A bias test for heteroscedastic linear least squares regression

A correlation between regressors and disturbances presents challenging problems in linear regression. Issues like omitted variables, measurement error and simultaneity render ordinary least squares (OLS) biased and inconsistent. In the context of heteroscedastic linear regression, this note proposes a bias test that is simple to apply. It does not reveal the size or sign of OLS bias but instead provides a statistic to assess the probable presence or absence of bias. The test is examined in simulation and in real data sets.