PS: Wouldn't it be reasonable to (just) create an overlap between the individual thresholded maps? This would imply that both contrasts entered need to be significant. Hence this would be a conjunction from my understanding. 

Hi Stefan,

I think FDR (at least as implemented by FSL) is a voxel-level test, so you could just threshold each image at q = 0.05 and then take the intersection. That's a how a conjunction with the "minimum statistic" would be implemented with threshold-free cluster enhancement (TFCE), which is also a voxel-level test. Cluster-based tests make things a bit more complicated (e.g., Woo et al., 2014, NeuroImage).

To clarify, though, are you saying you have two independent images: one that's a contrast (e.g., condition A - condition B); and another that's a different condition (or contrast between other conditions) covaried with a behavioral measure across subjects? If the two images/analyses aren't independent, then you may want to rethink the conjunction.

Cheers,

David

Thanks a lot, David, that were also my thoughts (see PS). The special script mentioned in the query might then exist because of the cluster-based issue or to do it without the FSL viewer. 

I will also consider your second point, because some of the effects I am interested are independent (different contrasts), others, however, refer to the same condition difference (e.g., general condition difference AND effect of behavioral covariate, e.g., correlation with behavioral change between conditions). 

Best, Stefan 

Ah, I missed the PS because I had already started responding. But yes, significant overlap is fine for voxel-level inferences. Tom's easythresh_conj script is for conjunctions on cluster-level inferences; in that case, you can't just look at overlap since the inference is on the size of the cluster.

Cheers,

David