Interdependent Preferences and Reciprocity

ju05_Article 2 6/10/05 1:40 PM Page 392

Journal of Economic Literature Vol. XLIII (June 2005), pp. 392–436

∗ JOEL SOBEL

1. Introduction I pay particular attention to how reciprocity influences decision making. Reciprocity uch of economic analysis stems from the refers to a tendency to respond to perceived joint assumptions of rationality and indi- M kindness with kindness and perceived mean- vidual greed. Common sense and experimen- ness with meanness and to expect this tal and field evidence point to the limits of this behavior from others. I introduce models of approach. Not everything of interest to econ- intrinsic reciprocity in section 3.4. omists can be well understood using these Intrinsic reciprocity is a property of prefer- tools. This paper reviews evidence that narrow ences. The theory permits individual prefer- conceptions of greed and rationality perform ences to depend on the consumption of badly. The evidence is consistent with the view others. Moreover, the rate at which a person that economic incentives influence decision values the consumption of others depends making. Hence there is a role for optimizing on the past and anticipated actions of others. models that relax the assumption of individual An individual whose preferences reflect greed. I discuss different ways in which one intrinsic reciprocity will be willing to sacri- can expand the notion of preferences. fice his own material consumption to increase the material consumption of others ∗ Sobel: University of California, San Diego. I present- in response to kind behavior while, at the ed a version of this paper at the First World Congress of same time, be willing to sacrifice material the Game Theory Society and to my colleagues at the consumption to decrease someone else’s Center for Advanced Study in the Behavioral Sciences. I thank Eli Berman, Antonio Cabrales, Miguel Costa- material consumption in response to unkind Gomes, Vincent Crawford, David M. Kreps, Herbert behavior. Gintis, Mark Machina, Efe Ok, Luís Pinto, Matthew It is more traditional to view reciprocity as Rabin, Paul Romer, Klaus Schmidt, Uzi Segal, Joel Watson, and Kang-Oh Yi for discussions, references, and the result of optimizing actions of selfish comments. I am especially grateful to two referees who agents. Responding to kindness with kind- supplied detailed, intelligent, and constructive comments ness in order to sustain a profitable long- on an earlier version of the manuscript and to John McMillan for his advice and encouragement. I worked on term relationship or to obtain a (profitable) this project while a Fellow at the Center for Advanced reputation for being a reliable associate are Study in the Behavioral Sciences. I thank my classmates at examples of instrumental reciprocity. the Center for conversations and encouragement and the Center for financial and clerical support. NSF funding is Economics typically describes instrumental also gratefully acknowledged. reciprocity using models of reputation and

392 ju05_Article 2 6/10/05 1:40 PM Page 393

Sobel: Interdependent Preferences and Reciprocity 393

repeated interaction. This approach is quite introduce and motivate the ideas I review in powerful as essentially all exchanges in natu- the manuscript. ral settings can be viewed as part of some 2.1 Descriptions long-term interaction. Consequently one could argue that the models of section 3.4 There are several ways to react to Paul’s are unnecessary. This essay presents the destructive activity. One response is to treat it counterargument that models of intrinsic as an emotional response not subject to eco- reciprocity can provide clearer and more nomic analysis. We should not give up so eas- intuitive explanations of interesting econom- ily. Paul may be crazy, but his former boss, ic phenomena. An openness to the possibili- Marsha, probably is not. Unless Paul’s actions ty of intrinsic reciprocity leads to a new and are completely unrelated to the environment, useful perspective on important problems. Marsha will want to understand how to The next section contains a stylized exam- reduce the chance of adverse behavior. ple that illustrates the limitations of standard Marsha may need to hire lawyers or psychi- models. I use the example to provide an atric consultants (instead of economists) to tell informal introduction to alternative theoret- her how to deal with Paul or reduce the risk of ical approaches and motivate the paper. costly outbursts by employees, but she should Section 3 introduces these models formally. evaluate her options using economic models. Section 4 describes some economic settings I will concentrate on descriptions that are in which the modeling approaches of section consistent with the hypotheses of optimiza- 3 may be particularly useful. Section 5 tion and equilibrium. Once we allow that reviews literature on the evolution of prefer- Paul maximizes something more that his own ences. Section 6 responds to stylized argu- monetary reward, there are many stories like ments against the approach and section 7 is this available. This section introduces some a conclusion. potential descriptions informally. Section 3 provides a more systematic treatment. 2. An Informal Guide to the Concepts 2.1.1 Static Income Maximization We regularly read accounts of dissatisfied Hypothesis. The narrowest version of or recently fired employees destroying prop- economic theory assumes that Paul seeks to erty, sabotaging computer files, or even maximize his utility and that his utility “going postal” and killing people at their depends on the quantities of the private workplace. The sense of outrage at an appar- material goods he consumes. In static ent injustice is real. Many people are willing income maximization, Paul balances the to take destructive actions as part of the out- immediate cost and benefits of actions rage. This kind of destructive behavior is rather than the long-term implications of unlikely to be in the material interest of a these decisions. In simple settings, this fired employee: it takes time, it is not com- hypothesis reduces to the assumption that pensated, and it carries the risk of criminal Paul maximizes his monetary income. penalties. How should we think about it? Analysis. Paul would carry out his For concreteness, imagine that Paul was a destructive action only if he imagined that it high-ranking executive who had worked for would lead to direct material gain. It is hard a company for more than ten years. He lost to rationalize Paul’s behavior under these his job when business turned bad. On his last assumptions. His actions may advance his day on the job, Paul destroyed vital company immediate material interests if Marsha gives documents and continued to sabotage com- him back his job or if he receives a payment puter files until he was caught six months to stop sabotaging the company, but a more later. In this section, I will use Paul’s story to elaborate description seems necessary. ju05_Article 2 6/10/05 1:40 PM Page 394

394 Journal of Economic Literature, Vol. XLIII (June 2005)

2.1.2 Interdependent Preferences security. By hurting Marsha, Paul gains because he positions himself to get another Hypothesis. Paul maximizes a utility job (which he may be less likely to lose) or sell function that depends on Marsha’s consump- a book about his experiences. Under these tion of material goods in addition to his own. circumstances, Paul may have preferences Analysis. If Paul’s utility is decreasing in defined over both his monetary wealth and the material wealth of Marsha, then Paul will his “sense of honor.” If the preferences are be willing to sacrifice his own material well increasing in both arguments, then he will be being to punish Marsha. These preferences willing to make material sacrifices in order to explain why Paul would wish to harm increase in honor. If Paul only cares about Marsha, but do not explain why he waits honor because it enables him to increase his until after he is fired to do so. There are two monetary payoff, then this explanation possibilities. In the midst of an ongoing reduces to income maximization. employment relationship, Paul does not Another possibility is that Paul takes pleas- harm Marsha because he fears that Marsha ure directly from the act of sabotage. That is, will fire him, which would be a sufficiently his preferences contain an additional argu- great punishment to deter him from hurting ment (“sabotage”). Paul will not maximize Marsha. his material payoff, but he is selfish and goal Alternatively, the marginal rate of substi- oriented. This explanation does not explain tution between Paul and Marsha’s material why Paul turns to sabotage only after he was income in Paul’s preferences may change as fired. Perhaps the advantage of maintaining a result of Paul’s termination. The impover- the employment relationship deterred his ished Paul is willing to sacrifice to make urge to destroy files until he was fired, but Marsha worse off. This explanation only this explanation suggests that the sabotage works if Paul’s income after being fired is levels would be the same whether the lower than after a voluntary separation (oth- employee was fired or separated voluntarily. erwise any separation would trigger Paul’s disruptive behavior). 2.1.4 Intrinsic Reciprocity For the example, it makes sense to assume Hypothesis. Paul’s utility depends on the that Paul’s utility is decreasing in Martha’s material wealth of Marsha. Moreover, Paul’s income. The interdependent preference perception of Marsha’s behavior determines approach permits Paul to be willing to sacri- the direction of Paul’s preferences. The mar- fice material welfare to decrease the income ginal utility of Paul with respect to an of others. increase in Marsha’s income increases when 2.1.3 Preferences over General Marsha is kind to Paul and decreases when Consumption Goods Marsha is unkind. Analysis. Paul believes that Marsha acted Hypothesis. Paul maximizes a utility unfairly when she fired him. Consequently, function that is a function of “consumption his preferences changed and he becomes goods” that are derived from marketed goods (more) willing to sacrifice his own income to through a personal production process. return the insult he received. Analysis. This approach generates several 2.1.5 Commitment possible stories. One possibility is that Paul’s behavior demonstrates that he has a mar- Hypothesis. Paul seeks to maximize a ketable characteristic—that is, he is not the utility function that depends only on his type of person who can be pushed around, he material consumption. He can commit him- is not afraid to stand up for injustice, or he is self to taking future actions that are in his capable of identifying weaknesses in a firm’s best interest when he makes his plans, but ju05_Article 2 6/10/05 1:40 PM Page 395

Sobel: Interdependent Preferences and Reciprocity 395

may not be in his best interest when he This hypothesis works like the commit- enacts his plans. ment hypothesis in this example, and it has Analysis. If fired workers retaliate, the same problem: Paul lacks incentives to Marsha might be reluctant to fire people or carry out his threat after he loses his job. she may offer attractive separation packages. For the repeated-game hypothesis to apply, These actions benefit workers, so if workers Paul must expect to receive benefits after he could commit to destructive actions after punishes Marsha that he would not receive being fired, then it might be in their interest otherwise. These benefits may come to do so.1 Marsha fired Paul because she because Paul receives rewards from third concluded that he would do more damage as parties (friends or future employers) after an employee than not, but there is no expla- he punishes Marsha. nation for why Paul carried out his threat 2.2 Sorting out the Explanations after he was fired. Standard equilibrium concepts rule out this kind of commitment. In this section, I have presented several For this reason, I do not treat commitment explanatory models in an informal context. At as a description consistent with equilibrium a conceptual level, it is sometimes difficult to and optimization. I discuss evolutionary differentiate the models. I make an attempt arguments for why individuals may maintain to do so in the next section, where I discuss commitment ability in section 5. the explanations more precisely in an explicit game-theoretic context. Nevertheless, the 2.1.6 Repeated Games distinction is often linguistic: The different Hypothesis. Paul seeks to maximize his approaches sometimes just use different material consumption, but he views the rela- terminology to describe the same thing. tionship as ongoing. More precisely, he is At an empirical level, the different engaged in a repeated game with Marsha, descriptions identify different reasons for and he seeks to maximize a discounted sum Paul’s behavior. Fully specified models will of single-period payoffs. lead to different ways to organize the work Analysis. Actions have implications for environment in response to the threat of dis- future payoffs in repeated games, so equilib- satisfied workers. On the other hand, given a rium behavior does not require short-term particular observation, it is usually possible maximization. It is natural to assume that to construct a “just-so” story from the per- Paul’s destructive action imposes a short- spective of one’s favorite class of descriptions term cost on Paul but an even greater cost that is consistent with the observation. My on Marsha, punishing her for firing him. taxonomy does not generate a fully specified Equilibrium strategies in repeated games testable model in each category, but rather a often specify that one player punish another family of models with the same underlying player following a deviation from equilibri- mechanism. Rejecting a model is easy, um behavior. In the simplest cases, potential rejecting an approach is nearly impossible. punishments deter the behavior that would Natural formulations of the different trigger them. So one would never see pun- hypotheses do have different implications in ishments. When Marsha is uncertain about some situations, however, in part because Paul’s willingness or ability to sabotage, it they have different conceptions of what the might be in her best interest to take actions benefit of Paul’s actions are. Factors that that lead to Paul’s destructive behavior. may distinguish the descriptions are how long Paul has worked for Marsha and the nature of their relationship, how widely 1 Alternatively, Marsha might institute tighter security or announce that she will seek severe punishments for known his destructive actions become, how destructive activities. much discretion Marsha has in her decision, ju05_Article 2 6/10/05 1:40 PM Page 396

396 Journal of Economic Literature, Vol. XLIII (June 2005)

and how old Paul is when he is fired. I close The different implications of the hypothe- the section with a few examples. ses imply that more precise versions of the In the repeated-game explanation, Paul stories can be supported or rejected by data. punishes Marsha to influence his future The different hypotheses also suggest differ- returns. If his future returns decrease with ent ways for Marsha to modify the environ- age and the cost of sabotage does not ment to improve outcomes. If Marsha change, then the older Paul is, the less likely thought that the repeated-game or general- he is to sabotage Marsha. It is not necessary ized consumption good hypotheses were the in the repeated-game story for Paul’s behav- best explanations of Paul’s destructive behavior to depend on the amount of severance ior, she would try to reduce sabotage by pay he received or the number of other manipulating Paul’s incentives after he leaves employees laid off. Simple stories based on the firm. If Marsha thought that Paul was intrinsic reciprocity would not predict Paul’s motivated by intrinsic reciprocity, then she behavior to depend on his age. Stories based would focus on changing behavior during the on interdependent preferences would pre- employment. dict that the wealth of Marsha and other employees would influence Paul’s behavior. 3. Models On the other hand, considerations based on reciprocity are largely retrospective. Paul Whenever a theory appears to you as the only would be less inclined to punish Marsha if he possible one, take this as a sign that you have thought that she had no choice but to fire neither understood the theory nor the problem him or if steps were taken to inform him and which it was intended to solve. reward him for service prior to separation. Karl Popper One would expect destructive behavior to Several different models have been devel- decrease with the amount of goodwill Marsha oped to describe and organize the evidence has accumulated during their relationship. of nonselfish behavior. No model provides a The consequences of Paul’s behavior after he complete description of the observational is fired are less important under the hypoth- findings. A sensible approach will take ideas esis of reciprocity than in the repeated-game from different models. This section parallels story or general consumption good story. section 2. It introduces formal models for The simplest explanation of Paul’s behav- the different descriptions of Paul and ior based on reciprocity involves only the Marsha’s conflict and the theoretical ques- relationship between Paul and Marsha. Paul tions raised by the use of these models. I destroys data to hurt Marsha. He may want maintain the assumption that individuals it known that something bad has happened have well defined preferences and they to the company because this revelation may behave to maximize their preferences sub- hurt Marsha, but he gains nothing from ject to resource constraints. For this reason, advertising his own connection to the crime. there is a clear sense in which all of the For at least some of the explanations based behavior I discuss is selfish. I will use the on generalized consumption goods or term selfish preferences in a more limited repeated games, it is important Paul’s associ- way to mean preferences that do not direct- ation with the sabotage becomes public. ly depend on the consumption of others.2 The interdependent-preference hypothe- Preferences are altruistic if they are increasing sis predicts less punishment if it is clear that in the material consumption of others. Marsha is suffering material losses when she fires Paul. It would be to Marsha’s advantage 2 The discussion of evolutionary models in section 5 to coordinate firings with reductions in pay permits a less arbitrary identification of selfishness with of employees under these circumstances. fitness maximization. ju05_Article 2 6/10/05 1:40 PM Page 397

Sobel: Interdependent Preferences and Reciprocity 397

The narrowest formulation of rationality an intrinsic preference for reciprocity will demands that an economic agent maximize a have different preferences over allocations utility function that depends only on his cur- depending on the context. Unkind behavior rent consumption of material goods. This of their neighbors induces destructive reci- formulation is easily refuted, but it is an procity while kindness induces constructive unnecessarily restrictive view of rationality. reciprocity. Section 3.4 examines models in General models of rational behavior permit which the process that creates outcomes a wider range of arguments to enter utility influences preferences. These models are functions. Somewhat arbitrarily, I classify formal representations of intrinsic reciprocity. the models according to the way they extend Section 3.5 briefly discusses how commit- preferences. Section 3.2 looks at models in ment power influences predictions. Section which an individual cares about people other 3.6 discusses how the theory of repeated than himself. Section 3.3 discusses models in interactions may lead selfish individuals to which preferences depend on more things behave as if they cared about the welfare of than marketed goods. others. This provides a foundation for theories Paul’s behavior toward Marsha illustrates of instrumental reciprocity.4 an apparent willingness to risk material well Throughout this section, I illustrate mod- being in order to damage someone else. Spite, els using the ultimatum game. The ultima- outrage, moralistic aggression, and the desire tum game provides a powerful challenge to for revenge are behaviors that are as familiar the hypothesis that income maximization to social scientists as they are inconvenient to and equilibrium describe economic interac- the economists’ narrow notion of self interest. tions. In the ultimatum game, two players But while people will go out of their way to bargain over the distribution of a surplus of harm enemies, they will make sacrifices to fixed size 1. The first player (proposer) can help their friends. I will call repaying unkind- choose any distribution of the surplus ness with unkindness destructive reciproci- s ∈ [0,1]. The second player (responder) then ty and repaying kindness with kindness either accepts or rejects the proposal. If the constructive reciprocity.3 Individuals with responder accepts the proposal s, then the proposer’s monetary payoff is 1 − s and the 3 Reciprocity has many definitions, so it is not surprising responder’s monetary payoff is s. Otherwise, that adjectives modifying reciprocity take on different both receive nothing. meanings depending on the author. Anthropologists Marshall Sahlins (1968, page 82) and Elman R. Service Game theory, assuming that players seek (Service 1966, pages 14 and 15) generously credit each to maximize their monetary payoff, makes other for definitions of generalized, balanced, and negative two predictions. First, in Nash Equilibrium, reciprocity. While economists do not use the terms generalized and balanced reciprocity, they use negative reciprocity all positive offers must be accepted. Second, to describe the tendency to punish people who treat you in subgame-perfect Nash Equilibrium, the badly. On the other hand, for Sahlins and Service negative proposer must offer s = 0 (or, if the set of reciprocity describes many standard economic “transactions = opened and conducted towards net utilitarian advantage’’ feasible proposals is discrete, either s 0 or 5 (Sahlins 1968, page 83). To avoid unnecessary interdiscipli- the minimum positive proposal). nary confusion, I propose the terms destructive and con- The ultimatum game is a beautiful subject structive reciprocity as alternatives to negative and positive reciprocity. This is not the only possible source of confusion. for experimental study. It is simple. The Richard D. Alexander (1987) uses the term direct reciprocity to describe bilateral exchanges of favors: the person 4 Ernst Fehr and Simon Gächter (2000) and Fehr and receiving a benefit compensates the person who generated Klaus M. Schmidt (2003) also review of the material in the benefit and indirect reciprocity when the return favor this section. comes from a third party. Robert L. Trivers (1971) uses gen- 5 It is a Nash equilibrium for the proposer to offer s > 0 eralized reciprocity in the sense that Alexander uses indirect and for the responder to accept any offer greater than or reciprocity. Herbert Gintis (2000) uses weak reciprocity to equal to s. This equilibrium fails to be subgame perfect describe reciprocal interactions that are instrumental and because it relies on the responder’s threat to reject positive strong reciprocity in the way that I use intrinsic reciprocity. offers less than s. ju05_Article 2 6/10/05 1:40 PM Page 398

398 Journal of Economic Literature, Vol. XLIII (June 2005)

combination of payoff maximization and payoffs. For simplicity, xi will refer to a rationality leads to sharp predictions. monetary payoff in this section. This model Experimental subjects repeatedly violate the is simple and leads to clear predictions. The theoretical predictions.6 The violations are predictions are systematically wrong in systematic: low (s < .2) proposals are rare; many interesting situations. It is this model, proposals are rejected; proposals rarely give combined with the assumption that players the second player more than half of the sur- use (subgame perfect) equilibrium strate- plus (so that s > .5 is rarely observed); and gies, that leads to the prediction that the equal or nearly equal splits (.4 < s ≤ .5) are first player demands essentially everything common, occurring more than half the time in the ultimatum game. in typical experiments. In addition, rejections 3.2 Interdependent Preferences from the responder decrease as s increases. This finding is consistent with, although This subsection describes models that much weaker than, the stark theoretical pre- assume individuals seek to maximize well- diction since subjects are more likely to defined preferences, and that base predic- choose actions that maximize their material tions on equilibrium behavior, but permit payoffs the larger the (material) benefit asso- preferences to depend on the consumption ciated with doing so. These results continue of others. to hold under a range of conditions. As in the case of income maximization, let = Before describing the models, I introduce O(s) (x1,…,xI) denote the outcome, where basic notation. Limit attention to a strategic xi is the material allocation of player i. With environment with I agents. The strategy set interdependent preferences, agents care

of agent i is si. Any strategy profile about the distribution of material goods and = ∈ = s (s1,…,sI) (where si Si for all i 1,…,I) not simply their own allocation. That is, determines an outcome O(s). Conventional ui(O(s)) may now depend nontrivially on xj, game theory adds to this formulation the for j ≠ i. assumption that agents have well defined Several authors have proposed specific preferences over outcomes. Assume that functional forms for interdependent prefer- agent i’s preferences can be represented by ences. For these authors, material allocations

a utility function ui(.) defined over out- are one dimensional—conformable to mone- comes. Without additional assumptions, self- tary payoffs in an experiment. To review ish behavior is not defined. The abstract these models, consider the utility function definition of outcome does not identify a ()=+∑λ () − consumption bundle for each agent. (1)uxii xij x i x j x j . ji≠ 3.1 Income Maximization The simplest form of interdependent = In this setting, O(s) (x1,…,xI), where xi preferences consistent with (1) arises when is an allocation to player i and ui(O(s)) is an ij is constant. A positive ij reflects increasing function of xi and independent of altruism (in the sense that an agent is willing ≠ xj for j i. That is, the outcome consists of to decrease his own consumption in order to private goods allocated to each player and increase the consumption of another agent); each player cares only about his or her own a negative ij reflects spite. consumption. In many applications the pri- Gary Charness and Matthew Rabin (2002) vate goods are one-dimensional monetary and Fehr and Schmidt (1999) offer specifications that are special cases of (1). These papers impose the further restriction that 6 See Werner Güth, Rolf Schmittberger, and Bernd Schwarze (1982) for early experiments and Güth (1995b) ij is independent of i and j and depends − and Alvin E. Roth (1995) for reviews. on only the sign of xi xj. ju05_Article 2 6/10/05 1:40 PM Page 399

Sobel: Interdependent Preferences and Reciprocity 399

Charness and Rabin (2002) opt for an his opponent in a dictator-game version of average of functional forms that place posi- the ultimatum game (in which the second tive weight on the selfish monetary payoff, player is required to accept any feasible pro- the monetary payoff received by the least posal). Allowing the sign of to change well off agent, and the total payoff. With depending on the income distribution is > 7 this specification, ij 0, but is greater consistent with this behavior. > when xi xj. That is, individual i always David K. Levine (1998) assumes that the places positive weight on the consumption extent to which agents care about another of others and places more weight on the player’s material utility is a weighted average consumption of individuals poorer than he of a pure altruism parameter and the altru- is than on richer individuals. ism parameter of the other player. Levine For the inequity aversion approach sug- assumes that people differ in the degree to gested by Fehr and Schmidt, ij is positive which they care about others and that peo- > < if xi xj and negative if xi xj. Under this ple care more about the material payoffs of specification, an agent cares about his own nice people. Formally, he assumes that i monetary payoff and, in addition, would like maximizes to reduce the inequality in payoffs between αβα+ +∑ iij the two players. Gary E. Bolton and Axel xxi j . ji≠ 1 + β Ockenfels’s (2000) ERC (for “Equity, i Reciprocity, and Competition”) model has a Here i is the altruism parameter of play- similar motivation, but proposes a utility er i and i is the weight player i places on j’s = function that is not in form (1). Instead, preferences. If i 0, then the weight player Bolton and Ockenfels assume agent i’s pref- i places on j’s material payoff is independent erences are an increasing (possibly nonlin- of j’s degree of altruism; otherwise, the

ear) function of xi and agent i’s relative weight is an increasing function of j’s altru-  x  ism parameter. This specification is a special income  i  . case of (1). In contrast to inequity aversion,  ∑N  j=1 x j the weight placed on the material payoff of The simplest model of interdependent another player depends on the identity of preferences provides the flexibility to that player. In Levine’s model, an individual describe some observed violations of income wants to be kind to a kind person. Levine maximization in the ultimatum game. For uses this model to describe experimental example, if the responder is spiteful, so that results. In order to do so, he assumes that =− players are uncertain about their opponent’s ij is constant and negative, then when she is offered the share s she will reject it if preferences and solves for the equilibrium s − (1 − s) < 0. If both players have ∈ (0,1), of incomplete information games. Agents then the unique equilibrium of the ultima- want to identify altruistic (high ) people so tum game would be for the first player to that they can be nice to them. Players draw γ inferences from the strategies of other people. This permits a form of reciprocity to offer the second player the share + γ , 1 arise in equilibrium. If agents with higher which is positive but less than one half. s choose nicer strategies, then players This resolution is unsatisfactory. Intuition suggests that at least some of the behavior in 7 Models of interdependent preferences predict all the ultimatum game comes from generosity positive offers are accepted when ij is nonnegative. and not fear of rejection. One would not Hence Charness and Rabin (2002) also combine their functional form with preferences that depend on context expect a spiteful individual to make charita- in order to explain some of the observed responder ble contributions or give a positive share to behavior in ultimatum games. ju05_Article 2 6/10/05 1:40 PM Page 400

400 Journal of Economic Literature, Vol. XLIII (June 2005)

place higher weights on the material pay- the subject’s utility function? Subjects have offs of people who play nice strategies, some idea that the experimenter is budget because playing nice strategies signals that constrained (or at least the money from the you really are nice. experiment comes from somewhere). So all Models of interdependent preferences allocations are just transfers. Since the total raise theoretical issues about how to deter- monetary payments are constant, utilitari- mine the arguments of the utility function. an objectives are not relevant. Also, the Imagine an agent who is motivated by the subjects must be aware that they make desire to maximize a utility function that more than one decision during an experi- reflects inequity aversion. Exactly whose util- mental session. Even if subjects do not ity enters into this function and how should bring their lifetime decision problem into this utility be measured? An experimental the lab, they may impose notions of fair- subject could try to maximize her monetary ness or social preferences across the entire payoff in the laboratory and then redistribute experimental session rather than just one her earnings to deserving people later. This decision at a time. behavior would be appropriate if the concern An optimizing agent whose utility function for inequity was “broadly bracketed” (Daniel places positive weight on the wealth of oth- Read, George Loewenstein, and Rabin 1999 ers could allocate income outside the lab and Richard H. Thaler 1999) in that con- carefully so that one could not infer the true cerns outside of the laboratory entered into nature of preferences from lab behavior. the decision making in the laboratory. If the This agent would optimize his interdepend- second player in the ultimatum game ent preferences before entering the labora- learned the proposer was relatively poor, tory, so at the margin he would be would she be willing to accept small offers?8 indifferent between allocating his laboratory If experimental subjects had to earn the right winnings to increase his personal income or to play a role in a game in which the equilib- to decrease inequality. If one accepts the rium prediction (assuming narrowly self- possibility that laboratory behavior takes into interested behavior) gave unequal payoffs account decisions made outside of the labo- would the effect of inequity aversion be ratory, then this argument suggests that reduced? Would it matter whether attractive experiments overestimate the amount of positions were allocated by scoring high on a (narrowly) selfish behavior, since selfish peo- test of general intelligence or by a pseudo- ple must be selfish in all situations and oth- random device, like being born on an odd ers may appear selfish in the laboratory in day of the year? Do winners of lotteries order to pursue their nonselfish interests attempt to find out the names of the losers in outside the laboratory more effectively.9 order to reduce inequality (or increase the There are tacit assumptions in the mod- wealth of the agent who did poorly on this els of interdependent preferences. The particular transaction)? modeler makes the assumptions when spec- Concerns that always arise in experimen- ifying the identities of the players in the tal settings are especially salient here. game and their initial wealth levels. In Should the experimenter’s payoff enter into applications, subjects are assumed to care only about the welfare of other active par- ticipants in the game and to make relative 8 Jacob K. Goeree and Charles A. Holt (2000) attempt to test for this effect in the laboratory. They vary the lump- sum payment received by subjects in a perfect-information 9 On the contrary, one could also argue that subjects bargaining game. Results from the experiment are consis- may have more incentive to curb their selfish instincts in tent with the hypothesis that subjects take into account the laboratory than in other settings if convincing fellow these payments, which are irrelevant to standard models of experimental subjects and experimenters that they are not the bargaining process. selfish is the best way to gain future riches. ju05_Article 2 6/10/05 1:40 PM Page 401

Sobel: Interdependent Preferences and Reciprocity 401

income comparisons based only on the (2) U(Z1,…,Zm) material payoffs of the game. It is typical where for i = 1,…,m (and necessary) to identify a “small world” in which to apply decision-theoretic argu- = ƒ (3) Zi i(X1i,…,Xni,t1i,…,t1i,S1,…,Sl,Yi) ments (particularly in models involving

choice under uncertainty). The problem Zi are the generalized consumption goods, ƒ seems especially critical when using inter- i is the production function for commod- dependent preferences, however. In decid- ity i, Xji is the quantity of the jth market ing how to interpret these models, one must good, tki in the time input of individual k,Sk understand why the subject cares about the is the human capital of the kth person, and

utility of other subjects, but not the utility of Yi represents all other inputs. Given wages the experimenter. In deciding how to apply and prices for the market goods, a house-

these models to a contracting problem, one hold selects Xji and tki to maximize (2) sub- must decide whether preferences are ject to (3). In any application, the definition defined over coworkers or just the parties to of market goods will not be controversial.

the contract. In deciding how to apply these The quantities Xji and tki will be observable. models to the labor market, one must In the general specification of the theory, decide whether workers care about inequity however, the levels of human capital, the across labor and management, across all functional form of the production functions, workers, or only across workers in similar and, indeed, the nature of generalized con- jobs. One must also decide whether to sumption goods can be freely selected by invoke an interdependent utility function to the modeler. determine decisions separately or whether While the approach of Stigler and Becker agents make decisions that reduce inequity is firmly grounded in ideas familiar to econ- over longer intervals. omists—budget constraints and individual maximization—it also reflects a common 3.3 Preferences over General Consumption view among cultural anthropologists. Sahlins Goods (1968, page 9), who provides a useful taxon- The “Chicago School” pursues the goal of omy of reciprocity, observes that “in an using the optimizing models of self-interest- uncommon number of tribal transactions ed agents constrained by a market environ- material utility is played down, to the extent ment—a limited endowment, existing that the main advantages appear to be social, prices, and economic institutions—to the gain coming in good relations rather than explain a broad range of economic phenom- good things.” Sahlins recognizes the need to ena. The approach is advocated powerfully look beyond immediate material gain to in the work of George J. Stigler and Gary S. understand the workings of simple Becker (for example, Stigler and Becker economies. While Stigler and Becker do not 1977). The theory explicitly exploits the pos- explicitly substitute “good relations” for sibility that self interest has a broad defini- “good things,” their formulation broadens tion. Preferences are not defined over the notion of consumption good and con- marketed goods, but general commodities cedes that economic exchange may be moti- that individuals transform into consumption vated by more than short-term material gain goods. Models in this tradition therefore do even in developed market economies. not require that experimental subjects base When stripped to its mathematical core, their decisions solely on their own monetary the Stigler–Becker model posits that decision- payoffs. makers have preferences over their choice Stigler and Becker’s posit that the decision- variables and that these preferences depend maker (household) has a utility function on something that is determined in part by ju05_Article 2 6/10/05 1:40 PM Page 402

402 Journal of Economic Literature, Vol. XLIII (June 2005)

choices and in part by other factors that are require that the outcome is a vector of left unconstrained in the basic formulation.10 observable private allocations. Models with Everything can be written generalized consumption goods permit some of the goods to be standard private goods, (4) u (O(s);(s;)) i but are flexible enough to include public = where s (s1,…,sI) is strategy profile (indi- goods. The arguments could also include vidual i chooses si), describes personal quantities that cannot be measured directly, characteristics, and is a parameter.11 In like a warm glow from giving. The second

making the transformation, si represents the difference is that in models of generalized choice variables of the household (quantities consumption goods preferences over out- of market goods and labor); denotes the comes may vary with parameters. In terms of additional parameters (human capital and the representation in (4), an individual’s “other inputs”); O(s) denotes those general- preferences over outcomes O can depend ized consumption goods whose production on the parameter , which is difficult to does not depend on and (s;) all other observe, let alone control. generalized consumption goods. The reduced-form description of the The description of the utility function in Stigler–Becker model also connects it to an (2) appears to rule out externalities, since an approach that, at least on the surface, agent’s utility is a function only of his own appears quite different. generalized consumption goods. The George A. Akerlof and Rachel E. Kranton reduced-form (4) allows utility to depend on (2000)’s formulation of identity posits that the entire outcome, so it permits interde- the decisionmaker has a utility function pendent utilities. Characteristics of other ε individuals in the economy enter through (5) Ui(ai,ai,Ii(ai,ai;ci, i,P)), the production function: it includes as arguments the human capital of all agents. where ai is the action of individual i, Ii is Further, since the general model makes no the individual’s identity, ci is the individual’s ε restrictions on the relationship between assigned social categories, i is individual i’s choices of other agents and their level of characteristics, and P are prescriptions that human capital on one hand and the defini- “indicate the behavior appropriate for peo- tion of generalized consumption goods on ple in different social categories in different the other, nothing prevents agent i’s opti- situations” (Akerlof and Kranton 2000, page ε mization problem from having an arbitrary 719). Given ci, i, P, and ai, individual k dependence on agent k’s market decisions. selects ai to maximize Ui. In this theory, What distinguishes this formulation from action choices are observable. the models of income maximization and Akerlof and Kranton’s model also interdependent preferences is that optimiz- reduces to (4) by denoting the action vari- ing decisions can be based on more than the able a by s, letting contain the variables distribution of material goods. There are two associated with social and individual char- aspects to this difference. First, the model of acteristics and prescriptions, and defining α = ε generalized consumption goods does not (s; ) Ii(ai,ai;ci, i,P). The models of Akerlof–Kranton and Stigler–Becker are thus mathematically identical. It is curious 10 In the Stigler–Becker formulation, the household utility function does not depend on the household, but that these formally equivalent approaches heterogeneity may enter through the human capital are associated with schools of thought that variables. 11 often are viewed as opposites. I use this formulation rather than simply Ui(s, ) in order to connect this model to explicitly strategic models The theories are identical because they are that I introduce in section 3.4. consistent with precisely the same set of ju05_Article 2 6/10/05 1:40 PM Page 403

Sobel: Interdependent Preferences and Reciprocity 403

observations. Any observation consistent interest in the future. The decision to sacri- with the first must be consistent with the sec- fice nonconspicuous consumption for ond and conversely. The theories have differ- increased status is a standard economic ent social scientific implications because they tradeoff. Proposers will not make low offers lead one to look for different ways to describe in the ultimatum game for the same reason observations. For example, the Chicago and also to avoid challenging the second school may assume that generous behavior is player’s status.13 Responders reject low a consumption good that directly enters the offers in the ultimatum game in order to sig- utility function (possibly because the appear- nal that small amounts of money are not ance of kindness that will be useful in the important to them. Voluntary contributions future). Akerlof and Kranton might posit that to public goods arise if status is enhanced by an individual’s identity required behaving contributing to charitable projects. according to accepted norms of fairness (and Fremling and Posner presumably wish to therefore the proposer loses utility if he maintain the Chicago tradition and base offers unequal divisions or the responder their explanations of differences in given loses utility if she accepts unequal divisions). status rather than preferences. They argue Gertrud M. Fremling and Richard A. that heterogeneous behavior arises because Posner (1999) provide a more elaborate different agents have different endowments example of this approach. They sketch a of status, but their formulation provides no model in which an individual’s utility way to measure endowments of status. depends on status and nonconspicuous con- Further, this position is a bit strained sumption. In this formulation, nonconspicu- response to heterogeneity of laboratory ous spending is the category that summarizes behavior, since experimenters make great expenditures on standard consumption efforts to suppress information about given items. Inserting a status argument in the util- status. ity function provides a reduced-form meant Fremling and Posner’s model is not fully to capture the instrumental value of increas- specified. They do not provide a complete, ing status. Agents have the same underlying operational definition of status. The status preferences, but differ in their given endow- argument that enters their utility function is ment of status.12 Individuals allocate their not observable. Hence there is no way in income over nonconspicuous consumption which it can be controlled in the laborato- items and expenditures that influence their ry.14 They do not provide an explanation of variable component of status. Because why status should enter the utility function, agents will forgo consumption to increase nor do they place substantial restrictions on their status, this formulation is sufficient to the form that it enters. Yet their motivation be consistent with many apparent departures from self interest. Fremling and Posner suggest that dictators will not take the entire 13 This effect should arise in the dictator game as well, surplus in order to signal that they are altru- so the difference in proposer behavior in the two games istic. Being known as a generous person must depend on some expectation that low offers will be rejected in the ultimatum game. enhances your status, which will put you in a 14 Alexander’s (1987) discussion of the evolution of better position to advance your material self morality contains discussions consistent with the view of Fremling and Posner. In his discussion of (apparently) altruistic giving on pages 159 and 160, Alexander raises selfish motivations for generous behavior. He is aware of 12 The fixed component of status derives from genetic the broadness of the theory and writes (page 160): “If endowment—inherited wealth, titles, race, and gender— such conditions seem to render the propositions virtual- in addition to other attributes obtained in the past. ly untestable, that is simply a problem that we must solve Decisions made in the current period cannot influence the if we are to deal in a better way with the unparalleled fixed component of status. difficulty of understanding ourselves.” ju05_Article 2 6/10/05 1:40 PM Page 404

404 Journal of Economic Literature, Vol. XLIII (June 2005)

is powerful and consistent with casual intu- identity and generalized commodities per- ition. There have been successful attempts mit preferences over outcomes to depend on to incorporate status concerns in prefer- the context in which the outcome was ences, which means that one can derive reached. Formally, this means only that preferences like the ones proposed by appears in the utility function. This subsec- Fremling and Posner from a more detailed tion describes a framework in which the model.15 Still, the lack of guidance about how preferences that players optimize in a game status influences preferences is disconcert- depend on the game itself. ing. Fremling and Posner describe situations Context matters in a strategic situation if in which conspicuous spending increases preferences over outcomes depend on the status and others in which conspicuous thrift game being played. The idea is best intro- (for example, buying a modest car or wear- duced by an example. Consider two versions ing unpretentious clothes) enhances status, of the ultimatum game. In the first version, which makes me suspect that ex post adjust- the proposer can make only two offers: an ments in their status variable will make their 80–20 split and a 20–80 split. In the second model consistent with any observation. version, the proposer can make three offers: While Fremling and Posner describe 80–20, 50–50, and 20–80. In both of these their model as a signaling model, it is not games, there exists an opportunity for the completely clear what is being signaled. responder to choose between 80–20 division Implicitly, status is important because it is and a 0–0 division. If the responder’s prefer- an observable way for third parties to learn ences depend only on the distribution of something important about an individual. In material payoffs available when she makes the formal model, the only characteristic her decision, then her decision after she has that may be hidden is an individual’s been offered 20 cannot depend on whether income. If status does signal income (as it player one could have proposed an equal does in many of the informal stories), then split. Intuition suggests that the additional presumably it is not status that enters the strategy might matter, with the responder utility function, but income (as perceived by more likely to reject the “unfair” 80–20 split third parties).16 The function that trans- when player one could have offered an forms investment in status into perceived equal division than when only unequal splits income would be determined as an equilib- are available. The experimental results of rium of a signaling game and need not have Armin Falk, Fehr, and Urs Fischbacher even the weak properties that Fremling and (2003) confirm this intuition.17 Posner posit. Assuming equilibrium behavior, there are two reasons why adding strategies might 3.4 Intrinsic Reciprocity influence the responder’s choice in the ulti- Game theory assumes that, in strategic sit- matum game. The first possibility is that uations, players act to maximize a preference adding strategies changes the equilibrium relation over outcomes. As a result the selection. This could only happen if the sub- process by which the outcome is reached games in which responder decides whether does not matter. I argued that models of to accept or reject a proposal have multiple equilibria. For appropriately defined prefer- 15 Andrew Postlewaite (1998) provides an overview of ences over outcomes, the responder may be one approach. This article also argues that there are advantages for explicitly incorporating the reasons why status (or indifferent between accepting or rejecting other intangible arguments) should matter into formal twenty units out of one hundred, but it is an models rather than relying on reduced-forms. 16 This interpretation suggests that differences in income across experimental subjects would be both rele- 17 Some of the experiments reported in Charness and vant for experimental results and difficult to control. Rabin (2002) have a similar flavor. ju05_Article 2 6/10/05 1:40 PM Page 405

Sobel: Interdependent Preferences and Reciprocity 405

implausible way to explain experimental In this expression, G is the game, 18 ∗ = ∗ ∗ = ∗ behavior. s (si,sj), s (si,sj), vi denotes player i’s The second possibility is that preferences utility function over outcomes, and O(s) is depend on more than final payoffs. In the the outcome obtained if the players play s. 19

ultimatum game, if player 2’s preferences Rabin interprets si as the strategy choice of ∗ depend only on final outcomes and she has a player i, sj as player i’s beliefs about player j’s ∗ strict preference between accepting the strategy choice, and si as what player i 80–20 split and turning it down, then she believes that player j believes about player must make the same choice on the equilibri- i’s strategy choice. In equilibrium, beliefs ∗ um path whatever the other strategies may are accurate, so that j actually plays sj and = ∗ G be available to player 1. Hence I concentrate si si. i is a function of the beliefs of on the possibility that preferences over out- player i. It is this feature that makes the comes at a decision point depend on more game a psychological game.

than just final payoffs. Player 2’s rejection of The utility function ui expresses i’s prefer- the 80–20 offer when the equal split is ences over his own strategies (si) conditioned unavailable is an example of destructive rec- on expected behavior (s∗). A player seeks to iprocity. In situations where beliefs about maximize a weighted average of his material the actions of others are irrelevant (for utility with that of his opponent. The weight example, when the second player decides G(s∗) depends on the game being played in whether to accept or reject an offer in the addition to the strategy profile. The natural ultimatum game), any descriptive model interpretation of G(s∗) is as a measure of the must permit allow preferences to depend on extent to which player i cares about player j’s more than the distribution of income. material welfare. The conventional formula- G ≡ Rabin (1993) was the first to propose a tion ( i 0) is a special case of this repre- G specific model of equilibrium behavior in sentation. When i is not constant, as is games where players take into account con- typically the case in Rabin’s model, player i’s text to determine their behavior. His formu- preferences over strategies depend on more lation uses the theory of psychological games than his preferences over outcomes: the introduced by John Geanakoplos, David strategic context matters. Rabin presents a Pearce, and Ennio Stacchetti (1989). specific functional form for the coefficient G Psychological games permit players’ beliefs i in (6). The form of the coefficient is G to enter into their preferences. In Rabin’s less important than its content. i is posi- model, the weight placed on an opponent’s tive if i thinks that j’s behavior is nice and material payoffs depends on the interpreta- negative if he thinks that the behavior is tion of that player’s intentions. He evaluated nasty. In this way, (6) provides a model of intentions by using beliefs (and beliefs about intrinsic reciprocity. Kind (unkind) treat- beliefs) over strategy choices. Rabin pro- ment raises (lowers) the weight placed on

posed that agent i pick his strategy si in a opponent’s material payoff is preferences, game to maximize a function of the form: making an agent more willing to sacrifice his own material payoff to increase (decrease) ∗ = + G ∗ (6) ui(si;s ) vi(O(s)) i (s )vj(O(s)). that of his opponent. The optimization problem (4) faced by 18 If the responder’s preferences over outcomes left her individual agents in the generalized con- indifferent between 0–0 and 80–20, then for all popular parametrizations of (1), she would have a strict preference sumption model appears to include the between 0–0 and 79–21. Consequently a model that explained the experimental results on the basis of equilibrium selection predicts that small perturbations in the set 19 Rabin (1993) assumes that the outcomes are distri-

of feasible offers would dramatically change experimental butions of money x and that vi(·) depends only on xi. outcomes. Charness and Rabin (2002) relax the second assumption. ju05_Article 2 6/10/05 1:40 PM Page 406

406 Journal of Economic Literature, Vol. XLIII (June 2005)

problem (6) as a special case. There is a sub- equilibrium outcome. When (Fight,Opera) tle difference in the way that one closes the is the expected outcome, the man believes models to compute equilibria, however. In that the women is planning to go to the the strategic models, agent i takes context as opera even though she believes he is going

given and chooses si to maximize (6) holding to the fight. In this situation, he thinks that s∗ fixed. In equilibrium, one must have s = s∗. she is being nasty and is willing to give up Models based on the generalized consump- the material value of coordination in order tion good idea often require a consistency to lower her payoff. condition (the condition may describe the Alternatively, one can view the payoffs evolution of human capital or the magnitude as the result of Akerlof and Kranton’s of a status argument), but in general the identity theory. Assume that preferences

decisionmaker controls si as it enters both have the same representation as before. directly and indirectly in (4). There is a tech- That is, they can be written in the form = + G ≠ G nical implication of the difference. Assuming ui(s) vi(s) i (s)vj(s), where i j, i is the that preferences over outcomes are linear in same as before, and the material payoffs

probabilities, the standard assumption, exis- (v1(s), v2(s)) are given in the payoff matrix. tence of equilibrium follows from standard Coordination permits the man to feel in arguments in the game-theoretic models. charge (if the outcome is (Fight,Fight)) or Existence would not be guaranteed in the thoughtful (if the outcome is (Opera,Opera)); Akerlof–Kranton or Stigler–Becker settings failure to coordinate leads to negative without assumptions that lead to quasicon- because the man does not want to be viewed cavity of the reduced-form preferences in as selfish (if the outcome is (Fight,Opera)) or (4). The following example illustrates part of confused (if the outcome is (Opera,Fight)). the problem. Now, the outcome (Fight,Opera) is not an Example 1. Consider the game: equilibrium because given that the column Fight Opera player is going to the opera, the man’s best response is to go there as well. Doing so rais- Fight 2,1 0,0 es his material payoff and also changes his Opera 0,0 1,2 identity ( increases from –.8 to .8). The difference between the two formula- The matrix describes the material payoffs tions is that, when considering a deviation, a for a standard battle-of-the-sexes game. player can change his “identity” but cannot Assume that payoffs can be written in the change his view of his opponent’s intentions. form (6) with vi as given in the payoff matrix In the example, Rabin’s model had a larger G = G = and i (Fight,Fight) i (Opera,Opera) .8 equilibrium set than Akerlof and Kranton’s, G = G =− and i (Flight,Opera) i (Opera,Fight) .8 but in general there is no relationship for i = Row and Column. These weights have between the two sets. a natural interpretation: a player who antici- The models of context-dependent prefer- pates coordination places positive weight ences include the interdependent prefer- on his or her opponent’s material payoff, ence approach. The underlying preferences

while one who anticipates a failure to coor- vi(·) in (6) are defined over outcomes. If an dinate blames the other player and places outcome specifies a material payoff to both

negative weight on that person’s payoffs. players, it is permissible for vi to depend on Assume that the players expect the outcome player j’s material payoff. Charness and s∗ = (Fight,Opera). Then each player will Rabin (2002) propose a functional form that

place a negative weight on the other player’s assumes that the vi are interdependent pref- payoff and will prefer playing as expected erences that place positive weight on j’s than deviating: That is, (Fight,Opera) is an monetary payoff (with the weight changing ju05_Article 2 6/10/05 1:40 PM Page 407

Sobel: Interdependent Preferences and Reciprocity 407

to reflect concern for the player with results is a tribute to their flexibility rather the lowest monetary payoff). Falk and than actual support for the formulation. 20 Fischbacher (2005) also attempt to separate Miguel Costa-Gomes and Klaus G. concerns for equity and concerns for inten- Zauner (2001) analyze data from ultimatum tions. Their model contains a “pure out- game experiments and estimate utility func- + λ come concern parameter.” This number tions of the form vi vj under the assump- measures the degree to which a player’s tion that agents play an equilibrium to a preferences in the game depend only on the game with perturbed preferences. Their outcome and not on the context in which it method contains an indirect test of the pure was obtained. At one extreme, a player interdependent preference approach. In all cares only about the outcome. In this case, terminal nodes in which the second player Falk and Fischbacher’s (2005) model has rejects the proposal, monetary payoffs are the flavor of the models of Bolton and the same (zero for each player). If the coef- Ockenfels (2000) and Fehr and Schmidt ficient λ does not depend on actions, then (1999). the estimated variance in the error term Since the approach outlined in this sub- should be the same after all rejections. This section generalizes the interdependent pref- is not the case, providing some evidence that erence approach, without restrictions, it has preferences depend not just on the outcome, broader descriptive powers. To the extent but on how the outcome was reached. that preferences over outcomes depend on There are other ways to demonstrate that the game, reciprocity models provide insight preferences depend on more than the final into the observations. distribution of wealth. Yoella Bereby-Meyer There is substantial experimental evi- and Muriel Niederle (2005) compare out- dence that the distributional approach is not comes of three-player games. The first play- sufficient to explain and organize experi- er proposes a division of a fixed quantity mental findings, which suggests that it would between himself and the second player. The be worthwhile investigating models of intrin- second player either accepts or rejects this sic reciprocity. proposal. The third player is nonstrategic. If Ken Binmore, John McCarthy, Giovanni the second player accepts, then the first and Ponti, Larry Samuelson, and Avner Shaked second players get paid according to the pro- (2002) compare one- and two-stage alterna- posal (and the third player receives nothing). tive-offer bargaining games in an effort to If the second player rejects, then (depending test whether experimental subjects obey on the treatment) either the first or the third backward induction for these games. The player receives a payment, while the other paper observes that subgame-perfect equi- two players receive nothing. Bereby-Meyer librium strategies played by agents with and Niederle find that the second player’s interdependent preferences satisfy back- behavior depends both on the quantity of ward induction. They discover systematic the payment given following a rejection and departures from backward induction, which who receives this payment. The second play- is evidence against the hypothesis of equilib- er is more willing to reject a small offer when rium behavior with interdependent prefer- she knows that doing so will not lead to a ences. Context-dependent models permit high payoff for the proposer (either because preferences over outcomes to depend upon the rejection payment is low or because the how one reached the outcome. Hence the third player receives the payment). To experimental results of Binmore, McCarthy, 20 Ponti, Samuelson, and Shaked do not con- Uzi Segal and Joel Sobel (2004) show that dominance arguments have power to rule out some predictions in tradict these models. The fact that these games governed by preferences represented by (6), but models are consistent with experimental these restrictions are weak. ju05_Article 2 6/10/05 1:40 PM Page 408

408 Journal of Economic Literature, Vol. XLIII (June 2005)

explain this behavior with some form of mover cares only about the distribution of interdependent preferences, the second payoffs, then contributions should be the player would not only need to have prefer- same in these two treatments. They are not. ences that depended non trivially on the The second mover tended to return more monetary payoffs of others, but she would money in Treatment A. The difference need to weigh an opponent’s monetary pay- between the amount returned is largest after off differently depending on his role in the large transfers, suggesting that in part the game. Models that permit preferences over second mover acted to reward intentionally strategies to depend on strategy choices pro- generous behavior. vide a convenient way to capture the intu- In contrast to the models that focus on ition that the second player might be willing context, adding strategies to a game does not to sacrifice her material payoff in order to influence a player’s preferences over the out- punish player one when the first player comes in the original game in Levine’s (1998) makes a small offer. Naturally, the ability to set up. When players are uncertain about punish (and therefore the second player’s their opponents’ preferences, however, play- action) depends on whether the first or third ers may try to use their actions to signal their player receives a payment after a rejection. preferences. The inferences one player James C. Cox (2004) compares the out- draws about his opponent’s payoffs depend comes of three related games designed to on the strategies available. Consequently, separate predictions from different models observed preferences over outcomes depend designed to identify the source of nonselfish on the strategic context in Levine’s model behavior in experimental trust games. In the (because changing the set of available strate- basic experiment (Treatment A), subjects are gies can change the signaling content of divided into two groups. Those in the first strategy choice). group receive $10 and decide how much to Expanding the set of arguments in the contribute to a member of the second group. utility function, as in the models of sections The member of the second group receives 3.2 and 3.3, is consistent with standard deci- three times the contribution. Finally, mem- sion theoretic methods. The new arguments bers of the second group can return any part in utility functions are externalities and fully of the transfer he or she received. (All trans- captured by expanding the definition of fers are anonymous.) In Treatment B, mem- commodity. The models in this subsection bers of the first group decide on transfers as are not traditional. One attempt to return in the Treatment A while members of the the models to the standard framework is to second group do not have a move. In the redefine the notion of an outcome. If the Treatment C, members of the first group do way in which one arrives at an outcome not move. Instead, experimenters make the influences preferences, then the outcome same contributions that were made in the should include that information. Formally, Treatment A (and subtract the appropriate an outcome would need to include a descrip- amounts from members of the first group). tion of the entire game and an anticipated Members of the second group receive trans- strategy profile. Specifically, the 80–20 offer fers as in the Treatment A, are told how the in the ultimatum game leads to an outcome transfers are generated (and how they influ- that not only specifies that player one ence the endowment of first movers), and receives 80 and player two receives 20, but then decide how much to return. The con- also the existence of other possibilities avail- trast between the results of Treatments A able to the first player. This transformation is and C present the evidence most relevant to logically possible, but hardly useful. Note the power of the interdependent prefer- further that the representation (6) specifies ences to describe outcomes. If the second preferences over player i’s own strategies ju05_Article 2 6/10/05 1:40 PM Page 409

Sobel: Interdependent Preferences and Reciprocity 409

conditional on a strategy profile s; it is not a result of incomplete information regarding utility function defined over outcomes. After the opponent’s characteristics rather than redefining the outcomes, one would need to due to conscious randomization on the part extend the preferences to this new space. of the opponent. In standard game theory, There are several problematic aspects of the characterization of equilibrium does not models based on context-dependent prefer- depend on the interpretation of mixed ences. The most basic problem is the speci- strategies. In games where a player’s prefer- fication of . Charness and Rabin (2002), ences depend on the intentions of his oppo- Martin Dufwenberg and Georg Kirchsteiger nent, the interpretation matters. A simple (2004), Falk and Fishbacher (2005), and example, taken from Segal and Sobel (2004), Rabin (1993) present explicit functional makes the point clearly. forms for , all motivated by plausible intu- Example 2. Consider the game: itive arguments and appeals to selected experimental evidence, but no one has AM PM described observable behavioral assump- AM 10,10 0,0 tions on that best describe behavior. Without a model of this parameter, Segal PM 0,0 10,10 and Sobel (2004) demonstrate that the theo- ALL 7,10 7,10 ry that makes few definitive predictions about behavior. Column is a plumber and Row is a home- Another problem is identification. It should owner with a leaky faucet. The plumber can be possible to separate preferences repre- come in the AM or in the PM while the

sented by vi and ui in (6) through revealed homeowner can arrange to be at home in the preference analysis. Observation of decisions AM, in the PM, or ALL day. The plumber

made in a nonstrategic setting determine vi. earns 10 if she coordinates with the home- Observations of decisions made in a strategic owner, but nothing otherwise. The home-

setting determine ui. Adding incomplete owner receives a payoff of 10 if he can meet information about preferences complicates the plumber and only cancel half of his this exercise, however. On the other hand, appointments; he receives 7 if he stays home many different combinations of material pay- all day; he receives 0 is he fails to coordinate offs vi and weight will lead to the same with the plumber. If players’ preferences behavior in strategic settings. Looking only at over strategies agreed with their preferences choice behavior in games it will not be possi- over outcomes, then the game has three ble to separate preferences for reciprocity equilibrium outcomes: (AM,AM), (PM,PM), from preferences over outcomes. and a continuum of equilibria in which the The third problem deals with the inter- homeowner stays home all day and the pretation of mixed strategies. The behav- plumber places probability of at least .3 on ioral interpretation of deliberate each pure strategy. randomization is somewhat strained because Now assume that the homeowner has it requires a player to select a precise weight preferences over strategies that lead him to on each pure strategy in spite of being indif- put a positive weight on the plumber’s payoff ferent over at least two pure strategies. in response to nice behavior (apparent coor- Hence it is often attractive to interpret dination) and a negative weight in response mixed-strategy equilibria as equilibria in to nasty behavior. (Assume that the plumber beliefs. The important idea is that in some cares only about her own payoffs.) Clearly, situations an agent must be uncertain about the two pure-strategy equilibria from the the pure-strategy choice of his opponent in game with standard preferences will contin- equilibrium, but that uncertainty may be the ue to be equilibria. But if randomization by ju05_Article 2 6/10/05 1:40 PM Page 410

410 Journal of Economic Literature, Vol. XLIII (June 2005)

the plumber is purposeful, then the home- of equilibria. Multiple equilibrium problems owner may well think that a plumber who are possible, but less pervasive for the para- randomizes equally between AM and PM is metric models of extended preferences used nasty because this behavior minimizes the in the literature. Coordination problems arise probability of coordination. With a suffi- when preferences are interdependent. Players ciently negative weight on the plumber pay- can get stuck in nasty equilibria in which they offs, the homeowner may prefer to play expect nasty behavior from their opponents either AM or PM rather than to stay at home and get it or in nice equilibria in which posi- all day. Consequently, there may be an equi- tive expectations are fulfilled.21 While the librium in which both the homeowner and existence of multiple equilibria may be useful the plumber randomize equally between in applications, the possibility broadens the set 1 + 1 AM and PM, while (ALL, 2 AM2 PM ) is of possible predictions from the theory so that no longer an equilibrium. equilibrium analysis places few restrictions on The above analysis depends on the inter- observable behavior. pretation of mixed strategies. In many appli- 3.5 Commitment cations, it is appropriate to treat the homeowner as if he is matched against a Allowing commitment changes the equi- population of plumbers, some with a ten- librium concept rather than the specification dency to come in the morning, others with a of preferences. A player with commitment tendency to come in the afternoon. If the power selects his strategy to maximize a homeowner does not attribute his uncertain- function of the form: ty to a deliberate strategy of the plumber, (7) u(s,BR (s)), then it is reasonable to assume that he does i i i i

not place a negative weight on the plumber’s where BRi(si) is the strategy profile of best payoff. In this case, however, there will be an responses for all other agents, taking i’s strat- equilibrium in beliefs in which the home- egy as given.22 Commitment models are con- owner always stays at home (and he believes sistent with more general specifications of with probability greater than .3 that the preferences. In the ultimatum game, if a self- plumber will come at any time). ish responder had full commitment power, Many of the motivations for the impor- then she would announce that she would tance of context and intentions are intrinsi- only accept s = 1 and the proposer could do cally dynamic, but dynamic considerations no better than to offer her the entire surplus. do not play a role in the basic formulation. Experimental results in the ultimatum game Dufwenberg and Kirchsteiger (2004) and are broadly consistent with the hypothesis Falk and Fischbacher (2005) study exten- that the responder has partial commitment sive-form games. They argue that the strate- power, but merely assuming commitment gic-form approach of Rabin does not lead to ability is not sufficient to handle a broader good predictions in extensive-form games range of empirical regularities. ∗ like sequential prisoner’s dilemma games. The solution si of problem (7) typically will Since it is possible to represent any exten- not be a best response to the strategy choices of sive-form game in strategic form, the model the other players. That is, commitment behav- applies to extensive-form games for standard ior is not consistent with Nash Equilibrium. I reasons. Even for standard game theory, there is dispute about whether this transfor- 21 Segal and Sobel (2004b) formalize the intuition that mation loses information. This issue is more allowing players’ preferences to depend on context generally complicated when intentions matter. enlarges the set of equilibria. 22 This formulation leaves open how the other I − 1 In standard examples, the notion of recipro- players arrive at best responses and what to do when the cal preferences tends to increase the number best response correspondence is not single valued. ju05_Article 2 6/10/05 1:40 PM Page 411

Sobel: Interdependent Preferences and Reciprocity 411

find it more useful to assume that all feasible In a repeated interaction, an agent exclu- commitment abilities are described in the sively interested in his material consumption specification of the strategic environment. could rationally forgo short-term utility in order to obtain future benefits. If an appar- 3.6 Repeated Games ently unselfish action is part of a repeated The theory of repeated games provides a interaction between patient agents, then the conventional framework in which to explain folk theorem of repeated games can explain apparently unselfish behavior. In an infinite- the observation as a part of equilibrium ly repeated game, agents play a given static behavior between selfish agents. The logic of game, observe its outcome, and then play the folk theorem is the logic of instrumental the game again and again, without end. reciprocity.24 Individuals forgo their short- Payoffs for the repeated game are discount- term selfish gains because being nice (or, ed sums of the individual static-game pay- more precisely and more generally, playing offs. Strategies are rules that specify how to their equilibrium strategy) will lead to nice play in each repetition as a function of past treatment in the future. Punishing nasty behavior. Repeated games typically have behavior serves to discourage nasty behavior, large sets of equilibria. The folk theorem of but punishment only occurs because players repeated games states, roughly, that any fea- fear that a failure to punish will lower their sible, individually rational payoff for the future payoffs. This argument requires that stage game can be obtained as a subgame- the actions an agent takes today influences perfect equilibrium payoff for the associated his future payoffs and that the influence is infinitely repeated game. sufficiently great to counter short-term To compare repeated games to the earlier incentives. approaches, consider a reduced-form descrip- Embedding an interaction into a repeat- tion in which player i selects a stage-game ed-game setting is a powerful and accepted

action si to maximize way to describe behavior that, when viewed with a static perspective, appears to be (8) (1 − )u(s,s ) + V((s)), i i i i inconsistent with selfish behavior. Like the

where ui(s) is player i’s stage-game payoff approach of generalizing consumption, function; ∈ (0,1) is a discount factor; repeated game theory forces the observer to describes the history of play; and Vi( ) gives ask: “What does that player stand to gain?” in the continuation (average) value given histo- a way that often leads to useful insights. ry . In this formulation, the continuation The approach has limits. I discuss some of value function is endogenous. Thanks to the them now. folk theorem, however, there are few restric- Because laboratory experiments carefully tions on Vi . So although it is not possible to control for repeated-game effects, these state that models of repeated interaction are results need a different explanation. formally identical to those based on general In order for conventional repeated-game consumption goods, it is clear that if Vi can arguments to apply, the future must be take an arbitrary individually rational and important. Agents must be patient and feasible payoffs and is close to one, there is there must be opportunities to reward and no reason that an equilibrium action choice punish today’s behavior. When these condi- for player i will be a myopic best response to tions fail, theory predicts a return to myopic the actions of his opponents.23 selfish behavior. Relationships do end, and

23 That is, if equilibrium strategies specify the actions s∗ 24 The anthropologists (for example Sahlins 1968 and ∗ in a given period, then there is no reason to expect that si Service 1966) studying exchange clearly recognize this ∗ solve: maxsiui(si,si). motivation. ju05_Article 2 6/10/05 1:40 PM Page 412

412 Journal of Economic Literature, Vol. XLIII (June 2005)

while it is easy to find evidence of myopic for all players. For example, the “grim trig- self interest at the end of relationships, it is ger strategy” specifies that players never also easy to cite examples of employees who again cooperate following a noncooperative do not shirk as retirement approaches and action in the prisoner’s dilemma. The princi- families that stand by dying relatives. pal of renegotiation proofness identifies a set Fitting this behavior into an individualistic of possible (renegotiation-proof) equilibri- repeated-game framework is possible, but um payoffs with the property that no payoffs awkward. in the set are Pareto-dominated by other Repetition increases the range of equilibri- payoffs in the set. The logic behind this def- um behavior because it creates the possibili- inition is that if players are able to “renegoti- ty of punishment. Punishment may be costly ate” at the beginning of each period on the for the punisher as well as the punished. If equilibrium that they will play, then they will so, the question arises: Why should anyone never agree to play an equilibrium that is punish? The theoretical answer is: People Pareto inferior to another equilibrium. punish because otherwise they will be pun- Renegotiation proofness says, in effect, that ished themselves. Of course, the argument history can influence future play, but no his- must be repeated to ensure that people are tory can induce players to select inefficient willing to punish the punishers of the pun- continuation. The idea is controversial (the ishers and so on. Theory provides elegant jus- weaker position that players will not play an tification for this answer.25 The theory is less equilibrium whose payoffs are Pareto domi- direct and, perhaps, less convincing than the nated by another equilibrium’s payoffs is answer supplied by models of intrinsic reci- already controversial for one-shot games). It procity: Punishment arises because people is not straightforward to define renegotiation get utility directly from lowering the welfare proofness in infinitely repeated games and of people who have hurt them. the restrictions imposed by renegotiation Repeated-game theory incorporates proofness do not lead to more descriptive strategic context, not by changing prefer- predictions.26 ences but by changing the way people play. Finally, repeated game theory provides In order to obtain equilibria distinct from predictions consistent with many observa- repetitions of equilibria of the underlying tions and also gives strong intuitions about static game, the history of play must influ- qualitative features that increase the possibil- ence future play. History does not influence ity of cooperation. But repeated games have preferences, but it does influence expecta- too many equilibria and the selection process tions about behavior. The principle of sub- is often tailored to particular examples. The game consistency would require play in a theory as it is typically used does not produce subgame to be independent of where the interesting refutable hypotheses. subgame arises in a larger game. One must abandon subgame consistency to in order to predict repeated-game behavior distinct 4. Using the Models from repetitions of static equilibria. This section describes several economic Conditioning on history is so descriptive settings in which narrow notions of self- that there is little resistance to abandoning interested behavior provide limited insight. subgame consistency. There is some support These examples provide further evidence in for adopting a weaker principal. Some histo- support of developing models that assume ries may trigger punishments that are bad extended preferences and illustrates the way

25 Classic treatments are Dilip Abreu (1988) and Drew 26 Abreu, Pearce, and Stachetti (1993) and Joseph Fudenberg and Eric Maskin (1986). Farrell and Maskin (1989) are two approaches. ju05_Article 2 6/10/05 1:40 PM Page 413

Sobel: Interdependent Preferences and Reciprocity 413

in which the different approaches of section income, but they do not care about the 3 provide alternative predictions. nature of transfers that determines income distribution.28 4.1 Charity James Andreoni’s (1990) warm-glow theo- It is difficult to rationalize charitable ry of charitable giving fits within the tradi- contributions as optimizing behavior from tion of preferences over general an individual who cares only about material consumption goods. Andreoni assumes that wealth. Nevertheless, people do make agents obtain utility by contributing to oth- charitable contributions. ers. Agent i’s utility depends on agent i’s Contributions appear to be sensitive to material consumption and how much he the economic environments in ways that are contributes to other agents. Preferences consistent with conventional theory.27 For depend on not just the distribution of mate- example, changes in tax laws that reduce the rial goods, but on how one arrives at the dis- marginal cost of giving increase the amount tribution of material goods. Andreoni’s of giving. Social psychologists have discov- model is qualitatively different from model- ered factors that influence contributions ing altruism by simply assuming prefer- that are less obvious consequences of stan- ences place positive weight on other dard economic assumptions. For example, people’s consumption as in the models of Robert C. Cialdini and Melanie R. Trost’s interdependent preferences. In distribu- (1998) review essay suggests that contribu- tional models, i cares about i’s income, but tions increase in response to small gifts from not the source of the income. If j’s income the charity. There is also evidence that con- increases, then (under the standard assump- tributions are an increasing function of the tion of diminishing marginal utility) i would contributions of others. These predictions be less willing to give money to j.In are less obvious consequences of standard Andreoni’s model, i derives utility from con- economic assumptions. tributing. In the purest formulation, i’s con- The models of section 3 suggest different tribution would be independent of i’s reasons why people give to charity. These income, in contrast to the prediction of models make different predictions about models based on altruism. what influences charitable giving and in prin- Robert Sugden (1984) provides a model of ciple can be distinguished. This section dis- charitable giving based on intrinsic construc- cusses some theoretical models of charitable tive reciprocity. Sugden assumes that agents giving and relevant experimental evidence. feel an obligation to contribute. With this Altruism, modeled as one agent placing (nonstandard) assumption, Sugden is able to value on the material welfare of others, pre- analyze his model using conventional meth- dicts positive contributions that decline with ods. The most interesting result is that, the contributions of third parties. This is the because obligations are assumed to be crowding-out effect. The crowding-out increasing functions of the contributions of effect should be strengthened if agents have others, there will be a positive relationship distributional preferences: If agents are between one’s contribution and the (expected) motivated to give because they wish to raise contribution of others. the income of the poor, then the more that Hence one obtains a different predic- the poor receive from others, the lower the tion about the relationship between the value of direct transfers. The distinguishing feature of this kind of explanation is that people care about the distribution of 28 If j gains income from third parties, the marginal value of a contribution by i to j should not increase. The literature assumes that contributions will actually decrease; 27 Andreoni (2001) is a survey. this prediction would not hold for linear preferences. ju05_Article 2 6/10/05 1:40 PM Page 414

414 Journal of Economic Literature, Vol. XLIII (June 2005)

contributions of others and one’s own con- the first agent). The second agent makes a tributions depending upon the underlying take-it-or-leave it offer to the first agent. theory. Altruism and more general distribu- Assuming that agents maximize their mone- tional approaches predict that an individual tary payoff, the subgame-perfect equilibri- contributes less when others contribute um predicts that there will be no investment. more. Warm-glow models predict that an The second agent will purchase the item at individual’s contribution does not depend the first agent’s reservation price. Typically, on the contribution of others. Sugden’s positive investment is efficient.31 The litera- model of giving based on reciprocity pre- ture focuses on institutional arrangements dicts that an individual’s contribution that reduce or eliminate the inefficiency. increases with the contributions of others. The standard analysis points out that if Empirical evidence from Cialdini and Trost agents can commit to a complete contract, (1998) and Rachel T. A. Croson’s (1999) the first agent can be given proper incen- experiments therefore provide evidence tives to invest. The literature on incomplete against the simplest models of altruism or contracts argues that complete contracts are the warm-glow hypothesis.29 infeasible if agents have asymmetric information or unenforceable if agents are free to 4.2 Incentives and Effort renegotiate and third parties lack informa- The hypotheses of greed and equilibrium tion needed to enforce the contracts. The lit- form the basis of powerful theories of con- erature then argues that ownership patterns tracting, which in turn make predictions and internal organization of firms arise to about how firms should design internal com- solve or lessen hold-up problems. Assuming pensation schemes and trade with suppliers. that the predictions of the hold-up model These predictions do not hold in simple with selfish agents are valid, the standard experiments designed to reflect natural set- approach provides a coherent framework in ting and do not appear to hold in many nat- which to study the nature of firm boundaries ural settings. Consequently there is a role for when complete contracting is not feasible. the kinds of models described in section 3. There have been many attempts to study properties of simple hold-up games in exper- 4.2.1 The Hold-Up Problem imental environments. The game underlying The hold-up problem has become an the hold-up problem has the same structure important toy model of bargaining between as the gift-exchange game introduced in arti- workers and firms.30 It has been used as the cles by Fehr, Kirchsteiger, and Arno Riedl foundation for theories of the internal organ- (1993) and Fehr, Erich Kirchler, Andreas ization of the firm. Yet there is substantial Weichbold, and Gächter (1998). In the first reason to be skeptical of the predictions of stage, the firm offers the wage w to the the model. worker. In the second stage, the worker can In the simplest hold-up model, the first either reject the offer (leading to zero mon- agent makes a costly investment. The invest- etary payoff to each player) or accept the ment increases the value of an object to the offer. If the worker accepts the offer, he second agent (but not the object’s value to must then choose a level of effort, e. There is a monetary cost associated with effort, but it 29 Andreoni (1995) and Thomas R. Palfrey and Jeffery E. Prisbrey (1997) study experiments designed to distinguish giving due to altruism, warm-glow effects, and mis- 31 These predictions are sensitive to modeling assump- takes. Both papers find evidence of warm-glow giving. tions. Tore Ellingsen and Jack Robles (2002) point out that Palfrey and Prisbrey’s design convincingly demonstrates adaptive dynamics do not avoid efficient outcomes. Faruk that apparently altruistic behavior is due to mistakes. Gül (2001) suggests that the standard results are sensitive 30 See Oliver Hart (1995) for an overview of the problem to the information structure and assumptions about the and its implication. bargaining process. ju05_Article 2 6/10/05 1:40 PM Page 415

Sobel: Interdependent Preferences and Reciprocity 415

increases the profits available to the firm. bonus. The agent receives her wage. After For example, monetary payoffs conditional observing the agent’s effort choice, the princi- on an accepted offer could be (v − w)e for pal decides whether to pay the bonus. Fehr the firm and w − c(e) for the worker, where and Schmidt assume that the principal must v is a redemption value set by the experi- pay an additional (small) cost to propose an menter and c is the worker’s cost of effort. explicit contract. Assuming that players maxi- If the worker maximizes his material payoff, mize their monetary payoffs, in nontrivial then he would choose e = 0, since the firm specifications subgame-perfect equilibrium cannot condition wages on output.32 predicts that the principal will offer an explic- Experimental subjects violate standard theo- it contract. If the principal offers an implicit ry. In the laboratory, wage offers are typical- contract, the agent will expect to receive no ly positive and effort supplied is an bonus independent of her effort choice, and increasing function of the wage. will therefore not agree to work. An explicit Charness and Ernan Haruvy (2002) contract typically can induce positive effort attempt to identify the cause of the increas- levels. Fehr and Schmidt’s principals opted for ing relationship between wages and effort. an explicit contract in fewer than 15 percent They compare the worker’s behavior when of the trials, with the frequency declining over wages are set by the firm, as in the original time. Principals made more money when they gift-exchange model, or by an external offered implicit contracts, which induced pos- process. In the treatments where wages are itive effort levels and positive bonuses. Fehr determined through an external process, and Gächter (2002) compare the performance they were drawn from a bingo cage or set by of explicit to implicit contracts in a similar the experimenter. In Charness and Haruvy’s environment, where the choice of contract is a (2002) study, effort increases with wages treatment rather than a choice variable of the when the firm sets wages, but is relatively flat principal. They find that the implicit contracts otherwise. This provides evidence that mod- perform better than explicit contracts in an els based on interdependent preferences are especially strong sense: agents cooperate less not sufficient to describe the worker–firm at a given level of compensation if they face a relationship. Preferences depend on the contract that fines them for shirking.33 process by which workers receive their The results of the contracting experiments wages—not on the wages themselves. are consistent with observational studies. These experiments suggest a reconsidera- George Homans (1953 and 1954) observed a tion of simple contracting environments. Fehr group of workers who systematically exceeded and Schmidt (2000) present an experimental minimum work standards without apparent study of a simple contracting environment. economic incentive for doing so. Truman F. The agent’s effort net of wages determines the Bewley (1999) conducted extensive surveys of principal’s monetary payoff. The agent earns managers. His informants (managers) empha- her wages net of effort costs. Fehr and size the importance of maintaining morale. Schmidt permit the principal to offer two dif- Reducing wages is costly to firms because it ferent kinds of contract. An explicit contract leads to lowered worker morale, which in turn specifies a wage, a target effort level, and a reduces workers’ productivity. Bewley argues fine. The agent receives her wage and if her that high-powered incentives (wages and effort is less than the target, then she pays the bonuses) are not an effective means of moti- fine with positive probability. An implicit con- vating workers. Bewley (1999, page 407) finds tract specifies a wage, a target effort, and a little support for the standard paradigm in his

32 When the monetary payoff to the firm is (v − w)e, the 33 Antonio Cabrales and Charness (2000) obtain similar firm’s choice of wage is not determined in equilibrium. results. ju05_Article 2 6/10/05 1:40 PM Page 416

416 Journal of Economic Literature, Vol. XLIII (June 2005)

interviews with managers. He describes the that explicit rewards may have the adverse hold-up model as “fanciful” and states bluntly consequence of crowding out the agent’s that “the hold-up problem hardly exists; . . . it intrinsic motivation for performing the task.35 is not a consideration in labor relations.” Part The social psychology literature contains of the explanation for the failure of the hold- experiments that demonstrate the possibility up model is that productivity depends upon that people provided with extrinsic rewards the attitudes that workers have toward each will devote less effort to a task than people other and to the firm, and that incentive not provided with explicit incentives. A rep- schemes are selected with this in mind. resentative study (Mark R. Lepper, David Employers opt for contracts that create the Greene, and Richard E. Nisbett 1973) meas- risk of hold up, but that workers do not take ured how the willingness of children to par- full advantage of this opportunity. Inefficiency ticipate in a drawing activity depended on arises, but not to the extent that theory pre- whether the activity had been rewarded in dicts. Furthermore, it does not pay to make the past. Children given extrinsic rewards for contracts as complete as possible. At the mar- the drawing activity were less likely than gin, there appears to be an important trade off other children to participate in the activity between contractual incompleteness and after the reward had been withdrawn. Social apparently unselfish behavior. psychologists interpret this result as a sign Theories of internal organization of firms that the existence of an explicit reward could benefit from an emphasis on inten- inhibits the subject’s ability to get intrinsic tions rather than on contracts and informa- enjoyment from the activity. The experimen- tion. Akerlof (1982) is a good example of the tal evidence on contracting is preliminary evi- potential of this approach. Akerlof uses dence that similar results arise in economic Homan’s (1953 and 1954) studies of clerical environments. workers in the early 1950s to motivate a Two approaches from economics provide model of labor markets that shares many fea- a framework for intrinsic rewards based on tures with the identity model that he later informational asymmetries. The work of developed with Akerlof and Kranton (2000). Bengt Holmström and Paul Milgrom (1991) In the model, the firm offers a fair wage; the on multidimensional agency problems sug- wage generates good feeling among the gests that high-powered incentive schemes workers, who in turn work more than the may encourage agents to devote effort into minimum required by the firm. Workers’ activities that lead to immediate or observ- utility may be increasing in effort in this able outputs. In this model, agents respond model due to a desire to conform to a norm to extrinsic rewards by allocating effort of behavior. The firm’s decision to pay a fair inefficiently. wage is a profit maximizing response to the Roland Bénabou and Jean Tirole (2003) strategic environment.34 assume that the principal has information relevant to the agent. Any incentive contract 4.2.2 Intrinsic versus Extrinsic Rewards offered by the principal has the potential to Economic theory emphasizes the impor- convey this information to the agent. For tance of incentives. Providing rewards con- example, if the principal has superior infor- tingent upon effort or positive performance mation about the difficulty of the job, then a directly encourages positive effort. While contract that promises a high-reward contin- there is substantial evidence consistent with gent on success of a project might convey the this point of view, there is also an argument message to the agent that the job is distasteful.

34 Robert M. Solow (1979) also argues that sociological 35 Edward L. Deci and Richard M. Ryan (1985) survey factors contribute to downward wage rigidity. the literature in social psychology. ju05_Article 2 6/10/05 1:40 PM Page 417

Sobel: Interdependent Preferences and Reciprocity 417

Bénabou and Tirole present models in which Direct exchange is standard monetary (extrin- the incentive offered by the principal may sig- sic) incentives. Indefinite exchange describes nal to the agent that the task is an onerous implicit, but material, payoffs. Internalization one. In their basic model, increasing compen- is a change in preferences brought about by sation increases the probability that an agent the work environment. If firms treat workers will supply effort, but also signals to the agent nicely, then the workers start to care about that the job is distasteful or that effort is the objectives of the firm. Employees become unlikely to lead to success. part of the team and work to advance the Uri Gneezy and Aldo Rustichini (2000a firm’s interests. Bewley suggests that firms do and 2000b) present evidence for explicit not lower wages because doing so would incentives having counterintuitive influences destroy the basis for voluntary cooperation. in laboratory and natural experiments. Intrinsic reciprocity focuses attention on the Gneezy and Rustichini (2000a) show that the role that incentive schemes play in preference imposition of an explicit penalty for failing to formation. Models based on intrinsic reci- pick up a child on time at a day-care center procity permit a firm’s behavior to induce its leads to a decrease in the number of people workers to obtain satisfaction from directly who pick up their child on time. Gneezy and contributing to the firm’s profitability. Rustichini (2000a) describe experiments 4.3 Markets and Selfishness that show the level of performance in a task is not monotonic in monetary rewards. Do markets cause selfish behavior? Some Gneezy and Rustichini suggest that a mech- have argued that they do. Samuel Bowles anism similar to the one described by (1998, page 89) observes that the more the Bénabou and Tirole is at work. The incentive “situation approximates a competitive (and scheme conveys information to the agent, complete contracts) market with many which leads to a counter-intuitive response. anonymous buyers and sellers, the less In the day-care setting of Gneezy and other-regarding behavior will be observed.” Rustichini (2000b), for example, imposing a In his comparative study of the development modest fine for late pick ups could lower the of markets in Indonesian villages, Clifford agent’s subjective probability that an even Geertz (1963, page 34) writes that “the gen- more severe penalty would be imposed. eral reputation of the bazaar-type trader for These models of negative aspects of ‘unscrupulousness,’ ‘lack of ethics,’ etc., aris- extrinsic rewards focus on information asym- es mainly from” asymmetry of roles in retail metries. In Holmström and Milgrom (1991), markets. Balance is more difficult to main- rewarding observable performance may tain in asymmetric transactions, so coopera- induce suboptimal allocation of effort. In tion based on reciprocity is more difficult to Bénabou and Tirole (2003), explicit incen- sustain. The reciprocity that facilitates coop- tives provide information that may lead to a eration in symmetric transactions does not rational agent to update his opinion about thrive in market settings.36 the attractiveness of a task. The psychology literature suggests that 36 On the other hand, the experiments that Joseph crowding out does not depend on the ability Henrich and his associates (2001) performed in different of incentive schemes to convey information societies led these authors to conjecture that cooperation about the task, but instead argues that incen- in simple experimental games is positively related to the degree of market integration. They argue that experimen- tive schemes change preferences in system- tal subjects apply strategies that succeed outside the labo- atic ways. Bewley (1999) and David M. ratory to novel experimental games. Subjects more Kreps (1997) support this point of view. exposed to situations where there are substantial gains from cooperation or where application of fairness norms Specifically, Bewley discusses three ways are important for success will be more likely to exhibit by which managers motivate workers. cooperative behavior in the laboratory. ju05_Article 2 6/10/05 1:40 PM Page 418

418 Journal of Economic Literature, Vol. XLIII (June 2005)

Results of experiments designed to model a subgame-perfect Nash equilibrium.38 competitive situations are consistent with Andreoni, Brown, and Vesterlund (2002) predictions of self interest. Vesna Prasnikar find that players make positive and roughly and Roth (1992) studied a variation of the equal contributions in treatment one, con- ultimatum game with many proposers. The tributions are positive for player one and proposers simultaneously make an offer. If slightly greater for player two in treatment the responder accepts the offer pi from pro- two, while in the third treatment experi- − poser i, then proposer i earns $10 pi, the mental behavior roughly conforms to the responder earns p, and the other proposers subgame-perfect Nash equilibrium theory earn 0. If the responder rejects all offers, (with the first player making no contribu- then all players earn nothing. Hence the tions). In the third treatment, the second design adds competition between pro- player is much less likely to punish a free posers. Subgame-perfect equilibrium pre- rider (by responding to a zero contribution dicts that the responder will receive (nearly) with a zero contribution) than in the game in all of the surplus. Experiments confirm this which contributions are additive. prediction. From the perspective of intrinsic reci- Andreoni, Paul M. Brown, and Lise procity, the third treatment is qualitatively Vesterlund (2002) identify a related experi- different from the first two. Since the sec- mental environment in which standard pre- ond player receives a higher marginal utili- dictions are confirmed and compare the ty from the public good, she can see that in result to similar games in which experimen- the third treatment any “sensible” contribu- tal results are not consistent with narrow tion made by the first player will be wast- versions of self interest. They consider a ed—the second player will want to two-player game in which the monetary pay- contribute even more, making the value of − + ƒ off of player i is T ci iG( (c1,c2)), where the first contribution zero. Hence the mag- T is an initial endowment; ci is player i’s con- nitude of player one’s contribution should > tribution; i 0 is a measure of a player’s not influence player two’s preferences over marginal utility of the consumption good; final outcomes. It is reasonable to expect ƒ aggregates the contributions; and G is player two’s material preferences to deter- an increasing, concave transformation of the mine her behavior. In the first and second total contribution. Andreoni, Brown, and treatment, player one’s strategy can influ- > 39 Vesterlund (2002) assume that 1 2 and ence player two’s material payoff. It is not distinguish three games. In the first, hard to specify preferences in the form (6) ƒ = + (c1,c2) c1 c2 and players move simultane- that are consistent with experimental ously. The second treatment uses the same results. Agents act as if they were selfish in ƒ but is a sequential game: player one the third treatment because the form of the moves first and then player two, knowing the game prevents player from exhibiting the first player’s contribution, moves second. kind of nice or nasty behavior that might The third treatment is like the second trigger deviations from selfishness. ƒ = 37 except that (c1,c2) max(c1,c2). In the first The evidence that markets behave as stan- treatment, the Nash equilibrium predicts dard theory predicts in experimental settings that the second player will contribute nothing, while in the second and third treat- 38 The predictions for the second and third treatments ments, the first player contributes nothing in require that 1 is not too much greater than 2. 39 Since the second player responds differently when player one fails to make a contribution in the second and third treatment, theories that rely on preferences that 37 This treatment is a variation on the best-shot game of depend only on the distribution of monetary payoffs will Glenn W. Harrison and Jack Hirshleifer (1989). not make predictions consistent with the experiments. ju05_Article 2 6/10/05 1:40 PM Page 419

Sobel: Interdependent Preferences and Reciprocity 419

motivates research to identify weaker market interactions (that make it difficult to assumptions under which the predictions of identify the kind and the unkind), the out- these models continue to hold. Becker’s line of a model of how markets might influ- (1962) observation that budget-constrained ence behavior begins to appear. Segal and individuals with randomly generated Sobel (2004a) provide such a model. This demands lead to downward sloping demand paper makes two contributions. First, it curves and Dhananjay K. Gode and Shyam shows that it in market environments more Sunder’s (1993) work on auction perform- general than the ones studied in experi- ance with “zero-intelligence” agents are ments, equilibrium outcomes are competi- examples. tive even when agents are willing to punish Bolton and Ockenfels (2000) and Fehr and nasty behavior and reward nice behavior. Schmidt (1999) show how the “competitive” The result follows because in a market, prediction of the game studied by Prasnikar agents can only influence the payoffs of oth- and Roth (1992) continues to hold under the ers by changing the equilibrium price and assumption that some individuals in the pop- there are limited opportunities to do this. ulation have utility functions that depend on When there is no opportunity to help or hurt the distribution of monetary payoffs. These others, narrow self interest determines results follow because, in the market, it only behavior. Second, it demonstrates for the takes two selfish bidders to drive the price up kind of separable functional forms used in to the competitive level. Assuming that some the study of interdependent preferences and of the proposers or the responder cares reciprocity ((1) and (6)), price-taking behav- about the distribution of payoffs does not ior cannot be distinguished from selfish change the equilibrium outcome. behavior. Hence agents will appear selfish The broad observations of Bowles and whenever the price-taking assumption Geertz and the specific experimental find- makes sense (for example, in large ings are also consistent with models of economies). intrinsic reciprocity. In these models, The existence of markets may not change unselfish behavior arises in response to the preferences, but it may remove incentives behavior of others. An agent may be gener- for reciprocal behavior. Indeed, the theory ous to another in order to avoid a spiteful suggests the possibility of substitution response or in order to provoke an altruistic between well organized economic institu- response. The power of spiteful behavior is tions and the observance of other-regarding weak in market environments. If a consumer behavior.40 As Partha Dasgupta (2000) refuses to buy an object because its price is notes, there is nothing mysterious in many “too high,” a firm will just sell to another cus- acts of reciprocity and “there is no reason tomer (and, in equilibrium, there will be to invoke the idea that there is greater another customer willing to pay the price). innate generosity and fellow-feeling among In some situations, the cost of generosity is high. In a competitive equilibrium, a firm that drops its price risks the possibility of 40 The interaction can go both ways. Toshio Yamagishi (1988) and Yamagishi, Karen S. Cook, and Motoki Watabe facing large excess demand. Meeting the (1998) demonstrate how well developed sanctions proce- demand might drive average cost above dures enhance trust in Japanese society. Fehr, Gächter, price. Hence, in certain market environ- and Kirchsteiger (1997) and Kevin A. McCabe, Stephen J. Rassenti, and Vernon L. Smith (1998) present additional ments it may be impossible to distinguish experimental evidence that providing more opportunities between the decisions of agents who are to punish increases cooperation in situations where self- maximizing their narrow self interest from ishness and subgame perfection predict that they would have no influence on behavior. It is possible that effective- unselfish agents who make the same deci- ness of punishment options differs systematically across sions. Added to the anonymous nature of cultures. ju05_Article 2 6/10/05 1:40 PM Page 420

420 Journal of Economic Literature, Vol. XLIII (June 2005)

poor people in poor communities than In the folk theorem, players must be for- exists among members of modern urban ward looking. In each period, there is typi- societies.” cally short-term benefit from cheating. Assuming that the predictions of standard Players refrain from cheating in order to gain models hold in more general settings, it future benefits. In reputation stories, players would be worthwhile investigating the extent use experience to determine whether they to which the properties of these outcomes believe their opponent is reliable. continue to hold. One can show the existence In the purest form, these stories differ in of competitive equilibrium in an environ- their predictions about breaches of coopera- ment where agents have exotic preferences, tion. In the simplest repeated-game stories, for example, but it is a separate question there are no deviations. (When there is a whether the unselfish behavior leads to good deviation, a punishment ensues.) In the rep- outcomes. The fundamental theorems of utation stories, a reputation develops only welfare economics provide conditions under because of learning. Learning takes place which selfish behavior leads to efficiency. only because cheating arises with positive There is no general result suggesting that probability. One expects to see breakdowns in outcomes in games played by agents with the some relationships (when the interests of the kinds of extended preferences described in players are not matched). Models of intrinsic this paper will be efficient—either with reciprocity provide a third way to describe respect to underlying material preferences cooperation in repeated interactions. After a or with respect to preferences that take into sequence of good outcomes, players’ interests account attitudes toward the behavior and become more closely linked. A history of pos- intentions of others.41 itive interaction with someone leads you to care about that person’s welfare. 4.4 Repeated Interaction Allowing players’ preferences to exhibit Game theory relies on two methods to intrinsic reciprocity typically enlarges the explain cooperative behavior in repeated equilibrium set in one-shot games. interactions. The folk theorem of repeated Therefore it is perhaps surprising that games provides a rich theory that is consis- repeated games played by agents with intrin- tent with all manner of behavior in repeated sic preference to reciprocate may have equi- interactions. If future gains are large enough librium sets that are smaller than when relative to the gains from cheating today, players have conventional preferences over then (assuming a mild technical assumption outcomes. Take a simple example. Let s∗ be holds) any individually rational, feasible pay- a strategy profile that maximizes the sum of off can be supported as an equilibrium pay- (material) payoffs in the stage game. ∗ off. The literature on reputations works Suppose that by playing s1 repeatedly, player differently. Players initially are uncertain 1 leads player 2 to play to maximize the sum about the motives of their opponent. of material payoffs. This could happen for Cooperation arises if players can infer from the appropriate specification of preferences. past behavior that their opponent is likely to It follows that if player 1 is sufficiently be trustworthy. patient, he can guarantee an average payoff equal to the outcome of s∗. Hence, under 41 Kranton’s (1996) theoretical work demonstrates the certain conditions, average repeated game stability of both market based and reciprocal systems of exchange. In her model, market-based exchange is more payoffs must be efficient (in the set of mate- efficient than bilateral reciprocal trading arrangements. rial payoffs) if individuals have intrinsic pref- Nevertheless, a system of reciprocal exchange may be self- erences towards reciprocity. On the other sustaining. When more agents opt for reciprocal exchange, markets thin. It becomes optimal for agents to engage in hand, if individuals have conventional pref- personal exchange. erences efficient payoffs are only guaranteed ju05_Article 2 6/10/05 1:40 PM Page 421

Sobel: Interdependent Preferences and Reciprocity 421

to be elements of a large set of equilibrium altruistic at the level of an organism will payoffs. The result requires strong assump- arise as the result of conflicting “self inter- tions and it suggests that agents will cooper- ests” of different genes. Traits that fail to ate in the final periods of long, repeated maximize the fitness of the organism are interaction, a prediction that is not consis- common. At this level of abstraction, theo- tent with all available evidence. It does pro- retical explanations of prosocial behavior are vide an alternative to conventional modeling easy to generate, but difficult to evaluate. In of repeated interaction. biology, reduced-form models sometimes abstract from gene-level conflicts and model evolution at the level of the individual. In 5. Origins economics, I am aware of no model that It is natural to ask where preferences treats selection at a level lower than the indi- come from. The question takes on a vidual. I therefore limit attention to selec- greater importance when one argues that tion for individual traits and describe incorporating extended preferences may conditions under which it is in the best long- lead to more useful models. A better term interest of an individual to maximize understanding of the origins of preferences something other than his short-term materi- may add structure to modeling efforts by al payoff. The interaction between individu- placing restrictions on the type of utility als and groups will play an important role in functions that people can have and the cir- this discussion. cumstances under which utility functions 5.1 Reciprocal Altruism can change. This section provides an overview of attempts to model the evolu- Trivers (1971) introduced the idea of tion of interdependent preferences and reciprocal altruism. His theory parallels reciprocity.42 standard repeated games arguments that In economic contexts, utility maximizing justify forgoing short-term gains for long- behavior can be confused with selfish behav- term benefits in repeated relationships. This ior. If one is intrinsically motivated by the type of behavior arises in nonprimates.44 desire to help others, then one can question There is a large literature on the evolution- whether making a material sacrifice to aid ary foundations of cooperative behavior in others is truly altruistic.43 The evolutionary repeated games. This literature provides a perspective might make it possible to distin- foundation for the appearance of behavior guish clearly between selfish behavior and that is inconsistent with short-term rational- more general preferences. There is no dis- ity. Reciprocal altruism is instrumental reci- pute that at the lowest level of selection, fit- procity. Actors may repay kindness with ness is the appropriate measure of success. kindness, but only because they anticipate Unselfish genes do not exist. The notion of future benefits in return. self interest leads to confusion even in bio- Arguments that generate cooperative logical models when one thinks about evolu- behavior in evolutionary settings using recip- tion of individual organisms (instead of individual genes). A trait that appears to be 44 See Lee A. Dugatkin (1997) for extensive examples. Frans B. M. De Waal (1996) describes more elaborate forms of reciprocity that arise in higher primates. Peter 42 Rajiv Sethi and E. Somanathan (2003) also review this Kropotkin (1902) presents many examples of apparently literature. The January 2004 issue of Journal of Economic nonselfish behavior in animals. Matt Ridley (1996), how- Behavior and Organization contains a review essay by ever, argues that elaborate forms of reciprocity are unique- Henrich (2004) on evolutionary models of prosocial prefer- ly human. Robert Boyd and Peter J. Richerson (1988) ences and a number of commentaries. briefly survey additional evidence supporting the position 43 Elliott Sober and David Sloan Wilson (1998) present that cooperation in large group settings is a characteristic a careful discussion of these issues. of human behavior. ju05_Article 2 6/10/05 1:40 PM Page 422

422 Journal of Economic Literature, Vol. XLIII (June 2005)

rocal altruism depend on the same kind of A well-understood limitation of this assumptions that are necessary for the folk- approach is that an individual who could theorem in repeated games. The future must grow a green beard without cooperating be important; there must be an ability and would have an advantage. This individual incentives to punish opportunistic behavior; would be able to reap the gains of generous and it must be possible to identify deviators. behavior when he meets others with green From the game-theoretic perspective, a beards without paying the cost. weakness of the repeated-game arguments is The same mechanism forms the basis for a multiplicity of equilibria. Indeterminacy general approach to the problem of selection arises in the evolutionary literature because of preferences in a strategic setting. The basic strategically stable strategies often fail to framework begins with a game. The payoffs of exist in repeated games. the game are assumed to be the players’ material payoffs (or, in a strict evolutionary 5.2 Green Beards framework, reproductive fitness). Players may Richard Dawkins (1982) describes a gener- alter their utility functions within a paramet- al mechanism that creates the possibility of ric class. For example, a player may replace cooperative behavior. Imagine a population his material payoff with a weighted average of that will play a prisoner’s dilemma game in his material payoff and the material payoff of pairs. A subset of the population has an observ- his opponent. Altering payoffs creates a new able feature (a “green beard”) that is perfectly game. Assuming equilibrium play in the new correlated with discriminatory cooperative game generates a predicted equilibrium out- behavior: people with green beards cooperate come, which in turn generates material pay- with other green beards, but with no one else. offs for the players. The evolutionary dynamic Individuals with green beards can thrive operates on the material payoffs. Players can because they gain the benefits of cooperative choose to play games using different strate- pairings without running the risk of getting gies than fitness-maximizing strategies. cheated. Robert H. Frank (1988) exploits a Further, the presence of non-fitness maximiz- version of this argument. He introduced a ers in the population may influence the model in which people decide whether to play behavior of others. Assuming that the fraction a prisoner’s dilemma game with an opponent of agents in the population with particular or opt out. The population contains one group preferences grows in proportion to relative of agent that always cooperates and another fitness, the literature asks whether there are that always defects.45 Group members give off situations in which non-fitness maximizers different signals. Frank shows that, if the sig- remain a positive fraction of the population. nals are sufficiently informative, players can There are many games in which a player use them to determine whether they wish to gains by acting as if he is not motivated by play with their opponent. A significant degree his material payoffs, provided that his oppo- of cooperation can be supported in equilibri- nents’ know his preferences. Güth and um provided that the signal that cooperators Menahem E. Yaari (1992)46 initiated the emit is sufficiently informative. The green- study of this process under the assumption beard argument provides a powerful reason, that strategy choice is observable.47 nicely modeled by Arthur J. Robson (1990), to believe that inefficient outcomes may fail to be 46 Güth (1995a), Güth and Yaari (1992), Joel M. evolutionarily stable. Guttman (2000), and Alex Possajennikov (2000) all provide results with the same general flavor, with an emphasis on social preferences broadly consistent with those introduced 45 Frank models this as a fixed characteristic of the to describe experimental outcomes. agent, but alternatively one could assume that different 47 Thomas Schelling (1960) gives vivid, early illustrations groups have different preferences. of the approach in the context of bargaining. ju05_Article 2 6/10/05 1:40 PM Page 423

Sobel: Interdependent Preferences and Reciprocity 423

These approaches study specific games; under which certain types of nonselfish they are open to the criticism that the suc- preferences can persist in a population cessful preferences are finely tailored to dominated by selfish individuals. The the strategic setting. Levent Koçkesen, Efe analysis does not provide a systematic theo- A. Ok, and Sethi (2000b) provide a version ry of the distribution of preferences that of these results that applies more broadly. would survive without a priori restrictions They identify a more general class of on preferences. Due to the complexity of games in which agents who choose a utility the analysis, the papers study the advan- function that is increasing both in their tages of a particular, intuitive sort of non- monetary payoff and their relative mone- selfish behavior, rather than identify the tary payoff (the ratio of monetary payoff to end result of an evolutionary process. This the average monetary payoff) receive high- approach is the appropriate first step in a er monetary payoffs in equilibrium than research program designed to provide a players who maximize monetary payoffs. rationale for social preferences, but sup- The critical assumption in Koçkesen, Ok, ports only the conclusion that evolutionary and Sethi (2000b) (and also in a related models do not demand all agents have fit- example of Possajennikov 2000) is an ness maximizing preferences. Future work assumption of supermodularity (or strate- may provide a more precise description of gic complementarity). To get an intuition stable preferences. for the result, imagine a symmetric two- A more basic problem is the central player game in which one player cares only assumption that preferences be observable. about monetary payoffs and the second In all of the models described, it is (at least cares about relative payoffs as well. weakly) to an agent’s advantage to convince Koçkesen, Ok, and Sethi (2000b) observe his opponent that he has nonselfish prefer- that for these games there is no equilibri- ences while actually having fitness maximiz- um in which the second player has a lower ing preferences. These agents gain the monetary payoff than the first player strategic advantages of commitment to non- (because by mimicking the first player, the selfish behavior (changing the behavior of second player increases both absolute and their opponents), but do not pay the cost relative payoff).48 Using similar tech- (making decisions that fail to maximize fit- niques, Sethi and Somanathan (2001) show ness). In this way, the models suffer from that preferences similar to those intro- precisely the same criticism as the green- duced by Levine (1998) can survive in a beard models. One would expect evolution- general family of games.49 As many public ary pressures to favor the kind of duplicitous goods games exhibit strategic substitutes, behavior found in standard models: Agents these results may help organize some would arise who look like nonselfish agents, experimental results. but whose true preferences are traditional. The approach does not provide specific These strategies are not feasible in commit- guidance about the nature of the interde- ment models because of the assumption pendent preferences that survive evolu- that there is complete information about tionary pressures. The analysis identifies preferences. circumstances in which commitment ability Ok and Fernando Vega-Redondo (2001) is valuable. The papers give conditions and Jeffry C. Ely and Okan Yilankaya (2001) examine the evolution of preferences when 50 48 The result requires more assumptions in n–player preferences are not directly observable. games. Koçkesen, Ok, and Sethi (2000b) prove a related result for games with strategic substitutes. 49 Koçkesen, Ok, and Sethi (2000a) contains a related 50 Although in Ok and Vega-Redondo (2001), players result. can infer it under some matching conditions. ju05_Article 2 6/10/05 1:40 PM Page 424

424 Journal of Economic Literature, Vol. XLIII (June 2005)

Consequently, these papers provide a formal in their abilities to deceive others and to be model that studies the critique of green- deceived.52 These characteristics vary with beard mechanism. The central idea is that individual preferences and have implications modifying preferences only increases fitness for the kinds of tasks people are suited to to the extent that the preference change can perform. modify the behavior of other agents. When The end results of these conflicting pres- preferences are not observable, there will be sures is not clear. Alexander (1987) and selective advantage to imitating the behavior Ridley (1993) argue that the pressures lead of self-interested agents. The evolutionarily to increases in social, mental, and emotional stable outcomes of the selection process must complexity.53 What remains is a complicated agree with the equilibrium outcomes of the picture of the preferences of individuals, but underlying game played by selfish agents. one which includes the possibility of non- These models therefore warn that the selfish behavior intrinsically included in assumption that preferences are observable preferences. in commitment models is critical. It provides 5.3 Kin and Group Selection a framework confirming the intuition that “green-beard” arguments rely on the assump- Arguments supporting altruistic behavior, tion. It naturally leads one to ask about the in the sense that individuals reduce their existence of mechanisms by which agents can own reproductive fitness to benefit others, credibly signal their true preferences. can be based on inclusive fitness. William D. As explanations for the existence of cooper- Hamilton’s (1964) notion of inclusive fitness ative behavior all of the green-beard models provides an explanation for altruistic behav- raise the same question: What prevents ior in animals. Hamilton shows that an defectors from learning how to fake signals? action that lowers the probability that an Even if the answer is “nothing,” it is likely that individual survives could increase that indi- selection will favor members of the popula- vidual’s total fitness (genetic contribution in tion who are capable of creating difficult to future generations) if it increases the proba- imitate signals or who are able to distinguish bility that relatives survive. Hamilton’s ideas sincere from insincere signals. Green-beard provide a way to understand a wide range of arguments have power if there is a hard-to- animal behavior. While these ideas provide break link between signal and behavior. While strong support for prosocial behavior when these links cannot be deduced from general individuals interact in small groups with principles, biologists argue that humans have closely related individuals, these conditions developed the capacity to evaluate the inten- surely do not apply to laboratory experi- tions of others.51 It seems likely people differ ments and are inadequate to explain the existence of prosocial behavior in common natural settings. 51 For example, Trivers (1971, page 5) states that “there is ample evidence to support the notion that humans There are coherent theoretical models respond to altruistic acts according to their perception of that extend kin selection to groups of unre- the motives of the altruist. They tend to respond more lated individuals. Sober and Wilson (1998) altruistically when they perceive the other as acting “gen- uinely” altruistic.” De Waal (1996, page 116) writes that “a describe how group-selection models can person who lies without blushing, who never shows remorse, and who grabs every opportunity to bypass the 52 Abraham Lincoln [or possibly P. T. Barnum] made rules just does not strike us as the most appealing friend or (most of) this point more gracefully: “You may fool all the colleague. The uniquely human capacity to turn red in the people some of the time; you can even fool some of the face suggests that at some point in time our ancestors people all the time, but you can’t fool all of the people all began to gain more from advertising trustworthiness than the time.” from fostering opportunism.” Paul Ekman (2001) argues 53 Alexander even conjectures that consciousness is the that there are characteristic facial expressions that provide outcome of the need to evaluate and interpret the motiva- credible signals of honest behavior. tions of others. ju05_Article 2 6/10/05 1:40 PM Page 425

Sobel: Interdependent Preferences and Reciprocity 425

lend support to nonselfish behavior (see also looks at three situations: two in which only Theodore C. Bergstrom 2002 and Paul A. one of the unselfish preference types is avail- Samuelson 1993). These models are natural able and one in which all three types of pref- generalizations of arguments based on inclu- erences may enter the population. Rewards sive fitness. Since Hamilton’s work, it has and punishments lower the material payoff of been apparent that nonselfish behavior can the follower. Herold investigates whether have selective advantage in closely related players with nonselfish preferences can be groups: One agent should be willing to sacri- evolutionarily stable in an environment fice individual fitness if by doing so there is where the game is played in (anonymous) a large enough increase in the fitness of small groups, fitness (material payoffs) deter- closely related individuals. In this way, the mines reproductive success, and groups re- individual’s genes (although not necessarily form after each generation. Players know the the individual) gain. More generally, one can distribution of preferences in their group, but imagine that nonselfish preferences in a sub- do not know the preferences of their partner. set of the group may make the average fit- Herold observes, as in Ok and Vega-Redondo ness of the group higher than that of an (2001) and Ely and Yilankaya (2001), the entirely selfish group. Within the group, self- leaders behave as if they maximize expected ish members do better than nonselfish mem- fitness in any stable outcome. Their behavior bers. This is the case for standard could involve giving gifts, but only if doing so public-goods models. If there is only one brings rewards or avoids punishment. There group, these conditions lead to the extinc- are stable outcomes in which a fraction of the tion of the nonselfish members. Imagine followers has interdependent preferences. instead that there is another population, Consider the case where only rewards are consisting entirely of selfish individuals. The possible. In a world in which every follower proportion of nonselfish agents in the entire is selfish, none of the leaders are generous. population may increase from one genera- Consequently, followers with preferences tion to the next if the relative increase of the for rewarding generous behavior can enter group containing nonselfish agents (at the the population without sacrificing fitness. expense of the other group) compensates for Moreover, if a positive fraction of the popu- the relative decline of the nonselfish agents lation of followers give rewards to generous within their own group. If groups remain leaders, then there is a positive probability stable over time, this argument only post- that a large enough fraction of them will be pones the extinction of the nonselfish agents concentrated in a group to induce the lead- as first the selfish group dies, and then the ers in this group to be generous. All mem- selfish agents take over the remaining group. bers of the cooperative group obtain higher If groups reform in each period, then there payoffs than the rest of the population and, are conditions under which nonselfish because there must be a high concentration behavior can survive. of followers who give rewards in this group, An attractive model using group selection the fraction of the entire population who arguments to explain the persistence of non- gives rewards can grow. selfish preferences in an economic environ- Next consider the case in which followers ment is due to Florian Herold (2003). Herold are either selfish or get utility from punish- examines a two-player game of perfect infor- ing. It is also possible for followers who like mation in which the leader can decide to punish to be represented in a stable pop- whether to give a gift to the follower. Giving ulation, but the argument differs in an the gift maximizes total surplus, but is costly important respect. It is costly for a follower to the leader. The follower can do nothing, who punishes greedy behavior to enter a reward, or punish the first player. Herold population in which all followers are selfish ju05_Article 2 6/10/05 1:40 PM Page 426

426 Journal of Economic Literature, Vol. XLIII (June 2005)

(because in this population leaders will not variation necessary for biological group be generous and therefore the punisher will selection arguments. Alexander (1987, page be required to engage in costly punish- 37 and pages 168–70) is representative of ment). Herold (2003) shows that in this set- the consensus.55 ting a monomorphic equilibrium of selfish While human groups are genetically simi- followers is stable, but there is also a lar, they are culturally diverse. These differ- monomorphic equilibrium with only follow- ences make the transmission of prosocial ers who punish. Selfish followers cannot behavior through cultural channels more gain a foot hold in the punishment equilib- probable than purely genetic transmission. rium because if there were enough selfish Christopher Boehm (1993 and 1997) pres- followers to create a group in which leaders ents and defends a mechanism that supports are greedy, then the followers in this the rise of altruistic behavior. Boehm argues group—with a disproportionate share of that human forager societies have character- selfish followers—would do badly relative to istic social organization that facilitates group the population and therefore be less com- selection for prosocial behavior. Boehm mon in the next generation. Finally, when (1993) cites evidence that foragers and other Herold permits all three types of prefer- small-scale societies create egalitarian cul- ences, a stable outcome in which all follow- tures in which all households have compara- ers punish exists and, under some ble social and economic status. This conditions, the equilibrium in which both structure comes about through the ability to some followers do nothing and some reward sanction both the shirkers who attempt to exists.54 share in the group’s resources without con- A straightforward analysis of the George tributing and bullies who attempt to monop- Price (1970) equation provides an under- olize the resources. A consequence of the standing of the mathematical conditions egalitarian structure is a reduction of the needed for a group selection argument to behavioral variation within groups. Boehm support the spread of prosocial behavior. then points out that there is variation across Henrich (2004) and Sober and Wilson groups in the way that they respond to emer- (1998) discuss these conditions carefully. A gency conditions (for example, famine). restrictive condition is that the variation Boehm (1997) argues that these features within groups is significantly smaller than of small-scale societies (in addition to the the variation between groups. If there is a lot ability and willingness of groups to sanction of variation within a group, then individuals deviations) strongly support the develop- who are not fitness maximizing within the ment of behaviors that favor group survival. group are at a large disadvantage relative to Because variation within a group is low and other group members. If different groups deviations from group behavior are sanc- are similar, then one group is unable to do tioned, there is within group pressure to significantly better than another. If there is retain traits that are good for the group. free mixing between groups, then the vari- Because different groups behave differently, ance between groups falls. successful groups grow. Hence the structure There is general agreement among biolo- imposed by human culture facilitates the gists that group selection depends on a degree of genetic variation across groups 55 Sober and Wilson (1998) argue that group selection is an important explanation for the evolution of prosocial that is not consistent with migration pat- behavior in humans. Barbara Smuts’s (1999) insightful terns. Relatively small amounts of intermar- review of Sober and Wilson’s book accepts the logical riage are sufficient to destroy between group validity of group-selection arguments, but argues that attempts to identify altruism in organisms should not lose sight of the fact that selection leads to fitness maximization 54 Gintis (2000) presents a related model. at the level of the gene. ju05_Article 2 6/10/05 1:40 PM Page 427

Sobel: Interdependent Preferences and Reciprocity 427

development of traits that are beneficial to Subjects are asked to turn over precisely group survival.56 those cards that need to be turned over to In summary, genetic group selection argu- determine the truth of the statement: ments are not likely to be an important reason for the development of altruistic If a card has a vowel on one side, then it behaviors, but that the ability of groups to has an even number on the other side. design cultural practices or institutions that Subjects make systematic errors in analyz- reduce intragroup variation and maintain ing conditional statements when given the intergroup variation does provide a power- problem in an abstract form.59 The error ful, consistent, and empirically justifiable rate goes down when the problem is refor- explanation of the development of prosocial mulated so that the task asks whether some- behavior. one has violated a social norm. In a 5.4 Evolutionary Evidence of Decision representative reformulation, the visible Biases faces of the four cards are: I have described evolutionary models (Beer) (24) (Coke) (17). based on strategic interaction. Evolutionary psychology suggests a variety of mechanisms Subjects know that each card has a beverage consistent with natural selection to describe on one side and an age on the other. social behavior. Some of this work provides Subjects are asked to turn over precisely reasons for action rules that violate standard those cards that need to be turned over to material-utility maximizing behavior in favor determine whether there are any violations of actions consistent with social norms.57 In of a law forbidding people under 21 from this section, I discuss one example that the drinking alcoholic beverages. routines are context specific. Leda Cosmides Cosmides and Tooby (1992) interpret and John Tooby (1992) review experimental these experiments as evidence that human evidence on Wason’s problem.58 In its origi- cognitive processes have evolved to identify nal form, the Peter C. Wason (1966) selec- violations of conditional statements when tion task, subjects examine four cards. On these statements can be interpreted as the visible side of the cards, they see: cheating on a social contract. They go on to speculate that people have mental algo- (E) (4) (K) (7). rithms that lead them to punish cheaters. Cosmides and Tooby’s experimental results Subjects know that each card has a letter on are stimulating, but alternative explanations one side and a number on the other. of the experimental findings that have no relationship to reciprocity or interdependent 56 Boyd and Richerson (1985) argue that cultural group preferences are available.60 selection can lead to the development and retention of prosocial behavior. They describe the importance of imita- tion as a mechanism that reduces intragroup variation, 59 Subjects should check the (E) and (7) cards to verify while maintaining intergroup variation. the statement. Most people examine either the (E) card 57 In a speculative essay, Francisco J. Varela (1999) sug- only or the (E) card and the (4) card. gests that effective people rely on “crazy wisdom.” He sug- 60 Patricia W. Cheng and Keith J. Holyoak (1985) and gests that ethical behavior requires a mixture of rational Paul Davies, James H. Fetzer, and Tom R. Foster (1995) calculation and spontaneity and that cognitive limitations offer critiques and alternative interpretations. Even if one increase the likelihood of ethical behavior. accepts Cosmides and Tooby’s interpretation of their data, 58 Social and cognitive psychologists conducted the their hypothesis only states that humans are equipped to experiments. The experimental designs violate the accept- identify certain types of cheating behavior. Standard mod- ed practices of experimental economics: the task was poor- els in economics typically assume that agents have this ly specified; subjects had no financial incentive for ability. Cosmides and Tooby’s experiments do not explain providing accurate answers; and controls on the context why agents would lower their material payoffs to respond were incomplete. to cheaters. ju05_Article 2 6/10/05 1:40 PM Page 428

428 Journal of Economic Literature, Vol. XLIII (June 2005)

5.5 Learning surrounded by supportive, cooperative informants. The hypothesis that they are Game-theoretic models of the evolution playing in cooperative strategic environ- of preferences take a simple view of the ments with cooperative players gets rein- evolutionary process. Reality is more com- forced. The existence of a supportive plicated. This subsection speculates on sev- environment is plainly essential: The child eral directions that may lead to alternative will die without food. The recognition of the explanations for nonselfish behavior. nature of the environment is also essential Animals (including humans) operate in a for the development of certain skills, notably variety of different strategic environments. language. The ability of humans to learn Cognitive constraints make it impossible for natural language presumably requires oper- anyone to optimize in every natural environ- ating under hypotheses that inputs are reli- ment. These limitations create at least three able. What mommy calls a ball really is a reasons why people would fail to exhibit ball. At least, it is something that everyone self-interested preferences. calls a ball. The ability to acquire language The first, and most obvious, observation is may be more important than the ability to that individuals will not be fully selfish avoid being duped in economic exchanges because they are unable to perform the later in life and (perhaps) biological hard- needed calculations. There is no doubt that ware may bias individuals to follow coopera- cognitive limitations prevent people from tive strategies even when they are not fitness solving complex optimization problems, but maximizing.61 in general the critique applies equally well Parents and schools attempt to teach chil- for all objective functions. Cognitive limita- dren to be nonselfish. Parents’ dominant tions are a good argument for some models position may enable them to induce their of bounded rationality, but without a theory children to internalize preferences that ben- of complexity, provide no systematic evi- efit the parents. Since conflicts of interest dence that people optimize interdependent exist between parents and children, it will preferences or incorporate intentions of not be in the best interest of parents to have others in optimizing behavior. selfish children. In particular, parents stand Another idea is that people have a limited to gain from having children who are willing ability to distinguish one situation from to repay kindness with kindness.62 To the another. Instead they use experience and extent that changes in their preferences easy-to-process signals to sort the problems operate against their self interest, children that they face into a small number of cate- should be expected to resist the changes. gories. For each category, they apply a pref- Since children have so much to gain from erence relationship (or behavioral rule of trusting their parents, and since the trust is thumb) that is well suited to representative frequently well placed, efforts to internalize members of the category. According to such preferences may be effective nevertheless. a view, an agent may have several prefer- Experimental evidence that preferences are ence relationships. Nonselfish preferences age dependent (for example the William would be likely to appear in some environments (for example, those resembling 61 Herbert A. Simon (1990 and 1993) argues that peo- repeated interaction with close associates) ple should be receptive to social influences because the benefits from access to cultural wisdom are greater than than in others. the costs associated with following society’s suggestions to Finally, preferences appear to be espe- help others. cially fluid during the earlier years of life 62 Alexander (1987, page 103) writes that a basis for moral systems is the strong incentive for individuals to and especially susceptible to the influence influence others to be “beneficent” to others. Parents may of others. Most children find themselves be particularly effective teachers. ju05_Article 2 6/10/05 1:40 PM Page 429

Sobel: Interdependent Preferences and Reciprocity 429

Harbaugh, Kate Krause, and Steven Liday Second, in all of the models, there is (2003) study of the ultimatum game) are scope for fitness maximizers. The crude consistent with the possibility that prefer- interpretation of this observation is that ences are fluid at some stages in life and that self-interested behavior, narrowly defined, people learn to adopt prosocial preferences. will always be with us. To the extent that other preferences appear in the population, 5.6 Summary they are balanced by traditional economic This section reviewed literature aimed at preferences.64 providing evolutionary foundations for non- That different people have different pref- selfish preferences. The results are mixed. If erences is hardly controversial, but it raises one takes the position that pressures that an important question.65 Under what condi- lead to selection of genetic material deter- tions on the strategic environment (or eco- mines an individual’s preferences, then there nomic institution) do standard predictions is no reason why different selection pres- remain valid when only a fraction of the sures faced by different genes in a single agents maximize material payoffs? individuals will direct the individual to max- Experimental results in auctions and best- imize fitness. The literature in economics shot bargaining suggest that standard pre- does assume that selection operates at the dictions continue to hold in competitive level of the individual. In this setting, the environments. This insight awaits a com- clearest theoretical setting for the survival of plete characterization. nonselfish preferences is an environment in Third, the evolutionary approach does not which preferences are observable or, more impose structure on preferences. The cau- generally, it is costly for selfish agents to tious conclusion from current research is appear otherwise. This condition sets the only that there is no strong argument for rul- stage for an evolutionary “arms race” in ing out all behavior that does not maximize which there is coevolution of both abilities to material payoffs. misrepresent preferences and to detect mis- representation. At the other extreme, when there is incomplete information about strategies, there is theoretical support for 64 A serious evolutionary study that permits hetero- the view that equilibrium behavior will coin- geneity of preferences must confront the issue of whether spatially isolated people facing different condi- cide with fitness maximizing equilibrium tions might evolve differently. People from different behavior for all agents. areas may have genetic predispositions toward different I draw three conclusions from the work preferences. For example, hunters face qualitatively different strategic situations than gathers. Hunters (of described. First, the analysis is consistent large game) must cooperate in order to gather food suc- with the idea that predictions based on the cessfully. Gatherers can usually collect food without assumption of maximizing material self- cooperation. One could imagine that if the style of food gathering was the essential arena for the evolution of interest need not be accurate in small-group preferences, then there would be different levels of fit- settings. In small groups, agents are more ness maximizers within a population of hunters than in a likely to know each other. Preferences are population of gatherers. On the other hand, to the extent that these differences ever existed, mixing of more likely to be observable (either directly populations would diminish them. Jared Diamond or through a signal that is linked to prefer- (1997) argues convincingly that it is unnecessary to ences). Sanctions based on expulsion are invoke genetic differences to explain broad patterns of 63 world history. feasible. These sanctions can reduce the 65 The proposition does shake the foundation of the fitness of selfish agents. methodology put forth in Stigler and Becker (1977), who argue that economists should seek to explain behavior under that assumption that “tastes neither 63 Expelled agents may have difficulty gaining entrance change capriciously nor differ importantly between to other groups. people.” ju05_Article 2 6/10/05 1:40 PM Page 430

430 Journal of Economic Literature, Vol. XLIII (June 2005)

6. Closing Arguments evidence that it fails to account for many interesting findings without being stretched If you would like to be selfish, you should do it and strained. Enlarging the toolkit to include in a very intelligent way. The stupid way to be more general models could expand the range selfish is…the way we always have worked seek- of useful economic analysis. ing happiness for ourselves alone….The intelli- Adopting a broader perspective also per- gent way to be selfish is to work for the welfare mits us to view apparent successes of stan- of others. The Dalai Lama dard theory as evidence that assumptions are too strong. When laboratory models confirm Economics, which frequently relies on the predictions of conventional models, for joint hypotheses of intelligence and self example in market settings, we can look for interest, should be open to models in which less restrictive assumptions that make the agents are selfish in intelligent ways.66 This same predictions. paper reviews reasons why more narrowly 6.2 Complexity conceived models may be insufficient, describes a variety of alternatives, and sug- Argument. Allowing extended or context- gests possible applications. This section con- dependent preferences leads to models that tains response to arguments against the impose unrealistic demands on agents’ abili- approaches discussed in the paper.67 ties to reason and modelers’ abilities to char- There are at least three problems with this acterize equilibria. list. First, it treats “standard theory” as a sin- Response. Traditional rationality assump- gle object, rather than many different tions do not impose any limits on the com- approaches. Second, I could not find clear putational abilities of agents. There is no statements of criticisms in the literature. I justification for imposing limits at an arbi- may have ignored or weakened the strongest trary point that separates traditional models criticisms. Third, I did not give the criticisms from the ones described in this paper. precise mathematical formulations. This There are several examples, notably makes the boundaries of the argument Bolton and Ockenfels (2000) and Fehr and unclear and makes it impossible for any of Schmidt (1999), of tractable models involv- the arguments or any of the counterargu- ing interdependent preferences. Arguably, ments to be decisive. Some of the criticisms none of the proposed models of reciprocity lend themselves to formal analysis. in games is as simple as these models, but further work may change that. To the extent 6.1 If It Is Not Broken, Do Not Fix It that the context-dependent models capture Argument. Standard theory works. There a genuine intuition about behavior and help is no need to change it. to organize observations and hypotheses, Response. Conventional theory does work they are useful. 68 well in many situations. The paper reviews 6.3 Only the Selfish Survive Argument. A careful study of the origin of preferences proves that only selfish agents 66 I found the Dalai Lama’s quotation in the review survive. essay on Sober and Wilson (1998) by primatologist Barbara Response.With sufficient freedom to Smuts (1999). She argues that many types of apparently unselfish behaviors are viable at the level of an organism. define “selfish” this statement is a tautology. 67 My list overlaps the list in section 3 of John Conlisk With a narrow conception of selfish behavior, it (1996), which provides a parallel defense of bounded- is possible to provide formal models that sup- rationality modeling. 68 See Edward Lazear (2000) for a compelling review of port the assertion, but the models rely on strong the successes of the Chicago school. assumptions. There are plausible reasons to ju05_Article 2 6/10/05 1:40 PM Page 431

Sobel: Interdependent Preferences and Reciprocity 431

believe that preferences for nonselfish 6.6 Definite Outcomes behavior and reciprocity have survival value Argument. Standard models provide and have survived. Until economics becomes clear predictions. Expanding the domain of a special case of molecular biology there will preferences makes it impossible to obtain be reason to examine reduced-form models definite answers to economic questions. that permit nonselfish behavior. Response. Traditional economic method- 6.4 Generality ology owes its power to its generality and its flexibility. In markets and strategic settings Argument. The standard tools of eco- indeterminacy is the rule. If preferences nomic analysis apply to a wide range of prob- need only satisfy standard neoclassical lems; no other approach has the same range. assumptions, then there are essentially no Or, in the words of Stigler and Becker (1977, restrictions on aggregate excess demand pages 76–77): “this traditional approach of functions and therefore no restrictions on the economist offers guidance in tackling market-clearing prices.69 In game-theoretic these problems and that no other approach environments, even when preferences are of remotely comparable generality and specified, multiple equilibrium problems power is available.” arise. There is no generally accepted way to Response. Relaxing the assumptions of select among multiple equilibria, especially the traditional approach creates a theory of in repeated-game environments where the more generality and more power. folk theorem places no serious restrictions 6.5 Discipline on what can be observed. Clear predictions come only as a result of imposing strong Argument. Economics needs the disci- assumptions on preferences or action sets. pline provided by the assumption of self- Not only is there indeterminacy once a interested behavior to generate behavioral model has been specified, there is no limit hypotheses and predictions based on well to the number of different models one can understood general principles. propose that have a plausible connection to Response. What Conlisk (1996, page an economic problem. Coming up with a 685) wrote in defense of bounded rationality model to explain observations is not difficult modeling is appropriate here: “Discipline ex post, the challenge is to come up with a comes from good scientific practice, not useful model that applies to more than one embrace of a particular approach. Any situation. approach . . . can lead to an undisciplined proliferation of hypotheses to cover all 6.7 Parsimony facts.” Even within conventional economic Argument. Models of extended prefer- theory, individual greed can mean many ences introduce too many free variables. The things. We are comfortable abandoning risk theory explains everything, therefore it neutrality to study gambling and insurance explains nothing. behavior. We are comfortable abandoning Response. It is important to distinguish myopic optimization to study dynamic inter- the set of all predictions that come from a the- actions. We are comfortable inserting ory from the set of predictions obtained from unmarketed goods into utility functions to a particular specification within the theory. model externalities. In the wrong hands, these modifications reflect a lack of discipline. Properly used, they reflect a willingness to extend and revise the principles of 69 For a careful statement of the results, see Gerard Debreu (1974), Rolf R. Mantel (1974), and Hugo equilibrium and optimization in order to Sonnenschein (1973). Werner Hildenbrand’s (1994) explain behavior. approach provides some limitations on aggregate behavior. ju05_Article 2 6/10/05 1:40 PM Page 432

432 Journal of Economic Literature, Vol. XLIII (June 2005)

Any of the approaches described in section 3 preferences. There is strong social pressure provide an imaginative modeler sufficient to internalize a preference for reciprocity. scope to summarize empirical regularities. There are plausible stories that describe For example, the folk theorem of repeated why these motivations persist. While the games guarantees that practically any out- body of evidence cannot establish the truth come can be supported as an equilibrium of a of the hypothesis, the evidence is sufficient- repeated interaction between patient players. ly strong and the advantages sufficiently While the theorem does have assumptions, it clear, to justify continued development of is hard to imagine any outcome from a the modeling tools that I have discussed. A dynamic interaction that could not be philosophical refusal to consider extended described as an equilibrium of some repeated preferences leads to awkward explanations game. We judge the value of a particular of some phenomena. It limits the questions repeated game model on its explanatory that can be asked and restricts the answers. power, generality, plausibility, and elegance. It is a handicap. But these criteria are informal, artistic cate- Extending the arguments of preferences gories rather than logical ones. A test of a and permitting the preferences to change class of models is its ability to provide useful with context in a systematic way enables the- descriptions and predictions. The new mod- orists to continue to use economic theory to els of interdependent preferences (Bolton predict and explain the impact of parameter and Ockenfels 2000 and Fehr and Schmidt changes, while expanding the scope of the 1999) or reciprocity (Charness and Rabin theory. These models promise a language for 2002, Dufwenberg and Kirchsteiger 2004, the study the effect of market institutions Falk and Fishbacher 2005, and Rabin 1993) and contracts on economic performance. put forth specific parametric versions of their To take the interdependent preference models. These models supply refutable theory seriously, work should proceed on implications. Parsimony demands that we three fronts. We need to develop founda- obtain our descriptions from a relatively small tional theory to identify general properties of collection of available parameterizations. extended preferences. We need to apply the theory to specific problems and develop restrictions leading to tractable models that 7. Conclusion efficiently summarize what we observe and The bully who boasts that he can beat his generate interesting hypotheses. We need foes with one hand tied behind his back will experimental work to investigate these prove his claim by picking his foes wisely. hypotheses and provide evidence about But he will look awkward when he wins and whether preferences are stable across look foolish if he loses. His boast is less a sig- games, roles, and individuals. nal of strength than an attempt to intimi- date. Restricting theory to use only a subset REFERENCES of available tools is not discipline. It is a Abreu, Dilip. 1988. “On the Theory of Infinitely Repeated Games with Discounting.” Econometrica, handicap. 56(2): 383–96. The hypothesis that reciprocity is an Abreu, Dilip, David Pearce, and Ennio Stacchetti. instrumental motivation for human behav- 1993. “Renegotiation and Symmetry in Repeated Games.” Journal of Economic Theory, 60(2): 217–40. ior is overwhelming. There are good reasons Akerlof, George A. 1982. “Labor Contracts as Partial to reciprocate in dynamic interactions as Gift Exchange.” Quarterly Journal of Economics, cooperation generates future cooperation 97(4): 543–69. Akerlof, George A., and Rachel E. Kranton. 2000. and retaliation may serve to inhibit exploita- “Economics and Identity.” Quarterly Journal of tion. There is strong evidence that the Economics 115 (3):715-53. desire to reciprocate is an intrinsic aspect of Alexander, Richard D. 1987. The Biology of Moral ju05_Article 2 6/10/05 1:40 PM Page 433

Sobel: Interdependent Preferences and Reciprocity 433

Systems. New York: Aldine de Gruyter. 817–69. Andreoni, James. 1990. “Impure Altruism and Cheng, Patricia W., and Keith J. Holyoak. 1985. Donations to Public Goods: A Theory of Warm-Glow “Pragmatic Reasoning Schemas.” Cognitive Giving.” Economic Journal, 100(401): 464–77. Psychology, 17(4): 391–416. Andreoni, James. 1995. “Cooperation in Public-Goods Cialdini, Robert C., and Melanie R. Trost. 1998. “Social Experiments: Kindness or Confusion?” American Influence: Social Norms, Conformity, and Economic Review, 85(4): 891–904. Compliance,” in The Handbook of Social Psychology. Andreoni, James. 2001. “The Economics of D. T. Gilbert, S. T. Fiske, and G. Lindzey, eds. Philanthropy,” in The International Encyclopedia of Boston: McGraw-Hill, 151–85. the Social and Behavioral Sciences. N. J. Smelser and Conlisk, John. 1996. “Why Bounded Rationality?” P. B. Baltes, eds. Oxford: Elsevier, 11369–76. Journal of Economic Literature, 34(2): 669–700. Andreoni, James, Paul M. Brown, and Lise Vesterlund. Cosmides, Leda, and John Tooby. 1992. “Cognitive 2002. “What Makes an Allocation Fair? Some Adaptations for Social Exchange,” in The Adapted Experimental Evidence.” Games and Economic Mind: Evolutionary Psychology and the Generation Behavior, 40(1): 1–24. of Culture. J. H. Barkow, L. Cosmides and J. Tooby, Becker, Gary S. 1962. “Irrational Behavior and eds. Oxford: Oxford University Press, 163–228. Economic Theory.” Journal of Political Economy, Costa-Gomes, Miguel, and Klaus G. Zauner. 2001. 70: 1–13. “Ultimatum Bargaining Behavior in Israel, Japan, Bénabou, Roland, and Jean Tirole. 2003. “Intrinsic and Slovenia, and the United States: A Social Utility Extrinsic Motivation.” Review of Economic Studies, Analysis.” Games and Economic Behavior, 34(2): 70(3): 489–520. 238–69. Bereby-Meyer, Yoella, and Muriel Niederle. 2005. Cox, James C. 2004. “How to Identify Trust and “Fairness in Bargaining.” Journal of Economic Reciprocity.” Games and Economic Behavior, 46(2): Behavior and Organization, 56(2): 173–86. 260–81. Bergstrom, Theodore C. 2002. “Evolution of Social Croson, Rachel T. A. 1999. “Theories of Altruism and Behavior: Individual and Group Selection.” Journal Reciprocity: Evidence from Linear Public Goods of Economic Perspectives, 16(2): 67–88. Games.” University of Pennsylvania Discussion Paper. Bewley, Truman F. 1999. Why Wages Don’t Fall During Dasgupta, Partha. 2000. “Economic Progress and the a Recession. Cambridge: Harvard University Press. Idea of Social Capital,” in Social Capital: A Binmore, Ken, John McCarthy, Giovanni Ponti, Larry Multifaceted Perspective. P. Dasgupta and I. Samuelson, and Avner Shaked. 2002. “A Backward Serageldin, eds. Washington, D.C.: World Bank, Induction Experiment.” Journal of Economic 325–424. Theory, 104(1): 48–88. Davies, Paul, James H. Fetzer, and Tom R. Foster. Boehm, Christopher. 1993. “Egalitarian Behavior and 1995. “Logical Reasoning and Domain Specificity: A Reverse Dominance Hierarchy.” Current Critique of the Social Exchange Theory.” Biology Anthropology, 34(3): 227–40. and Philosophy, 10(1): 1–37. Boehm, Christopher. 1997. “Impact of the Human Dawkins, Richard. 1982. The Extended Phenotype. Egalitarian Syndrome on Darwinian Selection Oxford: Oxford University Press. Mechanics.” The American Naturalist, 150: S100–21. De Waal, Frans B. M. 1996. Good Natured: The Bolton, Gary E., and Axel Ockenfels. 2000. “ERC: A Origins of Right and Wrong in Humans and Other Theory of Equity, Reciprocity, and Competition.” Animals. Cambridge: Harvard University Press. American Economic Review, 90(1): 166–93. Debreu, Gerard. 1974. “Excess Demand Functions.” Bowles, Samuel. 1998. “Endogenous Preferences: The Journal of Mathematical Economics, 1(1): 15–21. Cultural Consequences of Markets and Other Deci, Edward L., and Richard M. Ryan. 1985. Intrinsic Economic Institutions.” Journal of Economic Motivation and Self-Determination in Human Literature, 36(1): 75–111. Behavior. New York: Plenum. Boyd, Robert, and Peter J. Richerson. 1985. Culture Diamond, Jared. 1997. Guns, Germs, and Steel: The and the Evolutionary Process. Chicago: University of Fates of Human Societies. New York: Norton. Chicago Press. Dufwenberg, Martin, and Georg Kirchsteiger. 2004. “A Boyd, Robert, and Peter J. Richerson. 1988. “The Theory of Sequential Reciprocity.” Games and Evolution of Reciprocity in Sizable Groups.” Journal Economic Behavior, 47(2): 268–98. of Theoretical Biology, 132(3): 337–56. Dugatkin, Lee A. 1997. Cooperation among Animals: Cabrales, Antonio, and Gary Charness. “Optimal An Evolutionary Perspective. New York: Oxford Contracts, Adverse Selection, and Social University Press. Preferences: An Experiment.” Universitat Pompeu Ekman, Paul. 2001. Telling Lies. New York: W. W. Fabra Discussion Paper 478. Norton. Charness, Gary, and Ernan Haruvy. 2002. “Altruism, Ellingsen, Tore, and Jack Robles. 2002. “Does Equity, and Reciprocity in a Gift-Exchange Evolution Solve the Hold-Up Problem?” Games and Experiment: An Encompassing Approach.” Games Economic Behavior, 39(1): 28–53. and Economic Behavior, 40(2): 203–31. Ely, Jeffrey C., and Okan Yilankaya. 2001. “Nash Charness, Gary, and Matthew Rabin. 2002. Equilibrium and the Evolution of Preferences.” “Understanding Social Preferences with Simple Journal of Economic Theory, 97(2): 255–72. Tests.” Quarterly Journal of Economics, 117(3): Falk, Armin, Ernst Fehr, and Urs Fischbacher. 2003. ju05_Article 2 6/10/05 1:40 PM Page 434

434 Journal of Economic Literature, Vol. XLIII (June 2005)

“On the Nature of Fair Behavior.” Economic or Don’t Pay at All.” Quarterly Journal of Economics, Inquiry, 41(1): 20–26. 115(3): 791–810. Falk, Armin, and Urs Fischbacher. 2005. “Modeling Gode, Dhananjay K., and Shyam Sunder. 1993. Fairness and Reciprocity,” in Moral Sentiments and “Allocative Efficiency of Markets with Zero- Material Interests: The Foundations of Cooperation Intelligence Traders: Market as a Partial Substitute in Economic Life. H. Gintis, S. Bowles, R. Boyd, and for Individual Rationality.” Journal of Political E. Fehr, eds. Cambridge: MIT Press. Economy, 101(1): 119–37. Farrell, Joseph, and Eric Maskin. 1989. “Renegotiation Goeree, Jacob K., and Charles A. Holt. 2000. in Repeated Games.” Games and Economic “Asymmetric Inequality Aversion and Noisy Behavior 1(4): 327–60. Behavior in Alternating-Offer Bargaining Games.” Fehr, Ernst, and Simon Gächter. 2000. “Fairness and European Economic Review, 44(4-6): 1079–89. Retaliation: The Economics of Reciprocity.” Journal Gül, Faruk. 2001. “Unobservable Investment and the of Economic Perspectives 14(3): 159–81. Hold-Up Problem.” Econometrica, 69(2): 343–76. Fehr, Ernst, and Simon Gächter. 2002. “Do Incentive Güth, Werner. 1995a. “An Evolutionary Approach to Contracts Crowd Out Voluntary Cooperation?” Explaining Cooperative Behavior by Reciprocal University of Zurich Working Paper 34. Incentives.” International Journal of Game Theory, http://www.iew.unizh.ch/wp/iewwp034.pdf. 24(4): 323–44. Fehr, Ernst, Simon Gächter, and Georg Kirchsteiger. Güth, Werner. 1995b. “On Ultimatum Bargaining 1997. “Reciprocity as a Contract Enforcement Experiments-A Personal Review.” Journal of Device: Experimental Evidence.” Econometrica, Economic Behavior and Organization, 27(3): 329–44. 65(4): 833–60. Güth, Werner, Rolf Schmittberger, and Bernd Fehr, Ernst, Erich Kirchler, Andreas Weichbold, and Schwarze. 1982. “An Experimental Analysis of Simon Gächter. 1998. “When Social Norms Ultimatum Bargaining.” Journal of Economic Overpower Competition: Gift Exchange in Behavior and Organization, 3(4): 367–88. Experimental Labor Markets.” Journal of Labor Güth, Werner, and Menahem E. Yaari. 1992. “An Economics 16(2): 324–51. Evolutionary Approach to Explaining Cooperative Fehr, Ernst, George Kirchsteiger, and Arno Riedl. Behavior by Reciprocal Behavior in a Simple 1993. “Does Fairness Prevent Market Clearing? An Strategic Game,” in Explaining Process and Change: Experimental Investigation.” Quarterly Journal of Approaches to Evolutionary Economics. U. Witt, ed. Economics, 108(2): 437–59. Ann Arbor: University of Michigan Press, 23–34. Fehr, Ernst, and Klaus M. Schmidt. 1999. “A Theory of Guttman, Joel M. 2000. “On the Evolutionary Stability Fairness, Competition, and Cooperation.” Quarterly of Preferences for Reciprocity.” European Journal of Journal of Economics, 114(3): 817–68. Political Economy, 16(1): 31–50. Fehr, Ernst, and Klaus M. Schmidt. 2000. “Fairness, Hamilton, William D. 1964. “The Genetical Evolution Incentives, and Contractual Choices.” European of Social Behavior.” Journal of Theoretical Biology, Economic Review, 44(4-6): 1057–68. 7(1): 1–17. Fehr, Ernst, and Klaus M. Schmidt. 2003. “Theories of Harbaugh, William, Kate Krause, and Steven Liday. Fairness and Reciprocity—Evidence and Economic 2003. “Bargaining by Children.” University of Applications,” in Advances in Economics and Oregon Discussion paper. Econometrics: 8th World Congress. M. Dewatripont, Harrison, Glenn W., and Jack Hirshleifer. 1989. “An L. P. Hansen, and S. J. Turnovsky, eds. Cambridge: Experimental Evaluation of Weakest Link/Best Shot Cambridge University Press. Models of Public Goods.” Journal of Political Frank, Robert H. 1988. Passions within Reason. New Economy, 97(1): 201–25. York: Norton. Hart, Oliver. 1995. Firms, Contracts, and Financial Fremling, Gertrud M., and Richard A. Posner. 1999. Structure. Oxford: Clarendon Press. “Market Signaling of Personal Characteristics.” Henrich, Joseph. 2004. “Cultural Group Selection, University of Chicago Discussion Paper. Coevolutionary Processes and Large-Scale Fudenberg, Drew, and Eric Maskin. 1986. “The Folk Cooperation.” Journal of Economic Behavior and Theorem in Repeated Games with Discounting or Organization, 53(1): 3–35. with Incomplete Information.” Econometrica, 54(3): Henrich, Joseph, Robert Boyd, Sam Bowles, Herbert 533–54. Gintis, Ernst Fehr, Richard McElreath, and Colin Geanakoplos, John, David Pearce, and Ennio Camerer. 2001. “In Search of Homo Economicus: Stacchetti. 1989. “Psychological Games and Behavioral Experiments in 15 Small-Scale Societies.” Sequential Rationality.” Games and Economic American Economic Review, 91(2): 73–78. Behavior, 1(1): 60–79. Herold, Florian. “Carrot or Stick? Group Selection and Geertz, Clifford. 1963. Peddlers and Princes. Chicago: the Evolution of Reciprocal Preferences.” University University of Chicago Press. of Munich Discussion Paper. Gintis, Herbert. 2000. “Strong Reciprocity and Human Hildenbrand, Werner. 1994. Market Demand: Theory Sociality.” Journal of Theoretical Biology, 206(2): and Empirical Evidence. Princeton: Princeton 169–79. University Press. Gneezy, Uri, and Aldo Rustichini. 2000. “A Fine is a Holmström, Bengt, and Paul Milgrom. 1991. “Multitask Price.” Journal of Legal Studies, 29(1): 1–17. Principal-Agent Analyses: Incentive Contracts, Asset Gneezy, Uri, and Aldo Rustichini. 2000. “Pay Enough Ownership, and Job Design.” Journal of Law, ju05_Article 2 6/10/05 1:40 PM Page 435

Sobel: Interdependent Preferences and Reciprocity 435

Economics, and Organization, 7: 24–52. Rabin. 1999. “Choice Bracketing.” Journal of Risk Homans, George. 1953. “Status Among Clerical and Uncertainty, 19(1-3): 171–97. Workers.” Human Organization, 12: 5–10. Ridley, Matt. 1993. The Red Queen: Sex and the Homans, George. 1954. “The Cash Posters.” American Evolution of Human Nature. London: Viking. Sociological Review, 19: 724–33. Ridley, Matt. 1996. The Origins of Virtue. New York: Koçkesen, Levent, Efe A. Ok, and Rajiv Sethi. 2000a. Penguin. “Evolution of Interdependent Preferences in Robson, Arthur J. 1990. “Efficiency in Evolutionary Aggregative Games.” Games and Economic Games: Darwin, Nash and the Secret Handshake.” Behavior, 31(2): 303–10. Journal of Theoretical Biology, 144(3): 379–96. Koçkesen, Levent, Efe A. Ok, and Rajiv Sethi. 2000b. Roth, Alvin E. 1995. “Bargaining Experiments,” in “The Strategic Advantage of Negatively Handbook of Experimental Economics. J. H. Kagel Interdependent Preferences.” Journal of Economic and A. E. Roth, eds. Princeton: Princeton University Theory, 92(2): 274–99. Press, 253–348. Kranton, Rachel E. 1996. “Reciprocal Exchange: A Sahlins, Marshall. 1968. Tribesmen. Englewood Cliffs, Self-Sustaining System.” American Economic NJ: Prentice-Hall. Review, 86(4): 830–51. Samuelson, Paul A. 1993. “Altruism as a Problem Kreps, David M. 1997. “Intrinsic Motivation and Involving Group versus Individual Selection in Extrinsic Incentives.” American Economic Review, Economics and Biology.” American Economic 87(2): 359–64. Review, 83(2): 143–48. Kropotkin, Peter. 1902. Mutual Aid: A Factor of Schelling, Thomas. 1960. The Strategy of Conflict. Evolution. London: William Heinemann. Cambridge: Harvard University Press. Lazear, Edward. 2000. “Performance Pay and Segal, Uzi, and Joel Sobel. 2004a. “Markets Make Productivity.” American Economic Review, 90(5): People Look Selfish.” University of California, San 1346–61. Diego Discussion Paper. Lepper, Mark R., David Greene, and Richard E. Segal, Uzi, and Joel Sobel. 2004b. “Tit for Tat: Nisbett. 1973. “Undermining Children’s Intrinsic Foundations of Preferences for Reciprocity in Interest with Extrinsic Reward.” Journal of Strategic Settings.” University of California, San Personality and Social Psychology, 28: 129–37. Diego Discussion Paper. Levine, David K. 1998. “Modeling Altruism and Service, Elman R. 1966. The Hunters. Englewood Spitefulness in Experiments.” Review of Economic Cliffs, NJ: Prentice-Hall. Dynamics, 1(3): 593–622. Sethi, Rajiv, and E. Somanathan. 2001. “Preference Mantel, Rolf R. 1974. “On the Characterization of Evolution and Reciprocity.” Journal of Economic Aggregate Excess Demand.” Journal of Economic Theory, 97(2): 273–97. Theory, 7(3): 348–53. Sethi, Rajiv, and E. Somanathan. 2003. “Understanding McCabe, Kevin A., Stephen J. Rassenti, and Vernon L. Reciprocity.” Journal of Economic Behavior and Smith. 1998. “Reciprocity, Trust, and Payoff Privacy Organization, 50(1): 1–27. in Extensive Form Bargaining.” Games and Simon, Herbert A. 1990. “A Mechanism for Social Economic Behavior, 24(1-2): 10–24. Selection and Successful Altruism.” Science, Ok, Efe A., and Fernando Vega-Redondo. 2001. “On 250(4988): 1665–68. the Evolution of Individualistic Preferences: An Simon, Herbert A. 1993. “Altruism and Economics.” Incomplete Information Scenario.” Journal of American Economic Review, 83(2): 156–61. Economic Theory, 97(2): 231–54. Smuts, Barbara. 1999. “Multilevel Selection, Palfrey, Thomas R., and Jeffrey E. Prisbrey. 1997. Cooperation, and Altruism: Reflections on Unto “Anomalous Behavior in Public Goods Experiments: Others.” Human Nature, 10(3): 311–27. How Much and Why?” American Economic Review, Sober, Elliott, and David Sloan Wilson. 1998. Unto 87(5): 829–46. Others: The Evolution and Psychology of Unselfish Possajennikov, Alex. 2000. “On the Evolutionary Behavior. Cambridge: Harvard University Press. Stability of Altruistic and Spiteful Preferences.” Solow, Robert M. 1979. “Another Possible Source of Journal of Economic Behavior and Organization, Wage Stickiness.” Journal of Macroeconomics, 1(1): 42(1): 125–29. 79–82. Postlewaite, Andrew. 1998. “The Social Basis of Sonnenschein, Hugo. 1973. “Do Walras’ Identity and Interdependent Preferences.” European Economic Continuity Characterize the Class of Community Review, 42(3-5): 779–800. Excess Demand Functions?” Journal of Economic Prasnikar, Vesna, and Alvin E. Roth. 1992. Theory, 6(4): 345–54. “Considerations of Fairness and Strategy: Stigler, George J., and Gary S. Becker. 1977. “De Experimental Data from Sequential Games.” Gustibus Non Est Disputandum.” American Quarterly Journal of Economics, 107(3): 865–88. Economic Review, 67(2): 76–90. Price, George. 1970. “Selection and Covariance.” Sugden, Robert. 1984. “Reciprocity: The Supply of Nature, 227(5257): 520–21. Public Goods through Voluntary Contributions.” Rabin, Matthew. 1993. “Incorporating Fairness into Economic Journal, 94(376): 772–87. Game Theory and Economics.” American Economic Thaler, Richard H. 1999. “Mental Accounting Review, 83(5): 1281–1302. Matters.” Journal of Behavioral Decision Making, Read, Daniel, George Loewenstein, and Matthew 12(3): 183–206. ju05_Article 2 6/10/05 1:40 PM Page 436

436 Journal of Economic Literature, Vol. XLIII (June 2005)

Trivers, Robert L. 1971. “The Evolution of Reciprocal Routledge and Kegan Paul, 44–75. Altruism.” Quarterly Review of Biology, 46(1): Yamagishi, Toshio. 1988. “The Provision of a 35–58. Sanctioning System in the United States and Japan.” Varela, Francisco J. 1999. Ethical Know-How. Social Psychology Quarterly, 51(3): 265–71. Stanford: Stanford University Press. Yamagishi, Toshio, Karen S. Cook, and Motoki Watabe. Wason, Peter C. 1966. “Realism and Rationality in the 1998. “Uncertainty, Trust, and Commitment Selection Task,” in Thinking and Reasoning: Formation in the United States and Japan.” Psychological Approaches. J. Evans, ed. London: American Journal of Sociology, 104(1): 165–94.