Hi Dane,

I am a little unclear about what we should focus on, in
terms of what you view as unconventional. Is it that you
seem to be regressing only tasks of no interest, and not
those of interest?

The only thing that looks like a computational error here
is scaling after bandpassing (since you will probably have
zero-mean data at that point). But it does look like the
censor and bandpass steps are not being done well together,
either.

To bandpass and censor properly, using afni_proc.py might
be the way to go, since these steps do not look special
(except that scaling is being done across runs, whereas
afni_proc.py would do it per run).

Assuming the little "autocensoring" in step 1 is not
actually doing any censoring (since it is 3dDeconvolve),
then bandpassing is being done including censored TRs,
which is not good (the spikes will ring across time).

You mention 3dTproject too, so you seem to know what the
afni_proc.py steps are. It might be helpful to know why
you are currently choosing not to use afni_proc.py. What
are you trying to achieve that AP.py would not?

- rick