Estimation of mean response via effective balancing score

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Missing data is a common challenge in statistical analysis, potentially biasing results.
• Conventional methods for handling missing data often rely on specific model assumptions.
• Robust and efficient estimation methods are needed, particularly for high-dimensional data.

Purpose of the Study:

• To introduce effective balancing scores for estimating the mean response under a missing at random mechanism.
• To develop a robust, model-free estimation procedure that simplifies analysis.
• To assess the performance of the proposed method in terms of consistency and efficiency.

Main Methods:

• Construction of effective balancing scores through model-free dimension reduction.
• Development of three types of scores capturing missingness, response, or both covariate informations.
• Asymptotic and numerical investigations of the proposed estimator's properties.

Main Results:

• Effective balancing scores provide consistent estimation with minimal efficiency loss.
• The proposed estimator is more robust and less reliant on model specification than existing methods.
• The method performs well, especially in high-dimensional covariate settings.

Conclusions:

• Effective balancing scores offer a robust and efficient approach to handling missing data.
• The model-free nature of the method makes it highly applicable, particularly with complex datasets.
• The approach is demonstrated as effective using Human Immunodeficiency Virus disease data.