Robust Decision Support Methods: A Comparative Analysis

Methods for decision support in the context of deep uncertainty have been gaining interest in the context of complex and adaptive problems that are characterized by “tipping point” behaviours. Unlike a traditional “predict-then-act” methods, which determine policies based on specific predictions of future behaviour, these decision support methods seek policies that are robust to inherently uncertain future conditions. Three robust decision support methods have been identified: multi-objective robust decision making (MORDM), multi-scenario MORDM, and multi- objective robust optimization (MORO). These three methods are all based on the robust decision making (RDM) structure, but search for solution alternatives in different ways. While MORDM considers a single absolute performance measure when selecting solution alternatives, multi- scenario MORDM uses multiple references to determine performance in an effort to yield more robust solution alternatives. Finally, MORO directly optimizes alternatives for robustness to determine a set of potential solutions, making the most effort to consider robustness of all three methods.It has been previously assumed that the performance of each method is largely dependent on the behaviour of the problem and structure of the desired solution, but there has been no comparison made in literature to test that idea. This study proposes framework for comparing decision support methods for multiple problem and solution structures and uses that framework to compare the efficacy of all three robust decision support methods given a highly stylized problem that is commonly used in deep uncertainty research: the lake problem. Three solution implementation structures will be considered: a static structure known as the intertemporal variation, a dynamic structure known as the direct policy search variation (DPS), and a newly proposed variation named the planned adaptive DPS variation that aims to better capture real-world implementation properties.The results indicate that generally, independent of policy implementation structure, the more that robustness is incorporated into the early stages of analysis, the more robust the identified set of potential solutions will be. The increase in robustness comes with a sharply increasing computational cost, with MORO quickly becoming prohibitively expensive to run. The proposed planned adaptive DPS lake problem variant was also shown to yield a set of solution alternatives that do not weigh conflicting objectives as evenly as the other two variants, leading to an extremely conservative set of solution alternatives, especially in a multi-scenario MORDM analysis. Therefore, the development of a hybrid robust decision support method that can consistently find the most robust solutions while evenly weighing conflicting objectives and minimizing the computational cost is much needed. Until that time, analysts and decision makers will be able to use this framework to compare existing robust decision support and select the method that best fits the needs and limitations of a problem analysis.Engineering and Policy Analysi