On power comparison of normality tests of error terms in regression models

Abstract

Verification of regression models constitutes one of the most important steps in applied regression analysis and is primarily based on analysis of error terms. Some statistical procedures used in the testing of linear regression model such as t-test or F-test are based on assumption of normality of error terms. Failure to assess non-normality of the error terms may lead to incorrect results of usual statistical inference techniques. This contribution aims at assessment of a power of several robust and non-robust normality tests of error terms in regression models. For this purpose using a Monte Carlo simulation technique we simulate the dependent variable by p-location outlier models and estimate the ordinary least square residuals. Finally we test the normality of residuals to explore the power and robustness of selected robust and non-robust normality tests.