Hello,
What are my options when doing twin modeling on a multinomial variable (e.g. what is your favorite wine?: 1) red; 2) white; 3) rosé; 4) sparkling), and there is no hierarchical structure between the categories? That is, how do I best use the cross-twin information that co-twins of sparkling wine drinkers prefer rosé over red wine?
Dummy-coding is the only option I can think of, though that comes with some inherent negative covariances between the dummy-coded variables.
I seem to remember some of Michael Browne's work on structural models for preference data, where people are given a list of options and are asked to put them in order. I'll think more about that method and see if I can remember what it's called/give you a reference. As multinomial data can be seen as a censored version of that (what's your first choice? Oops, I forgot to give you the rest of the scale), it might be appropriate.
Hi Eivind
http://en.wikipedia.org/wiki/Cramér's_V
There's info on its standard error here too:
Liebetrau, Albert M. (1983). Measures of association. Newbury Park, CA: Sage Publications. Quantitative Applications in the Social Sciences Series No. 32.
Dummy variables, as Ryne says, might be exploited somehow. This slightly borders on the SEM analysis of ipsative data, which is no simple matter.