Dear Jennifer,
Given the usual experimental designs you probably already had some a-priori hypotheses about group differences (although possibly you did not know which regions to be affected). In that case you would in fact want to go with planned comparisons (directed or undirected), e. g. patient group 1 vs. controls, patient group 2 vs. controls. Some of the contrasts might be irrelevant (maybe patient group 1 vs. 2).
When looking at different textbooks you will find different opinions, some authors suggest it is necessary to look at the main effect first followed by planned comparisons for comparing the different levels, others suggest that it is unnecessary in case of specific hypotheses (so it also doesn't matter whether you detect any sig. effects in the F test) and that one can start with planned comparisons. In any case, you should always be able to set up separate two-sample t-tests, e.g. one for patients 1 vs. controls and another for patients 2 vs. controls (thus without relying on an ANOVA model at all). You will still have to consider how to control for multiple testing though.
In any case, I'd say you should first make sure about the hypotheses (is the main effect really what you are interetest in or not) and whether tests are planned or really only "post-hoc".
It's in the nature of post-hoc tests that you look more closely at results obtained from a previous analysis. You can't solve the issue by relying on e.g. an anatomical label instead of the cluster obtained from the ANOVA, as choice of label will always be guided by those previous results. In a strict sense, one might argue that you have to conduct a second study to test the more specific hypotheses (in your case, those for paired comparisons) as they were generated only after obtaining a sig. effect in the planned analyses of the first study (detecting the sig. main effect group). But as stated before, usually one does have more specific hypotheses.
Hope this helps as a start
Helmut
|