Dear Mark,
> I have 12 scans per subject and one confound variable per scan. There
> are 2 groups of subjects. I wanted to compare the 2 groups allowing
> the confound to be included in the analysis. I tried to run
> "multi-study: with replications and covariates" with 2 groups, 1
> condition per group, no covariate of interest, and one confound
> covariate. This didn't work, and produced lots of errors (log by
> zero). Obviously running "compare groups: 1 scan per subject" works
> but I can't enter the confound and it obviously massively decreases the
> degrees of freedom.
If the confound is the same for each subject (e.g. age) it is already
modeled implicitly in terms of subject effects. If it is scan-specific
(e.g. a VAS score after each scan) then the within subject variance of
this confound is orthogonal to the between subjects effect you are
trying to test for by comparing the groups. In this instance simply
take the average of the confounds for each subject and enter this as a
confound in the one-scan-per-subject analysis. The degrees of freedom
for the group comparison are necessarily much smaller than for
comparing condition effects using a fixed-effect analysis.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|