Validation and verification of agent-based models in the social sciences

Abstract

This paper considers some of the difficulties in establishing verificaction and validation of agent based models. The fact that most ABMs are solved by simulation rather than analytically blurs the distinction between validation and verification. We suggest that a clear description of the phenomena to be explained by the model and testing for the simplest possible realistic agent rules of behaviour are key to the successful validation of ABMs and will provide the strongest base for enabling model comparison and acceptance. In particular, the empirical evidence that in general agents act intuitively rather than rationally is now strong. This implies that models which assign high levels of cognition to their agents require particularly strong justification if they are to be considered valid. © Springer-Verlag Berlin Heidelberg 2009.