal that the author challenges. Majority turns out to be a slippery concept when there are three or more alternatives, and the difficulty pairwise methods have is based upon the lack of transitivity of aggregates of transitive voters. Suppose there are three candidates, \(A, B\) and \(C\), and three voters. One voter ranks the alternatives \(A>B>C\); one votes \(B>C>A\) and the third \(C>A>B\). Then \(A\) is ranked above \(B\) by two-thirds of the voters, and similarly, \(B\) is ranked above \(C\) by a two-thirds majority. However, the same voters also prefer \(C\) over \(A\) by a majority of 2 to 1. Since each alternative is ranked first by one voter, second by one voter, and third by one voter, any positional system will result in a tie. The author's insight is that the paradoxes afflicting pairwise methods, and the discrepancies between pairwise outcomes and Borda outcomes are due precisely to these `Condorcet effects'. The author's positive view is that addition and subtraction of `Condorcet cycles' as above does not affect Borda outcomes, but it does affect pairwise results. His normative view is that it shouldn't, that Condorcet cycles should cancel, and this presents his great break with majoritarian views. As the author discusses pairwise methods with reference to a baseline of positional systems, especially Borda count, he includes a number of results on positional systems, including the important notion of the universal kernel, the space of profiles where all positional systems give a complete tie. The author then introduces Basic profiles, which are essentially those that satisfy the Independence of Irrelevant Alternatives criterion, and the Condorcet subspace responsible for conflicts and paradoxes. The bulk of the paper is a deep analysis of how these conflicts and paradoxes occur, and their categorization and characterization. The paper contains numerous significant results.