Robust Statistical Inference via Model Uncertainty and Model Averaging in Risk Assessment

• Activity Number: 156
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Hojin Moon*+ and Steven B. Kim and Ralph L. Kodell
• Companies: California State University, Long Beach and University of California, Irvine and University of Arkansas for Medical Sciences
• Keywords: Benchmark dose ; diversity ; dose-response modeling

Abstract

The ultimate goal of this research is to provide risk assessors an algorithmic method for model subset selection and model averaging in toxicology and risk assessment to understand and prevent human and environmental risks. In chemical and microbial risk assessments, careful dose-response modeling is emphasized because a target risk level of infection or illness is often in the low benchmark response of 1% to 10%. Although multiple dose-response models may plausibly fit the data in the experimental range, each model behaves uniquely at low doses below the experimental range which lead to biased estimation of infectious/benchmark doses under model specifications. In this respect, model averaging is more robust than relying on any single dose-response model in the calculation of a point estimate and a lower confidence limit for an effective dose or equivalently benchmark dose of infection or illness. In model averaging, accounting for both data uncertainty and model uncertainty is crucial, and proper variance estimation is not guaranteed simply by increasing the number of models in a model space. This research addresses model uncertainty at low doses and enhances diversity and flexib