When clustering is endogenous

Title: When clustering is endogenous: (mis)interpretation of multilevel model estimates and alternate explanations

Abstract: In this talk, an evaluation of the current application of multilevel models in the field of education research is provided.  Specifically, the rise in the use of multilevel models in education over the past three decades is reviewed and, given the endogeneity of clustering, interpretations that have been made regarding estimates from some of these models in the recent applied literature is questioned.  Model assumptions, some testable and some untestable, that researchers should consider before making causal inference regarding cluster effects with observational data are highlighted.  Finally, alternative approaches for modeling, depending on the posed research question, are proposed and discussed.

Laura Stapleton is an Associate Professor in Measurement, Statistics and Evaluation in the Department of Human Development and Quantitative Methodology at the University of Maryland. Starting in 2013, she also serves as the Associate Director for Research for the Maryland Longitudinal Data System Center. Her research interests include the analysis of administrative and survey data obtained under complex sampling designs and multilevel latent variable models, including tests of mediation within a multilevel framework. She has served as Associate Editor for the Journal of Educational Psychology, has published widely in both methodological and educational policy areas and has co-edited a recent volume in advances in multilevel modeling for education research. Annually, she serves as an instructor for the U.S. Department of Education Institute of Education Science’s summer institute in cluster randomized trials. In the 1990s, she was an economist at the Bureau of Labor Statistics and, subsequently, conducted educational research at the American Association of State Colleges and Universities and as Associate Director of institutional research at the University of Maryland.