خویشتن بینی و تعادل اخلاقی: تجزیه و تحلیل تجربی

In our experiment, a dictator game variant, the reported outcome of a die roll determines the endowment (low/high) in a subsequent dictator game. In one treatment the experimenter is present and no cheating is possible, while in another subjects can enter the result of the roll themselves. Moral self-image is also manipulated in the experiment preceding ours. The aim of this experimental set up is to analyze dynamic aspects of moral behavior. When cheating is possible, substantially more high endowments are claimed and transfers of high-endowed dictators are bigger than when cheating is not possible (mediated by the preceding moral self-image manipulation). The preceding manipulations also have a direct effect on generosity, when subjects have to report the roll of the die truthfully. Moral balancing appears to be an important factor in individual decision making.

The dilemma between behaving morally and the tempting alternative that bends the social conventions to our advantage is a constant feature of everyday life. The study of potential factors affecting people's choices in such situations has been the topic of a large body of past and still ongoing research. While the role of outcomes, social interaction (in the form of intentions or emotions), or the situational environment the decision is taken in dominate the analysis of social preferences, our paper focuses on the dynamic aspects of moral behavior. Is an individual's tendency to behave pro-socially a constant, that is, will he act always in the same generous way in a given situation? Or is the decision affected by the context, specifically by the inter-temporal context? Imagine the following situation. A young man just left the subway train, heads up the station with the other passengers, and sees a woman with a baby buggy unable to get up the stairs on her own. Will he be more likely to offer assistance, if he just dodged the fare for the ride? On a more general level, do we have a built-in morality barometer that guides our behavior back to the level we appreciate most? Recently, self-image concerns gained increased recognition as a successful determinant of human choices, especially in the realm of moral behavior.1 While these models vary in their approach and terminology, their central message is arguably a common one. People desire to maintain a comfortable self-image. However, it remains an open question how they react, if there is a discrepancy between actual behavior and their self-image. Moral balancing theory (Nisan and Horenczyk, 1990) suggests that individuals keep account of their self-image over time. In line with the economics literature on self-image it also assumes that people wish at all times to keep their moral status on a level that they consider satisfactory. In addition, moral balancing proposes ways how people deal with deviations from their individual moral self-image. If one's moral self-image dropped below some standard, people would engage in moral cleansing to compensate. Likewise, when the moral self-image is above an ideal level, then people would have a tendency to behave immoral in an act of moral licensing. The aim of our paper is to study such dynamic aspects of moral behavior in an experimental design that endogenously manipulates subjects’ moral self-image. This allows us to analyze the effects of a variation of moral self-image on pro-social behavior. We test whether there are inter-temporal spillovers of (im)moral behavior within our experiment in which a dictator game's endowment depends on the roll of a six-sided die. Subjects who report an odd number receive a high endowment in the subsequent dictator game. Reporting an even number results in a low endowment. In one treatment (Open Roll) the experimenter is present and no cheating is possible, while in another (Hidden Roll) subjects can enter the result of the roll themselves. Therefore, in Open Roll the high endowment is legitimized by the transparent procedure. In contrast, subjects may cheat in Hidden Roll in order to claim a high endowment. As a consequence, average moral self-image is potentially lower in Hidden Roll and may lead to moral cleansing in the subsequent dictator game. Besides a stand-alone treatment our actual experiment was also conducted right after another experiment. This allows us to connect morally relevant information from the previous experiment to behavior in our experiment. In Kataria and Regner (2012), henceforth Philanthropy, a donation experiment involving a real effort task, moral self-image is supposedly low/high after one donated little/much in comparison to the other subjects. In Crosetto et al. (2012), henceforth VCG punishment, a voluntary contribution game with punishment, moral self-image is supposedly low/high, if one has been lucky/unlucky in the payment procedure. To the best of our knowledge previous studies on moral balancing used an exogenous variation (priming methods) to induce different levels of moral self-image.2 Instead, in our experiment subjects’ moral self-image is endogenously manipulated. The potential effect on the self-image is caused by the subjects’ own choice when they report the outcome of the roll of the die. In our condition Hidden Roll, they can report truthfully but they do not have to. Moreover, subjects are aware of the potential moral cleansing offered by the dictator game, when choosing whether to truthfully report the outcome of the die roll. Several recent studies used the self-reported outcome of a random event as a measure for cheating (see, for instance, Fischbacher and Heusi, 2008, Shalvi et al., 2010, Shalvi et al., 2011, Fischbacher and Utikal, 2011 and Shalvi and Leiser, 2013). Most comparable to our procedure are the following two studies. Bucciol and Piovesan (2011) used a binary event in a field experiment with children aged 5–15. The children were asked to toss a fair coin (black/white) in private. They knew that they would receive a reward only if they reported an outcome of white. Overall, 86% of the children reported the profitable outcome. Also Houser et al. (2012) used a binary cheating procedure. After playing a dictator game subjects were informed that they would get a chance to get an additional payment. They were told to flip a fair coin and report the outcome which determined the size of the extra payoff. Overall, 74.5% reported the high-payoff outcome. In our experiment, when cheating is possible (Hidden Roll condition) around 85% percent of subjects claimed a high endowment. In the stand-alone treatment we do not observe that subjects compensate for their dishonesty. Only when moral self-image is also manipulated in a previous experiment – and controlled for in the data analysis – transfers of high-endowed dictators are higher in Hidden Roll than in Open Roll (when cheating is not possible). Moreover, our results show that morally relevant variation in a previous experiment carries over and affects the decision making of dictators in Open Roll. The worse subjects performed in generating donations in the Philanthropy experiment, the more they transfer as a dictator. The more subjects earned in the VCG punishment experiment, the more they transfer. Finally, we find evidence for a Robin Hood effect. In the Philanthropy condition, when subjects previously took part in an experiment that involved donations, the rate of cheating is significantly higher, if a treacherously earned endowment could be shared with another participant instead of being directly appropriated. The paper is organized as follows. Section 2 describes the experimental design and develops behavioral predictions. Results are presented in Section 3 and Section 4 concludes.

Overall, our study of dynamic aspects of moral behavior provides evidence that people engage in moral balancing. Our experimental design employs an endogenous manipulation of subjects’ moral self-image as participants can choose themselves whether to be dishonest or not. A substantial fraction of participants make use of the opportunity to cheat and, at least partially, seize the opportunity to wash their conscience by donating more. In addition, also variations of moral self-image from the previous experiment have an effect on the dictator transfer. This highlights the relevance of contextual factors for the study of dynamic aspects of moral behavior. We conclude that moral balancing appears to be an important factor in individual decision making. In a modeling framework that centers around self-image concerns, moral balancing complements other context-dependent motivations that have been found to be of importance. While, for instance, social-image models consider the situational environment of a decision and reciprocity/emotion models take the social or inter-personal dimension into account, moral balancing addresses the intertemporal context.