Thanks Bob and all you wonderful afni folks.
There are many ways to skin this cat.
I have the means I need now.
How about a boot camp just on 3dcalc?
Wild horses could not keep me away.
So, now some philosophy (or substance):
Lets imagine I have 5 datasets from 5 people doing the same task.
The datasets are in percent change units.
I have individually masked each subject's data for statistically significant activation.
So subjects either have an estimate of percent signal change or 0.0 for each voxel.
What is the right thing to do when averaging?
Perhaps the zeros should be included in the average -- you are assuming that subjects with a 0.0 had no signal in that voxel.
Or perhaps the zeros should not be included in the average -- maybe you want an average of activated voxels, and maybe you think that setting a non-significant voxel to 0.0 is unreasonable.
In my current case, it made a huge difference.
I wonder how many people think of this distinction when publishing their results -- my guess: not many.
Lee