

LCA, count variables, overdispersion 

Message/Author 

John G. Orme posted on Thursday, December 15, 2011  10:25 am



I want to estimate a latent class model with count variables and for each variable the variance is much larger than the mean. I don’t think that the overdispersion is due to an excess of zeros. So, it doesn’t seem that the Poisson model available in Mplus would be appropriate. What would the best option be in this situation? Thanks for any help you can give me with this, and thank you for your answer to my earlier question concerning relaxation of the local independence assumption with count variables. 


I think a mixture Poisson is perfect for such a situation  and that is what you are doing when doing LCA and specifying the DVs as counts. 


I just wanted to check something regarding the output with a zero inflated Poisson mixture model. The means in the output are log means and would need to be exponentiated in order to interpret them as means. This would imply that extreme values like 15 would indicate a mean of 0. Is this correct? Thanks 


Yes. 

Back to top 

