Geometry of voting (Q1323813)

In a number of recent articles published in economics journals (Journal of Economic Theory, Social Choice and Welfare, Economic Theory etc.), the author has studied a class of voting methods generally called scoring rules. The ultimate goal of his research, which he attained, was to find ``everything that could ever occur'' with sincere voting. The mathematics used in these papers were rather advanced (at least they were advanced for economists and other social scientists). In the book under review, the description of voting rules are restricted in general to simple cases (two or three-candidate elections) so that the mathematical requirements are not too demanding. With three candidates, geometry provides a way to overcome complicated combinatorics. The book starts with a very didactive Chapter 1. It introduces several examples that are used throughout the book and it shows that the author is a remarkable story-teller. Chapter 2 is the central one. Consider there is a finite set of voters and three candidates. Each voter ranks the candidates according to his preference (there are 6 rankings when ties are not permissible and 13 otherwise; for simplicity, let's assume there are no ties). Each of the 6 rankings defines a type of voters. Let \(p_ j\), \(j= 1,\dots,6\), be the fraction of the voters having type \(j\). Then the vector \((p_ 1,\dots, p_ 6)\) defines what is called a profile, i.e., a profile is a rational point in the unit simplex of \({\mathbf R}^ 6_ +\). The election outcome is then a point in the unit simplex of \({\mathbf R}^ 3_ +\), and an election is a mapping associating an election outcome to a profile. The author then provides a triangle representing an election. Scoring or positional voting methods are opposed to so-called pairwise methods. For positional voting methods, one has a voting vector \(W= (w_ 1,w_ 2,w_ 3)\), with \(w_ i\geq w_{i+ 1}\), \(i= 1, 2\) and \(w_ 1> w_ 3\). In tallying a voter's ballot, \(w_ j\) points are assigned to the voter's \(j\)th ranked candidate, \(j= 1, 2, 3\). The ranking of the candidates is determined by the sum of the points assigned to each of them. In majority pairwise voting, a candidate \(a\) is ranked before a candidate \(b\) if the number of voters ranking \(a\) before \(b\) is greater than the number of voters ranking \(b\) before \(a\). For positional methods, everything that could possibly happen in the three-candidate case is described. For pairwise voting, the author introduces ``coordinate representations for profiles'' which make it possible to ``see'' all possible profiles supporting a particular election outcome. Based on this, it is shown why we can have serious doubt about using the Condorcet criterion (a candidate defeating every other candidate in pairwise voting must be the winner of the election), how we can have extensions of Black's single-peakedness (there is a candidate among the three candidates that is never ranked last by the voters), how we may uncover flaws with agendas etc. Chapter 3 deals with the comparison of positional methods and pairwise voting, and the comparison of different positional methods. A possible property of voting methods, neutrality, is particularly analyzed: if each voter interchanges his ranking of two candidates, then the election ranking of these candidates must be interchanged; if each voter completely reverses his ranking of all the candidates, then the election outcome must also be reversed. A very important result of the chapter concerns the Borda count, a positional method for which the voting vector is such that \(w_ 1- w_ 2= w_ 2- w_ 3\) (e.g. (2,1,0)). The Borda count is the only positional method satisfying the reversal symmetry property. The last two sections of the chapter consider other methods than positional or pairwise ones, for instance approval voting, scoring runoffs, cardinal procedures etc. In Chapter 4, interprofiles considerations are taken into account. What can happen if two subcommittees join, voters abstain, change their opinions or provide a strategic ballot (when they do not reveal their sincere rankings so that the election method gives a better outcome for them than the outcome resulting from their sincere rankings)? In passing, geometric proofs of Gibbard-Satterthwaite's Theorem and of Arrow's Theorem are provided. Though this book is basically a research monograph (though easily accessible with a knowledge of some of the mathematics generally used in economics, i.e., some linear algebra and convex analysis in Euclidean spaces -- there are five pages of introduction to these topics), it includes many exercises and consequently can be used as a textbook for courses on voting theory wherever such courses do exist. A minicourse was recently taught by the author at the graduate level for economics students at the University of Caen with great success. In the reviewer's opinion, Geometry of Voting is a great book, not only for its important contents, but also for its pedagogical quality.