Comparison of difference based variance estimators for partially linear models

In this research, we evaluated two difference based variance estimators: one by Gasser, Sroka, and Jennen-Steinmetz, and another by Hall, Kay, and Titterington for use in partially linear models. Under various settings, we compared power of tests for heteroskedasticity, and other finite population properties of the estimators using simulation studies. We also proved that under regularity conditions, the estimator from Hall, Kay, and Titterington provides larger power of the tests for heteroskedasticity. A real example is given to illustrate the usage of the estimators.