Causal Methods for Observational Data

Comparative effectiveness research often uses non-experimental observational data (like hospital discharge records or nationally representative surveys) to draw causal inference about the effectiveness of interventions for health. These ex post inferences require the careful use of specialized statistical methods in order to account for issues like selection bias and unmeasured heterogeneity. This paper briefly discusses the strengths and weakness of some of the most common causal methods for comparative effectiveness evaluation and provides instructions for using SAS® to implement propensity score matching, double difference, instrumental variables methods and regression discontinuity.