Discriminant Analysis with Highly Intercorrelated Variables.

Abstract

Discriminant analysis is considered for populations possessing multivariate normal distributions with possibly different means but a common variance-covariance matrix sigma. The form of the discriminant scores is derived for the case when sigma is singular and is used to justify a procedure for selection of meaningful discriminant scores when the estimate of sigma is 'near' singular (i.e., when the estimate of sigma has some 'small' characteristic roots). The procedure is based entirely on sample data and for appropriate cases will yield the usual discriminant scores. The primary application is to situations where the variables are highly intercorrelated (i.e., for multicolinear variables). For such cases, substantial improvement in the misclassification rates is sometimes possible. The procedure has additional applications illustrated herein and can be easily incorporated into a computer program for routine data analysis.

Document Details

Document Type
Technical Report
Publication Date
Nov 01, 1974
Accession Number
ADA012433

Entities

People

  • Donald A. Anderson
  • Lyman L. Mcdonald

Organizations

  • University of Wyoming

Tags

DTIC Thesaurus Topics

  • Computational Processes
  • Computer Programs
  • Computers
  • Computing-Related Activities
  • Covariance
  • Data Analysis
  • Data Mining
  • Data Science
  • Discriminant Analysis
  • Information Science
  • Mathematics
  • Normal Distribution

Fields of Study

  • Mathematics

Readers

  • Regression Analysis.