Comparison of outlier detection methods in standard 2 × 2 crossover design

This paper discussed methods for outlier detection in standard 2 × 2 crossover studies. Two outlier detection procedures were carried out based on residuals. Under a simplified model of 2 × 2 crossover design, we present the classical calculation of studentized residual (SR1) and propose a new studentized residual using median absolute deviation (SR2) to identify possible outlying subjects. The performances of both procedures in detecting subject outliers were compared. We show via simulation that a proposed procedure using SR2 is more powerful than that using SR1 for outlier detection. As an illustration, these procedures were applied to two real data sets from studies of bioavailability and kinesiology, respectively.