CALIBRATION OF QUALITATIVE OR QUANTITATIVE VARIABLES FOR USE IN MULTIPLE-GROUP DISCRIMINANT ANALYSIS

Abstract

Three limitations on the application of multiple discriminant analysis in certain statistical problems are discussed. These are: (1) difficulty in including qualitative antecedent variables in the analysis, (2) failure of quantitative variables to discriminate in a linear fashion, and (3) multivaluedness of antecedent variables within consequent groups. A technique for optimal assignment of numerical scale values to antecedent classes is proposed for overcoming these limitations. The technique evolves essentially as a calibration procedure. It depends only upon and exhausts all the information in the antecedent-class, consequentgroup contingency table. A straightforward procedure for carrying out the computations is outlined in detail. The problem of separate vs simultaneous calibration is resolved, and the implications of the use of calibrated variables in a screening (selection of antecedent variables) procedure examined. (Author)

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 1961
Accession Number
AD0251139

Entities

People

  • Joseph G. Bryan

Tags

DTIC Thesaurus Topics

  • Calibration
  • Computations
  • Data Science
  • Discriminant Analysis
  • Information Science
  • Mathematical Analysis
  • Mathematics

Readers

  • Calculus or Mathematical Analysis
  • Organizational Psychology.
  • Regression Analysis.