Is Multiple Comparison Correction necessary for multiple ANCOVA tests

Question

I'm working on fMRI data comparing brain activation from ten different brain areas across three groups after controlling for covariates such as age, sex, disease severity. I built an ANCOVA model with post hoc analysis with Bonferroni corrections including group as the independent variable, mean brain activation from a given brain area as the dependent variable and above-mentioned variables as covariates. I wanted to know if another multiple comparison correction is required to control for making 10 such models for different areas. I tried doing some literature search but couldn't get any confirmatory answer.

Answer 1

If you have the same dependent variable and you are testing your hypothesis in a single model (one ANCOVA/MANCOVA with several grouping variables like the brain regions you mention), then the typical way is to add pairwise corrections like Bonferonni. I note that this is the typical way because this is actually a contentious issue (see my answer here where I explain in the latter half of my answer some of the pros and cons of pairwise tests). The short answer is, perhaps they are useful if you had specific hypotheses by group if you want to sacrifice some power in doing so. But often I find ANOVAs are ran to test any difference and the pairwise tests are used in an exploratory way to "confirm" group differences.

This to me is a bit backwards (my personal opinion, so take that with a grain of salt). It would instead be better to either only create tests around these group comparisons of interest (using perhaps contrasts in a categorical regression) or look at the overall effect and model out group differences if they are not theoretically useful (such as a mixed model which only shows mean shifts by each group while looking at overall effects). I personally doubt neoropsychs or neurologists care about hypotheses about 10 regions of interest (even though they probably test them this way) and are instead concentrated on which areas are most crucial to their hypothesized models.