Promises & PartnershipGary Charness & Martin Dufwenberg*October 11, 2004TRUST, COMMUNICATION & GUILTFIGURE 22.3 Guilt aversionFIGURE 33. EXPERIMENTAL DESIGN AND HYPOTHESESTable 1: Payoff Outcomes with a (5,5) outside optionTable 2: Payoff Outcomes with a (7,7) outside optionSecond, whether or not communication is relevant, there may H2 will be tested against the alternative of positive correlEXPERIMENTAL RESULTSWe next visit several examples that illustrate how guilt-aveExample 1. Consider first the game in Figure 1. For simpli[A MessageGood luckProbit Regressions for Determinants of BehaviorPromises & Partnership Gary Charness & Martin Dufwenberg* October 11, 2004 Abstract: We examine, experimentally and theoretically, the impact of communication on trust and cooperation. We are especially interested in how problems of hidden action (as treated in contract theory) may be overcome. Our design admits observation of promises, lies, and beliefs. The evidence supports a theory according to which people strive to live up to others expectations in order to avoid guilt. The theory admits promises to enhance trustworthy behavior, which is what we observe in the experiment. Motivated by these results, we develop a notion of guilt-aversion equilibrium for extensive games. Besides explaining partnership interaction, the model sheds light on the role of language, discussions, agreements, and social norms more generally. The analysis also leads to some calls for more research in the field of psychological game theory. Keywords: Promises, partnership, guilt aversion, trust, lies, social preferences, psychological game theory, contract theory, behavioral economics, hidden action, moral hazard JEL codes: A13, B49, C72, C91, D63, D64, J41 We are grateful to Ninghua Du for excellent research assistance, Jon Baron, Paolo Battigalli, Jeanette Brosig, Steve Burks, Colin Camerer, Ernst Fehr, Ayelet Fishbach, Guillaume Frechette, Dan Friedman, Drew Fudenberg, Simon Gächter, Uri Gneezy, Brit Grosskopf, David Laibson, Rick Larrick, Dan Levin, David I. Levine, David K. Levine, Tanya Menon, Matt Parrett, Torsten Persson, David Reiley, Yuval Rottenstreich, David Strömberg, Richard Thaler, Dean Williamson, Bernd Wittenbrink, Bill Zame, seminar/conference participants at Antwerp, Berlin (Humboldt), Bonn, Chicago GSB, Cologne, Dortmund, Fuqua School of Business (Duke), Gothenburg, Harvard, Munich, Örebro, Oslo, Purdue, Santa Barbara, Stockholm, University of Arizona, 2002 GEW-Tagung (Wittenberg), 2002 Behavioral Research Council (Great Barrington), 2003 Public Choice (Tampa), the 2004 World Meetings of the Game Theory Society (Marseille), the SITE 2004 segment on Experimental Economics at Stanford, and ISNIE 2004 (Tucson) for stimulating discussions and helpful suggestions, and the Swedish Competition Authority and the Russell Sage Foundation for financial support. * Contact: Gary Charness, University of California at Santa Barbara, [email protected]; Martin Dufwenberg, University of Arizona, [email protected]. INTRODUCTION Much of human achievement is produced in partnerships. An extensive body of theoretical research – contract theory – is devoted to understanding which partnerships form, what contracts are signed, and what the economic consequences will be. Considerable attention has been devoted to environments with hidden action, where a party’s future choice is not contractible.1 Theorists have shown that if people are rational and selfish (caring only about own income), hidden action is a shoal on which efficient contracting may founder. We examine, experimentally and theoretically, the impact of non-binding pre-play communication (cheap talk) on cooperation in a simple one-shot trust game designed to capture the essence of hidden action. The conventional contract theoretic approach implicitly assumes that such communication is ineffective in promoting partnership formation and cooperation.2 Our study explores whether there are psychological aspects related to communication that conventional contract theory has not addressed. In particular, we incorporate a new theoretical perspective, building on psychological game theory (see Geanakoplos, Pearce & Stacchetti 1989, henceforth GPS), which furnishes a reason why communication may foster trust and cooperation. The basic idea, which we refer to as guilt aversion, assumes that decision makers experience guilt if they believe that they let others down. That emotion leads to a non-standard conception of utility (from the viewpoint of classical game theory), whereby a decision-maker’s preferences over strategies depend on his beliefs about the beliefs of others, even if there is no strategic uncertainty. Messages that may have seemed cheap talk gain cutting power; for example, guilt aversion permits a promise to feed a beneficial and self-fulfilling circle of beliefs about choices, beliefs about beliefs, trust, and trustworthiness. 1 This condition is often referred to as moral hazard. For entries to this literature, see e.g. Hart & Holmström (1986), Dutta & Radner (1993) and Salanié (1998, chapter 5). 2 Our focus on one-shot interaction is worth emphasizing. We do not consider repeated games (cf., for example, MacLeod & Malcolmson 1989), in which communication may serve as an equilibrium-selection device. 1In section 2, we consider the implications of guilt aversion for the specific games of our experiment. In the experiment, described in sections 3 and 4, we measure beliefs, record messages, and examine how ‘statements of intent’ correlate with subsequent choices. As it turns out, our theory does quite a good job explaining the data, even with respect to several alternatives that we consider. It turns out that a particular form of communication, namely promises to perform, can inspire an increased level of cooperation. We find, overall, that participants in both roles (principals and agents) make cooperative choices more than twice as often when the agent makes a promise to perform, with the corresponding rate of successful partnership formation more than quadrupling. In light of the experimental support for guilt aversion, we find it natural to ask what this might imply in other games. In section 5 we consider what kind of solution concept we get if use the GPS framework to incorporate guilt aversion in general extensive games. We provide an answer, which we call