594
Statistical Decision Theory
for Environmental Remediation
Lara J. Wolfson, Joseph B. Kadane, and
Mitchell J. Small
Abstract:
The use of loss functions for constructing a statistical decision
making framework is demonstrated through the case study of a former
battery recycling facility in Pennsylvania. Toxic lead contamination
of soil had occurred, and remediation was therefore mandated by the
EPA. A Bayesian model is proposed that uses covariate and prior
information to address the latent variable problem of distinguishing
background soil lead concentrations from plant contamination. The
results from this model are illustrated with a variety of loss
functions, formulated both from the perspective of the plant and of
the EPA, to create a framework that incorporates uncertainty for
deciding which properties near the facility are eligible for
remediation, while allowing each party in the decision-making process
to understand the implications of their decisions for the other
party. This approach can easily be adapted to many types of
environmental risk or similar public policy problems where uncertainty
is present and multiple stateholders have different perspectives on
potential losses or benefits of different decisions and outcomes.
KEYWORDS:
Loss Functions; Latent Variables; Lead Contamination;
Empirical Bayes; Laplace Approximation; Hazardous Waste; Site
Remediation.