Dear expert, 

Say I'm interested in visual stimulation (VIS) as compared to a control condition (CTR). If you run a meta analysis you get positive clusters (activation for VIS > CTR) in _pos.nii.gz and negative clusters (deactivation for VIS > CTR) in _neg.nii.gz. It is then implied that for CTR > VIS the same areas should be activated. Hence the same custer should be found as activation for CTR > VIS. 

I found out that a major difference of the new (PSI) algorithm as compared to the previous version seems to be that this is not the case anymore, because the sign flipping permuation test now needs to be performed twice (once for CTR > VIS and once for VIS > CTR). So If you analyse VIS > CTR you might not get any negative clusters, but if you analyse separately CTR > VIS you do get positive clusters. 

Is that correct? 

Thanks for the great software!

Dear Jorge, 

The permutation test saves the distribution of the maximum statistic (for _pos) and the minimum statistic (for _neg) simultaneously, and thus, it must be performed only once. That said, the permutations are random; thus, the threshold may vary if you conduct the test twice. This uncertainty may be relevant when the clusters are just above the threshold; you can increase the number of permutations to decrease this uncertainty.

Best regards, 
Lydia