Measurement Equivalence in Sequential Mixed-Mode Surveys

Many surveys collect data using a mixture of modes administered in sequential order. Al-though the impacts of mixed-mode designs on measurement quality have been extensively studied, their impacts on the measurement quality of unobservable (or latent) constructs is still an understudied area of research. In particular, it is unclear whether latent constructs derived from multi-item scales are measured equivalently across different sequentially-administered modes-an assumption that is often made by analysts, but rarely tested in practice. In this study, we assess the measurement equivalence of several commonly-used multi-item scales collected in a sequential mixed-mode (Web-telephone-face-to-face) survey: the Age 25 wave of the Next Steps cohort study. After controlling for selection via an extensive data-driven weighting procedure, a multi-group confirmatory factor analysis was performed to assess mea-surement equivalence across the three modes. We show that cross-mode measurement equiv-alence is achieved for nearly all scales, with partial equivalence established for the remain-ing. Although measurement equivalence was achieved, some differences in the latent means were observed between the modes, particularly between the interviewer-administered and self-administered modes. We conclude with a discussion of these findings, their potential causes, and implications for survey practice.