Candace Gail Hamner (Todd) - Dissertation 1983

Retyped Fall 1999

Abstract

Several graphical methods for representing multivariate data are evaluated in this study. Three face-type techniques and one non-face-type technique are used in clustering tasks and are evaluated on the basis of their ability to recover natural groupings in three data sets. Several hierarchical clustering algorithms and one ad hoc numerical technique for clustering are applied in one analysis and are ranked along with the graphical clustering methods on several evaluative criteria. The graphical techniques used and evaluated are line printer specific faces, a modification of the original Chernoff face, a recently developed totally asymmetric face, and polar plots of finite Fourier series. The plots for the methods are generated using an interactive graphics package called GRAPHPAK. Evaluative criteria used in the study include number of correct assignments, the Rand statistic, and other more recently developed measures of similarity between partitions of a set.