Guessing preferences: a new approach to multi-attribute ranking and selection

We consider an analyst tasked with using simulation to help a decision-maker choose among several decision alternatives. Each alternative has several competing attributes, e.g., cost and quality, that are unknown but can be estimated through simulation. We model this problem in a Bayesian context, where the decision-maker's preferences are described by a utility function, but this utility function is unknown to the analyst. The analyst must choose how to allocate his simulation budget among the alternatives in the face of uncertainty about both the alternatives' attributes, and the decision-maker's preferences. Only after simulation is complete are the decision-maker's preferences revealed. In this context, we calculate the value of the information in simulation samples, and propose a new multi-attribute ranking and selection procedure based on this value. This procedure is able to incorporate prior information about the decision-maker's preferences to improve sampling efficiency.