Dear all

I understood that NiiStat assumes lesions to be associated with decreased performance. In fact it is described that regions or voxels associated with better behavioral performance are ignored.

I first applied the FDR for correction of multiple comparisons and it described the confidence interval like this:
q=0.0500 FDR correction for PFT with 159 tests is z<-3.76274587, z>Inf
which I undestand as it is one-tailed.

Then I applied Normal Permutation for correction of multiple comparisons and it described the confidence interval like this:
p<0.050 permutation correction for PFT is z<-2.41413 z>3.73859
which confuses me as the Z-value can be positiv and means that the increase in lesion is associated with increase in performance.

Is it legitimate to interpret a positive z-value for a specific test in a specific ROI as an indication for increased performance in case of lesion in this area?

Best,
Magdalena

Yes Magdalena, this would be the correct interpretation for the data. 
Although we usually expect more lesion to be associated with worse performance, it is possible to find the opposite effects. 
Imagine comparing participants with ACA and MCA strokes. Also imagine that your primary DV is speech fluency. 
Having a larger lesion to the ACA might predict higher language fluency scores, because having a lesion to brain areas supplied by the ACA means you likely DON'T have a lesion  to brain areas supplied to the MCA (which matter more in terms of language function).
Hope that makes sense.
Best,
Roger

Dear Roger

Yes it indeed does make sense, thank you so much!

But why does the NiiStat manual say, that the statistical tests are one-tailed?
"NiiStat assumes that any possible relationship between brain data and behavioral scores is monotonic. For example, when analyzing lesion location, we typically make an a-priori one-tailed hypothesis: brain injury leads to impaired behavioral performance."
According to this description I did not expect NiiStat to even look for a positive effect of lesion on behavioral performance...


So in case I have these postive z-socres, before I can rely on them I should adapt p to be < 0.025?

Best,
Magdalena