An estimation method for the semiparametric mixed effects model

Area of Science:

• Statistics
• Biostatistics
• Data Analysis

Background:

• Mixed effects models are widely used for clustered or longitudinal data.
• Traditional models often assume Gaussian (normal) random effects, which can be restrictive.
• Analyzing data with non-Gaussian random effects requires advanced statistical approaches.

Purpose of the Study:

• To propose a semiparametric mixed effects regression model for continuous, ordinal, or binary outcomes.
• To relax the assumption of Gaussian random effects using a nonparametric density estimation method.
• To improve the accuracy of fixed effects estimators in mixed models.

Main Methods:

• Utilized a predictive recursion method for nonparametric estimation of random effects density.
• Developed a new strategy to accelerate the estimation algorithm.
• Employed Powell's conjugate direction search to maximize marginal profile likelihood for parameter estimation.

Main Results:

• Monte Carlo simulations demonstrated improved mean squared error for fixed effects estimators when random effects distributions deviate from Gaussian.
• The method effectively visualizes random effects densities, as shown in the Wisconsin Sleep Survey analysis.
• The proposed estimation procedure is computationally efficient for large datasets.

Conclusions:

• The semiparametric mixed effects model offers a robust alternative to traditional methods for complex data structures.
• Nonparametric estimation of random effects distributions enhances statistical inference.
• The approach is computationally feasible and practically useful for real-world data analysis.