Cluster Effects on Multilevel Latent Class Analysis

• Activity Number: 184
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Leila D Amorim*+ and Rosemeire L Fiaccone
• Companies: UFBA and UFBA
• Keywords: latent class analysis ; clustered data ; multilevel modeling

Abstract

Latent Class Analysis (LCA) yields definition of classes based on the probability of specific characteristics of the population according to association patterns across observed categorical variables. This methodology has been very useful to answer various research questions in social and behavioral sciences. More recently, there has been development of extensions of LCA that allows analysis of nested data structures. The Multilevel Latent Class Analysis (MLCA) accounts for clustering by introduction of random effects in the model to allow parameters to differ across groups. Although new, there is some discussion in the literature about framework and applications for MLCA. However, there is less debate about how cluster and sample sizes would affect the results for MLCA fit. Thus, the goal of this article is to discuss and illustrate the effects of clustering on MLCA. We consider two applications in Epidemiology: one related to a study that evaluates helminth infections in children from communities in Equator; and the second study contains factors associated to metabolic syndrome in students from different schools in North Carolina, United States.