Hello -
I have a set of subjects that, for various reasons, may have 'dead' voxels in different voxels in their EPI data, where dead voxels will very between subjects. (By 'dead' I mean unchanging and zero.) The dox for 3dttest++ say:
-mask mmm = Only compute results for voxels in the specified mask.
++ Voxels not in the mask will be set to 0 in the output.
++ If '-mask' is not used, all voxels will be tested.
++ HOWEVER: voxels whose input data is constant (in either set)
will NOT be processed and will get all zero outputs. This
inaction happens because the variance of a constant set of
data is zero, and division by zero is forbidden by the
Deities of Mathematics
It's not clear to me how to interpret the 'constant input data' clause when voxel A can be constant for some subjects, but not others. It could mean:
a) If one subject has constant data for voxel A, then that subject will not be considered as having provided any data for that voxel; however, statistics will be computed on that voxel using the the remaining subjects' data, resulting in one degree of freedom less owing to the omitted subject.
b) If one subject has constant data for voxel A, then no statistics will be computed for that voxel.
c) If one subject has constant data for voxel A, but other subjects have live data for voxel A, then AFNI won't notice that voxel A is dead for that subject, and will compute statistics as normal; except that they will be skewed by including a zero data point instead of live data.
d) something else.
I believe it is within my power to create some example datasets and test this out and find out the answer, but before doing that I figured that some kind soul might just be able to tell me.