Doubly Robust Estimation of Treatment Effects from Observational Multilevel Data
When randomized experiments cannot be conducted, propensity score (PS) matching and regression techniques are frequently used for estimating causal treatment effects from observational data. These methods remove bias caused by baseline differences in the treatment and control groups. Instead of using a PS technique or an outcome regression singly, one might use a doubly robust estimator that combines a PS technique (matching, stratification, or inverse propensity weighting) with an outcome regression in an attempt to address bias more effectively. Theoretically, if the PS or outcome model is correctly specified, a doubly robust estimator will produce an unbiased estimate of the average treatment effect (ATE). Doubly robust estimators are not yet well studied for multilevel data where selection into treatment takes place among level-one units within clusters. Using four simulated multilevel populations, we compare doubly robust estimators to standard PS and regression estimators and investigate their relative performance with respect to bias reduction.
KeywordsPropensity score Observational study Doubly robust estimator Multilevel modeling
The research reported here was supported by the Institute of Education Sciences, U.S. Department of Education, through Grant R305D120005. The opinions expressed are those of the authors and do not represent views of the Institute or the U.S. Department of Education.
- Angrist, J., & Pischke, J. (2009). Mostly harmless econometrics: An empiricist’s companion. Princeton, NJ: Princeton University Press.Google Scholar
- Keller, B., Kim, J.-S., & Steiner, P. (2013). Data mining alternatives to logistic regression for propensity score estimation: Neural networks and support vector machines. Multivariate Behavioral Research, 48(1), 165.Google Scholar
- Kim, J., & Seltzer, M. (2007). Causal inference in multilevel settings in which selection process vary across schools. Working Paper 708, Center for the Study of Evaluation (CSE), UCLA, Los Angeles.Google Scholar
- Kreif, N., Grieve, R., Radice, R., & Sekhon, J. (2011). Regression-adjusted matching and double-robust methods for estimating average treatment effects in health economic evaluation. Health Services and Outcomes Research Methodology, 13, (2–4), 174–202.Google Scholar
- Steiner, P., Kim, Y., Hall, C., & Su, D. (2015). Graphical models for quasi-experimental designs. Sociological Methods & Research, 0049124115582272Google Scholar
- Steiner, P., Kim, J.-S., & Thoemmes, F. (2013). Matching strategies for observational multilevel data. In JSM Proceedings (pp. 5020–5032). Alexandria, VA: American Statistical Association.Google Scholar