Designing, Formulating, and Communicating Agent-Based Models

A.J. Heppenstall et al. (eds.), Agent-Based Models of Geographical Systems, DOI 10.1007/978-90-481-8927-4_17, © Springer Science+Business Media B.V. 2012 Abstract Designing, formulating, and communicating agent-based models (ABMs) poses unique challenges, especially how to choose the right level of c mplexity and how to describe – and, even, think about – a model in a way that captures both its essential characteristics and its complete detail. Two techniques for dealing with such challenges have become established among agent-based modellers. The “ODD” (Overview, Design concepts, Details) protocol is a standard for describing ABMs in publications, but also provides design patterns for the model developer. ODD starts with an overview of what the model is and does, and then describes how the model implements ten “design concepts” that capture essential and unique characteristics of ABMs. Last come all the details needed to completely replicate the model. “Pattern-oriented modelling” (POM) is a set of strategies for using patterns observed in the systems to ensure that an ABM captures the right “essence” of the system. POM starts with identifying multiple patterns of behaviour in the real system and its agents that seem to capture the essential internal mechanisms for the problem being modelled. These patterns are then used to decide what kinds of entities, state variables, and processes need to be in the model; compare and test alternative “theory” for key agent behaviours; and filter potential parameter values to limit uncertainty. ODD and POM are important steps toward the acceptance of agent-based approaches as established, credible ways to do science.