[R-sig-ME] Estimate correlated non-nested RE?
Karl-Oskar Lindgren
Karl-Oskar.Lindgren at statsvet.uu.se
Tue Feb 20 06:00:37 CET 2007
Dear listusers,
I have a newbie question on lmer. Currently I'm working on a
networklike problem involving dyadic data. I would like to estimate a
crossed-random effects model in which I take account of the fact that
observations are most likely correlated both across rows and columns
of my dataset (the rows are elite representatives and the columns are
ordinary citizens, and the dependent variable is elite-mass policy
congruence). If I specify my model like this:
m2<-lmer(y~1+x1+x2+(1|column_id)+(1|row_id), family=binomial(link="probit"))
(where x1 and x2 are the fixed effects of interest) I get uncorrelated
crossed-random effects. But I would want to relax the assumption that
the unobserved heterogeneity across rows and columns are uncorrelated.
How do I do that? I guess this is really simple but I have failed to
accomplish this. All examples that I have come across on the web
either have correlation between slopes and intercepts or involves
nested random effects. But I would like to estimate a correlation
coefficient between two non-nested random effects. How do I do that?
Thanks in advance (and sorry if the answer is readily available in
some document that I have missed)
Best
Karl
More information about the R-sig-mixed-models
mailing list