I'm not 100% certain on the implementation of 3dMVM.

But the short answer for statistics of whether to use a series of t-tests vs. ANOVA framework comes down to "it depends" (Maxwell & Delaney, 2004, p. 165). If you are primarily interested in simple paired contrasts between groups, and feel confident that the assumption of homogeneity-of-variance is true (all three groups - A,B,C - have equal population variances), then the advantage to using an ANOVA framework is that you get a better estimate of the total variance that you can test against. The other major reason to use an ANOVA framework in this case is that you can make complex comparisons (e.g., -2 +1 +1). That said, most of my comparisons show almost identical results between 3dttest++ and 3dANOVAx/3dMVM. It's also noted in the help for 3dttest++ that the default is to use "pooled variance". I see one major advantage of 3dttest++ is that the processing time is considerably faster, particularly if you want to look at the impact of several different covariates separately.

From a reviewer's standpoint, I think you could easily justify either approach. Gang also has a nice thread about this, I can never find the original link, but here's the repost.



Edited 1 time(s). Last edit at 06/20/2013 02:16PM by Peter Molfese.