Evaluating propensity score weighting methods in heterogeneous population: a comparative study

Background: Propensity score weighting methods are frequently used in observational studies to estimate causal effects. However, researchers usually evaluate their performance only in homogeneous populations. When treatment effects are heterogeneous, it is unclear about the ability of commonly used weighting methods in estimating the average treatment effect (ATE) for the whole population. Methods: Two categories of propensity score weighting approaches are discussed to estimate the ATE for the whole population with heterogeneity: one in which the target of inference is the whole population, including Stabilized inverse probability weighting (SIPW), Empirical likelihood weighting (ELW), Covariate balancing propensity score weighting (CBPS); and the other whose target of inference is the specific population, including Inverse probability weighting with trimming (IPWT), Overlap weighting (OW). To assess the performance of these methods, a simulation study was conducted in terms of bias and root mean square error (RMSE), under six heterogeneity scenarios and three degrees of overlap in the propensity score distributions. Results: According to simulation results, approaches that target the specific population have a substantial bias that rises quickly with increasing heterogeneity. In contrast, approaches for the whole population perform well with relatively smaller bias and RMSE, where CBPS outperforms the others in all heterogeneity scenarios and overlap degrees. When methods are applied to Lindner data with heterogeneity, the results of approaches for the whole population are quite different from approaches for the specific population. Conclusions: When the estimand is the ATE for the whole populations with heterogeneity, applying weighting methods that target specific populations (e.g. IPWT, OW) to estimate it would lead to substantial bias. It is recommended to use methods targeting the whole population (e.g. CBPS, ELW) to achieve this goal.