Types of agent-based simulation models

A useful way of classifying agent-based simulation models is to consider their purpose and their reliance upon external data as, for example, done by Heath et al. [1]:

  • Generators: Theoretical models with the overall goal of confirming behavioural hypotheses and assumptions and identifying mechanisms
  • Mediators: Models that incorporate both behavioural assumptions and real-world data with the overall goal of identifying mechanisms to a certain extent and performing short-term predictions
  • Predictors: Purely data-driven models with the overall goal of performing medium- to long-term predictions

Social science research typically deals with generator models; classic examples are Schelling’s model of racial segregation [2] and Epstein’s Sugarscape model [3]. Simulations of technical systems such as robot swarms as well as models of consumer marketplaces which are highly driven by (historical) data, on the other hand, can be attributed to the third category. It is important to note that the categories are not mutually exclusive and should rather be seen as a continuous spectrum. Generator models typically have a strong explanatory flavour, whereas predictors—as the name suggests—mostly focus on prediction and forecasting.


[1] B. Heath, R. Hill, and F. Ciarallo. A survey of agent-based modeling practices (January 1998 to July 2008). Journal of Artificial Societies and Social Simulation, 12(4):9, 2009.

[2] T. C. Schelling. Models of segregation. American Economic Review, 59(2):488–93, May 1969.

[3] J. M. Epstein and R. L. Axtell. Growing Artificial Societies: Social Science from the Bottom Up. MIT Press, June 1996.