If you have a strong a-priori hypothesis about a particular brain region you can well run a ROI analysis, based on some functional localizer, an anatomical label derived from some brain atlases, some coordinates reported in meta-analyses, ... In your case it seems your hypothesis is about the FC between your seed region insula and another brain region. If that's the hypothesis, you can adjust your analyses accordingly, focusing on e.g. the average time courses of the two regions. However, you should definitely NOT look at your whole-brain results and define your ROIs a posteriori based on "interesting" trends. This is bad statistics / "double-dipping".
Having said that, it is common practice to "adjust" the hypotheses after some "preliminary" analyses. There are many papers in which authors give a nice literature overview in introduction and then infer that one or several brain regions would especially be interesting. Thus the ROI analyses seem to be justified, but if you have a closer look it often remains unclear why they focussed on this particular region. In other words, by citing other papers you would well have been able to define other/additional ROIs. Or have you ever wondered how they came up to run a GLM, plus some ROI analysis, then do some PPI on some funky region, plus another fancy advanced analysis? Looking at a final paper it is difficult to detect, but if you look at conference contributions you sometimes detect substantial changes in hypotheses / analyses ;-) This does definitely NOT mean that ROI analyses as such are bad. But often their application is flawed.
|