Hardev Ranglani, EXL Service Inc, USA

This paper introduces a novel method for estimating the Average Treatment Effect (ATE) in observational causal inference studies by calculating cluster-specific ATEs and taking a weighted average across clusters. Instead of directly applying Inverse Propensity Weighting (IPW), this approach leverages clustering to address issues such as positivity violations and extreme weights, which often arise when propensity scores are near 0 or 1. Each cluster is formed based on covariates, and the ATE is estimated for each cluster using propensity score weighting. The overall ATE is then calculated as a weighted average of the cluster-specific ATEs, where the weights are based on the average propensity score within each cluster. This method effectively captures treatment effect heterogeneity and mitigates the instability caused by extreme individual weights. Simulations on synthetic data and real-world datasets demonstrate the superiority of this method in producing more stable and reliable treatment effect estimates compared to traditional IPW.

Inverse Propensity Weighting, Clustering, Average treatment effects