> does AFNI take the absolute difference between two groups when
> computing a t-test when there is deactivation in one or both groups?
No, it takes whatever values from the input files without any manipulations.
> how do people handle negative values in their t-test results?
I don't know abou this.
> if there are voxels with zero values, and the t-test performs a voxel
> by voxel comparison, are some voxels with activation eliminated?
Not so sure what your question is about. Are you asking this: If those voxels with 0 values had nonzero values, might some of those voxels be identified as "activated"? If so, yes it is definitely possible.
AFNI Gang