Comparative Clustering for Contingent Collectives

Abstract

Clustering creates collectives of cases that have similar properties with a degree of distinctiveness. Clustering requires some composite measure of similarity or disparity, a criterion for conformity among collectives (linkage), and a strategy for configuring collectives. The collectives produced by a clustering method are conventionally called clusters. There are many methods of clustering, however, which typically differ to some degree in the groupings that result (Abonyi and Balaz 2007; Everitt et al. 2001; Kaufman and Rousseeuw 1990; Xu and Wunsch 2009). It is by comparing the collectives produced by different methods of clustering that one can gain insight from inconsistencies and have some confidence relative to consistencies. We call this comparative or complementary clustering and we use the term contingents (groups from groupings) for collectives of cases that emerge from this compound approach using cross-tabulations. Preliminary prioritization can be done among contingents and then progress to comparisons within contingents so that the computational complexities of comprehensive comparisons can be controlled.