Empirical Comparison of Binary and Continuous Proximity Measures for Clustering Occupational Task Data.

摘要

Thirteen binary and three continuous proximity measures were used to cluster-analyze job incumbent profiles of task inventory data. The results were compared: (1) to recommend a binary measure for programming into CODAP System 80, a software package used extensively by the military and many other organizations, and (2) to determine to what extent binary measures can produce cluster solutions similar to solutions based on continuous measures. Sixteen 250-by-250 proximity matrices were derived from each of three Navy occupational samples, and the clustering procedure in CODAP was applied to selected matrices. Proximity matrix and cluster solution comparison revealed that: (1) there was high variability among binary measures, (2) the Jaccard and Dice measures were the most powerful binary measures, and (3) there was high similarity between the Jaccard and distance measures. The implications of the findings are discussed with reference to the proportion of zero scores in task inventory data. The Jaccard measure is recommended for clustering binary data for tasks and for programming into CODAP System 80. (Author)