Mai Sherif posted on Monday, March 04, 2013 - 10:03 am
I am running an example similar to the one in Muthen and Masyn's (2005) paper: Discrete-time Survival Mixture Analysis, where I have both categorical items (y's) and binary survival indicators (u's). I am using WLSMV rather than ML for estimation though as I have a large number of latent variables/random effects. My question is: in that case how are the correlations between the binary survival items and the categorical items calculated? Is it using an underlying variable approach for both types of variables? If that is the case...does that mean that a probit rather than a logit is assumed for the hazard function?
WLSMV cannot be used with discrete-time survival analysis. ML is needed to give correct estimates.
Mai Sherif posted on Monday, March 04, 2013 - 7:09 pm
Thanks for your answer.
So in general, how are correlations calculated between binary and categorical items using WLSMV (not in a discrete-time survival analysis)? Is a continuous underlying variable assumed for both types of items?
Mai Sherif posted on Tuesday, March 05, 2013 - 1:10 pm
Can WLSMV be used if the model is "general" rather than "mixture"?