Jackknife empirical likelihood tests for error distributions in regression models

Regression models are commonly used to model the relationship between responses and covariates. For testing the error distribution, some classical test statistics such as Kolmogorov–Smirnov test and Cramér–von-Mises test suffer from the complicated limiting distribution due to the plug-in estimate for the unknown parameters. Hence some ad hoc procedure such as bootstrap method is needed to obtain critical points. Recently, Khmaladze and Koul (2004) [7] have proposed an asymptotically distribution free test via some Martingale transforms. However, the calculation of such a test becomes quite involved, which usually requires numeric integration when the Cramér–von-Mises type of test is employed. In this paper we propose a novel jackknife empirical likelihood method which is easy to compute and has a chi-square limit so that critical values are ready at hand. A simulation study confirms that the new test has an accurate size and is powerful too.