Article contents
The Deletion of Variables From Regression Models Based on Tests of Significance: A Statistical and Moral Issue*
Published online by Cambridge University Press: 28 April 2015
Extract
The purpose of this paper is to illustrate some of the dangers inherent in use of statistical tests as a criterion for deleting variables from regression models. The deletion of variables from regression models based on t or F tests of regression coefficients has been a procedure widely followed by applied economists and other researchers. When economic theory does not provide an adequate conceptual basis for rigorous a priori specification of the regression model, one approach to model specification has been to include in the regression equation all variables thought to be “somehow” related to the dependent variable of interest. Subsets of variables with statistically significant coefficients are identified, with the aid of a stepwise regression routine. Truncated models consisting of only those variables with statistically significant regression coefficients are sometimes presented in the published research without reference to the initial data dredging that took place.
- Type
- Research Article
- Information
- Copyright
- Copyright © Southern Agricultural Economics Association 1975
Footnotes
The authors are indebted for the assistance provided by E. W. Kehrberg, T. K. White, J. Havlicek, G. L. Bradford, Alan J. Randall, Eldon D. Smith and L. D. Jones. Any errors remain the responsibility of the authors.
References
- 3
- Cited by