Modeling cumulative count variables (0,1,2,3) in lavaan with WLSMV

[ahmad valikhani]

Hi,

I have two observed endogenous variables in my model—a mediator and an outcome—both measured as cumulative counts (e.g., number of medical conditions: 0, 1, 2, 3). Is it appropriate to model these variables in lavaan using the WLSMV estimator by specifying them in the ordered argument, or should they instead be treated as count variables following a Poisson distribution, which lavaan cannot directly estimate?

Best,

A

[Phil Wood]

It depends a bit on whether the variable is dispersed as a count variable. Not all counts are Poisson. For example, if you are measuring the number of drinks that someone reports consuming, it's very rare that people have only one drink. They usually either report no drinks, two drinks, three drinks and so forth. If that's the case, you're better off modeling it as an ordered category. FWIW. Phil