Louise Swift posted on Tuesday, September 04, 2012 - 4:14 am
When I fit a path analysis model i.e. all variables observed with a negative binomial dependent variable and using MLR ( I had difficulty using any other method as I wanted to estimate indirect effects), no chi squared orother fit index was produced. I note that the UG says on p533 that MLR produces ' chi -sq where applicable'. Am I therefore correct in thinking that it is not possible to produce fit indices when the DV is neg binomial? thanks/
Yes, there is no unrestricted H1 covariance matrix model with count outcomes, so no automatic overall test of model fit is available. What you can do is to run the model with all paths in the model and compare the loglikelihood of that model with that of the model you have. 2 times the difference is chi-square.