Hi
>
> 1) Say for subject1 I have run1 and run2: If I have a COPE value for the contrast A-B and I'd like to combine the two runs to create a single estimate for contrast A-B for subject1, can I just compute a new t-statistic by averaging across the two COPE and VARCOPE values and use the following formula: COPEaveraged/SQRT(VARCOPEaveraged)?
>
Yes, or simply concatenate within-subject and fit a single glm against the stacked design. That's a simple fixed-effects combination then
> 2) At the higher-level, I also have two experimental groups in the dataset. If I can accurately compute the t-stats in step 1, could I then test for group differences using an independent samples t-test across these new t-values?
Yes
>
> 3) Is there a better alternative for assessing how components vary by task conditions when the subject timing files differ (e.g., when I'm unable to use tensor ICA)?
>
Nope, not really. I think what you have is a pragmatic way of achieving the higher-level comparison so personally I'd go with this...
hth
Christian
|