The value of information from multiple sources of uncertainty in decision analysis

Abstract

In decision analysis, the expected value of information is a well known and used concept, where the value of information about sources of uncertainty are normally considered one at a time. When information is potentially available for more than one source of uncertainty in a decision problem and the decision maker must choose which sets of information to acquire, the nonlinearities and discontinuities which are inherent to the problem result in general non-additivity of information values. Additivity of the expected value of perfect information (EVPI) occurs only under certain conditions, and for a large proportion of decision problems, EVPIs are not even approximately additive. This paper presents an analysis of this class of problem, some fundamental results, practical implications and delineates a number of useful directions for further research.