Choosing marginal or random-effects models for longitudinal binary responses: application to self-reported disability...

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Longitudinal studies with binary repeated outcomes are increasingly common in epidemiology.
• Standard statistical methods often prove inadequate for analyzing such data.
• Challenges exist in modeling binary repeated responses effectively.

Purpose of the Study:

• To compare marginal and random-effects models for longitudinal binary data.
• To investigate the impact of choosing the time point origin in statistical models.
• To analyze self-reported disability in older women over a 6-year period.

Main Methods:

• Utilized a 6-year longitudinal dataset of self-reported disability in older women.
• Compared marginal and random-effects models, considering baseline response as a covariate.
• Analyzed the influence of time, age, and individual risk factors on disability.

Main Results:

• Significant differences were observed between parameter estimates from marginal and random-effects models.
• These discrepancies stem from conceptual differences between the models.
• The analysis revealed substantial between-individual heterogeneity in disability.

Conclusions:

• A random-effects model is identified as the most appropriate statistical approach.
• This model effectively captures the complexities of self-reported disability in older women.
• The findings highlight the limitations of standard methods for this type of data.