An approach to the use of packaged statistical programs for cluster,classification, and discriminant analysis of trace element data

The purpose of this paper is to present an approach to be taken when using packaged statistical programs for the evaluation of multivariate data, in particular those programs designed for clustering, classification, and discrimination purposes. A systematic step-by-step approach is outlined starting with those steps which should be conducted in order to assure that the assumptions underlying the statistical analyses are satisfied. The use of various methods and computational options in cluster, classification, and discriminant analysis is then discussed. Finally the need for an assessment of the results in view of the overall knowledge of the problem is pointed out.