CLUSTAR and CLUSTID: Computer Programs for Hierarchical Cluster Analysis
H. Charles Romesburg, Kim Marshall · The American Statistician · 1980
The computer program CLUSTAR performs hierarchical cluster analysis following the sequential paradigm described by Sneath and Sokal (1973): (a) A data matrix of measurements for each of t objects (items, individuals, etc.) over n attributes is read; (b) the data matrix is optionally standardized; (c) a (dis)similarity coefficient is used to compute a pairwise measure of (dis)similarity among the t(t + 1)/2 object pairs; (d) a clustering method is used to produce a dendrogram on a line printer showing the similarity/dissimilarity relationships among the t objects. Clusters, that is, groups of objects that are sufficiently similar to be treated as homogeneous units, are defined on the dendrogram by the user. The computer program CLUSTID can then be used to (a) identify which clusters new objects (i.e., those objects not used in the CLUSTAR run) belong to and (b) provide summary statistics for each cluster, that is, attribute means and standard deviations.