Examining Measurement Isomorphism of Multilevel Constructs: The Case of Political Trust

Abstract

Listen

Translate article

There is a growing recognition of the importance of testing for measurement equivalence when comparing latent constructs—such as political trust—across countries and over time. Indeed, equivalence of measurements across countries and time points is a precondition for making meaningful and valid comparisons of means, scores, and relationships between constructs. Until recently, most efforts in this area of comparative research have focused on establishing measurement equivalence across groups, such as countries or cultural regions. In contrast, scarce attention has been paid to examining the measurement equivalence of constructs across levels of data. Rather, most empirical studies that use cross-national survey data assume that the factorial structure of a given construct is isomorphic, in other words, similar in measurement and meaning at the individual and country level. This assumption is not always justified, as the dimensions found at the individual level of data do not always generalize to the country level. In such cases, the results and substantive conclusions based on the assumption of measurement isomorphism may be misleading. In this article, we emphasize the importance of examining measurement isomorphism when working with cross-national survey data and describe a testing procedure using multilevel confirmatory factor analysis. We illustrate the procedure by examining cross-level equivalence of the political trust scale included in the European Social Survey (2008–2009). The results of the analysis indicate that the structure of political trust factor can be considered configurally isomorphic across individual and country levels. In addition, the measurement scale across the two levels is found to be partially isomorphic. These findings could be regarded as an encouraging result for the applied researchers who use the aggregated individual scores of political trust. At the same time, we demonstrate that measurement isomorphism cannot be simply assumed and hence should be examined prior to analysis to ensure valid and meaningful results.