Multivariate Statistical Analysis for Discrete Data : An Approach to the Multinomial Classification Analysis

Abstract

In the field of industrial management, many of the data collected are sometimes of categorical or qualitative nature. But there is a tendency in analyzing such data composed of nominal or ordinal variables being to igonore their discrete characters and to proceed with continuous variale method as Fisher's linear discriminant function. This paper is concerned with the problem of classifying items which belong to one of two classes. Each classifying item is characterized by a discrete variable which has the multinomial distribution. We present a sample-based method for the two group multinomial classification and some brief comparison of this method to the other such as the use of linear discriminant function, distance method and so on. A procedure for the selection of subset of variables for the two group classification problem is also described.