Practicality of Agent-Based Modeling of Civil Violence : An Assessment

Joshua Epstein (2002) proposed a simple agent-based model to describe the formation and evolution of spontaneous outbursts of civil violence (such as riots or violent demonstrations). Epstein's model uses two types of agents (“populace” and “cops”); and each populace agent is either active or quiescent according to its level of “grievance”, its aversion to risk and the relative proportion of active agents and cops within a certain neighborhood of that agent. In this paper, we go a step beyond Epstein's study, and consider the application of the model to practical situations. First, we provide an improved mathematical characterization of the model dynamics that reduces the number of free parameters in the model. We show by simulations that there are no “tipping points” and the model behavior changes smoothly as a function of parameters. Using fundamental mathematical properties we explain the size and frequency distributions of outbursts, and confirm our results with simulation. We also argue that the model does not accurately reflect the mechanisms that lead to riot formation, and that nominal model parameters (such as “hardship”, “legitimacy”, and “average jail term”) are proxies for complicated social processes; hence we conclude that the model is better suited as a phenomenological description rather than a theoretical explanation of crowd violence. Finally, we present a method for fitting the model to practical scenarios based on the frequency and severity of outbursts