> could you please explain the reason why when run the afni ANOVA
> we need same group sizes? Is it a mathematical/statistical reason?
Yes, both 3dANOVA2 and 3dANOVA3 use the traditional approach to assessing the contributions from various factors and their interactions, which unfortunately requires equal number of subjects across groups.
For experiments with unequal number of subjects across groups, use R program 3dLME:
http://afni.nimh.nih.gov/sscc/gangc/lme.html
or Matlab package GroupAna:
http://afni.nimh.nih.gov/sscc/gangc/Unbalanced.html/
Gang