> With age and sex as covariates, would
> Duration*Sex*Age or
> Duration+Sex+Age
> properly exclude sex and age from any GLT results?
The word 'exclude' is not really accurate even though it's very popular in the context. I prefer to describe it as to "properly control for ...".
Which model to use? That is very difficult to a priori decide unless you have some prior information. For example, do you believe that the two sexes have different Duration or Age effect? Practically you may try
Sex*Duration+Sex*Age
which allows you to test those two interaction effects.
Also, it is more computationally economical to just use 3dttest++ for your data structure. And even 3dMVM is a little better than 3dLME in this case.
Gang