This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).
Scenario Discovery is a valuable approach to participatory, computer-assisted scenario development. It is useful for many hard, deeply uncertain policy problems such as climate change and energy planning. This project aims to advance scenario discovery methodology by applying novel algorithms that can provide significant new capabilities and then evaluating how decision makers respond to the results. It, thus, will simultaneously improve both the technical capacity of the modeling and the ability for decision makers to work effectively with the models.
To be more specific: Scenarios represent sets of plausible future states of the world that represent vulnerabilities of proposed policies, i.e., cases where a policy fails to meet its performance goals. Scenario discovery characterizes such sets by helping users to apply statistical or data-mining algorithms to databases of simulation-model-generated results in order to identify easy-to-interpret combinations of uncertain model input parameters highly predictive of these policy-relevant cases. The approach has already proved successful in several high impact policy studies, but its broad application to many policy areas is currently limited by the capabilities of current scenario discovery algorithms and the existing understanding of how decision makers interpret their results. This project will improve the algorithmic capabilities of scenario discovery by identifying and modifying existing adaptive sampling algorithms to address sparse sampling issues and by employing Principal Component Analysis methods to transform the simulation generated-databases into new coordinate systems that existing scenario discovery algorithms can interpret more successfully. The project will also improve the understanding of how decision makers can best use scenario discovery by conducting a series of workshops and interviews with decision makers to: i) assess the degree to which the proxy measures of interpretability used by the algorithms correspond to results that decision makers can actually interpret; and ii) evaluate decision makers? views on the usefulness of scenarios of different levels of complexity. The workshop results will guide the development of the algorithms considered in the project.
