Robust and efficient estimation of nonparametric generalized linear models

Area of Science:

• Statistics
• Data Analysis
• Statistical Modeling

Background:

• Generalized linear models (GLMs) are widely used but sensitive to model misspecification and outliers.
• Classical GLMs require correct parametric component specification and absence of atypical observations for reliable inference.

Purpose of the Study:

• To introduce a novel family of nonparametric spline estimators for GLMs.
• To develop estimators that are robust to outlying observations and maintain high efficiency with clean data.

Main Methods:

• The proposed estimators are derived from minimizing a penalized density power divergence.
• Both full-rank and lower-rank spline variations are investigated.
• The estimators are designed for ease of implementation.

Main Results:

• The nonparametric spline estimators demonstrate robustness against outlying data points.
• These estimators can be tuned for high efficiency when data are clean.
• Theoretical analysis shows fast convergence rates under weak assumptions.

Conclusions:

• The novel spline estimators provide a flexible and robust alternative to classical GLMs.
• These methods offer a practical solution for analyzing diverse datasets, including those with atypical observations.
• The study highlights the competitive performance of these estimators through simulations and real-world applications.