Nonparametric tests for the effect of treatment on conditional variance

Abstract

This paper proposes nonparametric tests for the null hypothesis that a treatment has a zero effect on conditional variance for all subpopulations defined by covariates. Rather than the mean of outcome, which measures to what extent treatment changes the level of outcome, researchers are also interested in how the treatment affects the dispersion of outcome. We use variance to measure the dispersion and estimate the conditional variances by series method. We give a test rule comparing a Wald-type test statistic with the critical value from chi-squared distribution. We also construct a normalized test statistic that is asymptotically standard normal under the null hypothesis. We illustrate the usefulness of the proposed test by Monte Carlo simulations and an empirical example that investigates the effect of unionism on wage dispersion.