Hi Pablo,
Although it is certainly possible to use subtractions to later study group and timepoint differences, as well as interactions, it becomes easily complicated for 3 timepoints (and 3 groups). I think for these cases, it is more straightforward to use the full design matrix, and deal with the repeated measurements using exchangeability blocks.
Please, see at the link below a mock-up for designs and contrasts (note that there are two sets) to test both within- and between-subject effects.
For design1, the hypotheses are for within-subject effects of time (C1 and C2), and interactions group vs. time (C3-C6). To run these in randomise, specify a file with the exchangeability blocks (EBs, with the option "-e"), so that permutations will happen within subject. This will assume compound symmetry, that is, the correlations between t1 and t2, t2 and t3, and t1 and t3 are all the same.
For design2, the hypotheses are for between-subject effects, i.e., group differences. Use the same EB file, but in randomise, include the option "--permuteBlocks", so that permutations will happen of these blocks as a whole.
The above is using randomise. If you'd like to correct for all these contrasts, it's possible to use PALM: you'd enter both designs (with "-d"), both contrast files ("-t"), both EB files ("-eb"), the options "-within" and "-whole", so that permutations will happen within-block and also whole-block, and "-corrcon", to correct across all contrasts (in these cases, not all permutations serve to test both designs simultaneously, but this is fine).
Hope this helps.
All the best,
Anderson
