What we found was that when we masked out any non-main effect A and B activity in datasets just prior to the ANOVA (%signal change Gaussian blurred data), we received different ANOVA results in each individual voxel. The difference was slight, but counter intuitive as I thought the ANOVA programs in afni processed on a voxel by voxel basis and there was no spatial variance pooling or anything of that sort.
ANOVA is indeed done on voxel-by-voxel basis with 3dANOVA programs. So the factor of interest has two levels, A and B? I have two questions here:
(1) How did you mask out your data with the two simple effects of A and B? Could you provide your script for the masking?
(2) During which step was the Gaussian blurring done, before or after 3dDeconvolve?
Gang