Comparison of regression models in case of non-normality and Heteroscedasticity

Abstract

In regression analysis, in case of comparing two regression models and coefficients where the distribution of variables in question is not known, generalized p values may be used. The generalized p value is an extended version of the classical p value which provides only approximate solutions. Use of approximate methods, generalized p value, has better results performance with small samples. In this study, the generalized p value which may be used alternatively when different assumptions aren't fulfilled is researched theoretically; a simulation is conducted and an application in regression analysis is given. It is concluded that in generalized p value works well for the comparison of regression coefficients both under non-normality and heteroscedasticity