Extended information criterion (EIC) approach for linear mixed effects models under restricted maximum likelihood...

Area of Science:

• Statistics
• Biostatistics
• Clinical Data Analysis

Background:

• Restricted maximum likelihood (REML) is standard for variance component estimation in linear mixed effects models.
• Comparing linear mixed effects models, especially with varying mean structures under REML, lacks straightforward methods.
• Existing approaches for comparing models with different mean structures under REML are limited.

Purpose of the Study:

• To propose a novel approach for comparing linear mixed effects models with different mean and covariance structures under REML estimation.
• To introduce the extended information criterion (EIC) as a bootstrap-based extension of AIC for model selection.
• To address the challenge of comparing models with differing mean structures within the REML framework.

Main Methods:

• Utilizing the extended information criterion (EIC), a bootstrap-based method.
• Applying EIC for the comparison of linear mixed effects models with diverse mean and covariance structures.
• Employing REML estimation as the underlying statistical framework.

Main Results:

• The proposed EIC approach facilitates the comparison of linear mixed effects models under REML.
• Simulation studies demonstrate the effectiveness of the EIC method.
• The approach is successfully applied to two real-world clinical datasets.

Conclusions:

• The extended information criterion (EIC) provides a viable solution for comparing linear mixed effects models with different mean and covariance structures under REML.
• The method offers improved model selection capabilities in clinical data analysis.
• EIC enhances the ability to choose appropriate models when mean structures vary.