Hi Matt,
I've found the bug, but I'll need some time pondering the solution.
> This seems to be working by using the groups as a covariate. In this situation,
> I would use -set instead of -groups, yes?
That's right. You pretend there is only one group of subjects.
> Also, how would I model the interaction? I had originally coded gender as 0 or 1
> (same with group). In order to do what you say, would I need to code it as 1 and 2
> or just code the interaction (for instance, SCZ males and CON females as one
> number and SCZ females and CON males as another)?
There are many (actually infinite) ways to code the factors, but 0/1 or -1/1 are the popular ones. With 0/1 coding, read the section How to use 3dMEMA to handle multiple groups on webpage [
afni.nimh.nih.gov]
Actually if you just want to control for the sex effect, you may want to code the sexes with -1/1 coding (without centering).
> I thought determining the mean of gender within the group was necessary.
> The value above 0.5 means that there are slightly more males in each group.
That would lead to interpreting the intercept as the average of all subjects, not really the group average, which might be fine for you. However, the intended difference for the two sexes would not be the real sex difference, contradicting the purpose of obtaining the contrast between the two sexes.
> If I didn't want to center the covariates, I would use -covariates_center gender = 0, correct?
That's right.
Gang
Edited 2 time(s). Last edit at 09/11/2013 09:42AM by Gang.