The Nexus of Multilevel Models and Causal Inference in Non-Experimental Multisite Research

• In Event: Advances in Propensity Score Analyses and Causal Inference

Abstract

The use of multilevel models is now commonplace in educational research to address the hierarchical, or clustered, structure of the education system. Similarly, causal inference under the potential outcomes framework is becoming standard in educational research. Understanding the nexus of these two methodological advances is still in its infancy, however. This paper pieces together key literature on causal inference and its application to non-experimental multilevel data to illuminate the key issues researchers must consider and how multilevel models can help address those issues. The discussion emphasizes multilevel complications to the key identifying assumptions of the potential outcomes framework, interpretation of multilevel regression-adjusted group differences as treatment effects, and the use of multilevel models with propensity score matching methods.

Author

• Jordan Rickles, University of California - Los Angeles