Composite Likelihood Estimation and Testing for Structural Equation Modeling

• Activity Number: 449
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Social Statistics Section
• Author(s): Irini Moustaki*+ and Myrsini Katsikatsou
• Companies: London School of Economics and London School of Economics
• Keywords: categorical data ; likelihood ratio test ; maximum likelihood ; pairwise likelihood estimation ; latent variable models

Abstract

Structural equation models and latent variables models are widely used in social sciences for measuring unobserved constructs such as intelligence, fear of crime, anxiety etc. In the last two decades, latent variable models have been extended to account for categorical responses, multidimensional latent variables, effects of explanatory variables, non-linear relationships, longitudinal data, missing values, outliers and complex survey data. Those extensions have led to complex models with many parameters in which estimation methods such as maximum likelihood is difficult if not intractable. In this talk, we discuss composite likelihood estimation methods and goodness-of-fit test statistics for dealing with the complexities of the models. Simulations and real applications will be used to illustrate the performance of the proposed methods.