Two visualization tools for analyzing agent-based simulations in political science.
Published in Computer Graphics and Applications, 2012
Abstract: Research in the social and behavioral sciences often uses agent-based modeling (ABM) to simulate and explore behaviors such as collaboration, conflict, violence, and population change. Researchers have also used agent-based models to identify a country’s political patterns, which might indicate the imminence of civil unrest and help predict catastrophic events. ABM represents a behavioral system as a collection of autonomous entities or agents. Each agent interacts with other agents according to a set of rules and goals, and over time it might influence and be influenced by the agents around it.
As increased computing power becomes more widely available, scientists can simulate more complex systems, which in turn generate substantially larger datasets that must then be analyzed and interpreted. Unfortunately, the methods and tools available to social scientists for analyzing simulation results can’t support datasets of such magnitude, making it difficult for scientists to effectively interpret and analyze the results. In addition, ABM is a stochastic simulation technique, using small random perturbations to the interaction rules and running each simulation hundreds or even thousands of times to avoid local minima and to generate a distribution of sample behavioral patterns. Analysts must therefore be able to compare simulated behaviors between and across distinct runs and to piece together many simulation runs into a single, cohesive overview.
To help address these challenges, we collaborated with domain experts to develop two visual analytics tools to support analysis of agent-based models in political science. With MDSVis (Multidimensional Scaling Visualization), analysts can explore the simulation space, finding similar patterns at an aggregated level and finding the dominant factors affecting agent behavior. With SocialVis, analysts can focus on one simulation run, exploring relationships between time steps and geographic regions.
Recommended citation: R. Jordan Crouser, Daniel Kee, Dong H. Jeong, and Remco Chang. Two visualization tools for analyzing agent-based simulations in political science. IEEE Computer Graphics and Applications, 32(1):67–77, 2012.