Related Experiment Videos
Approximate hierarchical modelling of discrete data in epidemiology
1Department of Biostatistics, University of Washington, Seattle 98195-7232, USA. norm@biostat.washington.edu
Statistical Methods in Medical Research
|April 9, 1998
Summary
Hierarchical models in epidemiology are useful for analyzing related risks. Penalized quasilikelihood methods offer better accuracy than empirical transform methods when dealing with small sample sizes in disease rate analysis.
Area of Science:
- Epidemiology
- Biostatistics
Background:
- Hierarchical models are essential for estimating and analyzing multiple, related relative risks in epidemiology.
- Applications include meta-analyses of 2x2 tables and mapping of spatially correlated disease rates.
Purpose of the Study:
- To compare the performance of empirical transform and penalized quasilikelihood procedures in hierarchical models.
- To evaluate these methods under varying cell frequency conditions, particularly small cell frequencies.
Main Methods:
- Utilized mixed model analysis programs to implement both empirical transform and penalized quasilikelihood procedures.
- Conducted simulation studies to assess the accuracy of variance component and regression coefficient estimates.
Main Results:
- Both empirical transform and penalized quasilikelihood provide satisfactory approximate inferences for large cell frequencies.
- Penalized quasilikelihood demonstrated superior performance in estimating variance components and regression coefficients under small cell frequencies.
Conclusions:
- Penalized quasilikelihood is a more robust method for hierarchical modeling in epidemiology when dealing with small cell frequencies.
- Standard mixed model software can be used for implementing these statistical procedures.