Before selecting an alternative based on attribute values that have been determined through a measurement process that has error, a decision-maker can, in some cases, collect additional data to reduce uncertainty. Unlike previous work in the area of ranking and selection, this study considered the problem of allocating limited data collection resources across multiple attributes rather than across multiple alternatives. In this work we assumed the multiple attribute values are measurements (samples) of physical characteristics and have a normally distributed measurement error. We conducted a simulation study to investigate how the sample allocation affects the likelihood that the best alternative will be selected and how this relationship is influenced by the relative importance of the attributes and the amount of attribute value uncertainty. The results suggested allocation rules based on the decision model and the general shape of the frontier. These rules were compared to a default rule that allocated the experimental budget evenly across the attributes. Better allocations increase the likelihood that the best alternative will be selected.
Proceedings Title: Resource Allocation for Selection Decisions with Measurement Uncertainty
Conference Dates: May 31-June 3, 2014
Conference Location: Montreal, Québec, -1
Conference Title: 2014 Industrial and Systems Engineering Research Conference
Pub Type: Conferences
Ranking and selection, experimental design, simulation, multicriteria decision making, decision analysis.