Majority cycles in a multi-dimensional setting (Q1865219)

The set \(X\) of alternatives is defined as a Cartesian product \(B_{1} \times \ldots \times B_{k}\) of finite sets. Let \(N\) = \{1, \(\ldots\), \(n\)\}, \(n\) odd, be the set of voters. A profile is an \(n\)-tuple of linear orderings on \(X\). A binary relation on \(X\) is called \textit{separable} if (roughly speaking), when facing the comparison between two alternatives in \(X\), an individual takes his decision on the exclusive basis of their differences. For instance, when comparing (\(a_{1}, b\)) and (\(a_{2}, b\)), the voter will decide on the exclusive basis of his preference with respect to \(a_{1}\) and \(a_{2}\). A linear ordering \(R\) on \(X\) is marginally single peaked with respect to a collection \(S\) = \{\(S_{1}, \ldots, S_{k}\)\} of orderings on \(B_{1}, \ldots, B_{k}\) respectively, if any restriction of \(R\) to some alternatives having all but one component \(i\) with the same value, is single-peaked with respect to \(S_{i}\). A binary relation \(Q\) on \(X\) is marginally transitive iff any restriction of \(Q\) to some alternatives having all but one component with the same value is transitive. The main result of this paper is the following: for any binary relation \(T\) on \(X\) that is complete, asymmetric, separable, marginally single peaked with respect to \(S\) = \{\(S_{1}, \ldots, S_{k}\)\} and marginally transitive, there is a profile \(\pi\) of \(n\) separable and marginally single peaked (with respect to \(S\)) preferences on \(X\) such that \(T\) is the result of applying the pairwise majority relation to \(\pi\). The converse also holds due to the properties of the majority rule. However, the pairwise majority relation might exhibit several cycles, in other words, marginal transitivity does not imply transitivity. The proof of the main result requires about six pages and is based on what the author calls the McGarvey principle. This result is interesting because of its relation to classical problems of multi-dimensional decisions such as logrolling and vote trading. The author also relates his result to a continuous version of it due to McKelvey.