Non-uniqueness and Inversions in Cluster Analysis

Care must be exercised when hierarchical methods of cluster analysis are used. Dendrograms may not be unique, and certain methods are prone to producing inversions. The nature and extent of these features are examined through two case-studies, and by applying seven methods to 20 data sets. Insufficient emphasis on the problems of non-uniqueness and of inversion is made in many text-books and also in computer packages and their manuals.