Propensity Score Matching Analyses: A Primer

As with any analytic approach, including randomized controlled studies, certain assumptions must be met for research to be considered internally (i.e., strength of causal inference) or externally (i.e., ability to generalize) valid.  The purpose of this presentation is to provide an overview of propensity matching and use examples from our three studies to illustrate design issues surrounding internal and external validity.  First, the presenter provides a brief overview of the potential outcomes framework (i.e., counterfactuals) undergirding all propensity matching analyses.  Then, the presenter reviews design features associated with higher internal and external validity, including:  the breadth of potentially confounding variables available for analysis, the degree to which there was overlap in values of confounding variables between the sample of interest and the matching sample, balance checking, and examination of both the region of common support and characteristics of unmatched cases.  Finally, the presenter briefly describes the different types of matching strategies used in these studies, which include propensity weighting and one-to-one matching.