A problem using the above method: it seems to yield percent changes of 20% and above in activated regions, which can't have anything to do with reality. Operating on only one run without concatenation, and determining the clip level as described above, my script is as follows:

# Calculate the mean for each run
3dTstat -prefix ejb1r1.mean ejb1r1.vo+orig

# Adjust each run to have baseline = 100
3dcalc -fscale \
-a ejb1r1.vo+orig \
-b ejb1r1.mean+orig \
-expr "(a/b * 100) * step(b-(2537))" \
-prefix ejb1r1.norm
end

# Create the HDF with the WAVER program:

waver -WAV -TR 4.0 -input /mnt/afnidisk/AFNI/t93_1111.1D \
> t93_1111_resp.1D

This is followed with the deconvolution routine. What am I missing?