I had a question about this too.
While I know that 3dAnova and 3dttest can accomodate different population sizes in each -dset, I'm looking for an honest way to present statistical maps to readers.
To date, I have had an equal sample size in each subject group such that the same t-value in a map of linear-contrast-elicted activations translates into the same "p" value for each of the two groups. Then, the report can show a color guide of "p" values (which I think are more conceptually interpretable to the reader) than T or F statistics.
If I have unequal sample sizes, would it thus be more honest- (i.e. present the reader with a more accurate comparison between the stat maps of the two patient groups) if I color coded the diagram based on actual voxel-wise t-statistics?
For that matter, how does a random-effect map of a single group calculated as difference from 0 in a one-sample t-test in 3dttest differ conceptually from the "mean" sub-brick of that same group in a 2dAnova of three patient groups? Would the t-stat subbrik of "mean" in 3dAnova be an apples-to-apples comparison between unequal sample sizes?
Jim B