An alternative empirical likelihood method in missing response problems and causal inference.

Area of Science:

• Statistics
• Biostatistics
• Econometrics

Background:

• Missing data is a pervasive issue in medical, social, and economic research.
• Complete case analysis can introduce bias when data is missing at random.
• Existing methods like inverse probability weighting and augmented inverse probability weighting have limitations.

Purpose of the Study:

• To introduce a novel empirical likelihood-based estimator for handling missing data.
• To evaluate the performance of the proposed estimator against existing methods.
• To demonstrate the application of the estimator in causal inference and real-world studies.

Main Methods:

• Developed a doubly robust and locally efficient empirical likelihood-based estimator.
• Utilized simulation studies to compare the proposed estimator with existing methods.
• Applied the method to an observational study on smoking within the Cardiovascular Outcomes in Renal Atherosclerotic Lesions trial.

Main Results:

• The proposed empirical likelihood estimator demonstrates improved performance when the propensity score model is correctly specified.
• The estimator possesses double-robustness properties, ensuring valid inference under certain model misspecifications.
• The method is effective for estimating average treatment effects in observational causal inference.

Conclusions:

• The novel empirical likelihood estimator offers a robust and efficient alternative for addressing missing data in statistical studies.
• The method shows particular promise when propensity score modeling is accurate.
• This approach has practical implications for analyzing observational data, including clinical trials.