SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Imprint / Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University

Contributions: Abstract

Assessing equivalence of measurement within the framework of Latent Class Analysis.

Guy Notelaers
Katholieke Universiteit Leuven
Jeroen Vermunt Marc van Veldhoven
University of Tilburg
The Netherlands

Cross-cultural survey research usually has to deal with a lot more methodological issues and problems than intracultural survey research. Perhaps the most prominent of these additional concerns is the problem of equivalence, or in other words: the problem of making valid comparisons across groups. Two closely related concepts play an essential role in cross-cultural comparisons, namely, equivalence and bias.

According to the hierarchy of the equivalence levels, construct equivalence is to be achieved first before going into item bias analysis. This is usually evaluated by means of structure oriented approaches. Van de Vijver & Leung (1997) see exploratory factor analysis, structural equation models as widely used techniques to investigate construct equivalence. Central is its operationalisation as factorial invariance. Less used to determine construct equivalence are multidimensional scaling techniques. With PINDIS it is possible to investigate the question whether there is construct equivalence. Even less used is latent class analysis or cluster analysis, as stated by van de Vijver & Leung.

In this contribution we assess equivalence for categorical data within the framework of latent class analysis of the Questionnaire Experience and Evaluation of Work that is widely used in Belgium to measure stress at work. Since the presumed strong Mokken model does not hold, the instrument has been remodelled with Latent Cluster Analysis to enable research to classify respondents properly.

For the description of the framework about equivalence we follow Hagenaars' (1994) conceptualisation of the relationships between latent and manifest variables that varied across Germany and Switzerland. We argue that testing construct equivalence and analyzing item bias analysis happens in the same cycle. Analogous to van de Vijver and Lueng three levels of equivalence are distinguished: strong equivalence, equivalence, and weak equivalence.