Parametric empirical Bayes estimation for a class of extended log-linear regression models
✍ Scribed by Wanzhu Tu; Walter W. Piegorsch
- Publisher
- John Wiley and Sons
- Year
- 2000
- Tongue
- English
- Weight
- 155 KB
- Volume
- 11
- Category
- Article
- ISSN
- 1180-4009
No coin nor oath required. For personal study only.
✦ Synopsis
This paper presents a fully parametric empirical Bayes approach for the analysis of count data, with emphasis on its application to environmental toxicity data. A hierarchical structure for the mean response is developed from a generalized linear model, based on a Poisson distribution. The linear predictor is embedded at the prior level of the hierarchy. This allows for enhanced ¯exibility when accounting for extra-Poisson variation, which is often displayed with count data from environmental bioassays. The model expands upon the traditional log-linear model in two dierent ways: (1) it extends the Poisson distributional assumption; and (2) it incorporates an extended family of link functions that includes the log link as a special case. The main advantage of this approach is that it combines relative computational simplicity with hierarchical modeling ¯exibility. In this paper, we emphasize the model's development and the practical issues related to the analysis. We describe an application of the proposed model to data from an environmental mutagenesis experiment.