A boosting method to select the random effects in linear mixed models

Area of Science:

• Statistics
• Statistical Modeling

Background:

• Linear mixed models are widely used in various scientific fields.
• Selecting appropriate random effects is crucial for model accuracy.
• Existing methods face challenges with non-convex objective functions.

Purpose of the Study:

• To propose a novel likelihood-based boosting method for random effects selection.
• To address the challenges posed by non-convex objective functions in model optimization.

Main Methods:

• Developed a boosting algorithm utilizing likelihood-based criteria.
• Incorporated directions of negative curvature alongside Newton directions for optimization.
• Applied the method to a simulated dataset and a real-world application.

Main Results:

• The proposed method demonstrates effective selection of random effects.
• The optimization strategy successfully navigates the non-convex objective function.
• Both simulation and real-data results confirm the method's good performance.

Conclusions:

• The novel likelihood-based boosting method offers a robust solution for random effects selection.
• The optimization technique enhances the reliability of fitting linear mixed models.
• This approach provides a valuable tool for statistical modeling and data analysis.