If by cluster-specific hierarchical modeling you mean multilevel modeling with random effects, then yes. GEE is not possible in Mplus, but for a comparison of GEE and the Mplus weighted least squares estimator for categorical outcomes, please send me an email requesting "paper #75".
Emily Blood posted on Saturday, September 13, 2008 - 2:35 pm
I have a two questions: 1. If I have a latent growth curve with a binary outcome, do you recommend the WLSMV estimator or the MLR estimator? Also, I assume the default is a logit link. If not, how do I specify a logit link?
2. If I have a time-varying covariate/predictor and repeated binary outcome, is it possible to model this with a logit model in Mplus? If so, can this also be done if there is a mediating variable included between the time-varying predictor and the binary outcomes?
1. You can use either estimator. If you want to include residual covariances in the model, it is more straighforward with weighted least squares. With maximum likelihood, each growth factor and each residual covariance is one dimension of integration so a model can quickly become computationally demanding.