Improving Causal Inference: Recommendations for Covariate Selection and Balance in Propensity Score Methods

Citations Over TimeTop 13% of 2017 papers

Abstract

Objective: This paper is a resource for applied researchers and journal reviewers who wish to learn more about propensity score methods, especially recent published recommendations for selecting and balancing covariates before conducting treatment effect analysis. Method: We reviewed and synthesized methodological studies of propensity score analysis from multiple disciplines in an effort to distill a cogent and defensible set of recommendations for research practice. Results: Selection of covariates for propensity score methods requires good understanding of empirical evidence and theory related to confounders of treatment assignment and the outcome, as well as clarity about the temporal relations among confounders, treatment, and outcome as measured in the data set in use. Some variables are not appropriate for inclusion in a propensity score model, and some data sets do not include sufficient variables for propensity score analysis. Achieving balance in the model covariates after conditioning on the propensity score should be an iterative process where models are respecified until proper balance is obtained. Important steps for propensity score methods include reporting results from iterative modeling and balance checks, and then defending the final sample conditioned on the propensity score before analysis of treatment effects. These steps are often overlooked in published research. Conclusion: Propensity score methods can be useful tools for reducing bias due to observed confounders. To credibly reduce bias, however, researchers must execute recommended practice for selecting and balancing covariates.

Related Papers

• → Comparison of Propensity Score Methods and Covariate Adjustment(2017)694 cited
• → The control of confounding by intermediate variables(1989)243 cited
• → Propensity Scores: Method for Matching on Multiple Variables in Down Syndrome Research(2009)25 cited
• [Application of propensity score matching in the design of an epidemiological study].(2009)
• → Propensity Scores and Propensity Score Matching for Assessing Multiple Confounders(2016)