Sufficient Dimension Reduction for Feasible and Robust Estimation of Average Causal Effect

Area of Science:

• Epidemiology
• Biostatistics
• Econometrics

Background:

• Estimating treatment effects in observational studies is challenging due to confounding.
• Existing methods like imputation and double robust estimators have different strengths and weaknesses.

Purpose of the Study:

• To develop a novel shrinkage estimator that combines the advantages of imputation and double robust methods.
• To improve the efficiency and robustness of treatment effect estimation in observational data.
• To assess the impact of maternal smoking on infant birth weight using the proposed method.

Main Methods:

• Utilized a semiparametric locally efficient dimension reduction approach.
• Integrated results using imputation, inverse probability weighting, and double robust augmentation estimators.
• Introduced a shrinkage estimator to combine imputation and double robust procedures.

Main Results:

• The proposed shrinkage estimator retains double robustness while improving variance when response models are correct.
• Simulated experiments and a real-world dataset demonstrated the estimator's performance.
• The study provides a more accurate estimation of treatment effects in observational settings.

Conclusions:

• The novel shrinkage estimator offers a superior approach for estimating treatment effects in observational studies.
• This method enhances statistical efficiency and robustness, particularly when response models are accurately specified.
• The findings have implications for research in public health, social sciences, and economics.