 |
 |
Dear Donald, What I had in mind was to conduct tests according to the initial hypotheses, in my example contrasting A and B, B and C, based on tests that fit the needs. In case the term "pair-wise comparisons" implies that one operates within the ANOVA model then sorry for the confusion & thanks for your clarification. Anyway. There might be instances in which one does not have explicit hypotheses, e. g. genotypes, maybe also phases of the menstrual cycle, different stages of a disease, but apart from that I would be surprised if one were to recruit three groups of subjects with no explicit hypothesis on why to recruit (and test) these three groups. In case of explicit hypotheses on between-group differences I would test accordingly from the beginning, "ignoring" the main effect as it is not of interest. In case of no such hypotheses one might test for a main effect and report group means and SDs, but one could argue to leave any additional inferential statistics to follow-ups. In some ways this reminds me of post-hoc tests after finding a significant 2x2 interaction. Looking at literature it is very common to conduct additional "post-hoc" analyses after finding a significant 2x2 interaction. Contrasting the levels A1, A2 for B1 and B2 or alternatively, contrasting cells can be informative, but these tests do not "explain" the 2x2 interaction. However, in case one considers these tests to be relevant then they would be additional hypotheses that could be tested independently of the outcome of the 2x2 interaction. Best regards Helmut