Simultaneous multiple comparisons with a control using median differences and permutation tests

Permutation methods using median differences for simultaneous pairwise comparisons with a control are investigated. Simulation results suggest that the permutation methods are generally more powerful than the Dunnett procedure when data are from nonnormal distributions. A new procedure is shown to provide strong control of the familywise error rate, and have highest power for detecting the treatment that differs most from the control, for certain nonnormal distributions. Step-down permutation procedures, which have greater power to detect treatment differences with the control, are also proposed and examined. The procedures are illustrated using an example from the applied literature.