Hi,
> Regarding my question #2 (If I understand you correctly, evaluating the change between time points would be an unpaired t-test once again where the design matrix would have the same number of rows as subjects. If I wanted to use include nuisance covariates (age, sex, etc) would I use the value from the first time point? I guess this question mainly concerns age...), I should have said:
>
> If I use the difference image of the two time points, then the design matrix I'd use would look LIKE a design matrix used in an unpaired t-test where there would be a column for each group (1 for membership in the group), a column for each covariate, and a row for each subject. I also wouldn't need to include a .grp file because it would be all 1s any how. As for the contrasts, I would use [1 -1 0 0...] to test if the change in the given DTI scalar value for controls is greater than the change for the group for column 2. Am I on the right track?
Yes, that's right - although as Anderson said, if your time intervals are very different then you have exchangeability issues within randomise. If you can select a subset where the intervals were al very close to one year, then that would be fine.
All the best,
Mark
|