Dear TBSS experts,
let's say I have two factors, AB (with levels A and B) and XY (with levels X and Y). Everything is fine if I only want to compare A with B or X with Y or if I use the full two-factor design: I simply copy all FA datasets in a directory, run the TBSS scripts and then apply randomise with an appropriate design.mat.
However, what if I want to compare subsets of these data only and omit the rest? For example, I have a sample of patients and matched healthy controls. The patients differ in respect to a particular feature, and I'd like to compare these two subgroups. In more abstract terms, I only want to compare levels A and B for those datasets in which the level of XY is X and not Y.
As far as I can tell, there's no way to tell randomise to omit cases, because it expects a design.mat of the length of all datasets present in the given directory. Or did I miss something in the documentation?
So what I have to do is create a new directory, copy only those FA datasets which should be included in the particular analysis (that is, those in which the level of XY is X and not Y) to this directory, run the TBSS scripts there again, and apply randomise with another design.mat constructed for this purpose.
I have two concerns regarding this approach:
1. It can become a bit tedious, especially if there are more factors or factor levels and thus more design cells for which separate comparisons are potentially to be conducted. For each of these, a separate directory and a fresh run of the TBSS scripts is needed. But I guess this can't be helped?
2. All of the individual analyses will be based on separate subsets of the data. This also means that the mean skeleton will differ, making it harder to compare the results from the different analyses. So I had the idea of using the skeleton from the full-sample TBSS in the randomise runs for the individual subset comparisons. In your opinion, would this be reasonable and methodologically sound?
Thanks in advance,