Thanks Gang,

I tried using 3dDeconvolve with -stim_times_AM2 and a trial type measure (I have individual trial information, so I will make that change later). The code is:

#!/bin/bash
# do amplitude modulated linear regression w/ tremor info as ABI

s=04
for t in posture; do # action posture; do
1dcat ${s}-${t}_pre_csf.1D ${s}-${t}_post_csf.1D > ${s}-${t}_csf.1D
3dDeconvolve -input ${s}-${t}_pre_scl+orig ${s}-${t}_post_scl+orig \
-jobs 3 -GOFORIT 10 -polort 1 -num_stimts 1 \
-stim_times_AM2 1 ${s}-posture_am2.1D 'WAV(1)' \
-xout -fout -tout -full_first \
-fitts ${s}-${t}_fitts -x1D ${s}-${t}_x1D.1D \
-bucket ${s}-${t}_bkt -cbucket ${s}-${t}_cbkt
done

and the 04-posture_am2.1D file has:
10*78.5:10 30*78.5:10 50*78.5:10 70*78.5:10 90*78.5:10 110*78.5:10 130*78.5:10 160*23.0:10 180*23.0:10 200*23.0:10 220*23.0:10 240*23.0:10 260*23.0:10 280*23.0:10

The measure before treatment was 78.5 and after treatment was 23.0, and the block design was:
tr block
--------
10 rest
10 task
10 rest
10 task
10 rest
10 task
10 rest
10 task
10 rest
10 task
10 rest
10 task
10 rest
10 task
10 rest

so the am2.1D file would look like two block design time series, where the blocks in the first half of the time series have a higher amplitude than those in the second half. Is there anything that you see wrong off the bat with the script? Thank you.