Incorporating Attribute Value Uncertainty into Decision Analysis

Attribute value uncertainty is present in a decision when the consequences are characterized by uncertain estimates of the underlying, unknown true values. Attributes defined based upon measurement data, experimental data, or survey data are subject to attribute value uncertainty. Uncertainty in decision modeling has received significant attention from the community, most notably the methods pertaining to risky decisions and decision ambiguity. These methods, however, proceed with only point estimates representing the attribute values, ignoring the uncertainty in the estimates, which can adversely impact the identification of the best decision alternative. This work describes a new decision analysis method that incorporates attribute value uncertainty by leveraging the systematic mechanism of the Monte Carlo method. This approach allows for the uncertainty in the attribute values to be propagated through the decision model and reflected in the resulting decision parameter. Several techniques, including stochastic dominance and majority judgment, can be used to identify the most desirable alternative based on the resulting uncertain decision parameter. The approach is illustrated by an example driven by a recent U.S. Department of Homeland Security program to investigate detection systems for screening individuals and their baggage for radioactive materials at U.S.-based international arrival airport terminals.