ABSTRACT
Chapter 10 presents propensity-score methods to adjust for confounding, explaining the dimension-reducing property of the propensity score – when H is a sufficient confounder and the model for the propensity score is correct, then the propensity score is itself a sufficient confounder. It emphasizes the importance of checking for overlap of the distributions of the propensity score corresponding to the two treatment groups. It illustrates using the propensity score to estimate a treatment effect conditional on the propensity score, as well as estimation via stratification on quintiles or quartiles of the propensity score. It presents standardization using the propensity score instead of the confounders, and then the chapter concludes with an example of matching on the propensity score to estimate the population averaged treatment effect and then average effect of treatment on the treated. Examples and R code are also provided.
