Agent-Based Modeling of Life or Death Decisions Following an Urban Biological Catastrophe

December 6, 2015 | Arlington, VA

Speaker: David Pynadath
Host: Society for Risk Analysis (SRA) 2015 Annual Meeting

To make informed policy decisions on how to both prepare and respond to disasters in urban environments, we need to understand the decision making of the local residents who are affected. Their responses to both the disaster and the policies implemented will shape both the immediate impact as well as the long-term prognosis for recovery. Simulation methods provide a powerful method for analyzing the possible outcomes to hypothetical scenarios, where analysts can explore the interactions among possible disasters, policy decisions, and resident behaviors. Agent-based modeling offers the promise of simulations that use computational models that capture the relevant uncertainties and objectives that drive resident decision-making in these scenarios. Unfortunately, constructing models that represent the complexity of human-decision making with any fidelity is often a time-consuming, trial-and-error process. In this work, we construct an automated pipeline that constructs agent models that capture the beliefs, objectives, and behaviors represented by people’s responses to surveys within a simulated disaster scenario. We use data gathered in a prior study using a video simulation of news reports on an evolving biological terrorist attack scenario. Our model construction process takes this data and extracts decision models in the form of PsychSim agents, using a multiagent framework for social simulation. PsychSim agents represent the beliefs and objectives of the survey respondents in decision-theoretic terms. We can thus compare the extracted PsychSim agent models against the original decision models manually extracted from the data. More importantly, we can then use the agents to conduct simulations and profile possible outcomes under hypothetical conditions not included in the original simulated scenario.