Identity for the NPMLE in censored data models

Area of Science:

• Statistics
• Biostatistics
• Survival Analysis

Background:

• Maximum likelihood estimation (MLE) is crucial for analyzing censored data.
• Nonparametric maximum likelihood estimators (NPMLE) and regularized MLEs are widely used but require rigorous theoretical justification.
• Proving consistency and efficiency of these estimators in complex censored data models can be challenging.

Purpose of the Study:

• To derive a novel identity for NPMLE and regularized MLEs in censored data models.
• To establish a unified framework for proving the consistency and efficiency of these estimators.
• To provide a general algorithm for estimating the limiting variance of NPMLEs.

Main Methods:

• Derivation of a key identity relating standardized maximum likelihood estimators to the standardized empirical process.
• Application of this identity to prove consistency and efficiency in specific censored data models.
• Development of a general algorithm for estimating the limiting variance of NPMLEs.

Main Results:

• A new identity for NPMLE and regularized MLEs in censored data models was successfully derived.
• The derived identity serves as an effective tool for proving estimator consistency and efficiency.
• The method was illustrated for univariate right-censored data, current status data, and mixture models.
• A general algorithm for estimating the limiting variance of NPMLEs was provided.

Conclusions:

• The derived identity offers a powerful and unified approach for theoretical analysis of NPMLE and regularized MLEs.
• This work advances the understanding and application of statistical inference in censored data settings.
• The provided algorithm facilitates practical implementation and variance estimation for NPMLEs in complex scenarios.