I got some seemingly odd results from the new 3dFDR algorithm. After flipping back and forth between the input and output dataset, I am suspicious that it is producing incorrect results. My input dataset has very few voxels set to a probability of 1 (within the masked region, the brain) and has some prounced clusters of activation (note compelling negative VMPFC, for instance). The output contains no significant voxels at any value of q that would be useful for making a statement about false discovery rate. If I run the old algorithm (-old), the output dataset includes the clusters at useful values of q. This "old" results seems to reflect the spirit of what FDR is intended to do better than the new results.
My suspicion, therefore, is that this is tripping over some bug in the implementation. Or do I miss understand what the new algorithm is trying to do?
I assumed positive regression dependency on subsets, as is default.
I would upload the input dataset in question, but the web server seems to check out when I attach it.
Best,
Brent