In regression analysis, classical estimations may be excessively influenced by a few atypical observations. We propose a Hausman-type test to balance robustness and efficiency and to check whether a robust method should be implemented. An economic application is presented.