

Estimation of logistic regression in LCA 

Message/Author 


I am using Mplus to estimate a logistic regression through LCA. The user guide indicates that the Maximum likelihood estimation with EM is used. However, I would like to know more. Is MMLE used with EM or other form of MLE? What optimization method does Mplus use in the Mstep? Are there other options available to estimate the model? I am writing a paper promoting the use of Mplus, and all these details have to be clarified. Also, is there any reference comparing the LCA in Mplus and MCMC? Thanks. Any comments are welcome! 


I think by MMLE you mean marginal maximumlikelihood estimation, and if so, yes, this is done in Mplus. We also use pseudolikelihood estimation when there are complex survey data features  see our special website section for that: http://www.statmodel.com/resrchpap.shtml especially the paper: Asparouhov, T. (2005). Sampling weights in latent variable modeling. Structural Equation Modeling, 12, 411434. For a description of algorithms used with mixture (LCA) modeling using ML, see page 615 of our Version 7 User's Guide which is available on our website. As shown in the UG examples, Mplus also does 2level LCA. You can also estimate the LCA model using Bayesian analysis. We have not written about ML vs Bayes comparisons for LCA. We do, however, have a paper on a specific use of Bayes for a generalized LCA: Asparouhov, T. & Muthén, B. (2011). Using Bayesian priors for more flexible latent class analysis. Proceedings of the 2011 Joint Statistical Meetings. This is posted under Papers, Latent Class Analysis. 

ChiaYi Chiu posted on Thursday, September 19, 2013  9:46 pm



Thank you very much. The response is very helpful! Another question. What optimization method does Mplus use in the Mstep for estimating the logistic function? Thanks!! 


In a regular Mstep we try first NewtonRaphson and then use QuasiNewton as a backup in case NewtonRaphson fails. 

Back to top 

