Hello,
> I want to make a T-test contrast between two groups (patients and controls)
> covarying with 2 covariates. I include two covariates (age and IQ) into the
> analysis as confounding factors to ensure that any observed difference of
> group on the images was independent of age and IQ-related changes.
>
> If I want to study that Group 1 > Group 2, wich contrast would be better for
> me?
> group 1 group 2 covariate1 covariate 2
>
> a) 1 -1 0 0
> b) 1 -1 1 1
> c) 1 -1 -1 -1
You want (a) (contrasts b and c don't tell you anything useful at
all.). Including the covariates in the model is what controls for
potential confounding effects, not the actual contrast; i.e., by
having columns for covariate1 and covariate2, the parameter estimates
for group1 and group2 are reflecting information that can't be
uniquely explained by the confounds. But you still want to just test
the contrasts of these (unconfounded) group parameter estimates.
Hope this helps,
Jonathan
|