A RANDOMIZATION TEST OF THE NON SPECIFICITY HYPOTHESIS IN NUMERICAL TAXONOMY

Abstract

This paper describes a method for the statistical analysis of the differences in phenetic relationships resulting from different sets of characters in numerical taxonomy. It is a statistical test of the non-specificity hypothesis of Sokal and Sneath (1963). Rohlf (1963a) first investigated the extent to which different sets of characters would yield different classifications when analyzed by numerical taxonomic methods. He found that a classification of Aedes mosquitoes based upon external morphological characters of the larvae was not entirely congruent with a classification based upon external morphological characters of the adults. A problem which resulted was whether the degree of difference between the two sets of phenetic relationships was greater than that to be expected due to chance differences in sampling from a homeogeneous collection of characters. Some approximate analyses by Rohlf (1963a) gave a crude measure of the reliability of the observed difference. These analyses were based upon assumptions known not to be entirely correct (independence of characters, normal distributions, and that the t(t 1)/2 pairs of relationships among the t species were independent). The present paper describes a randomization test which does not require these assumptions and hence is more valid statistically.