Dear André,
The contrasts [1 -1] or [-1 1] within the two-sample t-test models always test for differences between groups with regard to the forwarded con images, which might, depending on their creation, reflect the difference with regard to another difference (T1 - T2), thus a group x time interaction, or possibly, a three-way interaction or beyond (in case the con images on single-subject level test for e.g. (T1 Cond1 - T1 Cond2) - (T2 Cond1 - T2 Cond2) ), in any case always an interaction with group.
In contrast, depending on forwarded con images [0.5 0.5] or [-0.5 -0.5] might reflect main effect time (in case the con images from single-subject level are T1 - T2), or e.g. the interaction time x condition (in case the con images on single-subject level test for (T1 Cond1 - T1 Cond2) - (T2 Cond1 - T2 Cond2) ), but always averaged across groups.
For effects like G1T1, G1T2 and G2T1 and G2T2 you should set up the corresponding contrast vectors on single-subject level and forward this into one-sample t-tests. You could also go with contrast vectors [0.5 0] or [0 0.5] within the two-sample t-tests, but you would have too many df then due to the other subjects included in the model, so it's not the correct statistics.
Best
Helmut
|