Semiparametric Inference for Nonmonotone Missing-Not-at-Random Data: The No Self-Censoring Model

Area of Science:

• Statistics
• Biostatistics
• Data Science

Background:

• Multivariate data often suffer from missing values.
• Non-monotone and not-at-random missingness poses significant analytical challenges.
• Existing methods may lack robustness or require strong assumptions.

Purpose of the Study:

• To develop methods for identifying and estimating statistical functionals with non-monotone, not-at-random missing data.
• To establish theoretical guarantees for statistical inference under specific missingness assumptions.
• To propose a practical and robust estimation procedure.

Main Methods:

• Semiparametric modeling approach.
• Utilizing the
• no self-censoring
• or
• itemwise conditionally independent nonresponse
• assumption.
• Odds ratio parameterization of the joint density.
• Augmented inverse probability weighted (AIPW) estimation.

Main Results:

• Identification of statistical functionals under the specified missingness mechanism.
• Establishment of the semiparametric efficiency bound.
• Development of a practical AIPW estimator.
• Demonstration of double-robustness properties with always-observed covariates.

Conclusions:

• The proposed semiparametric approach effectively handles complex missing data.
• The novel estimator provides robust and efficient estimation.
• The methodology is validated through simulations and real-world data analysis.