 |
 |
You should think of the thresholds beforehand and not e.g. turn to FDR because clusters don't survive FWE. The same holds for regions of interest, if you have a-priori regions it's well reasonable to look at some composite score (average beta/con estimate, first eigenvariate) for them, independent of the whole-brain analysis. In contrast, deciding that you have "a-priori" regions once you've realized there are no effects on whole-brain level and/or that they don't survive certain corrections is bad practice. Modifying / adding additional analyses because reviewers suggest to do so based on the current results would also be bad practice if it's not explicitely stated that it's a mere exploratory analysis. Unfortunately, you can always write the paper as if you had those a-priori regions right from the beginning and/or search for arguments why you had always always always wanted to run the additional analyses. Accordingly, it would be very useful if study protocols were published beforehand. However, a workgroup with lots of funding might acquire the data beforehand nonetheless and then publish an "optimized" study protocol based on the results. Best Helmut