Related Experiment Videos
Marginal analyses of clustered data when cluster size is informative
John M Williamson1, Somnath Datta, Glen A Satten
1Division of HIV/AIDS Prevention, National Center for HIV, STD and TB Prevention, Centers for Disease Control and Prevention, MS E-37, 1600 Clifton Road, NE, Atlanta, Georgia 30333, USA. jow5@cdc.gov
Biometrics
|May 24, 2003
Summary
We introduce a novel weighted generalized estimating equation (GEE) method for analyzing clustered data with informative cluster sizes. This approach offers superior performance and efficiency compared to traditional methods, especially in smaller sample sizes.
Area of Science:
- Statistics
- Biostatistics
- Data Analysis
Background:
- Clustered data analysis presents challenges when cluster size influences outcomes.
- Traditional methods like unweighted generalized estimating equations (GEE) may not adequately account for informative cluster sizes.
Purpose of the Study:
- To develop and evaluate a new approach for fitting marginal models to clustered data where cluster size is informative.
- To compare the proposed method with existing techniques, including unweighted GEE and within-cluster resampling (WCR).
Main Methods:
- A novel GEE approach weighted inversely by cluster size was developed.
- The proposed method was compared to unweighted GEE and WCR using simulated data and a dental health dataset.
- Asymptotic equivalence with WCR was theoretically established.
Main Results:
- The proposed weighted GEE method demonstrated superior performance over unweighted GEE.
- Equivalence between the weighted GEE and WCR was observed for large sample sizes.
- The weighted GEE method outperformed WCR in scenarios with small sample sizes.
Conclusions:
- The proposed weighted GEE approach provides an efficient and effective method for analyzing clustered data with informative cluster sizes.
- This method offers a practical alternative to computationally intensive techniques like WCR, particularly in smaller datasets.