A test or estimator is robust if it is not sensitive to small discrepancies in assumptions, such as the assumption of the normality of the underlying distribution. A test is said to be robust to outliers if it is not unduly affected by their presence.