 |
 |
Dear Yolanda, the interaction contrast vector would be formally correct, but I'm not sure it's statistically valid. Donald McLaren's going to tell you probably, he typically answers messages like yours :-) The other contrasts are definitely incorrect, as [x x 1 -1 x x ...] means you're contrasting condition 1 with condition 2 averaged across time points. However, as the first two contrasts are no proper ANOVA contrasts anyway (instead of effect "session", effect "stimulus" and interaction), I'd say build the three contrasts on single-subject level and then run three separate one-sample t-tests on group level. Alternatively, you could go with GLM flex, which should give you statistical valid contrasts for "session", "stimulus" and the two-way interaction within a single model. In case of factors with two levels (like session and stimulus) the output is going to be a T statistic derived from the F statistic. As far as I could see it's absolutely equal to results from the one-sample t-test approach. Hope this helps for the moment, Helmut