 |
 |
Hello FSL experts, We have an experiment examining the effects of four separate groups on anxiety using fMRI. We examined changes in anxiety in response to a control, placebo, drug A, or drug B regimen. At the first level: we compared rest vs. manipulation (i.e., manipulation = drug, placebo, ect) in each subject. At the second level: I computed each group's average effect (across subjects from the first level) in parallel to regressing out the percent change in anxiety scores. So, EV1 is the group average of my two contrasts ( positive and negative contrasts = rest vs drug/or placebo) and EV2 is the percent change in mood ratings (positive vs negative contrasts). The data are beautiful. For instance, greater PFC predicts anxiety relief in drug A and greater DLPFC predicts anxiety relief for the placebo group. At the third level: I want to not only compare each group's main effect (i.e., drug > rest vs. placebo > rest), but more importantly I want to directly compare the brain regions associated with anxiety relief in each group BETWEEN groups. At the second and third level (for now), I have FLAME 1+2 checked. What makes the most sense to me is to run a higher level analysis comparing each respective group's .gfeat corresponding to the second level, however, when I have random effects turned on, the analysis "crashes" to where the post-stats tab is left empty. When I run it as a fixed effects, the analysis runs beautifully (i.e., finishes without crashing and has accompanying results). My main question for the experiment is: Are the brain mechanisms associated with drug a or b's anxiety relief distinct from those engaged by placebo? Am I justified to run the group regressor comparison (third level) with a fixed effects model since random effects was "turned on" at the second level? Do I need a significant pairwise comparison f-test to justify interpreting my apriori comparisons (third level)? Thank you so much for your time!