Bayesian modeling of heterogeneity for large scale comparative research

Josine Verhagen (PhD student)
Department of Research Methodology, Measurement, and Data Analysis (OMD)
Twente University

Project: Project at Twente University financed by NWO

Project running from: 1 May 2008 – 1 May 2012

Supervisors: Prof. dr C.A.W. Glas, dr ir G.J.A. Fox (Twente University)

Summary:
Inferences from large-scale (e.g., cross-national) studies have important implications for theory (e.g., causal relations between constructs, spurious relations, intervening variables) and practice (e.g., insights in policy related issues and malleable factors). The common item response theory models are not directly applicable to analyse large-scale survey data for comparative research. There are several measurement issues connected to comparative research that need to be addressed since ignoring them may lead to inferential errors. The approach is focused on delineating the source (i.e., individual or group differences in latent scores or in the way of responding to the questionnaire) and the direction of the significant differences in cross-national research. From a Bayesian point of view, (1) heterogeneity in the way individuals respond to the questionnaire is modelled. In addition, (2) a structural population model is built for the respondents’ latent scores which is focused on heterogeneity. Within this modelling framework, the Bayesian methodology allows the development of tools that can be used to account for errors related to the measurement issues.

Date of defence:  16 November 2012

Title of thesis:  Bayesian Item Response Theory models for measurement variance