Related Experiment Videos
Goodness-of-fit methods for generalized linear mixed models.
1Department of Biostatistics, University of North Carolina, CB 7420, McGavran-Greenberg Hall, Chapel Hill, 27599-7420, USA.
Biometrics
|January 13, 2006
Summary
We developed new graphical and numerical methods to check generalized linear mixed models (GLMMs). These techniques objectively assess model adequacy, ensuring reliable statistical analysis and leading to improved medical study models.
Area of Science:
- Statistics
- Statistical Modeling
Background:
- Generalized linear mixed models (GLMMs) are widely used but require rigorous adequacy checks.
- Assessing GLMM fit is crucial for reliable data interpretation and model-based predictions.
Purpose of the Study:
- To develop novel graphical and numerical methods for assessing the adequacy of generalized linear mixed models (GLMMs).
- To provide objective tools for detecting model misspecification in GLMMs.
Main Methods:
- Methods based on cumulative sums of residuals over covariates or predicted values.
- Utilizing zero-mean Gaussian processes and Monte Carlo simulation for null distribution approximation.
- Comparing observed residual processes with simulated realizations for visual and analytical assessment.
Main Results:
- Proposed goodness-of-fit tests demonstrate proper statistical sizes.
- Tests show high sensitivity to various forms of model misspecification.
- Methods are particularly effective for checking covariate functional forms and link functions.
Conclusions:
- The developed methods offer an objective approach to evaluating GLMM adequacy.
- These techniques can identify model misspecification, leading to more accurate statistical inferences.
- Application in medical studies resulted in demonstrably improved models.