I have the following issue with a paper, which I think is very common:
We are studying the effect of a certain disease on white matter microstructure. We want to know if there is an effect of this disease in both (or either) males and females. We have run a whole group analysis, and after that split the group by sex to test the association in each sex separately. Our results are as follows: there is no effect in the whole group, there is an effect in males but not in females.
However, as pointed out by one of our reviewers, the correct way of testing sex effect is really to do an interaction analyses, and only split by group if the interaction term is significant. In our case there is no significant interaction term. Thus, if we choose this option, we have no effect in the whole group and no interaction effect. However, if we choose such an approach, we "miss out" on the effect of the disease on white matter in the males.
I guess this boils down to what the question really is. Do we really need to test if males and females differ significantly from each other (which is what an interaction term tests) to know if disease X has an effect on the brain in either sex? Would it not also be accurate to just split and report that we find an effect in males, but not females, and add that we cannot draw a conclusion about the females (as they might have been to low N) and not about males vs females either?
I am curious what your views are on interaction vs splitting groups.