Accession Number:
AD0757430
Title:
A Subset Selection Procedure for Regression Variables,
Descriptive Note:
Corporate Author:
PURDUE UNIV LAFAYETTE IND DEPT OF STATISTICS
Personal Author(s):
Report Date:
1973-02-01
Pagination or Media Count:
17.0
Abstract:
Given a regression model with p independent variables, several methods are available for selecting a subset of size t p which gives an adequate description of the dependent variable. By using the capabilities of the computer, one can now determine the subset corresponding to the largest sample multiple correlation coefficient or equivalently the smallest residual mean square. Due to sampling variation, however, there is no guarantee that this corresponds to the smallest value of the expected residual mean square. A procedure is presented to determine a collection of subsets, each of given size t, having the property that the probability of including the subset corresponding to the smallest value of the expected residual mean square is bounded below by some prespecified constant, 1 - alpha. An example using real data is examined to illustrate the technique. Author
Descriptors:
Subject Categories:
- Statistics and Probability