I used both 3dANOVA2 and 3dttest to calculate t-stats for group averages and found that the t-stats were as much as 10x higher for anova2. This is a difference of about 10-3 in p-values.

For 3danova2, I used the mixed effects model with 2 different stimulus conditions as the fixed effect factor A, and 5 subjects as the random effect factor B. I think this is analogous to the exapmle shown on pg. 32 of the anova manual (4 different drugs and 20 different subjects).

For 3dttest, I ran 3dttest for each condition separately.

Looking at the source code for 3dANOVA2, it would seem that the higher stats are the result of dividing the stddev by sqrt(b * n) (number of subjects * number of total observations) instead of just sqrt(b)

First of all, is it valid to use 3dANOVA2 with mixed effects model for analyzing this type of experiment? Second, why should analyzing in this way produce so much more significant t-stats?