Doubly Robust Nonparametric Multiple Imputation for Ignorable Missing Data.

Area of Science:

• Statistics
• Biostatistics
• Data Science

Background:

• Missing data is a prevalent issue in medical and social sciences, complicating data analysis.
• Multiple imputation (MI) is a common technique for handling missing data by replacing missing values with plausible estimates.
• Existing methods face challenges with high-dimensional covariates and model misspecification.

Purpose of the Study:

• To propose a novel nonparametric multiple imputation (MI) method for estimating marginal means with missing data.
• To develop an approach that is robust to high-dimensional covariates and offers double robustness.
• To introduce a sensitivity analysis for evaluating working model validity and optimizing estimator efficiency.

Main Methods:

• A new nonparametric MI approach using two working models for dimension reduction.
• Development of imputing sets for missing observations.
• Incorporation of a sensitivity analysis to assess working model validity and select optimal weights.

Main Results:

• The proposed nonparametric MI method handles high-dimensional covariates effectively.
• The estimator demonstrates double robustness, remaining consistent if at least one working model is correct.
• The method is more robust to misspecification of both working models compared to existing doubly robust methods and avoids inverse-weighting issues.

Conclusions:

• The novel nonparametric MI approach provides a robust and efficient solution for missing data problems, particularly in high-dimensional settings.
• The method offers advantages over existing techniques in terms of robustness to model misspecification and sensitivity to missing probabilities.
• Simulation studies and application to a colorectal adenoma study support the favorable performance of the proposed method.