Technical Policy Briefing Notes - 3
Robust Decision Making
Description of the Method
Policy Briefs
Robust Decision Making
Description of the Method
Robust Decision Making (RDM) is a decision
support tool that can be used in situations of deep uncertainty. RDM is
premised on the concept of “robustness” rather than
“optimality”. The approach was developed to help
policymakers make more informed near-term decisions which have
long-term consequences.
RDM involves testing near-term strategies across a large number of plausible future states. The primary aim is to help policymakers anticipate or mitigate the negative impacts of possible future surprises resulting from the interaction of factors (exogenous uncertainties) outside of their control, with measures that are within their control. This is described as decision making under situations of deep uncertainty, i.e. when little or no probabilistic information is available.
The approach can be applied when traditional risk information (e.g. well-defined probability distributions) is not available, when there is no agreement on the conceptual models to use, or how to evaluate the desirability of alternative outcomes. RDM aims to help decision makers to take robust or resilient decisions today, despite imperfect and uncertain information about the future.
RDM has developed as an analytic, scenariobased approach for strategic decision-making. The formal application of the approach involves the combination of both qualitative and quantitative information through a human and computer-guided modelling interface (Lempert et al, 2003: Groves and Lempert, 2007). This powerful combination surpasses the analytical power of traditional qualitative and quantitative decision support tools. This computer based analysis allows RDM to evaluate how different strategies perform under large ensembles, often of thousands or millions of runs, which reflect different plausible future conditions (Lempert et al, 2003). Iterative and interactive techniques are then applied to “stress test” different strategies, identifying potential vulnerabilities or weaknesses of proposed approaches (Dessai et al., 2009). The formal application of the approach is characterised by the use of data mining algorithms which carry out vulnerability-and-response-option analysis (Groves and Lempert, 2007).
The formal application of RDM involves a series of steps, set out in Figure 1 below (Groves et al., 2008). RDM analysis begins by structuring the problem, but instead of characterising key uncertainties (or more accurately, risks) as a prelude to optimally ranking strategies, the analysis proposes alternative strategies. Uncertainties associated with the parameters defining these strategies are then characterised, assigning a range of uncertainty values for each variable using stakeholder consultation or other approaches.
Each strategy is then assessed over a wide range of (computer-generated) scenario futures. Combinations of uncertainty parameters that are most important to the choice between strategies are statistically derived and a summary of key trade-offs among promising strategies developed (Groves et al, 2008).
Figure 1. The RDM process
Source: Groves et al., 2008
The analysis of strategies across the scenario futures is measured with performance measures, which are used to measure pre-selected, desirable outcomes. Ideally, an RDM analysis helps to identify a robust strategy – one that performs well over a very wide range of scenario futures. In the event that a robust, well performing strategy is not identified, the iterative process of strategy reformulation begins again with stakeholders.
The formal application of RDM tests many strategies against computer-simulated future scenarios, considering exogenous factors (outside the decision maker’s control) and policies or options (within their control), linking these with functional relationships, and assessing strategies against the performance measures in quantitative terms. The former method involves application of statistical or datamining algorithms.
RDM involves testing near-term strategies across a large number of plausible future states. The primary aim is to help policymakers anticipate or mitigate the negative impacts of possible future surprises resulting from the interaction of factors (exogenous uncertainties) outside of their control, with measures that are within their control. This is described as decision making under situations of deep uncertainty, i.e. when little or no probabilistic information is available.
The approach can be applied when traditional risk information (e.g. well-defined probability distributions) is not available, when there is no agreement on the conceptual models to use, or how to evaluate the desirability of alternative outcomes. RDM aims to help decision makers to take robust or resilient decisions today, despite imperfect and uncertain information about the future.
RDM has developed as an analytic, scenariobased approach for strategic decision-making. The formal application of the approach involves the combination of both qualitative and quantitative information through a human and computer-guided modelling interface (Lempert et al, 2003: Groves and Lempert, 2007). This powerful combination surpasses the analytical power of traditional qualitative and quantitative decision support tools. This computer based analysis allows RDM to evaluate how different strategies perform under large ensembles, often of thousands or millions of runs, which reflect different plausible future conditions (Lempert et al, 2003). Iterative and interactive techniques are then applied to “stress test” different strategies, identifying potential vulnerabilities or weaknesses of proposed approaches (Dessai et al., 2009). The formal application of the approach is characterised by the use of data mining algorithms which carry out vulnerability-and-response-option analysis (Groves and Lempert, 2007).
The formal application of RDM involves a series of steps, set out in Figure 1 below (Groves et al., 2008). RDM analysis begins by structuring the problem, but instead of characterising key uncertainties (or more accurately, risks) as a prelude to optimally ranking strategies, the analysis proposes alternative strategies. Uncertainties associated with the parameters defining these strategies are then characterised, assigning a range of uncertainty values for each variable using stakeholder consultation or other approaches.
Each strategy is then assessed over a wide range of (computer-generated) scenario futures. Combinations of uncertainty parameters that are most important to the choice between strategies are statistically derived and a summary of key trade-offs among promising strategies developed (Groves et al, 2008).
Figure 1. The RDM process
Source: Groves et al., 2008
The analysis of strategies across the scenario futures is measured with performance measures, which are used to measure pre-selected, desirable outcomes. Ideally, an RDM analysis helps to identify a robust strategy – one that performs well over a very wide range of scenario futures. In the event that a robust, well performing strategy is not identified, the iterative process of strategy reformulation begins again with stakeholders.
The formal application of RDM tests many strategies against computer-simulated future scenarios, considering exogenous factors (outside the decision maker’s control) and policies or options (within their control), linking these with functional relationships, and assessing strategies against the performance measures in quantitative terms. The former method involves application of statistical or datamining algorithms.
download this briefing note