Graphical model checking with correlated response data

Area of Science:

• Biostatistics
• Statistical modeling
• Biomedical data analysis

Background:

• Correlated response data are common in biomedical research, necessitating robust analytical methods.
• Generalized Estimation Equation (GEE) is a standard approach for analyzing such data.
• Existing methods for checking GEE model adequacy are limited.

Purpose of the Study:

• To propose novel graphical methods for assessing the adequacy of regression models within the GEE framework.
• To introduce a bootstrap approach for quantifying uncertainty in marginal mean function comparisons.
• To adapt these methods using generalized additive models (GAMs).

Main Methods:

• A graphical method based on Cook and Weisberg's marginal model plot.
• Application of a bootstrap method to generate reference bands for uncertainty assessment.
• Extension of the methodology using generalized additive models (GAMs).

Main Results:

• The proposed graphical methods are easy to implement using existing software for independent data.
• The methodology was successfully applied to a correlated binary data set from the Lung Health Study.
• The methods provide a reliable way to assess statistical uncertainties in comparing marginal mean functions.

Conclusions:

• The developed graphical methods offer practical solutions for evaluating GEE model fit in biomedical studies.
• These approaches enhance the reliability of regression analyses involving correlated data.
• The integration of GAMs provides a flexible alternative for model adequacy checks.