Hi Janina,
For example, note that applying bandpassing between
frequencies 0.01 and 0.1 on a 200 TR time series with
TR = 2s would cost approximately 120 degrees of
freedom. It would then be like starting with only 80
(before censoring, motion, polort, or anything else).
Compare 80 with 3 DoF for the 3 principle components.
That is relatively free. The single mean CSF would
be just one DoF.
Keep in mind, 1 regressor = 1 Dof.
Censoring depends strongly on the type of subjects
that one has. I think you mentioned having higher
motion subjects. If that is the case, you almost
certainly need to use a higher censor limit than 0.3,
or too many subjects would be lost.
If you already have motion files for many subjects,
consider Example 18 from "1d_tool.py -help". That
shows how you could evaluate the effects of various
censor levels (e.g. 0.3, 0.5, 0.75, 1).
Sure using 25% censoring as a cutoff for dropping
subjects seems fine.
Regarding use of WMe along with WMeLocal, I agree
with you. That addition to the example was really
showing the sorts of options one has. Though some
people have used it (also part of why I included
it in the example). In any case, that regressor
is probably not going to have a big effect either
way.
- rick