Hi Giuseppe,

The proper way to estimate the smoothness has always been from
the residual time series. Gang made a temporary suggestion of
using the stdev for the case that the residual time series data
was not available, but a better alternative is to measure it
from the pre-regression data, instead.

Keep in mind that the point is to measure the smoothness in the
data. That comes from 2 main sources: directly from the scanner
and whatever is added during processing.

Smoothness should have little to do with contrasts or t-stats.

---

Stick to the -errts dataset created by 3dDeconvolve (when using
that option). For example, review how afni_proc.py estimates it
in the "blur estimation" block after 3dDeconvolve in this sample
processing script:

[afni]
[afni]

That script will output a 'blur_est*.1D' file containing blur
estimates for that subject. Average them across subjects for a
robust blur measure that you can apply in every AlphaSim command.

- rick