A consistent test of significance

This paper presents a test of significance consistent under nonparametric alternatives. Under the null hypothesis, a regressor has no effect on the regression model. Our statistic does not require to estimate the model on the alternative hypothesis, which is left unspecified. Hence, no smoothing techniques are required. The statistic is a weighted empirical process which resembles the Cram~r-von Mises. The asymptotic test is consistent under Pitman's alternatives converging to the null at arate n-1/2. A Monte-Cario experiment illustrates the performance ofthe test in small samples. We also inelude two applications involving biomedical and acid rain data