Agent-Based Modeling and Simulation in the Social Sciences

The Field of ABM

ABM is a methodology for the study of social phenomena by exploring, through the use of computational models, the relations connecting microbehavior with macro patterns. This methodology is increasing in popularity for three reasons. The first is the increasing technical advances in computing capability since the last decades of the 20th century. The second is the flexibility of programming languages compared with mathematical and logical formalisms. The third is the growing interest in features such as complexity, emergence, self-organization, nonlinearity, and path dependence in the social sciences.

ABM is framed in a wider approach to social phenomena: computational social simulation. Computational simulation allows one to avoid both the vagueness and ambiguity of verbal description and the stringent assumptions of purely mathematical description. It works through the design of a model intended to mimic or replicate the desired [Page 15]social phenomenon in a particular way. The model can then be subjected to experimentation by the manipulation of prespecified parameters.

Computational models to study social phenomena can be divided into three types: macro-simulation, micro-simulation, and agent-based simulation. In the first, models are used to explore system dynamics, a name often used to describe this type of simulation. Such models use sets of difference or differential equations to estimate the behavior of system variables over time. The macro prefix is used because this estimation is performed from a top-down perspective: The target system is simulated as a whole. Individuals are not taken into account in the model but only population-level attributes. These attributes are thought of as causally connected, and causal relations are represented in the model's equations. During the simulation, the equations calculate the values for each variable at every time step, according to the previous value of the variables with causal influence.

In micro-simulation, the estimation also depends on sets of equations, but the opposite approach, bottom-up, is employed. This means that estimation is based not on the values of system variables but on the attributes of the basic units (individuals, households, organizations, etc.). Micro-simulation works by applying transition rules that change these individual attributes as time passes. The individual results are then aggregated into a hypothetical sample that represents the overall population once the simulation is over.

Although different in their approach, micro- and macro-simulations are both simulation techniques in which models are fed with empirical data and in which the main concern is prediction. ABM is grounded in a different principle. As with micro-simulation, ABM uses a bottom-up approach. However, in ABM, the basic units have the possibility to interact. This important feature has channeled the discussion about ABM into two specific topics: (1) the characterization and role of the basic units—that is, the agents in the models—and (2) the way in which the tension between explanation and prediction in social science is approached with ABM.

...