Dear Mike,
All four approaches are bad practice: Having failed to detect sig. effects on whole-brain level you start to consider ROI analyses, which is not identical but similar to just re-running the whole-brain analysis with a lower threshold (which is what most of the small volume corrections are about - these SVCs are conducted on one or a limited set of a-priori regions after a whole-brain analysis in which other a-priori regions had already reached significance. Rather, one would have to run a SVC on all a-priori regions, then followed by the whole-brain or better, a "rest-of-the-brain" analysis).
In case the two components EV1 and EV2 can be separated properly (which can be tricky without partial / half trials consisting of EV1 component only), and assuming there's another factor task (control, task 1, ... task m), I would opt for a 2 x m design. Then test for the two main effects and the interaction. The latter allows to conclude whether the effect during EV1 differs significantly from the effect during EV2 or not, which I assume to be an interesting question in that context. Otherwise you might just end up with having a sig. effect during EV1 and none during EV2, which does not imply that the effect differs between EV1 and EV2.
In case of proper a-priori hypotheses on those regions I would conduct ROI analyses based on (2) or (3). You can then conduct tests, plot average response of the region, provide effect sizes.
In case the hypothesis is not about the average response of the hippocampus, amygdala, but rather about "somewhere within those regions", you should go with a small volume correction. Throw together all your a-priori regions, then focus on the mapping (where do I detect sig. voxels). If there are no effects in some instances, well, that's it then.
Best regards
Helmut
|