Bioequivalence and Bioavailability Forum

dear atish !

the choice of a post hoc test depend on the hypotheses you are interested in; here are some basic procedures based on individual p-values (e.g Wilcoxon-rank sum test):

- in the case of many-to-one comparisons I would suggest to use Hommel’s procedure
- in the case of all-pairwise comparisons I would suggest Holm’s procedure

more powerful post-hoc tests (for many-to-one, all-pairwise and any other contrasts) can be performed for example by SAS PROC MULTTEST (see http://support.sas.com/kb/22/addl/fusion22950_1_multtest.pdf for more information)

in the case that you can assume a monotone relationship between outcome and group (e.g. dose-response) I recommend the following paper:

Tamhane AC, Hochberg Y, Dunnett CW (1996). Multiple Test Procedures for Dose Finding. Biometrics, 52:21-37.

although this paper is based on normal distributed data – you can apply the presented approaches using permutation tests (e.g by SAS PROC MULTTEST with option=PERMUTATION and using the CONTRAST statement for your specific hypothesis) to get the p-values for the hypothesis you are interested in.

hope this helps

martin

PS.: please note that a statistically significant kruskal-wallis tests may not lead to a statistically significant individual hypothesis and vice versa. For this reason, I recommend not to use any overall tests when you are interested in individual hypotheses (refer to Hochberg and Tamhane (1987). Multiple Comparison Procedures. Wiley).