Experimental Appraisal of Certain Procedures for the Classification of Data

Abstract

An experimental appraisal of 2 of the commoner techniques for empirical classification, namely with Sneath Q-study and Williams and Lambert normal analysis, is presented. It is shown that the methods have unresolved problems associated with them and that results obtained by their use are not necessarily liable to effective interpretation. A number of objections are raised and it is suggested that biologists using the methods should reconcile their conclusions with these objections. Data are presented for the first time on a range of relevant features including matched results, comparisons of split data, the influence of random variation, the relative performance of X2 versus percentage similarity, the nature and stability of percentage similarity, and on situations where the techniques as they stand are very suitable.