06-21-2013 04:15 PM
Currently I am working on a random parameters bivariate ordered probit model in SAS and I used the following command to work on the fixed parameters:
proc qlim data=a method=qn;
model y1 = x1 x2;
model y2 = x1 x2;
endogenous y1 y2 ~ discrete(d=normal);
But I am not sure how to treat the independent variables as random parameters for this type of a model(for example, the MDC procedure in SAS allow you to do Mixed Logit)
I would appreciate if anyone could help.
06-24-2013 08:56 AM
I don't think there is a an available PROC that does this. When you say treat the independent variables as random parameters, I assume that one of x1 or x2 is such that it represents a sample that you want to use to infer to a broad inference space? Is that correct? If there is a correlation between y1 and y2, there might be a possible approach using PROC GLIMMIX. Check out Example 41.5 Joint Modeling of Binary and Count Data as a start, and then see if you can fit two different probits.
I really don't have much else.
06-24-2013 12:21 PM
Thank you for your reply. Actually the issue that I referred is the same as the Mixed Logit model (different from the regular Multinomial Logit Model), where we treat some of the parameters to vary across observations. I will check the GLIMMIX procedure as you mentioned.