Investigating the Relative Impact of Different Sources of Measurement Non-Equivalence in Comparative Surveys

An Illustration with Scale Format, Data Collection Mode and Cross-National Variations

Abstract

Different factors are known to affect the comparability of multinational, multicultural and multiregional (‘3MC’) survey data. These include factors relevant to the design of the questionnaire in different contexts (such as cultural differences in how a concept is understood, inaccurate or approximate translations of concepts, and variant adaptations to question formats). Others include factors relating to the survey design in general and how it is implemented across contexts (such as sample design, choice of mode(s), and contact strategies). Together, they contribute item, method and construct biases that can affect the invariance of composite measures. While research to date has looked at the effects of these factors on measurement invariance individually, there have been few attempts to compare them directly and assess their relative impact. To illustrate how this can be done, the present paper tests for measurement invariance in a subjective wellbeing measure across question formats, modes, languages, and countries, combining European Social Survey data from designed and natural experiments (resulting from the use of variant question formulations and translations) from Germany, Switzerland and France. Overall, we find translation errors, language and culture to be bigger sources of non-equivalence than question format and mode. The findings have implications for both survey designers making decisions about optimal resource allocation in the design of 3MC studies, as well as for comparative analysts interested in comparing countries with shared languages and interpreting cross-group differences.