Accession Number : AD0424824

Title :   ON THE PERFORMANCE OF THE LINEAR DISCRIMINANT FUNCTION,

Corporate Author : HARVARD UNIV CAMBRIDGE MASS

Personal Author(s) : Cochran,William G.

Report Date : 1963

Pagination or Media Count : 13

Abstract : The relation between the discriminating power of linear discriminant function (LDF) and the discriminating powers of the individual variates used in the function is studied. The extensive literature on the LDF gives a little guidance on how to select variates for constructing a discriminant. From his knowledge of the problem, an investigator can sometimes suggest a small number of varieties that he thinks will have good discriminating power. To these he can often add a second and larger list of variates that might be helpful, though he is less sure of this. A meterologist, for example, could probably supply two such lists of weather variables if the purpose were to predict whether it would rain or not on the next day. When measurements are made on the second list of variables in the two populations between which we wish to discriminate, it may be found that a number of them show little separation between the two populations. In this event it would be useful to have a simple rule by which such variates can be discarded before computing the LDF, on the grounds that their inclusion is unlikely to produce a material increase in discriminating power. Given a knowledge of the discriminating power of each of a set ovf variates, it is sometimes helpful to be able to make a rough estimate of the discriminating power of the LDF without going to the trouble of computing it. (Author)

Descriptors :   (*PROBABILITY, STATISTICAL ANALYSIS), MEASUREMENT, POTENTIAL THEORY, FUNCTIONS(MATHEMATICS), STATISTICAL TESTS

Distribution Statement : APPROVED FOR PUBLIC RELEASE