Error in Decision Analysis:How to Create the Possibility of Large Losses by Using Dominated Strategies.

摘要

This report examines some concepts, sources, and possible consequences of error in decision analysis. Recent articles on the possibilities for error in decision analysis showed that under some relatively mild assumptions deviations from optimal decision strategies or from optimal model parameters will lead only to minor losses in expected value. This 'flat maximum' property of decision analytic models applies, however, only to admissible decisions. By inadvertently selecting a dominated (inadmissible) decision, the decision maker creates the possibility for large expected losses. Usually dominance can be recognized and losses can be avoided by elimination of dominated decisions. Unfortunately, for a large class of errors the discovery of dominance is difficult if not impossible. These errors consist of failing to use information or using it inappropriately in decision strategies. The main point this report makes is that such errors can, and typically will, lead to dominated strategies, and so can lead to substantial expected losses.