> Are you suggesting that you can skip an ANOVA and just do t-tests when you
> have two groups and multiple factors
Yes, you can.
> if you had three groups you would still recommend an ANOVA?
I don't think I said that. If you really want an F-test for the group main effect (whether there are differences across multiple groups), go with an ANOVA. Otherwise you don't have to stick with ANOVA since any comparisons with two groups can be obtained through 3dttest/3dttest++/3dMEMA. And 3dMEMA can analyze multiple two-group comparisons with one model. More specifically an F-test is indeed informative when there are more than two levels for a factor including the situation with more than two groups; however, such an F-stat is not something we have to have. To me pair-wise comparisons are more important and more specific.
Gang