Giving in dictator games: regard for others or regard by others?

• Author: Koch, Alexander K.

1. Introduction

   Experimental studies have challenged the notion that human behavior can be well approximated with models assuming narrow self-interest. Strong evidence for other-regarding behavior comes from dictator game experiments (Kahneman, Knetsch, and Thaler 1986; Forsythe et al. 1994). One player (dictator) decides how to allocate an amount of money between himself and a passive player (receiver). Narrow self-interest would call for the dictator to keep the entire amount, leaving the receiver with nothing. Forsythe et al. (1994) find that 70% of the dictators actually do not grab the entire pie for themselves, leaving the receiver with on average 24% of the total amount. Similar patterns have been documented in many different dictator game settings (see, e.g., the survey of Camerer 2003).

   But what is the source of seemingly altruistic behavior? The two fundamental views on this ancient question are pitted against each other in Plato's (360 BC) Republic. Glaucon claims that "[...] a man is just, not willingly or because he thinks that justice is any good to him individually, but of necessity, for whenever any one thinks that he can safely be unjust, there he is unjust." He illustrates this with the tale of the shepherd Gyges, who finds a ring that enables him to become invisible and act unjustly without anybody ever knowing. His brother Adeimantus concurs, saying that otherwise "[...] we should not have been on the watch to keep one another from doing wrong, but every one would have been his own watchman" (Book II). In contrast, Socrates claims that an individual who acts just for its own sake rather than for its appearance is truly serving his own interest, justice being concerned "[...] not with the ontward man, but with the inward [...]" (Book IV).

   This paper looks into whether external forces--concerns for the "outward man"--are needed to motivate dictator giving or whether internal mechanisms--concerns for the "inward man"--underly such behavior. It has been argued that giving in dictator games might arise from a desire of subjects to appear to be acting in a socially appropriate way because their individual behavior is observed by others. Hoffman et al. (1994, p. 349) suggest that "fairness" may not be "'own' preference, but a derivative of judgement by others." Because subjects' behavior might also be influenced by expectations about the experimenter's judgement. Hoffman et al. (1994), and Hoffman, McCabe, and Smith (1996) run a double-blind treatment (DB1) to ensure that the experimenter cannot personally attribute subjects' decisions. Incidence of positive offers (36% of dictators) and average donations (9.2% of pie) drop significantly relative to Forsythe et al. (1994). This vindicates Glaucon and Adeimantus' view that external enforcement of norms is important. Reducing observability of actions along the dimension of increasing anonymity leads to less seemingly other-regarding behavior. (1)

   Nevertheless, 36% of the dictators in the DB1 experiment still donate some money (mean gift 25% of the pie). Does this imply that these subjects have pure other-regarding preferences? Hoffman et al. (1994, p. 371) argue that the DB1 results "may approach the appropriate indicator of fairness as a pure preference phenomenon." However, there is another dimension of observability. Giving by a dictator might not be driven by true regard for the receiver's welfare but rather stem from a desire to avoid that the receiver discovers the deed of violating a social norm (rather than the perpetrator as an identifiable individual). Then, even in completely anonymous settings, regard by others would still serve as an external enforcement device of norms.

   Recent bargaining experiments provide some evidence that an important driving force in seemingly altruistic behavior is whether or not the responder can observe a proposer's action. Introducing an information asymmetry about the pie size to be divided by the proposer, the typical finding is that subjects are concerned about being perceived by others as abiding by social norms (even under intersubject anonymity) but otherwise exploit their informational advantage in a selfish way. (2) Guth, Huck, and Ockenfels (1996, p. 597) conclude that "complying--at least superficially--with a (fairness) norm has some intriusic value: you feel better when others do not know how greedy you are."

   Dana, Cain, and Dawes (2006) investigate the dimension of receiver knowledge in a dictator game setting without introducing experimenter anonymity. After dictators have made their allocation decision they are given an exit option. If the option is chosen, the dictator gets $9 and the receiver gets nothing but remains ignorant of the game. If the exit option is not chosen, the receiver gets his part of the $10 pie, along with the dictator's instructions (treatment 1) or without becoming aware of the game (treatment 2). The pre-exit option pattern in treatment 1 replicates the Forsythe et al. (1994) results: 70% of dictators offer more than nothing, receivers get 25% of the pie on average. The exit option is chosen by 33% of subjects, bringing down the incidence of positive dictator offers actually carried out to 43% (receivers in the experiment get 15% of the pie on average). This is close to the levels in the Hoffman et al. (1994), and Hoffman, McCabe, and Smith (1996) double-blind treatment (DB1) experiment (36% positive offers, the average receiver share is 9.2% of the pie).

   In sum, eliminating only one of the two dimensions of external enforcement of norms, regard by the receivers (Dana, Cain, and Dawes 2006) or regard by the experimenter (the Hoffman et al. [1994], and Hoffman, McCabe, and Smith [1996] double-blind procedure), appears to affect subject behavior to a similar extent. Our contribution is to test for the importance of receiver knowledge in the absence of the second dimension of external enforcement of norms. The experiment therefore eliminates reputational and other strategic considerations with a design that guarantees experimenter anonymity and precludes any possibility of future interaction between dictators and receivers. Paraphrasing Adeimantus, we ask to what extent individuals are effective in policing themselves in completely anonymous situations. Or put differently, does it matter for an individual in a completely anonymous setting that those affected by his or her actions are able to assess whether these actions conform to established social norms or not?

   The treatments vary the information that receivers have. The A ware Receiver (AR) treatment provides receivers with the dictator game instructions. Common knowledge of the game enables receivers to become aware of being treated kindly or unkindly by the dictators. Such knowledge by receivers is crucial for external enforcement to be effective. In contrast, it is irrelevant for dictator behavior if pure other-regarding preferences (or any other internal mechanisms) are at work. Therefore, we compare behavior with that in the Unaware Receiver (UR) treatment, which provides no information to receivers about the source of dictator offers. In fact, receivers do not even know that they participate in an experiment. In all treatments, we control for experimenter...