Robustness of statistical tests