MEG - MEGACLUSTER ANALYTIC STRATEGY FOR MULTISTAGE HIERARCHICAL GROUPING WITH RELOCATIONS AND REPLICATIONS

An analytic and computer strategy is introduced and demonstrated for m ultistage Euclidean grouping (MEG). The procedure sequentially produce s first-stage clusters for independent data blocks; second-stage, high er order clusters based on a full similarity matrix for first-stage cl usters; and third-stage clusters that allow case migration to relocate prior misassignments and to optimize within-cluster homogeneity. The process is facilitated by special SAS computer codes and, in addition to conventional SAS cluster output, produces special fusion statistics , plots of all fusion statistics, and indices of homogeneity within cl usters and within profile variables. The program also reports replicat ion rates for final clusters.