Experimental economics and agent-based models

Validating agent-based models is challenging due to the emergent properties of complex systems. Defining system-level patterns and determining which underlying processes created them is difficult because of the irreducibility of these systems. This paper describes methods to empirically calibrate and validate models, specifically using laboratory experiments with human subjects. Using experimental techniques, well-defined decision situations can be reproduced, and behaviours of human participants revealed. The outcomes of the experiment can be used to calibrate an agent model, offering a way to capture heuristics of decision making and reveal strategies that humans use in dealing with complex situations. This presents an alternative to representing agents as fully informed and rational decision makers, yet offering a defensible alternative that is grounded in empirical data. An example of using this technique is presented for agricultural agents acting in a cap and trade system for water quality management. Human participants are immersed in a running agent model where they control the decisions of one agent. The data revealed by the participants can then be used to calibrate artificial agents. The objective of this paper is to contribute to improved calibration and validation through participatory modelling, and a technique of combining agent-based models with experimental economics is presented as an empirically grounded method of defining agent behaviours.