Regression analysis of longitudinal binary data with time-dependent environmental covariates: bias and efficiency.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Regression Modeling

Background:

• Generalized estimating equations (GEE) are standard for marginal regression parameter estimation.
• Valid GEE inference necessitates modeling the response as a function of all covariate values (past, present, future).
• Focusing on 'cross-sectional' models with single covariate lags can lead to biased parameter estimates for time-varying covariates.

Purpose of the Study:

• To investigate the bias-efficiency trade-off in working correlation choices for GEE with binary response data.
• To identify data characteristics influencing parameter estimate performance under different weighting schemes.
• To provide guidance on selecting appropriate covariance models in GEE analyses.

Main Methods:

• Simulation study evaluating parameter estimates under various working correlation models.
• Analysis of data characteristics including cluster size, response association, and covariate distribution.
• Comparison of bias and efficiency across different weighting schemes (working covariance models).

Main Results:

• The choice of covariance model significantly impacts GEE parameter estimates.
• Data characteristics, such as cluster size and association structure, dictate the performance of different weighting schemes.
• No single working correlation model is universally optimal; performance is data-dependent.

Conclusions:

• Careful consideration of data features is crucial when selecting working correlation models in GEE.
• Ignoring the full covariate history can lead to bias, while inappropriate correlation models cause inefficiency.
• The study highlights the importance of understanding the bias-efficiency trade-off for robust GEE analysis.