Some Alternative Methods to Stepwise Regression for the Screening of Variables.

Abstract

This paper addresses the problem of screening potential variables for entrance in a linear multiple regression setting. The purpose of the work presented here is to propose two screening methods, both of which have roots in principle component analysis, and which evaluate a combination of variables in an efficient enough manner so that enumeration of all combinations is feasible even when the number of potential variables is quite large. Using the square of the multiple correlation coefficient as the criterion, the selections made by these methods in several test cases are evaluated, and compared with the selections made by the methods of total enumeration and stepwise regression. The paper concludes with overall evaluations of the two methods and suggests directions for further study.

Document Details

Document Type
Technical Report
Publication Date
Dec 01, 1974
Accession Number
ADA005470

Entities

People

  • Dennis George Lambell

Organizations

  • Naval Postgraduate School

Tags

DTIC Thesaurus Topics

  • Coefficients

Fields of Study

  • Mathematics

Readers

  • Organizational Psychology.
  • Systems Analysis and Design