Sorry for being slow on this...
Yes, 3 seconds is a bit short for a GAM function. You might get a better feel for some of the basis functions using "timing_tool.py -help_basis", with help for plotting them being shown at the end.
The BLOCK function seems to be a good model, so you can compare how it would model the response to a 3 second event with a GAM function. You can also convolve GAM with a 3 second boxcar. For example, plot GAM, a 3-second GAM, a 1-second BLOCK and a 3-second BLOCK as follows:
3dDeconvolve -nodata 200 0.1 -polort -1 -num_stimts 4 \
-stim_times 1 '1D:0' GAM \
-stim_times 2 '1D:0' 'GAM(8.6,0.547,3)' \
-stim_times 3 '1D:0' 'BLOCK(1,1)' \
-stim_times 4 '1D:0' 'BLOCK(3,1)' \
-x1D X.xmat.1D -x1D_stop
1dplot -one X.xmat.1D
or
1dplot X.xmat.1D
The short functions, GAM (black) and BLOCK(1,1) (green) peak at a similar time, though BLOCK(1,1) is wider. The 3-second GAM (red) peaks about 3 seconds later than GAM, while BLOCK(3,1) (blue) peaks about 2 seconds later than GAM.
We would tend to suggest using the blue BLOCK(3,1). This should help clarify the differences.
- rick