Longitudinal measurement (non-)invariance in latent constructs : conceptual insights, model specifications and testing strategies

Kurzfassung/Abstract

Many key concepts in the social sciences (e.g., personality traits, values, beliefs, attitudes, cognitive competencies) are not directly observable but represent latent variables, measured by a set of observable variables called manifest indicators (Bollen, 2002). Furthermore, major contributions to longitudinal statistical modeling and the improved availability of large-scale panel data gave rise to increased interest from a developmental research perspective, focusing on intra-individual change of latent variables over time. The validity of time-related comparisons of latent means (and covariance structures) presupposes that a set of testable assumptions regarding the underlying measurement models—subsumed under the term measurement invariance (MI; e.g., Meredith, 1993; Millsap, 2011; Vandenberg and Lance, 2000)—is satisfied. However, methodology reports of relevant panel studies usually do not contain systematic assessments of MI for key concepts over time, but put the responsibility on the individual data users (Seddig and Leitgöb, 2018a). Thus, the chapter aims at providing a guideline on how to (i) correctly specify longitudinal confirmatory factor (CFA) models with autocorrelative structures, (ii) systematically test for measurement (non-)invariance in latent variables over time, and (iii) proceed if the strong assumption of exact MI (all equality assumptions are satisfied) cannot be sustained.