Key points are not available for this paper at this time.
Six cluster methods were subjected to empirical trials evaluating their ability to solve a "dummy" problem, that of detecting an underlying taxonomy of biological sex. Each cluster method was applied to four sets of sex-discriminant self report personality-interest questionnaire items. The validity of each item-indicator was empirically determined using biological sex as a criterion variable. The four sets of item-indicators of biological sex were constructed so as to vary in their average item-indicator validity. Three of the cluster methods detected accurately the biological sex taxonomy in most of the trials; the other three methods were seldom accurate. It is argued that for detecting real empirical classes, the cluster methods are of questionable value since we typically lack assurance that the clusters are likely to be accurate and not spurious. It is suggested that use of internal validity or 'consistency' tests could eliminate this shortcoming.
Golden et al. (Wed,) studied this question.