Stochastic dominance with linear partial information

Abstract This paper is concerned with expanding the range of application of stochastic dominance as a basis for choosing between alternative decision strategies. Unlike most work in this area, it does so on the assumption that the decision maker is not willing to specify a unique subjective probability distribution for future states of nature, but is only able to articulate a fuzzy, inexact set of beliefs, which may be summarized by appropriate linear constraints on probabilities of events. It is shown that the concept of stochastic dominance readily transfers to this decision environment and that the relevant calculations are quite straightforward. Additionally, the requirements for stochastic dominance and for statistical dominance are compared; the existence of the former is shown always to imply the latter.