Interactive computer graphics in multivariate statistical research

Abstract

The value of interactive computing and of interactive computer graphics is discussed with particular reference to the study of statistical data. A system designed for interactive graphics for multivariate statistics is described, including a command language to be used as part of this system.
The system is in two parts: administrative programs and statistical programs. Types of statistical problems suitable for interactive graphical analysis are given, with the multivariate techniques used in the solution of these problems: principal components, factor analysis, cluster analysis and discriminant analysis. Each form of analysis is described with an outline of the program which implements the technique, including interactive aspects and forms of graphical output provided.
The latter part of this thesis covers the use of these programs. Principal components analysis is used for two sets of dataj a selection of results is shown and discussed. Results obtained from factor analysis of data collected for a psychological study are similarly given. For a fourth data set, initially a cluster analysis solution is found. The graphical presentation of results led to the identification of a set of variables which dominates this initial solution. The interactive procedure for identifying this and further sets of variables is described. Finally, results obtained from the initial cluster analysis solution are input to the program for discriminant analysis.