> Gang, what do you mean by the 'effect of covariate'?
What I mean by covariate effect is the parameter 'b' (or slope) in the model
y = a + b * x + e
And it indicates the amount of response (y) per unit of the covariate (x).
> As far as I understood, the sub-brick with the covariate shows results with a covaried variable controlled for.
I'm not so sure what you mean by that. If you're looking for the group average effect while controlling the covariate at a specific value (e.g., mean), then that would be the parameter 'a' (intercept or constant) *if* you center the covariate properly.
> So, for example, if I want to compare two groups with and without depression,
> the first two sub-bricks would tell me the difference between the two groups
> and the other two sub-bricks would shown me the results if I take the impact
> of depression out... Is that correct?
I've only focused on the discussion so far for the case with only *one* group of subjects. With two or more groups, the situation is slightly more complicated because of the centering issue and the complication of whether the two groups have the same or different covariate effect.
There is a brief coverage of the subtle issues in the group analysis part for the AFNI workshop:
[
afni.nimh.nih.gov]
When a covariate is modeled, I don't think that it's an accurate description to state that one can "take the impact of the covariate out" even though such a statement seems to be prevalent. More specifically, there are two goals in covariate modeling: 1) It is not to take out the covariate effect, but instead to control for the covariate variability and interpret the group effect at a *specific* covariate value. 2) Sometimes one is interested in the covariate effect itself.
> since Jessica is talking about connectivity, the output is interpreted differently?
Connectivity measure or brain response, that is just the variable y in the model, and the subtleties remain the same.
Gang
Edited 1 time(s). Last edit at 05/24/2013 02:07PM by Gang.