seems to me that many things could occur - I have no definitive answer
here but here are some thought
if you read a bit the spm_spm script you will see the the mask is used
in a way that affect different things. For instance if the non
sphericity is not specified SPM computes an omnibus test to get some
'activated' voxels .. so I guess if different mask then maybe different
nb of 'activated' voxels ; same for the estimation of the smoothness etc ...
> 2) Case 1: Normalize First - I normalized the first EPI to the EPI template,
> and applied the transformation to the EPI time series, and my mask_original
> to create "mask_normalized". I then ran the model, using "mask_normalized"
> as my explicit mask. SPM created a mask that is the union of
> mask_normalized and the internally derived implicit mask. I'll call that
> union mask "normalized_mask_A", since it's already in MNI space.
> 3) Case 2: Normalize After - I took my raw EPI images and modeled them,
> using "mask_original" as my explicit mask. SPM created a mask that is the
> union of "mask_original" and the implicit mask. I'll call that "mask_B".
> Then, I normalized the first EPI to the EPI template, and applied the
> transformation to the con images and to "mask_B" to produce
> "normalized_mask_B". Through trial and error, I've figured out that this
> produces a con image that need to be masked, and the subsequent masked con
> contains the zero values that have to be converted to NaN before it will
> work properly in the group model, but I don't think that's the root of the
> Now, my question is this: Shouldn't "normalized_mask_A" and
> "normalized_mask_B" be the same? They aren't! Am I doing something
> incorrectly here?
in addition to the diff in stats that could arise for the reasons
mentioned above I think (you will need to check) to the difference comes
from the implicit mask created by SPM. In case 1 SPM creates a mask from
the normalized EPI time series whereas in case 2 it is from the
non-normalized images. These masks are different. I can't see other
differences .. but I may be wrong here.
Hope this helps
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.