Lukociene, Olga

Performance of latent class analysis based random coefficient models

Project executed at: Department of Methodology, Faculty of Social Sciences, Tilburg University

Project running from: 1 September 2004 – 1 September 2008

Project financed by: NWO

Promotor:   Prof. dr J.K. Vermunt

Summary:
The two basic assumptions underlying standard linear random-effects models – normal errors and normal random effects – may be unrealistic in social science research. Outcome variables of interest are very often categorical variables, which makes it necessary to use non-linear mixed models. Also the distributional assumptions about random effects are not realistic in most applications. Latent class regression analysis provides an alternative nonparametric approach that relaxes this assumption and that makes it straightforward to deal with categorical outcome variables. The objective of this project is to provide a systematic comparison between parametric and nonparametric random-coefficients models.

Date of defence:   26 March 2010

Title of thesis:   Latent class models for categorical data with a multilevel structure