COVARIATE DECOMPOSITION METHODS FOR LONGITUDINAL MISSING-AT-RANDOM DATA AND PREDICTORS ASSOCIATED WITH...

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Longitudinal data analysis is crucial for assessing changes over time.
• Missing data and correlated predictors are common challenges in such studies.
• Existing methods like maximum likelihood have limitations with MAR data or correlated random effects.

Purpose of the Study:

• To develop robust statistical methods for longitudinal data analysis.
• To address inconsistencies in existing methods arising from missing data and predictor-covariate correlations.
• To provide an easy-to-use approach for unbiased estimation of change.

Main Methods:

• The study employs theoretical analysis, simulation studies, and fits to example data.
• It focuses on decomposition methods for generalized linear mixed models.
• The methods are designed to handle missing at random (MAR) data and correlated random effects.

Main Results:

• Decomposition methods using complete covariate information yield consistent estimates.
• In practice, these methods often only require observed covariates, simplifying application.
• The proposed approach effectively mitigates bias from both cluster-level confounding and MAR missingness.

Conclusions:

• Decomposition methods offer a reliable solution for longitudinal data with missingness and confounding.
• These methods provide consistent estimates for assessing change, even when random effects correlate with predictors.
• The findings present a practical and accessible approach for unbiased longitudinal data analysis.