Abstract:
|
Propensity scores are commonly used to identify the causal effect of treatment in an observational study. The propensity score estimates the probability of receiving treatment as a function of pre-treatment covariates. Under the assumption of strong ignorability and the stable unit treatment value assumption, using the propensity score to analyze the data (via matching, stratification, or weighting approaches) will reduce bias due to differences in the covariate distributions of the treatment and control group. Here, motivated by a re-analysis of an observational study on prostate cancer patients, we consider ways in which the propensity score can also be used to identify treatment effect heterogeneity across the covariate space.
|