Online Program

Saturday, October 22
Knowledge
Community
Influence
Sat, Oct 22, 4:30 PM - 5:15 PM
Carolina Ballroom
Poster Session 6

A Comparison of Multilevel Mediation Modeling Methods (303494)

*Christina Kelsey Zigler, University of Pittsburgh 

Multi-level or nested data is commonly encountered in practice, especially in educational research, and often violates the assumptions needed for traditional mediation analysis. Currently, there are three proposed methods for modeling mediation in multi-level data that go beyond aggregation/disaggregation; conflated multilevel modeling, unconflated multilevel modeling, and multilevel structural equation modeling (MSEM). Since all three methods are being employed in current peer reviewed journals, practical guidelines for researchers are needed to choose an adequate method based upon theoretical models and actual data characteristics. Because researchers often do not know the underlying structure of their real data, simulated data for this study were generated using two basic models; a model that only simulated within-pathways and a model that directly simulated separate between- and within- pathways. The data generation model was found to influence bias, type 1 error rate, and power of the between indirect effect. We offer recommendations for applied researchers on when to implement each modeling method.