Marginalized models for moderate to long series of longitudinal binary response data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Marginalized models allow likelihood-based inference for longitudinal binary response data.
• Existing models include marginalized transition and latent variable models, capturing serial or exchangeable dependence respectively.

Purpose of the Study:

• To extend marginalized models by proposing a unifying model for both serial and long-range dependence.
• To provide a flexible framework for longitudinal analyses with numerous repeated measurements per subject.
• To identify and model both serial and exchangeable forms of response correlation.

Main Methods:

• Development of a single, unifying marginalized model.
• Application of maximum likelihood and Bayesian approaches for parameter estimation and inference.
• Investigation of large sample operating characteristics under model misspecification.

Main Results:

• The proposed unified model effectively describes both serial and long-range dependence in longitudinal binary data.
• The study provides insights into the performance of the model under different dependence structures.
• Analysis of the Madras Longitudinal Schizophrenia Study demonstrates practical application.

Conclusions:

• The unified marginalized model offers a powerful tool for analyzing complex correlation patterns in longitudinal binary data.
• This framework enhances statistical inference capabilities for studies with substantial repeated measures.
• The findings have implications for understanding disease progression and treatment effects in longitudinal studies.