NCJ Number
91969
Date Published
Unknown
Length
59 pages
Annotation
This paper describes a form of cluster analysis designed to maximize within-set homogeneity which can be used as an alternative to traditional methods, which have some shortcomings when applied to binary data, such as test items.
Abstract
The procedure presented is agglomerative and non-hierarchical, based on a particular form of average-link methods. The distance measure used is any one of several forms of association index that are believed to be appropriate for binary data. The procedure includes a type of 'second-order' clustering designed to identify the most consistently forming clusters. The procedure was successful in identifying the clusters in artificial data and seemed very satisfactory in identifying an appropriate cluster solution in several sets of empirical data. Tabular data and 23 references are provided. (Author abstract modified)