Algebraic structures for pairwise comparison matrices: consistency, social choices and Arrow's theorem (Q2054755): Difference between revisions

The authors present the algebraic structures behind the approaches used to work with pairwise comparison matrices (PCMs) and the representation of preferences. They obtain a general definition of consistency and a universal decomposition in the space of PCMs, allowing them to define a consistency index. Finally, they present Arrow's theorem in a general algebraic form. Let \(G = (G, +)\) be an abelian group and \(\mathcal{M}_n\) be the set of all \(n \times n\) matrices \(A = (a_{ij})_{ij}\) whose entries belong to \(G\). For instance, \(1, 2, \ldots, n\) might be alternatives and \(a_{ij}\) might express the degree (in terms of a natural number) by which an agent prefers alternative \(i\) to alternative \(j\). \(A = (a_{ij})_{ij}\) is called a pairwise comparison matrix (PCM). An \(n \times n\) matrix \(A = (a_{ij})_{ij}\) is said to be \textit{skew-symmetric} if \(a_{ji} = - a_{ij}\) for every \(i, j = 1, 2, \ldots, n\). Consequently, in any skew-symmetric matrix \(a_{ii} = 0\) for every \(i = 1, 2, \ldots, n\). Example: \(A = \left( \begin{array}{ccc} 0 & -1 & 2 \\ 1 & 0 & 4 \\ -2 & -4 & 0 \end{array} \right)\). \\ A skew-symmetric matrix \(A\) is \textit{consistent} if \(a_{ik} = a_{ij} + a_{jk}\) for all \(i, j, k = 1, 2, \ldots, n\). In the context of preferences this seems a reasonable property: the degree by which an agent prefers alternative \(i\) to \(j\) plus the degree by which he prefers \(j\) to \(k\) equals the degree by which he prefers \(i\) to \(k\). The matrix above is not consistent, since, for instance, \(-1 = a_{12} \neq a_{13} + a_{32} = 2 - 4 = -2\). Here is an example of a consistent skew-symmetric matrix: \(A_1 = \left( \begin{array}{ccc} 0 & 1 & 5 \\ -1 & 0 & 4 \\ -5 & -4 & 0 \end{array} \right)\). \\ A vector \(v = (v_1, v_2, \ldots, v_n)\) in \(G^n\) is said to be \textit{coherent} for a matrix \(A\) if \(v_i - v_j = a_{ij}\) for every \(i, j = 1, 2, \ldots, n\). For instance, (6, 5, 1) is a coherent vector for \(A_1\). The authors show among others that 1) any two vectors coherent for \(A\) differ by a constant \(c \in G\); 2) if \(v\) is a cohorent vector for \(A\), then \(A\) is consistent; 3) if \(A\) is consistent, then each column vector of \(A\) is coherent for \(A\); 4) a matrix \(A\) is consistent iff there is at least a coherent vector for it. A matrix \(A\) is said to be \textit{totally inconsistent} if for each \(i = 1,2 , \ldots, n\), \(\Sigma_{j = 1}^{n}a_{ij} = 0\). The authors show constructively that every skew-symmetric matrix can be decomposed into the direct sum of a consistent and a totally inconsistent matrix. The authors define a consistency index \(I_{C}(A)\) of a skew-symmetric matrix \(A\), not necessarily consistent, in terms of \(e_{ijk}^{(A)}\), where \(e_{ijk}^{(A)} = a_{ij} + a_{jk} + a_{ki}\). Notice that \(A\) is consistent iff \(e_{ijk}^{(A)}\) = 0 for every choice of \(i, j\) and \(k\). Let \(\mathcal{S}\) be the set of all skew-symmetric \(n \times n\) matrices, \(A = (a_{ij})_{ij} \in \mathcal{S}\), \(G = (G, +, \leq)\) a partially ordered abelian group and \(v = (v_1, v_2, \ldots, v_n) \in G^n\). \(v\) is called an \textit{ordinal evaluation vector} for \(A\) iff for every \(i, j = 1, 2, \ldots, n\): if \(a_{ij} > 0\), then \(v_i > v_j\) and if \(a_{ij} = 0\), then \(v_i = v_j\). For instance, (2, 1, 0) is an ordinal evaluation vector for \(A_1\). And a skew-symmetric matrix \(A\) is \textit{weakly consistent} if for every \(i, j = 1, 2, \ldots, n\): if \(a_{ij} > 0\), then for all \(k = 1, 2, \ldots, n\), \(a_{ik} > a_{jk}\) and if \(a_{ij} = 0\), then for all \(k\), \(a_{ik} = a_{jk}\). Consistency implies weak consistency, but not conversely. Next the authors formulate Arrow's conditions in the partially ordered space setting. Let \(\mathcal{T} \subset \mathcal{S}^m\), where \(m\) is the number of voters. Then a \textit{profile} is an element of \(\mathcal{T}\) and an \textit{aggregation procedure} on \(\mathcal{T}\) is any function \(\Phi: \mathcal{T} \rightarrow \mathcal{S}\). The present reviewer would have liked to see more context or background information with more examples in order to make the abstract algebraic presentation more accessible.