condition contrast at first level or second level?


I have one statistic question:
Say I have a within-subject repeated measure design, with two conditions A and V, I want to test if there are any regions more involved in A comparing to it in V.

Two approaches in my mind:

  1. each condition has one regressor in the first level GLM matrix, resulting in beta for A and beta for V, for group level analysis, do paired t-test to test the difference of beta-A against beta-V. (i.e. the example script s5.ttest.paired in AFNI_data6/group_results)

  2. add a -gltsym ‘SYM: A -V’ -glt_label 1 A-V, to test the difference of A against B at subject level, resulting one beta indicating the difference, then do one-sample t-test to test this difference beta value against zero.

Could you share your mind on this topic, which approach is more recommended, or it depends on different conditions?

Many thanks!


Meng, the two approaches should end up with roughly the same results. So, either one is fine.