This paper proposes a powerful alternative to the t-test of the null hypothesis that a coefficient in linear regression is equal to zero when a regressor is mismeasured. We assume there are two contaminated measurements of the regressor of interest. We allow the two measurement errors to be...