skip to main content skip to footer

A Simplified Procedure for Multiple Discriminant Analysis

Saunders, David R.
Publication Year:
Report Number:
ETS Research Memorandum
Document Type:
Page Count:
Subject/Key Words:
Discriminant Analysis, Scores


Multiple discriminant analysis, as used in this paper, refers to any method for determining from a set of scores which of two or more given groups of individuals a new individual most closely resembles. There are practical situations in which the psychometrist would like very much to have such information for an individual however, until quite recently this was the end of the story. In 1950, Bryan described the first feasible solution to the problem for the case in which there are both many groups and many measured variables involved. However, the computational solution that Bryan's work makes available still has limitations mitigating against its wide application. In general, the procedure calls for the computation of a good many intermediate matrices, and employs processes that are unfamiliar to many computers. It would appear to be desirable to ask whether some of the elegance of this method might better be sacrificed, on occasion, in favor of a more economical and readily comprehended computing plan, that would produce results adequate for many purposes. It is the purpose of the present report to describe a simplified procedure for multiple discriminant analysis which seems to have these properties. This paper was read before the Psychometric Society on September 6, 1952. A more complete report of the work is in preparation, but has been delayed by higher priority work.

Read More