Hi

I have a dataset of twin pairs with many comparisons of interest. One of these is comparing between twins with active disease (in a flare) and twins with inactive disease (in remission).

I have about 4 twin pairs where one has active disease, and the other doesn't. I also have about 10 pairs where both are in remission and about 5 pairs where both have active disease.

I would like to include all the data ideally, since I'll have much less power if I only use the pairs that have one representative with each disease status. I know that duplicateCorrelation has been recommended for use with incompletely paired designs like this one. But I'm wondering if this type of design is too far beyond what it's designed for, since one might expect that the correlations between twins where both have the same status might be very different than the correlations between twins where they have different status. Or is the method robust enough to handle this kind of situation?

Thank you

You ask a good question. Twin pairs are generally a very powerful way to control for genetic and environmental differences. duplicateCorrelation will certainly do the right thing for you in a qualitative sense. It will use all the data while giving most weight to comparisons within twin pairs, because the within-twin comparisons are the most precise. The correlation within twin pairs should be pretty high, and you should check that.

If half or more of the twin pairs were discordant, I would probably recommend that you drop the other pairs so as to have a clear cut comparison using twin pairs only. That is not your situation though. Given that less than a quarter of the pairs in your data are discordant, I would generally recommend the duplicateCorrelation approach.

These are general comments only of course. I have not seen your data, so cannot advise an analysis specifically for it.