You can still use the Matlab package GroupAna for group analysis with the number of fixed factors up to 4 and for unbalanced designs (unequal number of subjects across groups). You can also use 3dRegAna to handle missing data or designs with covariates. However if you don't have Matlab available or if your design falls beyond the capability of GroupAna/3dRegAna/Yourself, then this R package, 3dLME, might be the program to try your luck. Unlike Matlab, R is an open source platform with continuing support from a huge pool of statisticians and programmers. 3dLME adopts the linear mixed-effects modeling approach in R (similar to its counterparts in SAS, SPSS, and other statistical platforms) for FMRI group analysis.

Advantages:

a. Unbalanced designs

Like GroupAna, 3dLME can run analysis of unbalanced designs with unequal number of subjects across groups. However unlike GroupAna in which Type I sum of squares was adopted, the approach of linear mixed-effect modeling in 3dLME is quite different. And more options may be added in the future.

b. Missing data

Previously if a few beta values are missing from a subject, that subject may have to stay out of the whole group analysis. Now this is not an issue in 3dLME. Whatever partial information available can contribute to the analysis.

c. Unlimited number of factors and covariates

The only upper bound is the computer power and patience.

d. Bringing individual beta's instead of AUC to group analysis

If each regressor is modeled with multiple basis functions at individual subject analysis level, the traditional approach for group analysis is through AUC (Area Under the Curve). With 3dLME, it is possible to carry individual basis function coefficients to group analysis.

e. Model fine-tuning through various diagnostic methods

Some model diagnostic tools will be added to improve a model or compare multiple models. Even if your design can be dealt with 3dANOVA2/3dANOVA3/GroupAna, 3dLME has the flexibility to model heteroscedasticity across groups, or variance-covariance structures.

Bye, 3dRegAna! Well maybe not quite yet. If you can run a one-sample or two-sample t test with 3dttest without the covariate(s), 3dRegAna is still a handy program by following some examples. However anything beyond that could get very tedious and unwieldy with 3dRegAna: If you ever tried to create a design matrix for 3dRegAna modeling random effects of subjects, you should know what I mean!

Downsides:

a. High computation cost

Simple designs should be done within an hour, but sophisticated ones may take a couple of days or even more.

b. Sometimes difficult to compare the results with the traditional ANOVA

You don't really need to grasp anything about R except for a few simple commands unless you want to explore the R world further. Right now 3dLME only provides F values for main effects and interactions. Contrast testing (and runtime reduction) is in the pipeline.

See more here:
http://afni.nimh.nih.gov/sscc/gangc/lme.html

Gang