Jan Ivanouw posted on Monday, September 14, 2009 - 3:26 am
I have heard and read the presentations for topic 2 and also the web talk, and I have two questions regarding the (always) zero-class in zero-inflated Poisson variables during CFA
1) What is ‘means’ in the output. 2) How does the categorical part of the output variables (always zero vs. the Poisson variable) contribute to the latent variable in CFA? Is it treated as an dependent bivariate categorical output variable contributing to the latent variable (and in that case how do I get the loadings in the Mplus output?)? Or does it ‘delete’ the cases belonging to the zero-class for the purpose of estimations for the CFA? Or?
Any litterature references for the treatment of the zero-class in CFA?
1) "means" refers to any location parameter. For ZIP it could be the inflation logit when there is no covariate for it. See UG ex3.8 for an explanation.
2) Mplus does not default to also regressing the inflation part on the factor. This is in line with how Mplus operates for regression. Instead the inflation is a constant (a location parameter). But a person could add that regression, in line with how it is done for ex3.8. So such loadings would come from ON statements such as u# ON f. See also ZIP growth modeling where the inflation part can be modeled as a growth model - or not.