03-07-2014 12:45 PM
I'm hoping someone can suggest procedures for running a Probit model using panel data (aka longitudinal data, cross-sectional time series data, etc.). From what I understand, PROC PROBIT doesn't handle a random effects necessary for dealing with the correlated data across time and cross section. And PROC PANEL doesn't have options for binary choice dependent variables. I've looked into using PROC QLIM, PROC NLMIXED and PROC NLIN, but I figured I would ask the denizens of the SAS forums what procedures you would recommend to address this problem.
Thanks in advance!
03-10-2014 03:04 PM
If you are confident in programming likelihood functions, then NLMIXED is a strong possibility.
However, you can use PROC GLIMMIX with joint distributions for the variables in the panel (I assume there is more than one dependent variable). As an example, check out Example 41.5 Joint Modeling of Binary and Count Data in the SAS/STAT12.3 documentation for PROC GLIMMIX. Although the dependent variables here are from distinctly different distribution families, the method could be generalized.