Elande Baro posted on Saturday, July 12, 2014 - 3:13 pm
I was wondering if you are aware of any way to estimate a LCA model with less than 4 binary indicators(responses). Also, if you can point me to some reference paper that would be great. Ideally, I would like to be able to fit LCA with just one binary indicator, as I am working in a causal inference context where there is usually a single response; I know the identifiability conditions are not met in this case, but wondering if there is any way to go around that.
If you have covariates you can identify latent classes with at least the logit intercept varying. Topic 5 shows an example of that called "Trypanosomes" with a reference. It is harder to identify a class-varying slope.
In the case of a randomized study, there is a large literature on CACE mixture modeling with classes of shows and no-shows representing compliance classes. There are CACE examples in the UG.