Abstract
The most popular approach to the assessment of the cross-national equivalence of concepts in surveys is multiple-group confirmatory confirmatory factor analysis (MGCFA). By this method, an increasingly restrictive set of constraints is imposed across groups, defined by countries, in a stepby- step manner. The logic of this approach is that the more restrictive the model achieving conventionally acceptable measures of global fit, the more parameters can be validly compared across countries in the structural part of the model. Of particular concern to us here is that, for comparisons to be made between means of latent constructs, it is necessary to establish crossnational equality of the latent intercepts of observed items, so-called ‘scalar invariance’. Where scalar invariance does not hold, conventional wisdom suggests that this is due to inherent differences in the meaning of concepts across cultures and/or faulty translation of survey questions. Our aim in this paper is to make two related claims. First, that failure to observe scalar invariance in a cross-national context may be due, at least in part, to social rather than cultural or linguistic heterogeneity. And, second, that in order to reject the null hypothesis that the measurement properties of survey questions do not differ across countries, social-psychological and demographic differences in the social composition of countries should be taken into account. We illustrate our argument with substantive examples based around the measurement of social and political trust in the second wave of the European Social Survey.