A Bayesian Decision Theoretic Approach for Sample Design to Support Risk Assessement. P. K. Black, Newton, Neptune and Company, 1505 15th Street, Suite B, Los Alamos, NM 87544; J. L. Lewis, Radian Corporation, 1821 Carlisle, Albuquerque, NM 87110; and C. A. Newton, Los Alamos National Laboratory, Los Alamos, NM 87544
The case studies presented in this paper provide examples of how expert opinion can be characterized so that it is available for statistical design in support of risk assessment decisions. Expert judgment concerning the distribution of contaminants is modeled formally, through a Bayesian statistical model, to provide a rational, defensible basis for decision making. Statistical models of expert judgment are developed through an elicitation session, in which the decision makers and technical experts provide information at a cognitively appropriate level. The required design inputs fulfill the needs of the prior distribution for the parameters of interest, the associated loss function, and the cost of sampling function. This approach provides the necessary information, through application of Bayesian statistical decision theory, for determining the need for sampling and the optimal number of samples. The optimal sample size is achieved when the value of information gained from taking one more sample is outweighed by the cost of taking that sample. This approach is applied to several site case studies in which human health risk is evaluated from metals contamination of soil at the Los Alamos National Laboratory.