Hi Ian,
I replied to that old post (a little late). Thanks for pointing it out.
It seems likely that Gang used 1s as basically the greatest common divisor of 3s (stim) and 2s (TR), so that the resulting oversampled timing would indeed be "TR-locked" at 1s. Personally, I like to upsample more than that, as it makes the deconvolution/reconvolution steps close to inverses of each other. And that comes with nice properties rather than the side-effects that come with a slow filter. There is a similar but still different PPI analysis at
AFNI_data6/FT_analysis/PPI. README.txt is actually the main script.
Does that seem reasonable?
- rick