Pitfalls in the Development of Agent-Based Models in Social Sciences: Avoiding Them and Learning from Them

Abstract

The references on the principles and methodology for developing agent-based models of social phenomena usually describe general principles and illustrate the process using worked examples, but seldom focus on the pitfalls and errors that make practical model building a tortuous and difficult task. This chapter contains a discussion of the positive and negative aspects of my personal experience in a PhD work on simulation of large scale social conflict. The purpose will be to describe the process from the initial plan to the final dissertation, analyze the pitfalls and their overcoming in terms of principles of model development, and summarize the ideas that I found useful for practical development of agent-based models of social phenomena. The most serious pitfalls usually occur at the conception and design stages, when seemingly trivial points can be easily overlooked. These include starting with excessive ambition but unclear ideas on whether the purpose is understanding or prediction (i.e. what is the level of abstraction), poor knowledge of the relevant theories, and failure to identify which entities, variables and mechanisms must be considered. Several practical hints for avoiding these issues are presented, such as writing a reduced version of the “Overview, Design Concepts and Details” template that includes the bare minimum of items for a first working version, and devising efficient strategies for exploring the parameter space. This chapter will be of interest to MSc and PhD students working on social simulation, and to researchers developing projects on agent-based modeling of social phenomena, either individually or in teamwork.