Bayesian (Non)Linear Random Effects Mediation Models: Evaluating the Impact of Omitting Confounders

• In Event: Educational Statisticians SIG Paper Session

Abstract

Often in educational and psychological research, researchers are interested in understanding the mediation mechanism of longitudinal variables (repeated measures). Longitudinal mediation models in the literature mostly stem from structural equation modeling framework and thus, cannot directly accommodate intrinsically nonlinear functions (e.g., exponential, logistic). Furthermore, no research has assessed the impact of omitting confounders when modeling mediation effects using intrinsically nonlinear functions. Thus, the current study aims to develop a Bayesian random effects mediation model to directly estimate intrinsically linear and nonlinear functions ((N)BREMM). Additionally, motivated by an empirical example, we investigated the impact of omitting confounders on model estimation of (N)BREMM. Monte Carlo simulation studies were conducted to evaluate the model performance of (N)BREMM, with and without confounders.

Authors

• Ziwei Zhang, University of Minnesota

• Nidhi Kohli, University of Minnesota

• Eric F. Lock, University of Minnesota