Hi. I’m wondering what the relative advanatages and disadvantages are for -cenmode ZERO vs. -cenmode KILL in 3dTproject. The default with 3dTproject is KILL, but when it’s called from afni_proc.py, it appears to default to ZERO. KILL makes more sense to me intuitively, but I assume there’s a good reason why afni_proc.py likes ZERO.
ZERO is used for consistency across programs. If the
residuals come from either 3dDeconvolve or 3dREMLfit,
they will behave as with “3dTproject -cenmode ZERO”,
giving zero volumes when censoring occurred. Without
the zeros, a time series would not match the original
one in length, which is probably why 3dDeconvolve and
3dREMLfit were written that way.
Also, the zero volumes do not affect computation of a
subject correlation volume (since the errts has a mean
Potential downsides of the zeros are when running
other types of analyses, such as a subsequent ICA
or something. In those cases, the zeros should be
Note the examples (in the proc script) of removing
the zero volumes based on running:
1d_tool.py -infile X.xmat.1D -show_trs_uncensored encoded ...
That is done for blur estimation, for example.
Does this seem reasonable?
That makes sense. I’m just doing correlations, so ZERO should be fine for me. Thanks.