Seminar
Bengt Muthen
A Simple Estimation Procedure for Multilevel Models with Categorical Outcomes and Many Latent Variables
Multilevel analysis often leads to modeling with multiple latent variables (random effects, factors)
on several levels. While this is less of a problem with Gaussian observed variables, maximum-likelihood (ML)
estimation with categorical outcomes presents computational problems due to numerical integration with many
dimensions. We describe a new method which compared to ML is both computationally efficient and has similar
MSE. The method is an extension of the Muthen (1984) weighted least squares (WLS) estimation method to
handle multilevel multivariate latent variable models for any combination of categorical, censored, and
normal observed variables. Using the Mplus program, we compare MSE and computational time for the ML and
WLS estimators in a simulation study and present multilevel analyses of mental health data. Examples
include random effects modeling of categorical longitudinal data in cluster samples, two-level factor
analysis, and two-level logistic regression with latent variables among the covariates.
