Prediction of random effects in linear and generalized linear models under model misspecification

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Random effects models are widely used for longitudinal and correlated data.
• These models often assume random effects follow a Gaussian distribution.
• The impact of violating this assumption is not fully understood.

Purpose of the Study:

• To investigate the effect of misspecifying the random effects distribution in statistical models.
• To assess the impact on prediction accuracy and recovery of true distributions.
• To determine if standard statistical approaches remain valid under assumption violations.

Main Methods:

• Theoretical calculations
• Numerical calculations
• Simulation studies
• Analysis of real-world data (Heart and Estrogen/Progestin Replacement Study)

Main Results:

• Predicted values can differ based on the assumed distribution.
• Prediction accuracy, measured by mean square error, is robust to mild-to-moderate misspecification of the random effects distribution.
• Standard statistical modeling approaches are often adequate.

Conclusions:

• The Gaussian assumption for random effects in longitudinal models is not always critical for prediction accuracy.
• Mild-to-moderate deviations from the assumed distribution do not significantly impair results.
• Existing statistical software and methods are likely sufficient for many applications.