CLASSIFICATION OF PRISON INMATES BASED ON HIERARCHICAL CLUSTER ANALYSIS

Abstract

Six clustering methods (Ward's method, average linkage clustering, complete linkage clustering, single linkage clustering, nearest neighbor, and k-means) were applied to the matrix of Euclidean distances among inmates in the 6-dimensional space defined by the following variables: 1) age, 2) length of the internment already fulfilled, 3) paranoid ideations and behavior, 4) hardness, 5) psychotic ideations and behavior, and 6) means-ends interpersonal problem solving skills. The classification determined by each of the clustering procedures was matched with the actual classification of the inmates into those who were and those who were not granted outside permits, by computing unconditional and conditional classification rates. Ward's method, complete linkage and the k-means procedure were able to match satisfactorily the qualitative classification of the inmates. However, these methods were more efficient in determining which inmates would be denied an outside permit than in determining those that would be granted an outside permit. This also occurs in the qualitative classification: it is easy to determine which inmates will not make good use of an outside permit; it is much more difficult to determine which inmates will make good use of it.