آیا شما پاداش و تنبیه می کنید به طریقی که شما فکر می کنید دیگران از شما انتظار دارند؟

This paper addresses three questions: (1) When deciding on whether to reward or punish someone, how does how you think others expect you to behave affect your decision? (2) Does it depend upon whether others expect you to reward them vs. punish them? (3) What is the interpretation of such a causal effect? We investigate these questions using a modification of the lost wallet trust game (a simplified version of the trust game) that permits punishment. Like previous studies, we collect data on what second-movers think that first-movers expect them to do by directly eliciting the second-movers’ expectations. Unlike previous studies, we ensure exogeneity of these expectations by instrumenting for them. The instrument is the expectations of neutral observers which are disclosed to second-movers prior to the elicitation of second-movers’ expectations. We find that what you think others expect you to do has a zero causal effect on both reward and punishment decisions. We also find that it is important to instrument for second-order expectations because they are endogenous. We interpret these findings in terms of models of guilt-aversion and intentional reciprocity.

Reward and punishment are critical to regulating economic relationships, even in one-shot settings. Investigating their determinants – especially the potentially controllable ones – is an important step toward understanding optimal incentive schemes. One such malleable determinant is what you think others expect you to do. In this paper, we address the following empirically driven questions. 1. In the trust game, when deciding on whether to reward or punish someone, how does how you think others expect you to behave affect your decision? 2. In the trust game, does it depend upon whether others expect you to reward them vs. punish them? In addition to these primary goals, a secondary goal is to answer the following theoretical question. 3. In the trust game, if there is a causal effect of what you think others expect you to do on your behavior, what is its interpretation? Previous studies include the lost wallet version of the trust game (Dufwenberg and Gneezy, 2000; see Berg et al., 1995 for the trust game), shown in Fig. 1.The sender starts with $10 and the responder starts with nothing. The sender can either keep the $10, ending the game, or she can send the $10 to the responder. If the sender decides on sending the $10, they are tripled, and the responder unilaterally decides how much of the $30 to return to the sender (y). Let y′ denote the sender's expectation of y. Let y″ denote the responder's expectation of y′, referred to as the responder's second-order expectation of y: y′ is what the responder thinks that the sender expects the responder to send back. By eliciting y″ in simple variants of the trust game, Dufwenberg and Gneezy (2000) and other studies find a positive relationship between y″ and y. 1 This is an example of behavioral confirmation: you are more likely to behave in a certain way if you think that others expect you to behave in that way. This is to be contrasted with its obverse: behavioral disconfirmation. Behavioral confirmation and disconfirmation are descriptions of the causal effect of second-order expectations on behavior. The aforementioned studies interpret this instance of behavioral confirmation as resulting from guilt-aversion (which is what their designs are seeking to test): when y″ exceeds y, the responder is failing to fulfill what she believes to be the sender's expectations. The responder feels guilty about disappointing the sender, and will feel guiltier the larger the difference between y″ and y. A guilt-averse responder will therefore send back more money when she thinks that the sender expects more back. In the trust game, the responder's only alternative to a materialistic best response is to reward the sender.2 What if the responder is also allowed to punish the sender? In punishment decisions, do we expect behavioral confirmation or disconfirmation? We here consider an extended form of the trust game that allows for punishment. In addition to extending the scope from reward to punishment decisions, we also investigate the interpretation of the data. In particular, there are theories that predict behavioral disconfirmation in the trust game, such as the model of intentional reciprocity.3 Observing behavioral confirmation in the trust game is evidence in favor of guilt-aversion. However we explore why observing behavioral confirmation in the trust game is not necessarily evidence against models that predict behavioral disconfirmation in the trust game, and, crucially, why observing neither in the trust game (which we find) is not necessarily evidence against guilt-aversion. A final issue that we explore is how second-order expectations are observed. Dufwenberg and Gneezy (2000) elicit second-order expectations directly from the responders. Since this explanatory variable is not randomly induced by the experimenter, the design risks endogeneity bias; Charness and Dufwenberg (2006) refer to this possibility as a form of the false-consensus effect. To estimate the endogeneity bias, we transmit the expectations of non-playing observers to responders and then elicit the second-order expectations of responders. We then use the expectations of the observers as instruments for the second-order expectations of the responders. In a final treatment, to permit comparability with Ellingsen et al. (2010) and Reuben et al. (2008), we transmit the expectations of each sender to her partner prior to the responder making her choice. Our results are as follows. When using observer expectations as an instrument for second-order expectations, second-order expectations have no effect on both reward and punishment behavior, i.e., we find neither behavioral confirmation nor behavioral disconfirmation. Similar results are obtained when we transmit expectations to each sender's partner. When we elicit expectations directly, we find behavioral confirmation in both reward and punishment decisions, suggesting that elicited expectations are endogenous.

This paper addressed three questions in the context of the trust game. 1. When deciding on whether to reward or punish someone, how does how you think others expect you to behave affect your decision? 2. Does it depend upon whether others expect you to reward them vs. punish them? 3. What is the interpretation of such a causal effect? The manipulability of second-order expectations and the importance of reward and punishment decisions mean that these are significant questions. One should note that we attempted to answer these only within the context of one variant of the trust game (the judgment game) in a laboratory setting (see Gneezy and List, 2006 for a field reciprocity game). Nevertheless we were reassured by the congruency of our results with previous laboratory studies (or data garnered from previous studies). We found that second-order expectations had an economically and statistically negligible causal effect on both reward and punishment decisions. We also found that it is important to instrument for second-order expectations because they are endogenous. From the perspective of optimal incentive schemes, this suggests that there is no gain to exogenously manipulating second-order expectations: in one-shot environments, changing how others think you are expecting to get rewarded or punished has no impact on the incidence of either. This is consistent with Charness and Dufwenberg's (2006) finding that messages sent by senders did not affect responders’ behavior. Since such manipulation can be costly, resources devoted to improving incentive schemes are best directed elsewhere. In light of the substantial evidence on the importance of both guilt-aversion and intentional reciprocity to reward decisions (much of which is unrelated to second-order expectations), we interpreted our results as implying mutual cancellation of the two mechanisms.10 One need not treat the two explanations as mutually exclusive. A within-subject design where we could observe how a responder reacts to different first-order expectations would be a useful extension for parsing the two models. In punishment decisions, generating testable predictions from the models requires ad hoc assumptions, and so explaining the data is harder. Ultimately, since punishment is costly to the punisher, there is no intuitively obvious answer to the question: “If I think that you are expecting me to punish you, does that make me more or less likely to punish you?;” this remains an interesting avenue for future research. Our contributions have been extending the analysis of the causal effect of second-order expectations on behavior to the domain of punishment choices, and, in the context of reward behavior, asserting the mutual inclusivity of guilt-aversion and intentional reciprocity in explaining the data. Finally, to the best of our knowledge, we are the first to use instrumental variable methods to ensure exogenous variation in second-order expectations.