Bayesian Approaches to Analyzing and Incorporating "Soft Evidence" in Risk Analysis. R. M. McDowell, USDA-APHIS, Risk Analysis Systems, Riverdale, MD; and S. Kaplan, Bayesian Systems, Inc., Rockville, MD
The determination of both model structure and parameter values in quantitative risk and safety analysis relies on the evaluation of a wide variety of evidence. This evidence ranges from "hard" data —such as well-known physical constants and experimental or sample data derived using statistically appropriate study designs—to "soft data" which may be incomplete, non-quantitative, anecdotal, or otherwise not suitable to conventional statistical analysis. The appropriate treatment of hard data is well-defined; the appropriate treatment of "soft data" is not, yet in many cases soft evidence comprises much of what we know of particular systems. In a variety of fields including engineering, hydrology, and medicine, a Bayesian approach has led to the development and adoption of methods to incorporate data of varying types, resulting in a quantitative representation of our total knowledge of the evidence. The Bayesian approach provides a logical and transparent way to incorporate hard and soft data in our risk assessments. Further, it forces analysts to describe explicitly (and quantitatively) our state of knowledge and likelihood of specific pieces of evidence, given this knowledge, and provides the calculus for fusing these into a posterior distribution. Approaches to developing distributions that specify our prior knowledge and likelihood functions in soft data environments are discussed in reference to agricultural safety assessment. The software environment for Bayesian analysis is also discussed.
Authors thank Ali Mosleh for his contributions.
Go to . . .
1999 SRA Table of Contents
1999 SRA Author Index
Main Abstracts Menu Page
RiskWorld Home Page