Abstract

Characterizations of Decision Analytic Models Restricting reasoning to only the structure of influence diagrams yields models that may be too weak to make many decisions of interest. A few research efforts have explored intermediate levels of representation that give more than simply structure but demand less than full numbers. Wellman [38] defines qualitative influences among random variables in terms of stochastic dominance of their distribution functions, and defines a qualitative version of synergy and anti-synergy among joint influences on a variable. Because these definitions capture only the direction of influence, not its magnitude, the resulting models are clearly not useful to resolve detailed tradeoffs among competing decision policies. Inferences with the qualitative models can, however, derive the kinds of qualitative results needed in planning or critiquing plans. For example, Wellman's planner is able to reject large regions of the space of all treatment plans by recogn...

Citations