I conducted a LPA with five continuous indicator variables. I am happy with my results and now would like to analyze the relationship between my latent class variable and auxiliary variables, three of which are counts (e.g., count of how many drinks consumed in a week). I was planning to use the 1-step (pseudo class) method because my entropy is very high (.98). What I am wondering is…
Can I use this auxiliary approach with count variables? If so, how do I specify what variables are counts?
Alternatively, can I use Poisson or negative binomial, within auxiliary statements?
Finally, should I transform these counts (square or log) instead and use them in the aux?
If you have a different approach, I am open to your suggestions and help.
If the counts are seen as outcomes, that is, predicted by the latent class variable, you define them as count variables and use any of the various models for counts. If the counts are seen as predictors of the latent class variable you may want to take the log of them.