One-shot decisions under linear partial information (Q2368038)

The paper discusses the following decision problem: Given is a set \(\{x_ i\}\) of possible strategies, a set \(\{z_ j\}\) of relevant states, a set \(\{p_ j\}\) of probabilities associated with the states and a set \(\{u_{ij}\}\) of utility payoffs defined over strategies and states. An ``optimal'' strategy is to be chosen. The authors consider the problem under the assumption that the decision maker may neglect states with small probabilities and may accept a strategy leading to a high payoff with a probability less than one, but still high. Three cases are considered: (1) The values of \(\{p_ j\}\) and \(\{u_{ij}\}\) are known; (2) The values of \(\{p_ j\}\) are unknown (e.g. because the decision situation is unique, without any precedents -- ``a one-shot decision''). The only information available is a set of linear inequalities satisfied by \(\{p_ j\}\) (``linear partial information''); (3) The values \(\{u_{ij}\}\) are unknown. The only information available is a linear preference ordering defined over the outcomes. The considerations are accompanied by several easy-to-follow examples.