Using item response theory (IRT) to evaluate the equivalence of surveys translated into different languages: An illustrative empirical example

Cross-cultural research offers an approach to understanding the prevalence and course of disease across cultures. However, cross-cultural comparisons often involve translating existing survey instruments. Methodological hazards abound when translation occurs, e.g., idiomatic meanings, and the validity of comparisons based upon translations that have not been empirically examined is in doubt. This issue is particularly salient in any multilingual nation. Differential item functioning (DIF), a problem with nonequivalent translations, presents when individuals equal on a construct but from different groups do not identical observed scores probabilities. For example, because of inadequate translation, an individual responding to a survey in the translated language, e.g., French, Spanish, etc., may respond differently than one responding in English despite equivalence in their true values. Item response theory (IRT) offers a method for assessing DIF, also known as measurement bias. The method allows increased validity when comparing responses and disease estimates across translations and cultures. The current study provides an empirical example, assessing the equivalence of an English to French translation of a survey assessing depression in a large (n = 8,688) sample of adults surveyed in the US and Canada as part of the Joint Canada/US Survey of Health, 2002-2003. The presentation describes the statistical model for assessing equivalence and illustrate its application. Findings describe statistically significant DIF for several items. Discussion emphasizes interpreting findings such as these, addresses generalizations to other surveys, highlights interpretation, and empirically demonstrates how translational DIF can affect validity.