This repository contains the accompanying code to the course "Comparative Social Research with Multi-Group SEM" for the GESIS Spring Seminar 2023 by Daniel Seddig, Eldad Davidov, Peter Schmidt and Yannick Diehl.
Determining whether people in certain countries, or at different time points score differently in measurements of interest, or whether constructs relate differently to each other across nations can indisputably assist in testing social sciences theories and advancing our knowledge. However, meaningful comparisons require equivalent measurements of these constructs. This is especially true for subjective attributes such as values, attitudes, perceptions, or opinions. In this course, we first discuss the meaning of cross-group measurement equivalence, look at possible sources of nonequivalence, and suggest ways to prevent it. Next, we examine the social science methodological literature for ways to empirically test for full or partial measurement equivalence using multiple group confirmatory factor analysis (MGCFA). In addition, we discuss how to test equivalence of regression coefficients and/or latent means and variances using MGCFA and multiple group structural equation modeling (MGSEM). We present such tests using the software environment R (e.g., lavaan). Furthermore, we consider what may be done when exact and partial equivalence is not supported by the data. We discuss strategies based on the less strict assumption of approximate equivalence, such as alignment optimization and the Bayesian estimation procedure. These methods offer exciting directions and solutions for future research in cross-group measurement equivalence assessment when exact equivalence is not supported by the data. Finally, we will address the analysis of categorical data. During the exercises, participants will have the opportunity to conduct these tests using data on human values from the European Social Survey, and if time allows data on biodiversity and intentions to behave in an environmentally friendly way with a reasoned action approach. We also encourage participants to bring their own data and apply the methods discussed in the course to their data.