Cluster Analysis in a Parolee Sample

Abstract

This study extends earlier work concerned with the classification of individuals. The method used, called Association Analysis, has received widespread attention since its first use by Williams and Lambert.¹ Ordinarily, in Association Analysis, use has been made of the statistic Chi-square (or of the Phi coefficient) to measure the associations among attributes. This report discusses the use of two different statistics (Cramér's Statistic ϕ' and the multiple correlation coefficient) for the purpose of classification according to a hierarchical subdivision process like that previously used.

Although the procedure is strictly one of classification, not necessarily for purposes of prediction, the attributes by which an individual is classified are those often found associated with parole outcomes and used in the Uniform Parole Reports.² To explore the use of this technique in parole prediction, a pooling procedure was applied to the subgroups obtained, resulting in groups of paroled prisoners, each with significantly different parole out comes.