It's standard in PPI analyses to use the same dataset for both defining subject-wise functional ROIs and doing the PPI analysis.
Without further elaboration, a (conservative) statistician outside the neuroimaging community might frown on this.
Suggestions that it's actually OK to "double-dip" in some situations can be found in Friston et al., "A critique of functional localisers," which focuses on factorial designs.
The double-dipping article itself ("Circular analysis in systems neuroscience: the dangers of double dipping," Kriegeskorte et al. alludes to a common criterion: you can use the same dataset if the contrast for defining the functional region is statistically independent of the contrast used in further analyses. That criterion is valid. An often-used proxy for that criterion is that the contrasts be orthogonal. As pointed out on p. 536 of Kriegskorte et al., that simpler criterion isn't quite right in most cases, though one could argue that the resulting bias is small.
Bottom line: many or even most referees won't fault you for double-dipping, but it's possible some might. (I.e., using functional localizers in PPI is fairly widely but not uniformly accepted in the neuroimaging community). There's also your own preferences as to how statistically certain/conservative you want to be.
Ideally, of course, these issues would never come up because we'd be able to do so much scanning that we could always use independent datasets, but the community as a whole has implicitly decided the expense isn't worth it.
|