Time-varying coefficient model estimation through radial basis functions

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Time-varying coefficient models are crucial for analyzing dynamic changes in longitudinal studies.
• Accurate estimation of these dynamic parameters is essential for understanding complex biological and clinical processes.
• Existing methods may have limitations in flexibility and accuracy for capturing time-dependent effects.

Purpose of the Study:

• To propose and evaluate a novel method for estimating dynamic parameters in time-varying coefficient models using radial kernel functions.
• To compare the performance of the proposed radial kernel function approach with traditional regression splines.
• To investigate both Frequentist and Bayesian estimation and inference strategies.

Main Methods:

• Utilizing a linear combination of weighted radial kernel functions with a specified bandwidth.
• Implementing Frequentist estimation via weighted least squares and bootstrap methods.
• Employing Bayesian inference using Markov chain Monte Carlo (MCMC) and variational methods.

Main Results:

• The proposed radial kernel function method demonstrates performance comparable to, or exceeding, regression splines in extensive simulations.
• The method's effectiveness is validated across various scenarios, including different sample sizes and correlation structures.
• The methodology is successfully applied to real-world data from AIDS clinical studies.

Conclusions:

• Radial kernel functions offer a robust and effective approach for estimating dynamic parameters in time-varying coefficient models.
• The proposed method provides a valuable alternative to existing techniques, particularly for complex longitudinal data.
• The findings have implications for the analysis of dynamic processes in various scientific fields, including clinical research.