> Or, do you believe a statistical test is necessary or more convincing than this pure illustrative form?
Well, the informative value of showing the intersection of sig. activations (and/or reporting corresponding cluster sizes and such) is certainly very limited. Corresponding figures provide a nice overview of the different activation maps, but that's it then.
> However, is there a way to statistically test this?
I'm not aware of any scripts that solve the issue for the current models. Instead of the two-stage approach (single-subject models followed by a Flexible factorial) you could try to go with MFX models (there might be problems due to memory requirements though). Conjunctions on between-subject effects should be valid in that case (as they should have the correct error terms). If this doesn't work and if you still want a valid test then you would probably have to turn to a different software.
> I know that reviewers might lprefer a statistical test, but maybe it is not the case?
Well, I don't see any use in providing results from a statistically invalid test. From a methodological perspective, in case the reviewer asked you to run a conjunction analysis within that model I would respond that within SPM and given your model, the results would be invalid. If you're working on the draft it might be wise to add a footnote to avoid corresponding questions during the review. From a practical perspective, why not go with invalid results like others do...
Best
Helmut
|