afni.nimh.nih.gov

> I have discovered that the results I get from running 3dttest versus contrasts
> in the anova are slightly different; the betas are the same but the t statistic is
> not (presumably because of the differences in the error term that goes into the design)?

Yes, that's exactly the difference. Basically GroupAna assumes the variability is the same across multiple conditions, thus is pooled together.

Gang