> I'm curious if the coefficient from sub-brick 0 in spmg3 can be treated as a
> percentage change value as I would have done for the GAM-based coefficient.
No, the SPMG3 basis functions are not normalized, and they don't have a peak of 1. You can check them out by running the following two commands:
3dDeconvolve -nodata 300 0.1 -polort -1 -num_stimts 1 -stim_times 1 '1D: 0' SPMG3 -x1D spmg3
1dplot -one spmg3.xmat.1D &
So you may have to scale your original beta values by ~6 before you could interpret them as percent signal changes.
HTH,
Gang