On the Importance of Frailty in Social Science Theory (and other lessons of agent-based modeling)

Abstract

This paper is about the theoretical implications of agent-based modeling exercises. Construction of an agent-based model challenges a social scientist to formalize many concepts and relationships that would have remained implicit or unrecognized. While formalizing these "unimportant" assumptions can be a nuisance, it can also have substantial theoretical payoffs. In order to fill the gaps of the model, the researcher is forced to confront the gaps in the theory that motivated the model in the first place. Using examples drawn from several large political science simulation models, the paper argues that frailty, defined as unpredictability in the behavior of agents, is often required in order to bring closure to the modeling exercise. It is difficult (or impossible) to square the dynamic or aggregate implications of the agent-based model with observations without placing a substantial amount of emphasis on frailty. Hence, the component in behavior that we often treat as "error" in empirical analysis is actually a vital part of the glue that makes the many different moving parts of a social system interact in coherent ways. The example models were developed with the Swarm simulation system (http://www.swarm.org) during the last decade.