Methodological Challenges of Complex Latent Mediator and Moderator Models

Applied research in the social sciences goes far beyond the analysis of linear relationships. In recent years, mediation and moderation analyses are gaining popularity as a way to model complex relationships in the data. Due to their popularity, these research methods are developing rapidly. Consequently, users can easily lose track of the newest developments, improvements, or recommendations. Purely theoretical articles take a long time for new methodological developments to be noticed and used by applied researchers. Therefore, applied researchers need more guidance about new methodological developments and their applicability. In two recent articles (Cortina et al. 2021; Cheung et al. 2021) the authors have stated, that in nearly all applications employing latent variables, researchers use structural equation models for testing the additive parts of their models but go back to regression models when interaction or nonlinearity is involved, hence, ignoring the measurement structure of the data. In addition, many substantive applications with latent variables contain only simple moderations between two latent variables. Typically, the number of product terms is limited to one or two, and the investigated nonlinear effects are often limited to moderator effects of exogenous latent variables on endogenous latent variables, thereby ignoring possible mediator and moderated mediator effects between endogenous and exogenous variables. Finally, a systematic comparison of the different approaches in terms of bias, efficiency, complexity of handling, and run time as guidelines for applied researchers is missing.