Novel mixture SEM methods for comparing structural relations among many groups
Many important questions researchers aim to answer concern relations between unobservable or latent variables such as beliefs, traits, attitudes. For instance, which factors predict parental stress or compliance with social distancing during a pandemic? Structural equation modeling (SEM; Kline, 2015) is the state-of-the-art for properly modeling relations between latent variables, also called structural relations. For comparing structural relations across a large number of groups, such as countries, existing SEM approaches fall short, especially when the measurement of the latent variables is not perfectly invariant across the groups. In this project, we develop novel mixture-based SEM methods that use parsimonious approaches for capturing measurement non-invariance invariance (Kim et al. al., 2017; De Roover, 2021; De Roover et al., 2020), aiming to cluster the groups based on their structural relations specifically. Since parsimony tempers overfitting and lowers sample size requirements, building on a parsimonious approach for capturing measurement non-invariance may yield a better recovery of the clustering of the groups according to their structural relations than using group-specific parameters for capturing differences in measurement.
Dr.K. de Roover
Prof. dr. J.K. Vermunt
October 2022 – October 2026