Robust multiple comparisons against a control group with application in toxicology
The Dunnett procedure compares several treatment or dose groups with a control group, while controlling the familywise error rate. When deviations from the normal distribution and heterogeneous variances occur, the nominal $α$ level may be violated, and power may be reduced. Various robust modifications are discussed, whereby the novel most likely transformation (MLT)-Dunnett version is recommended as almost always appropriate by means of a simulation study. The MLT-Dunnett is especially useful because it can jointly and comparably analyse differently scaled endpoints. Furthermore, a related multiple endpoints test is proposed using the odds ratio as a common effect size. With the statistical software R, the method is readily applicable using the CRAN libraries \verb|multcomp| and \verb|mlt|, real data can be easily analyzed.