Computer simulation is blurring the long-standing distinction between two broad categories of science, based on the relation between theory and data.

Some sciences, such as molecular biology, solid-state physics, and chemistry, can move in both directions between data and theory: theory is developed as an abstraction of observed data, and then experiments are devised to gather more data under constraints that test the theory. Other sciences, such as historical geology, meteorology, and evolutionary biology, move from observed data to theory, but cannot easily construct experiments. In principle, one can construct experiments to test sociological theories that involve a few people, but such experiments often run afoul of ethical concerns, and are impractical if one is studying the dynamics of large firms or cities. Thus, sociology has tended to be more like meteorology than solid-state physics.

Computer simulation enables researchers to construct and explore “possible worlds” in domains where conventional experimentation is not feasible. This approach emerged first in domains whose behavior can be expressed analytically. A classic example is the set of meteorological differential equations by Edward Lorentz, in 1960, that marked the birth of the modern interest in chaos theory. In aggregate, one can model social systems with differential equations (the heart of the systems dynamics approach), but it is much more natural to model the behavior of individuals with rules than with systems of flows. The advent of object-oriented, and later agent-oriented, programming, and the increased power and lower cost of computer hardware have led to an explosion in modeling techniques for social and other problems.

The first edition of this work [1] was an invaluable roadmap through the jungle of computer modeling and simulation. The authors distinguished seven different techniques on the basis of whether they captured the interaction of multiple levels (essential for modeling emergence), whether they allowed distinct agents to communicate, the complexity of agents, and the number of agents. These techniques were system dynamics, microsimulation, queueing models, multilevel simulation, cellular automata, distributed artificial intelligence (AI), and learning models. The authors devoted a chapter to each of these. The book’s emphasis was very much hands-on, directing the reader to sources for free or inexpensive modeling software, and presenting sample code that the reader could modify to gain experience with a range of applications.

This second edition preserves the proven organization and practical emphasis of the first, with four changes, all for the better. First, the references to “distributed AI” in the first edition have been replaced with “multiagent systems,” following the evolution of terminology in the research community. Second, to the original chapter that describes multi-agent models, the authors have added a chapter on how one develops such a model. The additional emphasis given to this class of model reflects its particular importance for the sociological domain. Third, the authors have changed the environment in which they present the examples in the latter half of the book (those dealing with cellular automata and multiagent systems). In the first edition, they encouraged readers to install and use Lisp-Stat, a free dialect of Lisp intended for statistical computation, but capable of constructing reasonable displays and user interfaces for social models. The second edition translates these examples to NetLogo, a much friendlier and more powerful environment (that is, interestingly, also based on Lisp). (The book uses version 2 of NetLogo. Interestingly, the current version 3 includes a systems dynamic modeler that could also be used to implement the examples in chapter 3.) Finally, the bibliography has grown by more than 30 percent, to include work published since the first edition appeared.

With this updated edition, this work will continue to be the preeminent manual on computer modeling for social science, appropriate for classroom use and individual study.