Related Experiment Videos
Effects of variance-function misspecification in analysis of longitudinal data.
1Department of Statistics and Applied Probability, National University of Singapore, 3 Science Drive 2, Singapore 117546. stawyg@nus.edu.sg
Biometrics
|July 14, 2005
Summary
Generalized estimating equations (GEE) analysis is sensitive to variance function specification. Correctly specifying the variance function improves estimation efficiency for quantitative responses, even with misspecified correlations.
Area of Science:
- Statistics
- Biostatistics
- Quantitative Methods
Background:
- Generalized estimating equations (GEE) extend generalized linear models.
- GEE models within-subject correlations using a working correlation matrix.
- Variance function is typically assumed to be a known function of the mean.
Purpose of the Study:
- Investigate the impact of misspecifying the variance function in GEE.
- Analyze effects on mean parameter estimators for quantitative outcomes.
- Evaluate the interplay between variance and correlation structure misspecification.
Main Methods:
- Numerical simulations were conducted to assess estimation efficiency.
- Generalized estimating equations (GEE) were applied to model quantitative responses.
- A real dataset on cow growth was used for illustration.
Main Results:
- Correct variance function specification enhances estimation efficiency, irrespective of correlation structure accuracy.
- Misspecifying the variance function disproportionately affects within-cluster covariate estimators compared to cluster-level covariates.
- Accurate correlation structure selection may not improve efficiency if the variance function is misspecified.
Conclusions:
- The variance function is a critical component in GEE modeling.
- Careful consideration of the variance function is essential for robust estimation.
- Findings highlight the importance of appropriate variance function selection in GEE analyses.