Conferences

When the comparability of attitude scales in international surveys is to be assessed, sociologists habitually think of structural equation modelling (SEM) as the method of choice. Alternative methodological approaches such as the Rasch model, however, offer some conceptual and practical advantages of their own. The present paper will demonstrate how Rasch models can be used to investigate measurement equivalence in comparative research settings.

Among survey researchers, the basic Rasch model is (in)famous for rarely fitting ‘real world’ data. Still, several different routes can be taken to apply Rasch scaling fruitfully in the social sciences even when a straightforward application fails, without giving up too much of their theoretical promise. One of these routes is to simply adopt the Rasch perspective on measurement, and to accept that not every respondent may be ‘measurable’ with sufficient reliability by the instrument at hand. A second, somewhat different route is to allow for some heterogeneity in each of the population(s) surveyed and to scale different (sub-) populations with similar, but not identically parametrized Rasch models, using manifest criteria to distinguish the populations. Obviously, samples from different nations provide a manifest grouping criterion on which an analysis of Differential Item Functioning can be based. As a third route, with the advent of discrete mixture models, ‘latent’ subgroups with different item parameter sets can be distinguished in a more exploratory fashion as well.

These strategies will be briefly demonstrated for a well established scale, namely that of “general national pride”, as measured in ISSP 2003. For the sake of demonstration, the analyses will be limited to a few national samples.