Social position and distributive justice: experimental evidence.

Our experiment also contributes to the understanding of behavior in dictator experiments in general. While the extent of giving in dictator experiments varies considerably across studies, several consistent behavioral effects have been demonstrated. Self-interested behavior increases with anonymity or social distance (Hoffman et al. 1994; Hoffman, McCabe, and Smith 1996) and when dictators bargain over earned wealth (Hoffman et al. 1994; Ruffle 1998; Cherry 2001; Cherry, Frykblom, and Shogren 2002). While we maintain anonymity, we demonstrate the importance of changes in endowments (and power) that are unearned. Unlike Hoffman et al. (1994), who used the language "provisionally allocated to each pair" to avoid perceptions of property rights, we explicitly allocate the right to be the decision maker and an unearned endowment to one party or the other to reinforce and investigate this effect. We are unaware of other studies that explicitly examine the behavior of "giving" dictators to that of "taking" dictators.

3. Results

Table 1 presents the experimental results. (4) We compare subjects' choices (as represented by the mean payoff to Player 1) using one-tailed Mann-Whitney tests; we compare the frequency of equal splits and $20/$0 splits using a one-tailed proportions test. The results are generally consistent with the hypothesis that subjects' perceptions of distributive justice are influenced by unearned social position. Decision makers (Player 1) in GD were in the strongest position of power and resources. These subjects chose an income distribution that yielded the greatest monetary benefit (mean = $16.17) to themselves, with the lowest rate of equal splits (24%) and the second highest number of $20/$0 splits. However, compared to other similar experiments, subjects in GD made positive offers more frequently (59% of the time compared to around 40% in DB2 in Hoffman, McCabe, and Smith 1996). This difference may be partially explained by the lack of a show-up fee. If dictator subjects in our GD treatment made $0 offers, the recipient truly left the experiment with $0 rather than a $5 or $10 show-up fee. Subjects may factor show-up fees into their choices, which would impact their perceptions of relative earnings.

Decisions in PD (mean = $15.30, with 29% equal splits and 45% $20/$0 splits) were statistically no different than in GD. We conclude that because players knew their decision would only be relevant if they were chosen to be the decision maker, they made their decisions as if they were in the position of full power and resources.

However, shifting resources from Player 1 to Player 2 (giving the initial $20 to Player 2) caused an economically and statistically significant decrease (one-tailed significance = 0.02) in Player l's monetary demand in TD ($14.03) compared to GD ($16.17). Subjects also chose equal splits more frequently (39% compared to 24%) and made statistically significantly fewer $20/$0 splits (16% compared to 41%; one-tailed significance = 0.016). Unlike in Cherry (2001) and Cherry, Frykblom, and Shogren (2002), the difference between these treatments was unearned. Cherry (2001) finds that only 24% of dictators made positive offers from "earned" endowments, whereas 74% made positive offers from "unearned" endowments. When Player 2 received the "unearned" endowment in TD, 86% of dictators effectively made positive "offers." Ruffle (1998) finds that dictators reward "deserving" recipients, those whose skill led to a larger pie size for the pair. The modal offer in that case was half of the pie, with 21% of offers actually being greater than half.

It must be noted that given our anonymity procedures, it is possible in TD for Player 2 subjects to transfer less than the requested amount, in which case the recording experimenter adjusted the contents of the envelope to match the requested amount. This provides an interesting experimental variable and occurred in 12 of the 31 cases. However, if Player 1 subjects expected that Player 2 subjects were less likely to comply with large demands, this may bias the demands downward. The fact that Player 2 returned $0 in four out of five cases when $20 was demanded indicates that this may be a real concern. In contrast, Player 2 subjects from whom $10 was requested transferred all $10 on 10 of 12 occasions. Further experimentation with added controls would be necessary to determine the significance of this effect on demands.

As expected, the Player 1 average payoff in VD was significantly lower than in GD ($9.80 vs. $16.17, one-tailed significance = 0.00). Subjects chose equal splits 43% of the time, which is significantly more than the 24% of equal splits that occurred in GD (one-tailed significance = 0.061). However, preferences were less risk-averse than hypothesized by Rawls. The standard deviation of Player 1 payoffs was greater in VD than in all of the other protocols. Furthermore, nearly one in four subjects chose a $20/$0 split (despite the lack of a show-up fee), a number which, while significantly fewer than observed in GD (one-tailed significance = 0.069), is greater than in the TD and HD protocols. Clearly, some subjects were willing to gamble on being the one to get the higher payoff.

Finally, the decision makers in HD, who are in potentially the weakest positions of power and resources, indicated (hypothetically) that they would have chosen an income distribution that was even more egalitarian (an average payoff to themselves of $11.93, with equal splits 52% of the time) than any treatment other than VD. Only 7% of subjects indicated they would have taken the entire $20. Compared to the GD and TD protocols, the Player 1 (hypothetical) payoff decrease in HD is both economically and statistically significant (one-tailed significance = 0.00 and 0.04, respectively). Interestingly, the difference between what these subjects predicted the GD decision makers would choose ($15.38) and what the GD Player 1 subjects actually chose ($16.17) is not statistically significant, although they underestimated the propensity to choose equal splits (10% predicted vs. 24% actual). Similarly, Ruffle (1998) found that, when asked what offer they hypothetically would have made in their counterpart's position, the unskillful recipients (those whose effort on a general knowledge test led to the smaller of two pie sizes for the pair) would have offered substantially more than their counterparts actually did.

It is unclear, of course, how reliable any hypothetical response can be in this situation. One alternative explanation for the generous responses in HD is that some subjects prefer to maintain a good self-impression (e.g., Murnighan, Oesch, and Pillutla 2001) even if they cannot maintain a good social impression as a result of anonymity conditions. That is, being generous in a hypothetical response makes them feel good about themselves. If self-impression concerns are significant when real money is being transferred, we would expect an even stronger impact on hypothetical responses, because maintaining a good self-impression has no cost. Unfortunately, we have not been able to design an experiment in which participants who cannot influence the final distribution must make a decision other than a hypothetical one.

4. Conclusions

We used slight variations in language and procedures to generate differences in social position, as measured by resources and power, in a simple, double-blind dictator experiment. These variations may also have changed the social distance between subjects, a variable that has been identified as an important explanatory variable for behavior (Hoffman, McCabe, and Smith 1996). While it is difficult to fully disentangle social distance from social position experimentally, our results are consistent with Rawls' (1971) assumption regarding distributive justice in that unearned differences in a subject's social position, as measured by power and resources, affect expectations regarding a just income distribution. While all individuals in such experiments must balance their self-interest with their personal views of justice and fairness, individuals appear to develop a sense of entitlement to a higher payoff when granted initial property rights or the power to influence the final distribution. As other experiments have demonstrated, making the social positions earned further strengthens the effect. This is particularly important given that unearned differences (such as the wealth of the family into which one is born) can subsequently lead to earned differences (such as obtaining a college degree) that are the result (at least partially) of the individual's own efforts.

These results are in stark contrast to the predictions of standard game theory and theories of pure altruism or inequity aversion, which predict identical outcomes in each of our "unveiled" protocols. However, even in a "veiled" decision-making position, subjects in our dictator experiment failed to unanimously choose an equal income distribution, as a strict interpretation of the Rawlsian hypothesis would require. Their preferences were less risk-averse and had greater variance than Rawls hypothesized. Some subjects appeared willing to accept an outcome where some players had more and some had less, relative to the egalitarian outcome, as long as there was a fair chance of being the one at the desirable end of the distribution. Perhaps there is a testable lesson here that distributive justice is not necessarily a matter of the final wealth distribution, but rather of the probability of reaching a desirable position in the distribution.

Received February 2006; accepted October 2006.

References