Hi -
I'd appreciate your opinion of the attached model that I would like to use with DTI data
and randomize. I searched through prior messages and didn't see anything quite like it.
Attached (PRB.zip and PRB_interaction.png) is the GLM tool output of a simplified version.
Essentially I have three groups (A, B and C), age as a covariate and the group*age
interaction terms. The age covariate is of limited interest - all other comparisons are of
interest. I used your "triple T-test" as a model.
EV1 = overall mean
EV2 = models A-B
EV3 = models A-C
EV4 = age
EV5 = A-B*age interaction
EV6 = A-C*age interaction
EV2-6 were orthogonalized wrt the overall mean (EV1) and thus demeaned
The contrasts are (as I understand):
C1 = where A > B
C2 = where A > C
C3 = where age correlates with FA
C4 = where the influence of age on the FA values differ between groups A and B
C5 = where the influence of age on the FA values differ between groups A and C
The F tests are (as I understand):
F1 = where there is any effect of group
F2 = where there is any effect of age
F3 = where there is any interaction
My primary question: Is this a conservative model and are the F test truly representing
what I stated above?
Thanks in advance.
Paul Borghesani
University of Washington, USA
|