Dear Helmut,
Thanks a lot again, it's clear now. However, what do you mean with contrast at the group level are F contrasts? Can I not use two t contrasts, 1 -1 and -1 1? Finally I need to know the direction of the effect, which group has higher/lower activity?
Cheers
André
> Am 22.09.2015 um 18:41 schrieb H. Nebl <[log in to unmask]>:
>
> Dear André,
>
> Your contrast vector is meaningful = changes over time, but you can already look at the main effect "Time" within the Flexible factorial based on the two con images reflecting the contrast for session (time) 1 and session (time) 2. The interaction "Group x Time" within the Flexible factorial is also valid.
>
> In contrast, the main effect "Group" within the Flexible factorial is invalid due to error term issues. That's why a second model is needed. To test for main effect "Group" you have to set up a contrast that averages across time [1/8 1/8 -1/8 -1/8 ... 1/8 1/8 -1/8 -1/8 ... 1/8 1/8 -1/8 -1/8 ... 1/8 1/8 -1/8 -1/8 ...]. The corresponding con images from group 1 go as "Group 1 scans" in the two-sample t-test, same for group 2. Within the two-sample t-test go with [1 -1].
>
> Instead of this approach with two different models (Flexible for main effect "Time" and interaction "Group x Time"; Two-sample t-test for main effect "Group") you could also go with the following:
>
> 1) Two-sample t-test based on [1/8 1/8 -1/8 -1/8 ... 1/8 1/8 -1/8 -1/8 ... 1/8 1/8 -1/8 -1/8 ... 1/8 1/8 -1/8 -1/8 ...] to test for group effects via [1 -1]; BTW to test for average "condition" effect go with [0.5 0.5]
> 2) Two-sample t-test based on [1/4 1/4 -1/4 -1/4 ... 1/4 1/4 -1/4 -1/4 ... -1/4 -1/4 1/4 1/4 ... -1/4 -1/4 1/4 1/4 ...] to test for time effects via [0.5 0.5]; to test for interactions go with [1 -1] within the very same model.
>
> Actually this might be preferable, as this way, you don't have to wonder whether variance is equal or unequal for "Time". Accordingly, there should be some differences between the "two two-sample t-test" approach and the "one flexible factorial and one two-sample t-test" approach, although hopefully no relevant ones.
>
> Just to make sure, contrasts on single-subject level are T contrasts, contrasts on group level are F contrasts. If you want to go with two directed T contrasts instead you have to test "into the other direction" as well, thought it might be bette rto keep in view when staying with the F contrasts.
>
> Best
>
> Helmut
|