Dear Ben,
To begin with, I think if it were common to report beta/con estimates (also for whole-brain analyses) then post-hoc tests would be more "straightforward". It's a little funny to provide plots only on post-hoc level, as it always feels somewhat "weird" then.
Leaving this aside, the approach should be alright in general, except if you want to prove that there is a group difference for A and no group difference for B (= proving the null - instead of concluding there is a group difference for A and no evidence for a sig. group difference for B).
Instead of the sphere one could extract / report / plot estimates from the entire cluster(s) - this way one would rely on the sig. cluster and not a subset of voxels (and possibly the sphere also includes non-sig. voxels). The post-hoc tests would still be based on an aggregate score (in contrast to the initial whole-brain analysis) though, e.g. the average across all voxels.
To avoid this you could run your two-sample t-test but use whole-brain analyses for the post-hoc tests (thus conducted within SPM) and provide beta/con maps and (unthresholded) T maps for voxels within the sig. cluster(s) of the interaction contrast. This way, one could present data for every single voxel within those cluster(s).
Finally, depending on hypotheses one could also just start with a set of whole-brain models for the planned comparisons, present contrast and T maps for Con A, Pat A, Con B - Con A, Pat B - Pat A as within-group contrasts, followed by Con A - Pat A, (Con B - Con A) - (Pat B - Pat A). This certainly is a lot of information (which could in part go into supplementary material maybe), but it would provide an extensive overview of the data & might be useful in some instances.
Best regards
Helmut
|