Heterogeneity and Network Structure in the Dynamics of Contagion Comparing AgentBased and.pdfVIP

Heterogeneity and Network Structure in the Dynamics of Contagion: Comparing Agent-Based and Differential Equation Models Hazhir Rahmandad hazhir@mit.edu, John Sterman jsterman@mit.edu System Dynamics Group, MIT Sloan School of Management, Cambridge MA 02142 Abstract When is it better to use an agent based (AB) model, and when should differential equation (DE) models be used? We compare and contrast the dynamics of AB models with those of the corresponding mean-field DE model, using the common and important context of the spread of contagious disease as an example. We compare the well-known SEIR (Susceptible-Exposed-Infected-Recovered) model of contagion, a lumped nonlinear DE system, to an explicit AB model of the same system. We examine both the impact of heterogeneity in agent attributes and the impact of different network structures for the interactions among the agents, including fully connected, random, Strogatz-Watts small world, scale free, and ring-lattice networks. We further show how agent based models can be formulated in continuous time while preserving the stochastic character of state transitions, allowing AB and DE elements to be combined in the same model. Sensitivity analysis demonstrates the conditions under which the added detail of an AB representation leads to different conclusions compared to the aggregated DE model. Introduction Spurred by growing computational power, agent-based modeling (AB) is increasingly applied to problems previously modeled with nonlinear differential equations (DE). Scholars interested in the dynamics of complex systems can now choose from highly disaggregate, agent-based representations to highly aggregated models. Each approach has strengths and weaknesses. AB models can readily include heterogeneity in the attributes of the agents, and in the network structure of their interacti