Detecting univariate outliers