A Study of Robustness of Preliminary Tests in Conditionally Specified Regression Models
Abstract
**Please note that the full text is embargoed** ABSTRACT: The selection of variables in regression analysis is usually done by using
preliminary tests. When it is doubtful whether the regression coefficients of a set of variables are equal to zero or not, one may use a significance test. When the test rejects the null hypothesis, the set of variables is retained in the model otherwise the set is deleted from the model. The preliminary test is usually an F test based on normal theory. This paper examines the robustness of the preliminary test in model selection by a Monte Carlo study.