Hi FSL Experts,
I had a question about using the GLM gui with a set of our TBSS data. I have one group of 70 subjects that I've taken through the TBSS processing stream. One major question of interest that I have is testing a possible interaction effect between sleep efficacy (i.e. sleep quantity expressed in minutes) and age on white matter integrity (i.e. FA or whichever DTI measure).
So I was wondering how can I model a contrast with the GLM gui on FSL. I previously read a post where the following design has been suggested (considering an interaction effect between memory performance and age):
EV1: intercept
EV2: age (continuous values, all subjects, mean-centered).
EV3: memory (continuous values, all subjects, mean-centered)
EV4: product EV2*EV3.
The interaction is tested with:
C1: [0 0 0 1]
C2: [0 0 0 -1]
This is a good design, but in my opinion it still presents a problem. How can I account for the opposite effect the variable age and the variable sleep quantity have on white matter, by simply modeling their interaction as a product? I mean, I ended up having the same product with completely different meanings (e.g. 2 hours of sleep x 80 years =160 and 8 hours of sleep x 20 years=160). Even if the variables are mean-centered the problem persists. So, if either C1 or C2 are significant, they are ambiguous to interpret.
How could I create my interaction design with factors that have opposite effects on white matter?
Thank you in advance for your help!
Elisa melloni
|