> why would that difference in code (group*age vs. group+age) cause the model with
> the glt to fail but not the model without, when the only difference really between what
> I'm getting out is that the basic model is giving magnitude of difference and the glt is
> giving me magnitude and direction?
The two operators, + and :, between two variables A and B are used in the following sense: A+B is the sum of the effects of A and B without interaction, while A:B is their interaction.
In your original model specification "group+age", no interaction was assumed. In that case, it's meaningless to compare the age effect between the two groups. More specifically, it would render a situation pretty much like 0 divided by 0. And that was why 3dMVM was unhappy.
On the other hand, with a full model of "group*age", you are considering both main effects and interaction. In other words, the difference of the age effect between the two groups is modeled, and it becomes natural to test that GLT.
Gang