Propensity Score Matching

[Page 1276]Propensity score matching is a research design used to minimize selection bias when comparing two or more nonrandomly assigned groups on an outcome of interest. The propensity score is the predicted probability that observations will be assigned to each treatment group, conditional on observed pretreatment characteristics. Through the process of propensity score matching, researchers identify virtually identical observations in different treatment conditions for comparison on outcomes of interest. Observations with similar propensity scores are matched for comparison in subsequent analyses. Ideally, treatment and control units could be matched on all relevant covariates so the comparison groups are identical—this perfect case would allow for a causal estimate of the treatment effect, as the main effect can be isolated from any selection bias. The practical implementation ...