Hi Tom and Lukas,
OK, see if I can get this straight.
Since you want a quick test without worrying any potential statistical 'penality', you'd probably have to comvert the two t subbriks into z (normal distribution) by running 3dmerge with its -1zscore option.
Because the information about standard deviation is not stored in the ANFI dataset, we have to do a little calculation:
Z_1 = M_1 / S_1
Z_2 = M_2 / S_2
where Z_1 and Z_2 are the coverted z scores of the two sets, M_1 and M_2 group means, S_1 and S_2 standard deviations. Among these variables, only S_1 and S_2 are unknown.
Now we want to get (M_1 - M_2) and the corresponding z score.
3dcalc can generate (M_1 - M_2), and its standard deviation can be estimated as sqrt(S_1^2 + S_2^2). Therefore its z score would be
Z = (M_1 - M_2) / sqrt(S_1^2 + S_2^2)
= (M_1 - M_2) / sqrt((M_1/Z_1)^2 + (M_2/Z_2)^2)
Then run 3dcalc again with the above expression.
Is this something you want?
I am not so sure about your scenario, but the more accurate and more legitimate way is to obtain the comparison test while you run group analysis such as 3dttest or ANOVA or 3dDeconvolve.
Gang