Dear Alfonso,
I conducted an FC-MVPA, trying to identify predictors of treatment
response. I would like to implement cross-validation (using
different clusters) but I'm a bit lost.
1-Using spm_crossvalidation, I get results for each component (10)
and clusters: is there a recommended way to simplify or aggregate
the results across components?
2- What is the next step? I think I should run a logistic
regression (I am trying to classify response vs nonresponse), look
for corrected permutation p values, etc. but I'm wondering if you
have any guidelines on how to do this.
3- I read I should conduct it both on the MVPA results as well as
in the SBC (post-hoc) results. Should I conduct two different
cross-validation approaches? (one from MVPA and one for SBC
results?)
Thank you!
Sorry Alfonso, I know another solution could be to use the
calculator (i thought about running spm_crosscorrelation and
importing those values when looking at my post-hoc analyses).
However, that way I am unsure how to evaluate accuracy when
predicting group allocation.
Hope this makes sense,
Thank you!