12-21-2016 02:47 PM
I am trying to estimate a bivariate probit model with random effects.
In my data, I have two binary outcomes (d and p) and only two continues predictors (x and y). d is a function of x and p is a function of y. I also have a variable called group that has two levels: one and two. the observations in each of the two groups are correlated.
If I ignore the group and correlation between the variables, I can use proc qlim as the following:
proc qlim data=test; model d = x; model p = y; endogenous d p ~ discrete; run;
But how can I model the random effects and consider for the fact that observations within the same groups are correlated?
12-27-2016 02:37 PM
You may wish to consider PROC GLIMMIX, with two dependent variables. Multiple dependent variables can be handled using the dist=byobs() syntax.
If you can recast your model as something like:
proc glimmix data=hernio_uv; class dist; model response(event='1') = dist dist*age dist*OKstatus / noint s dist=byobs(dist); run;
then you will be well on your way. For more info, check the GLIMMIX documentation, and in particular the section following DISTRIBUTION=keyword.