ABSTRACT
Agglomerative methods of clustering begin with the calculation of a distance matrix. Similarities such as correlations are converted to distances by subtracting each value from a constant such as 1.00. Ward uses the sum of squares as the basis for inclusion of objects in a group. Ward relates that for job classification and task analysis, percentage of overlap is used more often than correlation as a measure of similarity. Dunn-Rankin, Shimizu, and King studied the reward preference of fifth and sixth-grade children. They had students respond to a paired comparison task involving five kinds of reward using a Reward Preference Inventory (RPI). PROC Cluster in SAS starts with a distance matrix. If the data are similarities they must be converted to distances prior to the analysis. In Johnson's nonmetric clustering method similarities are converted to distances. Distances are either measured to the closest member in a cluster or the farthest member in a cluster.
