Cross-Cluster Propensity Score Matching With Cluster-Level Treatment Assignment

• In Event: Methods for Complex Hierarchical Structures

Abstract

Educational interventions may be administered at the cluster level (school, classroom) such that all students within a given cluster are either in the treatment or control group. One-to-one propensity score matching of individuals across clusters with cluster-level treatment assignment was evaluated in a simulation under varying conditions of number of clusters, within-cluster sample size, ICC, and treatment effect magnitude with regards to average bias, root mean square error, and balance of covariates. When a quasi-experimental method is necessary and assignment is at the cluster-level, one-to-one propensity score matching with sufficient sample size remaining after matching and propensity score estimation with a covariate highly correlated with the outcome and moderately so with treatment assignment is recommended to improve treatment effect estimation.

Authors

• Peter Boedeker, Baylor College of Medicine

• Marcus Fagan, University of North Texas

• Darrell M. Hull, University of North Texas

• Qi Chen, University of North Texas