Hello,
I have multi-category choice data that involve several dimensions of complexity. The data are a panel (varying both cross-sectionally and across multiple time periods), and there are frequently multiple choices made per chooser in a single time period. The possible responses are a set of seven unordered categories.
I have been using PROC LOGISTIC, and link=glogit to estimate a multinomial logit model, and including a time identifier directly as a variable in the regression. The latter is a cheap remedy to account for the time-series element of the dataset, but no account is made in this context for the non-independence of certain observations in the same time-period. Ideally, I think, I would use a panel specification, and cluster on the chooser ID (is this right?). However, I have not come across a way to define strata for an unordered multinomial logit model, nor have I seen an explicit way to distinguish the dataset as a panel in SAS.
Does anyone have insight into either of these issues?
Many thanks,
R.