Apologies if these questions are addressed in the help files - I've spent quite a bit of time looking through them but I'm still not confident in the correct way to proceed.

I have a task with four major events (cue, anticipation, stimulation, and rest) in which I will be entering regressors for cue (always 3s), anticipation (variable between 3s, 6s, 9s, and 12s), and stimulation (always 6s), letting rest be modeled as baseline.

I had been using the following lines in my afni_procy.py script, but I suspect this may be incorrect:

-regress_basis GAM \
-regress_stim_times ${regressors} \
-regress_stim_types AM1 \
-regress_stim_labels ${reglabels} \

My regressor files were each a single column of times onset:offset.

A few major questions:

Most examples with regressor files that take into account duration are formatted as a single column of times onset:duration - should I swtich to durations? Why didn't 3dDeconvolve fail with the offsets if it was reading them as durations since this would have exceeded the length of the runs?

Should I use -regress_basis 'dmBLOCK' instead of GAM? How will this change my results?

Is the combination of dmBLOCK and AM1 the most appropriate method for modeling events of variable duration without amplitude modulation? With regard to another task I am analyzing, what about when the durations are fixed?

Thank you very much for your time and assistance.

Sincerely,
Kristin