Semi-Parametric Estimation of Potential Outcome Distributions and General Causal Estimands by Borrowing Information from Both Treatments and Controls
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Estimation of treatment effects is of great importance and has drawn much attention in various areas.In the potential outcomes framework, since potential outcomes refer to the same characteristic under different treatment assignments, they usually share certain similarities.Such similarities are largely overlooked in the literature, leading to potential efficiency loss.In this paper, we introduce a semi-parametric proportional likelihood ratio model (SPLRM) to jointly model the conditional distributions of potential outcomes through a shared baseline distribution, fully utilizing information from both treatment and control groups.We estimate all underlying parameters and general causal estimands by maximum empirical likelihood estimation.An iterative empirical likelihood algorithm is developed for parameter estimation, and a simple likelihood ratio test is introduced to assess the distributional treatment effect.We show that the proposed estimators for various treatment effects are asymptotically normal, and the likelihood ratio test statistic follows asymptotically a central Chi-square distribution when there is no distributional treatment effect.This approach improves efficiency and robustness compared to traditional methods that separately estimate treatment effects.Simulations and an analysis of the National Supported Work Demonstration dataset demonstrate the practical applicability and advantages of the pro-Statistica Sinica: Newly accepted Paper
Fields of Study & Subjects
Advanced Causal Inference Techniques
Statistical Methods and Inference
Statistical Methods in Clinical Trials
Publication details
- Journal / source
- Statistica Sinica
- Publication date
- 31 March 2026
- Language
- EN