Dear mailing list,
I am interested in assessing group differences (healthy controls versus patients) in terms of the correlation between a physiological variable (i.c. circulating testosterone level) and brain activation.
I've set up a 2 sample T test design, modeling the 2 groups (first 2 columns in the design matrix), and the covariate of interest (as 2 separate columns in the design matrix, consisting of the covariate values for each group, padded with zeros for the other group).
Now, I've looked at the interaction contrast, say, 0 0 1 -1, which presumable gives me the brain regions that show a positive correlation in group 1, and a negative correlation in group 2. Interesting. But as I understand it, it does not allow me to identify regions where say, group 1 has a positive correlation, and group 2 also has a positive correlation, but significantly reduced compared to group 1. Or vice versa for negative correlations in both groups.
Should I run an analysis with testosterone as a single covariate of interest (i.e. containing all the values for both groups), and the assess these contrasts: 1 -1 1 for group differences in positive correlations? and 1 -1 -1 for group differences in negative correlations?
Thanks for any assistance you can provide!
Ans
|