Related Experiment Videos
Comments about Joint Modeling of Cluster Size and Binary and Continuous Subunit-Specific Outcomes.
1Division of Biostatistics, Department of Epidemiology and Public Health, Yale University School of Medicine, 60 College Street, New Haven, Connecticut 06520, USA. ralitza.gueorguieva@yale.edu
Biometrics
|September 2, 2005
Summary
This study presents a maximum likelihood approach for jointly modeling clustered binary and continuous outcomes, offering a simpler alternative to Bayesian methods. It resolves discrepancies in previous analyses and highlights the importance of accounting for cluster size in statistical modeling.
Area of Science:
- Biostatistics
- Statistical Modeling
- Longitudinal Data Analysis
Background:
- Joint modeling of clustered binary and continuous outcomes is crucial in longitudinal studies.
- Bayesian approaches (Dunson, Chen, and Harry, 2003) exist but can be complex.
- Standard software offers a more accessible maximum likelihood estimation method.
Discussion:
- This note demonstrates using SAS PROC NLMIXED for maximum likelihood estimation in a joint model for clustered outcomes.
- A more general model with additional random effects is proposed for improved data fit.
- Discrepancies between Bayesian (DCH) and frequentist (Catalano & Ryan, 1992) estimates are reconciled.
Key Insights:
- Maximum likelihood estimation via standard software simplifies joint modeling of mixed-type clustered data.
- Ignoring cluster size can lead to biased inferences, particularly for dose-effect relationships.
- The proposed maximum likelihood approach is broadly applicable and requires less specialized programming.
Outlook:
- This method facilitates robust analysis of complex clustered and longitudinal data.
- It provides a practical alternative for researchers without extensive Bayesian programming expertise.
- Further applications in developmental toxicity and other fields are anticipated.