Dear Tom, dear Mark,
I do not fully understand your post on longitudinal designs for vertex analysis...
Could you help me with these questions?
> There perhaps is some confusion because there are 2 different ways you can construct a 2 sample t-test in randomise: You can model the mean difference with one
> column consisting of -1's and 1's, and then a contrast of c = [ 1 0 0 0 ...], or you can use two columns columns consisting of 1's & 0's (each marking the 'pre' and 'post'
> status of each observation) and a contrast of c = [-1 1 0 0 0 ... ]. You of course need to also include indicator 'dummy variable' predictors for each subject. (For the
> second option, if you include Nsubject dummy variables, you will have a rank-deficient design, but since the contrast is valid ('estimable') there won't be any problems.)
> The *crucial* thing with randomise and paired t-test, is that you must also supply the exchangeability block indicator file, which indicates which observations are paired
> together.
Let's assume I have 4 subjects at two different timepoints
My design matrix is:
group EV1 EV2 EV3 EV4
1 1 0 0 0
1 0 1 0 0
1 0 0 1 0
1 0 0 0 1
-1 1 0 0 0
-1 0 1 0 0
-1 0 0 1 0
-1 0 0 0 1
Contrast: c = [1 0 0 0 0]
- Do I need a F contrast for c = [1 0 0 0 ...] for this design?
- What do you mean exactly with "indicator 'dummy variable' predictors for each subject" and with "exchangeability block indicator file"? Is this the same? I assume, it refers to my EV1-EV4 - is that right?
Thanks you very much!
Viola
|