Decision Are Preferences for Allocating Harm Rational? Alexander L. Davis, John H. Miller, and Sudeep Bhatia Online First Publication, April 3, 2017. http://dx.doi.org/10.1037/dec0000076 CITATION Davis, A. L., Miller, J. H., & Bhatia, S. (2017, April 3). Are Preferences for Allocating Harm Rational?. Decision. Advance online publication. http://dx.doi.org/10.1037/dec0000076 Decision © 2017 American Psychological Association 2017, Vol. 1, No. 999, 000 2325-9965/17/$12.00 http://dx.doi.org/10.1037/dec0000076 Are Preferences for Allocating Harm Rational? Alexander L. Davis, John H. Miller, and Sudeep Bhatia Carnegie Mellon University The allocation of nonmonetary harm is an important—yet understudied—domain of choice. Using a modified Dictator Game, we asked 27 participants to allocate a harmful event (time of putting their hand in ice water) between themselves and an anonymous stranger. We found substantially less coherent, and more egalitarian, preferences compared to other studies that ask participants to allocate monetary endowments. Specifically, 26% of participants made choices inconsistent with utility maximization, and 78% of participants behaved in an egalitarian manner. In comparable studies of monetary gains, only 2% were inconsistent and 30% egalitarian. The results suggest that the focus on monetary gains likely overestimates the rationality of other-regarding preferences and underestimates egalitarianism. Keywords: altruism, prosocial behavior, rationality, revealed preference Research in the last two decades has illumi- an individual going off to war, to the more nated our understanding of costly, other- mundane, like taking out the trash. regarding behavior (Andreoni & Vesterlund, We emphasize nonmonetary bads for a vari- 2001; Benabou & Tirole, 2006; Fehr & ety of reasons. First, as discussed above, it is a Schmidt, 2006; Fisman, Kariv, & Markovits, domain where decisions are frequently made, 2007; Forsythe, Horowitz, Savin, & Sefton, including examples that can be quite common 1994; Frey & Bohnet, 1995). Most of this work (like chores), to rather astounding, such as when has used experiments like the Dictator Game an individual risks his life to save the lives of to show that, in general, individuals are willing strangers (see, for just one example, Dovidio, to share a monetary windfall with an anony- Piliavin, Schroeder, and Penner’s (2006) dis- mous stranger in a way that is consistent with cussion of Paul Rusesabagina who helped Tut- some set of well-behaved (utility-maximizing) sis and Hutus escape the Rwandan genocide by preferences (Andreoni & Miller, 2002). This letting them stay in his hotel). Between these prior work focused almost exclusively on the extremes include the donation of blood and allocation of monetary gains. Here we examine organs, serving in the armed forces, participat- how individuals allocate nonmonetary bads, ex- ing in a revolution, or even becoming a suicide amples of which range from the heroic, such as bomber. Notwithstanding the importance and real-life consequences of nonmonetary bads, lit- tle systematic study has been devoted to under- standing behavior in this domain. Second, there are rarely markets or prices for these bads, allowing us to test whether existing models of Alexander L. Davis, Department of Engineering and Pub- This document is copyrighted by the American Psychological Association or one of its allied publishers. lic Policy, Carnegie Mellon University; John H. Miller and other-regarding preferences generalize to a This article is intended solely for the personal use of the individualSudeep user and is not to be disseminated broadly. Bhatia, Department of Social and Decision Sci- novel context. Third, decisions about allocating ences, Carnegie Mellon University. harm likely invoke types of reasoning (e.g., Sudeep Bhatia is now at the Department of Psychology, moral-deontological) that may be inconsistent University of Pennsylvania. We would like to thank Terence Einhorn, John Sperger, with utility maximization. Finally, previous and April Jintao for their valuable assistance in helping work in psychology has found that people are conduct the research. willing to experience harm in place of another Correspondence concerning this article should be ad- person. For example, a study by Batson et al. dressed to Alexander L. Davis, Department of Engineering and Public Policy, Carnegie Mellon University, 5000 (1983) found that 65% of participants volun- Forbes Avenue, Pittsburgh, PA 15213. E-mail: alexander.l teered to take electric shocks in place of another [email protected] participant (who was actually an experimental 1 2 DAVIS, MILLER, AND BHATIA confederate), and on average agreed to take to the pair. As a result, a participant with Le- more than 50% of the shocks (although none of ontief preferences tries to equalize the outcomes the shocks were actually implemented). In com- for herself and the other participant, behaving in parison with the typically more modest rates of an egalitarian manner. Finally, the utility func- generosity observed in Dictator Games using tion for a dictator with Perfect Substitute pref- ϭ ϩ monetary gains, these proportions raise the pos- erences is U(xs, xo) xs xo, meaning the sibility of greater generosity for allocations of outcomes for herself and the other person are nonmonetary harm. However, differences in ex- treated interchangeably. A participant with Per- perimental design, including the use of decep- fect Substitutes preferences will maximize the tion, lack of anonymity, and available informa- total amount of the good obtained by the pair. tion about the victim, limit comparability across these studies. In this work we provide a more direct comparison using a Dictator Game with Method allocations of nonmonetary harm. Our research strategy uses an approach cre- Participants ated by Andreoni and Miller (2002), where par- ticipants are asked to make a series of choices In advance we aimed to collect 25 pairs of that involve allocating a resource between dictators and receivers, a number we considered themselves and another person. In our modifi- large enough to detect interesting effects given cation, each choice involves a different tradeoff our budget constraints. Twenty-seven pairs of between harming another person and oneself. participants were ultimately recruited through Given these choices, we test whether the partic- the Center for Behavioral Decision Research ipant’s behavior is consistent with the General- website. Sessions were run in groups of four to ized Axiom of Revealed Preference (GARP; Af- eight participants (two to four pairs). The aver- age age of the dictators was 24 years (with a riat, 1967; Varian, 1982). GARP captures the 1 intuition that inconsistency in observed choices range from 18–59), 14 were women. can reveal irrational decision making. For ex- ample, if a decision-maker chooses a bundle of Procedure goods A when an alternative bundle B could Participants were randomly assigned an iden- have been chosen at the same price, the choice tification number unknowable to others using reveals that A is preferred to B (or indifferent to shuffled opaque sealed envelopes. After random B). If we then observe the same decision-maker assignment, participants were told that they choosing bundle B when A is cheaper than B,an would be paid $10 in cash at the end of the inconsistency would arise. Remarkably, a for- experiment. They were informed that they malization of this simple idea captures all of the would be split into two rooms (based on observable consequences of the standard utility whether they had an even or odd identification theory used by economics. number), and that each odd-numbered partici- In addition to examining the consistency of pant (dictator) would make a decision that choices, we also explore the diversity of pref- would affect an even-numbered participant (re- erences that arise in this context. Specifically, ceiver), but that even-numbered participants Andreoni and Miller (2002) found that partici- would not make any choices that affected the pants’ preferences are well captured by three 2 This document is copyrighted by the American Psychological Association or one of its allied publishers. odd-numbered participants. All participants types of functions: Selfish, Leontief, and Perfect This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. were assured that their choices throughout the Substitutes. If we call the dictator’s utility func- experiment would be anonymous. Before mak- tion U(x , x ), where x is the amount of good s o s ing their choices, participants came to the “Ice allocated to the dictator, and xo is the amount of good allocated to the other person, then a Self- 1 ish participant has the simple function U(xs, All de-identified data, R code for analyses, and materi- ϭ xo) xs, which is only dependent on the payoff als associated with the study are available through the Open to herself. The utility function for a dictator Science Framework at http://osf.io/82jpk. ϭ 2 We note that even though the receiver was anonymous, with Leontief preferences is U(xs, xo) min(xs, there was exactly one receiver, implying the existence of an xo); that is, the utility experienced is only as identifiable victim effect. However, such an effect is also large as the minimum amount of good allocated present in other Dictator Game studies. ALLOCATING HARM 3 Table 1 Parameters for the Eight Choices Made by Each Dictator, and Choice Patterns of Time Not in Water (xs,xo) for Each of the Preference Types ´ Decision Tokens (M) Hold (ms) Pass (mo) M p Selfish Leontief PerfectSub 1 60 1 1 60 1 (60, 0) (30, 30) (0–60, 0–60) 2 60 1.5 1 60 1.5 (60, 0) (24, 24) (60, 0) 3 60 2 1 60 2 (60, 0) (20, 20) (60, 0) 4 60 3 1 60 3 (60, 0) (15, 15) (60, 0) 5 60 1 1.5 40 .67 (40, 0) (24, 24) (0, 60) 6 60 1 2 30 .50 (30, 0) (20, 20) (0, 30) 7 60 1 3 20 .33 (20, 0) (15, 15) (0, 20) 8 80 1 1 40 1 (40, 0) (20, 20) (0–60, 0–60) Water Experience” section of the questionnaire, Each dictator’s allocation of tokens was trans- where they were asked to raise their hand to formed into a period of time she and the re- notify the experimenter.