Abstract Mediation analysis in large-scale assessments often involves a multilevel structure, where students are nested within classrooms or schools. In such a context, multilevel structural equation modeling (MSEM) provides a flexible framework for estimating and testing the mediation process. Plausible values (PVs), however, present unique challenges for mediation analysis in large-scale assessments, yet methodological guidance remains limited. In particular, standard pooling procedures complicate the inference of indirect effects, which relies on the construction of confidence intervals. To address these gaps, we conducted a Monte Carlo simulation study comparing three modeling methods (aggregation, two-step approach, and MSEM) and three confidence interval methods (delta, distribution of the product, and Monte Carlo) in the context of 2–2–1 mediation with PVs. We evaluated their performance in terms of relative bias, confidence interval coverage, and power across a range of realistic conditions. Simulation results suggest that the MSEM-Monte Carlo combination performs best when sample size requirements were met. An empirical example is also provided to illustrate the practical implementation of 2–2–1 mediation analysis with PVs.
Li et al. (Tue,) studied this question.