Abstract

Pursuing Causal Inferences in the Absence (or Failure) of Random Assignment: An Introduction to Propensity Score Analysis

Randomized experiments are the gold standard for making causal inferences — without randomization, post-treatment group differences may be confounded by pre-existing group differences (i.e., selection bias). But randomization is not always possible, practical or ethical in the social, behavioral and education sciences. Alternative, quasi-experimental methods are needed in the absence — or failure — of randomization.

Propensity score analysis (PSA) is a broad collection of methods that aims to statistically equate two or more groups on a set of observed covariates as a means for minimizing selection bias. This presentation will provide an introduction to PSA, with emphasis on propensity score matching, subclassification and weighting. Empirical examples will be provided to demonstrate the methods’ practical application.