10-04-2011 12:05 PM
I need to fit a discrete choice model to a set of data in which individuals pick from one of four alternatives. I have data both on the individuals and the alternatives. To relax the independence from irrelevant alteratives (IIA) assumption, I will probably need to fit a nested logit or mixed logit model. Based on my research of SAS, I understand that I need to use PROC MDC for this.
Also, my data is from a national survey with a complex design. My understanding is that PROC SURVEYLOGISTIC is the way to go when analyzing survey data to get the proper standard errors (I have the replicate weights). However, it appears that SURVEYLOGISTIC does not handle the fitting of discrete choice models. In addition, it appears that PROC MDC cannot handle weighted analysis, let alone using replicate weights to get the proper standard error estimates. So, I'm confused as to which way to go from here. I'd appreciate any thoughts on whether my assumptions regarding the limitations of the two procedures are correct and, if so, any ideas on which direction I should pursue.