Validating agent-based marketing models through conjoint analysis ☆ Rosanna Garcia a, ⁎ , Paul Rummel b , John Hauser c a Northeastern University, United States b Comdel, Inc., United States c Massachusetts Institute of Technology, United States Received 1 March 2006; received in revised form 1 December 2006; accepted 1 February 2007 Abstract Agent-based modelers in the field of marketing research have paid little attention to validation issues. This paper provides a definition of validation relevant for this community of modelers. On the basis of a history-friendly model for simulation calibration [Malerba, F., Nelson, R., Orsenigo, L., and Winter, S. (1999). ‘History-friendly’ models of industry evolution: the computer industry. Industrial and corporate change, 8(1), 3–40.], the authors demonstrate how conjoint analyses can be used to instantiate and calibrate an agent-based marketing model. Methods for model instantiation using conjoint partworths and model calibration using the conjoint first-choice rule are demonstrated. When the model matches the results of the first-choice rules for consumer preferences, the modeler can feel more confident that calibration has been achieved. When verification replicates stylized facts on a macro-level, the model is one step closer to validation. Because conjoint data results are meaningful on an individual level as well as on an aggregate level, this type of empirical data collection is ideal for agent-based marketing models. © 2007 Elsevier Inc. All rights reserved. Keywords: Conjoint analyses; Agent-based modeling; Validation; Calibration; History-friendly model 1. Introduction Validation of computational models is an area of concern to simulation modelers (Conway, 1963; Knepell and Arangno, 1993; LeBaron, in press). Different types of validation (Knepell and Arangno, 1993), different levels of validation (Carley, 1996) as well as different methodologies for conducting validation (Carley, 1996; Fagiolo et al., 2005) have resulted. These differing studies, along with feedback loops, path dependencies, sensitivities to internal conditions, and the unpredictability of agent adaptation (Fagiolo et al., 2005) associated with empirically based agent-based models (ABMs), easily confound the task of validation. Important questions asked by agent-based modelers, especially those investigating real-world systems, are: which methods of validation are best? Which levels should be considered? How does one know a model is correct? Agent-based modelers in the field of marketing research have paid little attention to these validation issues. The goal of this paper is therefore two-fold: to provide a definition of validation relevant for agent-based modelers in marketing research and to introduce a calibration method based on conjoint analysis that incorporates real-world data into a marketing-oriented agent-based simulation. The paper first provides a definition for validation and for validation levels that are important to this community of agent-based modelers. Drawing upon reviews grounded in agent-based computational economics (ACE) by Carley (1996) and Fagiolo et al. (2005), the authors then briefly present three methodologies used to seek validity in ABM simulations. This foundation serves to demonstrate how conjoint analysis can be used to calibrate an agent-based marketing model. Journal of Business Research 60 (2007) 848 – 857 ☆ The authors acknowledge the support of Northeastern University's Institute for Global Innovation Management for this research. They also wish to thank the participants of the Agent-based Models of Market Dynamics and Consumer Behavior Symposium 2006 at the University of Surrey, Guilford, as well as Walter McManus and Colette Friedrich for their comments on earlier versions of this paper. Finally, the authors offer their sincere thanks to Olivier Toubia, who was a critical contributor to the conjoint analyses. Send correspondence to Rosanna Garcia, Assistant Professor of Marketing, Northeastern University, Boston, MA USA 02115 (r.garcia@neu.edu). ⁎ Corresponding author. E-mail address: r.garcia@neu.edu (R. Garcia). 0148-2963/$ - see front matter © 2007 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2007.02.007