Exploiting moral wiggle room: Illusory preference for fairness? A comment

Abstract

We designed an experiment to test the robustness of Dana, Weber, and Kuang’s (DWK), 2007 results. DWK observed that, when participants were given a “costless” way — the click of a button — to ignore the consequences of their actions on others’ payoffs, they chose to remain ignorant and fair behavior diminished. By implementing a double-blind experiment together with a design that controls for alternative explanations for the observed behavior, we confirmed DWK’s findings.

1 Introduction

Previous works have indicated that, although individuals often act altruistically (e.g., Becker 1974; Andreoni 1989, 1990) they also engage in strategic ignorance and thus avoid situations where altruism might be expected or required of them (Bolton, Katok, & Zwick, 1998; Konow, 2000; Carrillo & Marriotti 2000; Dana, Cain & Dawes, 2006). In a recent paper, Dana, Weber & Kuang (2007) — henceforth DWK — presented results of an experiment where participants were given a choice of either remaining uninformed about the effects of their actions on someone else’s earnings, or clicking a button on the computer screen to learn about how their decisions could affect another participant’s earnings. Despite this “costless” mechanism, half of the participants in their experiment chose strategic ignorance and, subsequently, selfish actions. In contrast, when participants’ did not have a choice to remain ignorant, behavior was mostly altruistic.

Based on these data, DWK argue that the generosity observed in many experiments may not be due to a preference for a fair outcome, but may be a result of “…people feeling compelled to give due to situational factors, while not really valuing the corresponding outcome” (p. 77). Indeed, DWK observed that, when the one-to-one mapping between actions and outcomes was relaxed, fair behavior diminished. Given the potential significance of these findings in reshaping the discussion of the motives for fair and altruistic behavior and the role that situational factors may play in influencing them, here we check the robustness of DWK’s results to an alternative experimental design.

In DWK’s experiment, subjects were asked to click on a button on the screen to make the consequences of their actions transparent. In addition to a preference for strategic ignorance, there are alternative reasons for why subjects in their experiments may have chosen not to click the button (remain ignorant) and behaved selfishly. First, subjects may have perceived the click of a button as costly, not costless. An extra click requires an additional effort (the click). Secondly, the click may have been avoided because it generated a delay in consumption. In DWK’s experiment, a participant who chooses to reveal the other’s payoffs is forced to wait to learn about the other’s earnings before being able to consume her own payoff. Finally, and more importantly, it is possible that omission bias (i.e., the tendency to judge harmful actions as worse, or less moral than equally harmful inactions — see Ritov & Baron, 1995) may have caused the curious results. Indeed, one can reasonably expect more generosity in an experimental design where transparency is the default choice and its avoidance the “click” option.

In this paper, by altering the way we ask participants to reveal or not reveal information about others’ earnings, we control for decision costs, consumption delay, and (with some caveats) omission bias. Indeed, in our experiment, to reveal or not reveal information about the other’s payoffs required the same amount of effort and time, and both options required an action or commission. ¹ We also implement a double-blind experiment to generate the most reliable environment to test participant’s decisions. Interestingly, overall, our results confirm the results of Data, Weber & Kuang. We find that individuals engage in strategic ignorance and, when the consequences of their actions are not fully transparent, fair choices diminish.

2 Experimental design

2.1 Subjects and earnings

A total of 54 college students were recruited for a thirty-minute experiment. They were recruited from various departments at Emory University through public announcements and e-mails. There were a total six sessions; at the beginning of each session, each subject was seated in an isolated booth and was assigned a role of either the dictator (Player X) or the receiver (Player Y) by drawing slips of paper from an envelope. The dictator determined his own payoffs and the payoffs that the second player received. In addition to the earnings determined by the dictators, all participants received a $5.00 show-up fee. The average earnings in the experiment (including the show-up fee) were about $10.

2.2 Transparent treatment (baseline)

Our baseline treatment was similar to DWK’s, but we did not use computer terminals and our experiment was double-blind. We implemented a double-blind decision to eliminate potential experimenter induced biases. Participants were given a payoff identical to the table shown below. At the beginning of our baseline sessions, the dictators made their choice of either Action A or B privately and anonymously. In our experiments, subjects made their decisions by opening one of two folders labeled A or B that contained envelopes with payments for both players. They were then asked to get their payment from the envelope and leave B’s payment inside the envelope. Subsequently, subjects dropped the envelopes containing B’s payments inside a box, which we later recovered and distributed among Y players. At no time did we know who, among the X players, had chosen A or B (see instructions for this experiment in Appendix A).

In the baseline treatment, the earnings for both dictators and receivers were common knowledge, thus the consequences of the dictators’ actions were transparent. Notice that B resulted in both players making $5, but A rendered one additional dollar to Player X at the expense of Y, who would get only $1. We refer to this game as the conflicting interest game (CIG). After making a choice, the dictators left the experiment with their payment, which was paid to them anonymously and in cash. Receivers were then randomly matched with a dictator’s decision and also paid anonymously in cash before they left.

2.3 Non-transparent treatment

The non-transparent treatments were also run double-blind and without computers, otherwise, they closely followed the experimental design of DWK (see Appendix B). Below are the payoff tables of two Dictator Games. The upper payoff table contains payoffs where Player X’s best interest is in conflict with Player Y’s. Again, we call this game conflicting interest game (CIG) — notice that this game is identical to that found above. The lower payoff table, on the other hand, shows payoffs where Player X’s best interest is in line with Player Y’s. We call this the similar interest game (SIG); indeed, both the dictator and the receiver earn the most when Action A is chosen.

In our non-transparent treatments, subjects did not know which game they were playing before making their choices, but they were informed that a coin flip, before the session began, had determined which game they would play. Note that, as shown below, in both games Player X’s payoffs were the same, but Player Y’s payoffs depended on which one of the games was being played.

Before deciding on an action (A or B), the dictator had to determine whether or not he wanted to know which of these games was being played. Each dictator was faced with two folders labeled REVEAL PLAYER Y’S PAYOFFS and DO NOT REVEAL PLAYER Y’S PAYOFFS. Dictators were instructed to open the folder corresponding to their choice to either reveal or not reveal which game was being played. If the dictator chose to reveal the game, he opened the REVEAL folder, where he saw the appropriate game (determined before the session began by a random draw), then he chose either Action A or Action B. If a dictator chose to not reveal which of the two games was being played (remaining strategically ignorant), he opened the DO NOT REVEAL folder, where he saw the game matrix as shown below. This table does not tell which game is being played, as the receiver’s payoffs are blocked with a question mark. The dictators then chose either action A or action B.

Payoff table when the “not reveal” option was chosen

To keep it double-blind, the choice of an action was implemented by allowing subjects to select a labeled envelope that contained an amount of money equal to the sum of X and Y’s payoffs for that game. For example, if the conflicting interest game was being played, inside the folder, there were two envelopes labeled A and B with $7 and $10, respectively. Subjects would take the money from one of the envelopes, keep her payment, place B’s payment in a blank envelope, and move to a private area, where the blank envelope was dropped inside a box. The envelopes with B’s payments were later picked (but never opened) and distributed to the receivers. In contrast to the design of DWK, here, both options: to reveal or not reveal, required the same amount of effort (opening a folder), rendered equal consumption time delay, and required a positive action (commission).

Finally, in the non-transparent treatment, while all the dictators were making their decisions, receivers were asked to predict dictators’ choices, to report what they would choose if they were a dictator and what they hoped their matched dictator would choose. ² Clearly, the answers to these questions had no effect on the dictators’ actions, but enabled us to confirm the clarity of the instructions, and the participants’ understanding of the experiment and its consequences.

3 Results

Overall, our results are similar to DWK’s results. As shown on Table 1, in the CIG baseline treatment, 78% of our dictators chose Action B, giving up one dollar of earnings to increase the receiver’s earnings by four dollars. In comparison, DWK, found that 74% of the dictators behaved altruistically; these proportions are statistically equal (Fisher Exact Test, p=1.00).

Table 1: Dictators’ decisions in all treatments.

In contrast to our baseline treatment, in our non-transparent treatment, the percentage of players choosing B (the altruistic choice) when the conflicting interest game (CIG) was played was only 22%. The null hypothesis that the proportion of dictators that acted selfishly when playing the CIG in the baseline and non-transparent treatments is the same can be rejected (Fisher Exact Test, p=0.028). Again, this result is similar to the findings in DWK, who report that only 37% of the subjects behaved altruistically in the CIG non-transparent condition as compared to 74% in the CIG baseline treatment.

As in DWK’s experiment, about half of our subjects chose to remain ignorant about the consequences of their actions. In fact, 53% of our subjects chose not to reveal information. As in DWK, among those who chose not to reveal information about the other’s earnings, 100% chose Action A (the selfish action). Finally, and in contrast to DWK, in our non-transparent treatment most of the participants who chose to reveal information in the CIG also chose the selfish option (Action A). The difference, however, is not statistically significant. With a bigger sample, we speculate that the difference may become significant. It is possible that participants who reveal and choose Action A feel less guilty about being selfish when they know that the receivers do not fully know whether the payoff matrix had been revealed to them (the dictators). Thus, even when one knows the consequences of one’s actions, the uncertainty regarding whether the receiver knows this may give the dictator some wiggle room to be selfish.

The existence of strategic ignorance was further reflected to us in the self-reported responses of receivers. Receivers were asked to imagine the motivation of dictators in their choice of folders. We transcribed some of the writings, most of which provided very similar arguments. For example, one participant said: “by looking in the do not reveal folder the decision will then be easier.” Another argued that a likely motivation for looking in the “do not reveal” folder would be “…to maximize payoffs…to remain blind to what they are doing to the other person.” Finally, someone suggested that “…it is better to remain in denial about the repercussions of the decisions one is making.”

Overall, as show by the last two columns in Table 1, we evidently observe a clear tendency to be more selfish in the non-transparent treatment.

4 Discussion

In this comment we investigated the robustness of Dana et al.’s (2007) results to changes in the implementation of the experiment. In particular, we were concerned that in their design the revelation of information required an additional action (clicking a button on the computer screen). We believe that such action was not costless. Indeed, it is possible that a preference for strategic ignorance and the observed selfish behavior was motivated by the avoidance of an additional effort, by the avoidance of delayed consumption, and more importantly, by a preference for the current state (omission bias).

We designed an experiment that would control ³ for these possible confounds — in addition to making it double-blind — to see if we could replicate DWK’s main findings. Consistent with their results, we found that, when given a chance, most people would prefer to stay ignorant about the consequences of their actions on others’ earnings. We also observed an increase of selfish behavior in the non-transparent setup. Thus, DWK’s results are not an artifact of their experimental design. Our replication of their findings may further challenge experimental and behavioral economists into thinking about the motives for observed altruistic and fair behavior in the lab, and the role that situational factors may play in influencing them. Indeed, it does seem that participants do not particularly care much for the other’s well-being when the outcomes of their actions are not fully transparent.