I have been creating group comparisons with fMRI data. I am using basic models, no grand mean scaling, no masking of images, and no global calculation. The group analyses contain about 12 subjects.

I have noticed that if I set the threshold at corrected 0.05 I get almost an entire brain of activation, which doesn't reflect any of the individual subject data. If I set the threshold at uncorrected 0.001 I get very small amounts of activation more in line with the individual subject data.