Hello there!
Thanks for all your help last time. I have another question to ask. We are calculating group data for 3 different models (1) with no motion correction (2) with motion correction but no motion parameter regressors and (3) with MC and motion parameters. I have the question as to how to threshold properly to reduce false positives without too many false negatives. The reviewers' comment that we need to have a more conservative statistical correction for reducing the likelihood of false positives. They say that several clusters are (1) found only in a single model and (2) are only 3-5 voxels in volume. I decided to look into both AlphaSim and FDR corrections. I tried AlphaSim and get that one needs ~20 voxels to get alpha<.05 for whole brain using the current single voxel threshold of .001. This I think would be overly conservative. I also noted the q-value from the GUI is always <.05 in our case but the MDF is 90% or higher. I am confused since according to FDR the false positives should be fine for the whole brain but how and why is the MDF so large--is it a problem?. My question is how does one pick a proper threshold in this case? when one knows we are trying to compare clusters and their volumes between models. Are the reviewers correct that those voxels are false positives or are they from the fact we are over conservative on false positives since we actually don't even detect most true positives?
Thanks for your help ahead of time.
