BEHAVIORAL FOUNDATIONS OF RECIPROCITY: EXPERIMENTALECONOMICS AND EVOLUTIONARY PSYCHOLOGY

ELIZABETH HOFFMAN, KEVIN A. MCCABE and VERNON L SMITH*

Laboratory experiments have generally supported the theorem that, in classicalproperty rights environments, noncooperative behavior in markets yields efficientsocial outcomes. Experiments, however, regularly fail to support the game theoreticprediction of noncooperative behavior in small-group strategic interaction and inlarge-group public good environments. In these two types of experiments subjectsfrequently achieve more efficient social outcomes—they collect more money fromthe experimenter—than noncooperative game theory predicts. Many subjects inthese experiments exhibit reciprocity even in single-play games. Evolutionary psy-chologists hypothesize that humans have evolved mental algorithms for identifyingand punishing cheaters in social exchange. {JEL Al 1, A12, B41, C70, C72, C92)

I. INTRODUCTION

Theorists have long studied the fundamen-tal problem that cooperative, socially efficientoutcomes generally cannot be supported asequilibria in finite games. The puzzle is theoccurrence of cooperative behavior in the ab-sence of immediate incentives to cooperate.For example, in two-person bargaining exper-iments, where noncooperative behavior doesnot result in efficient outcomes, we observemore cooperative behavior and greater effi-ciency than such environments are expectedto produce. Similarly, in public good experi-ments with groups varying in size from fourto 100 people, the participants tend to achievemuch higher payoff levels than predicted bynoncooperative theory. Moreover, examplesof cooperative behavior achieved by decen-tralized means have a long history in the

•We are grateful to the National Science Foundationfor research support under NSF #SBR-9210052 to the Uni-versity of Arizona.Hoffman: Provost and Vice Chancellor for Academic

Affairs, and Professor of Economics, History, andPsychology, University of Illinois at Chicago, 111.Phone 1-312-413-3450, Fax I-312-413-3455E-mail ehorrman@uic.edu

McCabe: ESL Senior Research Scholar, Professor ofEconomics, and IFREE Distinguished ResearchScholar, Economic Science Laboratory, University ofArizona, Tucson, Phone 1-520-621-3830Fax 1-520-621-5642E-mail lcmccabe@econlab.arizona.edu

Smith: Regents' Professor and McClelland Professor ofEconomics, Economic Science Laboratory, Universityof Arizona, Tucson, Phone 1-520-621-4747Fax 1-520-621-5642E-mail smith@econlab.arizona.edu

human experience. Anthropological and ar-chaeological evidence suggest that sharing be-havior is ubiquitous in tribal cultures that lackmarkets, monetary systems, or other means ofstoring and redistributing wealth (see, e.g.,Cosmides and Tooby [1987; 1989]; Isaac[1978]; Kaplin and Hill [1985]; Tooby and DeVore [1987]; Trivers [1971]).

In this paper we draw together theoreticaland experimental evidence from game theory,evolutionary psychology, and experimentaleconomics to develop a reciprocity frameworkfor understanding the persistence of coopera-tive outcomes in the face of contrary individ-ual incentives. The theory of repeated gameswith discounting or infinite time horizons al-lows for cooperative solutions, but does notyield conditions for predicting them(Fudenberg and Tirole [1993]). Recent re-search in evolutionary psychology (Cosmidesand Tooby [1987; 1989; 1992]) suggests thathumans may be evolutionarily predisposed toengage in social exchange using mental algo-rithms that identify and punish cheaters. Fi-nally, a considerable body of research in ex-perimental economics now identifies a num-ber of environmental and institutional factorsthat promote cooperation even in the face of

ABBREVIATIONSCT: Cosmides and ToobyFHSS: Forsythe, Horowtiz, Savin and SeftonHMSS: Hoffman, McCabe, Shachat and SmithKKT: Kahneman, Knetsch and ThalerPD: Prisoner's DilemmaVCM: Voluntary Contribution Mechanism

335Economic Inquiry(ISSN 0095-2583)Vol. XXXVI, ;uly 1998, 335-352 OWestem Economic Association International

336 ECONOMIC INQUIRY

contrary individual incentives (Davis and Holt[1993]; Isaac and Walker [1988a,b; 1991];Isaac, Walker and Thomas [1984]; Isaac,Walker and Williams [1991]). Moreover, theseexperimental results indicate that trust andtrustworthiness play a much greater role thanthe evolutionary psychologists' punish-cheat-ers model would suggest. We hypothesize thathumans' abilities to read one anothers' minds(Baron-Cohen [1995]) in social situations fa-cilitates reciprocity.

II. REPEATED GAMES

Repeated-game theory offers two explana-tions of cooperation based on self-interest:self-enforcing equilibria and reputations.Self-enforcing equilibria are based on the ideathat players can credibly punish noncoopera-tive defections. The nagging problem withself-enforcing cooperative equilibria is thatthere are many equilibria in such games withcooperation being only one possibility.

Experiments demonstrating that subjectscooperate in games with repeated play and rel-atively short finite horizons (Selten and Stoec-ker [1986]; Rapoport [1987]) suggest reputa-tions are important in games with incompleteinformation (Kreps et al. [1982]). The idea isthat if players are uncertain about otherplayers' types, then the possibility emergesthat players will mimic (develop a reputationas) a type different from their own. In circum-stances where cooperation is mutually benefi-cial players have an incentive to mimic coop-erative behavior.

In the examples given by Kreps et al.[1982], players rationally compute strategiesbased on (utility or payoff) type uncertainty.They cooperate from the beginning until nearthe end of the game, and then defect. This isnot, however, the pattern observed in experi-ments, where it is common for cooperation todevelop out of repeated interactions; also, de-fection near the end is often not observed.

The strength of the theory is that it is basedon individual (but longer run) self-interest,and is parsimonious. Its weakness is that itadmits many possible equilibria without sug-gesting why cooperation is the most likelyoutcome. Moreover, for reputation-basedequilibria, people must entertain beliefs aboutcertain types of people.

But where do these beliefs come from? Weintroduce the hypothesis that types emerge

from the evolutionary fitness of certain cog-nitive abilities which predispose many peopletowards reciprocity. Actual circumstances andexperiences may lead to reciprocal behaviorby many persons. Not everyone has to be aparticular type; variability is the stuff fromwhich selection occurs and which allows na-ture to adapt to change. But the type mustexist in sufficient numbers for people to be-lieve that reciprocity pays. And if reciprocitypays, culture and norms develop to specify theforms that reciprocity will take.

III. MENTAL ALGORITHMS FOR SOCIALEXCHANGE: STRATEGIES IN HUMAN

COGNITION THAT SUPPORT COOPERATION

The complex organization of the humanmind is thought to be the product of at leasta few million years of evolutionary adaptationto solve the problems of hunting and gather-ing.1 Evolutionary psychologists hypothesizethat these problems were solved not only byneurobiological adaptations, but also by adap-tations in human social cognition (see Cos-mides and Tooby [1992], hereafter CT, and thereferences therein). The idea is that humanshave special and highly developed cognitivemechanisms for dealing with social exchangeproblems: that is, mental modules for solvingsocial problems are as much a part of theadapted mind as our vision and hearing-bal-ance faculties.2

Examples of mental "computational" mod-ules that solve specialized design problems in-clude vision, language and "mind reading."The mechanism which constitutes vision in-

1. But see Rice [19%] for an experiment in whichfemale fruit flys are prevented from coevolving with males.After only 41 generations male adaptation leads to a re-duction in female survivorship in the genetic battle of thesexes.

2. Research by neuroscientists on the amygdala, an al-mond-sized structure deep in the temporal lobe of the brain,has shown that it is directly involved in the perception ofsocial signals. That the amygdala participates in the socialcognition and behavior of animals has been known formany years, but recent studies have shown that these find-ings extend to humans (Allman and Brothers [1994]Adolphs et al. [1994]). Thus, subjects with damagedamygdalas are unable to recognize or distinguish expres-sions such as fear, surprise and anger on faces in photo-graphs of people. In one study, the subject had great dif-ficulty determining whether individuals were looking ather or away from her. The amygdala operates precon-sciously: "the evidence ... clearly indicates that theamygdala is involved in the evaluation of complex stimulilong before they are completely analyzed cognhively, andprobably long before they enter awareness" (Halgren[1992, 194]).

HOFFMAN, MCCABE & SMITH: FOUNDATIONS OF RECIPROCITY 337

volves neural circuits whose design solves theproblem of scene analysis (Marr [1982]). Thesolution to this problem employs specializedcomputational machinery for detecting shape,edges, motion, bugs (in frogs), hawks (in rab-bits), faces, etc. Just as we leam by exposureto see and interpret scenes without beingtaught, we learn to speak without formal train-ing of any kind.

Although culture is known to operate onour mental circuitry for language learning, thedeep structure of language is common acrosscultures (Pinker [1994]). Normal English-speaking preschoolers can apply mental algo-rithms to root words to form regular noun plu-rals by adding "s" and the past tense of regularverbs by adding "ed" (Pinker [1994, 42-43]).The preschooler even "knows" that you cansay that a house is mice-infested but neverthat it is rats-infested, that there can beteethmarks but never clawsmarks—the mentalalgorithm here allows compound words to beformed out of irregular plurals but never outof regular plurals. This is because of the waythe unconscious brain works: regular pluralsare not stem words stored in the mental inven-tory, but words derived algorithmically by theinflectional rule to add "s." Preschoolers in alllanguages automatically make these kinds ofdistinctions without being taught by theirmothers or their teachers (Pinker [1994, 146—47]).

That the mind contains blueprints for gram-matical rules is further indicated by a lan-guage disorder in families which appears tobe inherited like a pedigree with a dominantgene. English speakers afflicted with this dis-order are unable to inflect root words to formderivatives such as the "s" rule for obtainingplurals.

"Mind reading"—the process of inferringthe mental states of others from their wordsand actions—facilitates "social understand-ing, behavioral predictions, social interaction,and communication" (Baron-Cohen [1995,30]). Autism in children makes them mindblind—they are not automatically aware ofmental phenomena in others, and cannot"mind read".3 A genetic basis is suggested by

3. Pinker [1994, 227] for example provides the follow-ing exchange: Woman: "I'm leaving you." Man: "Wlio ishe?" If you are not autistic you know what this conversa-tion means.

its greater risk in identical twins and biologi-cally related siblings. Baron-Cohen [1995,88-95] implicates the amygdala and relatedareas of the brain as jointly controlling theability to detect eye direction in others and tointerpret mental states (have a theory of mind)in others. Other detector mechanisms appearto include "friend or foe"—cooperation is notautomatic for foes—and the "fight or flight"response to sudden danger.

The hypothesis that our minds are also pre-disposed to learn behavioral responses thatpromote cooperative outcomes does not meanthat we are born with such behavioral re-sponses. We only need to be born with thecapacity to learn such responses developmen-tally from social exposure, much as we areborn with the capacity to learn any languagebut not with the ability to speak any particularone. A capacity for the natural learning ofstrategies that induce cooperation in social ex-change has fitness value. But the im-plementational form of what is learned varieswidely, depending upon the environment, ac-cidents of nature, and how parental, familial,and societal units organize exchange pro-cesses. Consequently, "culture" is endlesslyvariable, but, functionally, reciprocity is uni-versal.

Naturally selected fitness strategies are hy-pothesized to be embodied in the designs thatmodulate reasoning about social exchange. Ananalysis of these strategies allows one to de-duce the behavioral characteristics of the as-sociated mental algorithms. This analysis alsoallows predictions about human responses inreasoning experiments of the kind that wesummarize below. These psychology experi-ments are of particular interest to experimen-tal economists because they complement sub-ject behavior in many games of strategic in-teraction.

Consider the standard two-personPrisoner's Dilemma (PD) game, but think ofthe entries corresponding to C (cooperate) orD (defect) for the row and column players asnet benefits and net costs measured in unitsthat increase (or decrease) the individual's in-clusive fitness. C might represent the strategy"trade," while D might represent "steal." Asdiscussed above, game theory predicts thatmutual cooperation will not emerge in a sin-gle-move game—people are all self-interested"foes."

338 ECONOMIC INQUIRY

Imagine a tournament that matches pairsfrom a large population of organisms so thatthe same two individuals are never matched asecond time. Each member is matched, repro-duces itself, and dies. The offspring inheritsthe strategy choice propensity of the parent,and the number of offspring is proportional tothe payoff gains of the parent in its matchedplays of the game. Each generation repeatsthis process.

Repeated-game principles can be used toanalyze equilibrium outcomes in such a game.Repeat interaction is a prominent characteris-tic of social exchange. Needs are rarely simul-taneous. But, long before human societies in-vented a generally accepted medium of ex-change, various cultural mechanisms providedsocial adaptations which allowed delayed mu-tual benefits to be gained: I share my meatwith you when I am lucky at the hunt, andyou share yours with me when you are lucky.Although this is commonly referred to as re-ciprocal altruism, we prefer to call it reciproc-ity. I am not altruistic if my action is basedon my expectation of your reciprocation.

Reciprocity leads naturally to propertyrights. If I grow corn and you grow pigs, andwe exchange our surpluses, then we each havean interest in the other's property right in whatis grown. If either of us plays "steal," thatends the trading relationship. Hence, mutualrecognition and defense of informal propertyright systems need not require the pre-exis-tence of a Leviathan.

But how might such mutual cooperationemerge in a repeated PD game? We knowfrom the work of Axelrod and Hamilton[1981] that strategy C cannot be selected forin repeated play, but that the contingent coop-erative strategy, T (tit-for-tat), can be selectedfor. In general any strategy, including T, cansuccessfully invade a population of defectorsif (and only if) it cooperates with cooperatorsand punishes defectors (Axelrod [1984]). Asnoted by CT [1992, 176-77], it is an empiricalissue to determine which strategy, out of thisadmissible set, is actually embodied in humancognitive programs.

The need to solve the PD problem toachieve cooperation provides an abstractschema for organizing our thoughts about co-operation beyond immediate kin. However,simply referring to the motivating example ofthe PD will not carry us to a full understand-

ing of human social exchange. In particular,it will not help us understand cooperative be-havior toward anonymous strangers whenthere is no prospect for punishment. This isan anomaly in the CT evolutionary paradigm.

An important question for the evolutionaryparadigm is whether the mental algorithms forsocial exchange consist of a few content-freegeneralized rules of reasoning, or whetherthey consist of designs specialized for solvingsocial exchange problems. Economic/gametheory is driven by the principle that humansnaturally use content-free generalized rules ofreasoning in solving decision problems. If thisis so, why is economics so hard to teach? Ifthese rules come only from culture, wheredoes culture come from?

CT [1992] argue that the evolutionary per-spective favors specialized over generalizedrules. General rules, applicable to any subjectmatter, "will not allow one to detect cheaters... because what counts as cheating does notmap onto the definition of violation imposedby the propositional calculus. Suppose weagree to the following exchange: 'If you giveme your watch then I'll give you $20.' Youwould have violated our agreement—youwould have cheated me—if you had taken my$20 but not given me your watch. But accord-ing to the rules of inference of the proposi-tional calculus, the only way this rule can beviolated is by your giving me your watch butmy not giving you $20" (CT [1992, 179-80]).That is, the way you falsify "if P, then Q,"statements is to look for "P, not Q," evidence.In this example, giving me your watch is theP statement; my not giving you $20 is thenot-Q statement. If such rules were the onlyones contained in our minds, we would haveno special ability to detect cheating.4

4. Unlike deductive logic, a cheater detection mecha-nism must account for intentionality. In the CT experimentsexchange is sequential: first, I give you the watch, thenonly later do you pay the $20. Here the clear interpretationis that the second mover has cheated if he or she does notpay the $20. This rules out the use of the biconditionalstatement, "You give me your watch," if and only if, "Igive you $20," as a substitute for the conditional. Sincethe biconditional has an ambiguous intertemporal interpre-tation, it is less clear that a contract is implied Suppose Igive you $20, but you don't give me your watch. Thebiconditional is clearly false even if I haven't cheated you;when, for example, I give you the $20 altruistically. Notewe can write the more complicated logical statement, if(we agree to P i f f 0 , then (P iff 0 , to give the biconditio-nal the correct intertemporal interpretation without com-mitting to the order of trade.

HOFFMAN, MCCABE & SMITH: FOUNDATIONS OF RECIPROCITY 339

One theme in the CT research program isto design experiments that will test thesekinds of propositions (CT [1992, 181-206]).The selection task that CT employ was devel-oped by Wason [1966], whose motivation wasto inquire as to whether the ordinary learningexperiences of people reflected the Popperianhypothesis-testing logic outlined above. Theprocedure uses four cards, each carrying oneof the labels, P not-P, Q, noi-Q on the sidefacing up, and another of the same four labelson the side facing down. Each card corre-sponds to some situation with one of the la-beled properties. The rule is violated only bya card that has a P on one side and a noi-Qon the reverse side.

Subjects are asked to indicate only thecard(s) that definitely need to be turned overin order to see if any cases violate the rule.The correct answer is to indicate the cardsshowing P (to see if there is a noi-Q on theother side) and noi-Q (to see if there is a Pon the other side). In one example, asecretary's task is to check student documentsto see if they satisfy the rule: If a person hasa "D" rating, then his document must bemarked code "3." Four cards show D, F, 3,and 7, and subjects should indicate the cardsshowing the letter D and the numeral 7. Lessthan 25% of college students choose both ofthese cards correctly.

Now consider a law which states that "If aperson is drinking beer, then he must be over20 years old." Out of four cards which alsoinclude "not drinking beer" and "25 yearsold" the correct response is to choose the card"drinking beer" and the card "16 years old."In this experiment about 75% of college stu-dents get it right. Why the difference from theprevious example?

Although people do better in more familiarexamples such as: "If a person goes to Boston,then he takes the subway," less than half getit right. A survey of this literature (Cosmides[1989]) suggests that "Robust and replicablecontent effects were found only for rules thatrelated terms that are recognizable as benefitsand cost/requirements in the format of a stan-dard social contract" (CT [1992, 183]). Six-teen out of 16 experiments using social con-tracts showed large content effects. Fourteenout of 19 experiments which did not use con-tract rules produced no content effect, two

produced a weak effect, and three produced asubstantial effect.

These findings launched a number of stud-ies designed to separate the social contract hy-pothesis from confounding interpretations,such as familiarity, or that the social contextmerely facilitates Popperian reasoning. CT re-port that the alternative hypotheses have notsurvived experiments designed to separatethem from the cheater-detection hypothesis.5

IV. OBSERVABILITY, COMMUNICATION, ANDINTENTIONALITY SIGNALING

If humans are preprogrammed to learn toachieve cooperative outcomes in social ex-change, then factors that facilitate the opera-tion of these natural mechanisms should in-crease cooperation even in the presence ofcontrary individual incentives. For example,cooperation should increase if individuals canobserve and monitor one anothers' behaviors,even if there are no direct mechanisms for en-forcing specific behaviors. In Baron-Cohen's[1995] model of mind reading, the eye direc-tion, shared attention, and intentionality de-tectors are used to identify and ratify the vo-litional states of others. Observation and mon-itoring activate one or more of these detectors.Moreover, if it is possible for agents to di-rectly punish cheating by other agents, coop-eration should increase even further.

Similarly, if agents can communicate withone another, they can frame a group decisionas a social exchange problem and ratify oneanothers' volitional states, thus activating nat-ural inclinations to cooperate for increased in-dividual gain. Thus, communication can in-crease cooperation even if there are no effec-tive mechanisms for monitoring and punish-ing cheaters.

5. Other experiments have examined violations of so-cial contracts when they do not involve cheating(Gigerenzer and Hug, cited in CT [1992, 195]. Only 44%correctly solve the no-cheating version, while 83% get thecheating version correct. Cosmides and Tooby (in prepa-ration, cited in CT [1992, 198]) have examined social con-tract problems which distinguish violations due to cheatingfrom violations due to innocent mistakes. The cheatingversion is correctly solved by 68% of the subjects, but themistake version is only solved by 27% of the subjects.Other social contract reasoning tasks asked subjects to de-tect altruists instead of cheaters. People are not good atdetecting altruists. In fact where the rule was a social law(public good) more people detected cheaters than altruists(CT [1992, 193-95 and footnote 17]).

340 ECONOMIC INQUIRY

Voluntary Contribution Experiments

The standard environment for studying thefree rider problem in the allocation of publicgoods is the voluntary contribution mecha-nism (VCM), extensively studied by Isaac andWalker, and their coauthors (Isaac, McCueand Plott [1985]; Isaac, Schmitz and Walker[1989]; Isaac and Walker [1988a,b]; Isaac,Walker and Thomas [1984]; Isaac, Walker andWilliams [1991]). In a VCM experiment, eachsubject is given a set of tokens at the begin-ning of each period. The subject may investtokens in an individual exchange, with a fixedmonetary return per token, and/or a group ex-change, which returns money to the subject asa function of the total contributions of all thesubjects in the experiment.

Typically the individual incentives are de-signed to make strong free riding, or zero con-tributions to the group exchange, the domi-nant strategy for each subject. On the otherhand, the highest joint payoff for all subjectsis achieved when all subjects contribute 100%of their tokens to the group exchange.

Isaac and Walker and their coauthors, ascited above, find that contributions to thegroup exchange are sensitive to differences inthe rules of message exchange that relate toour previous discussion of cognitive mecha-nisms for social exchange. With subjectgroups of four or ten subjects, if subjectsmake contributions in private, if there is noidentified target level of contributions, and ifthey do not communicate with one another atany time during the experiment, then contri-butions to the group exchange decline fromabout 40% of tokens in period one to about10% of tokens in period 10 (Isaac and Walker[1988a]; Isaac, Walker and Thomas [1984]).These results extend to large groups of 40 or100 people, but per capita contributions actu-ally increase relative to groups of size four orten in some treatments.

In the same experimental environment,however, if subjects can talk with one anotherfor a short period before each decision, con-tributions to the group exchange quickly riseto almost 100% of tokens, even if actual in-vestment decisions are made in private (Isaacand Walker [1988b]). These results illustratethe importance of "cheap talk" communica-tion in creating an environment in whichagents expect one another to behave coopera-

tively and they abide by the reinforced normeven when all decisions are made in privateand no individual's defection can be detectedby others.

The results can also be interpreted in a sig-naling context. During the communicationphase, individuals verbally signal that theywill behave cooperatively and that they expectothers to reciprocate. During the decision-making phase, individuals generally abide bythe norm reinforced by the signal, and a co-operative outcome is achieved. While no di-rect punishment can be inflicted by other sub-jects in the event of defection, other subjectscan exact general punishment by defectionagainst other subjects in future rounds.

In other experiments (Isaac, Schmitz andWalker [1989]), the experimenter establishesa minimum provision-point contribution tothe group investment. Comparing results withand without a provision point, and allowingno communication, contributions to the groupaccount increase with the provision point.When the provision point is 100% of tokens,contributions rise even further, although manygroups fail to attain it.

From a signaling perspective, the provisionpoint signals an expected joint level of con-tribution to the group account, and helps toinduce common expectations of substantialcontributions to the group account. With equalendowments the implied signal is that eachsubject should contribute (l/«)th of the an-nounced provision point.

Ultimatum and Dictator Experiments

Ultimatum and dictator experiments illus-trate the importance of observability, sharedexpectations of social norms, punishment, andsignaling in enforcing reciprocity behavior. Inan ultimatum game, player 1 makes an offerto player 2 of SX from a total of $M. If player2 accepts the offer, then player 1 is paid $(A/— X) and player 2 receives $X; if player 2rejects the offer, each gets $0. In the dictatorgame, player 2 must accept player 1 's offer.

Under the usual rationality assumptions thenoncooperative equilibrium of the ultimatumgame is for player 1 to offer player 2 thesmallest dollar unit of account, and for player2 to accept the offer. In the dictator gameplayer 1 offers player 2 nothing. In the ulti-matum game, however, player 2 can punish

HOFFMAN, MCCABE & SMITH: FOUNDATIONS OF RECIPROCITY 341

player 1 for "cheating" on an implied socialnorm of reciprocal sharing across time, in so-cial exchange, by rejecting player l's offer.That response is a dominated strategy, ifviewed in isolation, since both players wouldbe financially better off even with avanishingly small offer. But, in the absence ofcommon knowledge of self-interested behav-ior, the possibility of punishment may changeplayer 1 's equilibrium strategy.

In Kahneman, Knetsch and Thaler [1986](hereinafter KKT), players 1 and 2 in an ulti-matum game are "provisionally allocated"$10 and player 1 is asked to make an initialoffer to "divide" the $10 between the twoplayers. Player 2 may veto the division, inwhich case they both get $0. Kahneman andhis coauthors find that most often player 1offers $5 to player 2; offers of less than $5are sometimes rejected. Although there aresome differences, the general features of theseresults have been replicated in cross-culturalcomparisons suggesting that the results arenot strongly culture-specific (Roth, Prasnikar,Okuno-Fujimara and Zamir [1991]). This sug-gests that the explanation may transcend cul-ture.

Forsythe, Horowtiz, Savin and Sefton[1994] (hereinafter FHSS) replicate KKT's re-sults from the ultimatum game, and also studythe dictator game. They find that about 20%of dictator player Is offer nothing to theirplayer 2 counterparts, as noncooperative gametheory would predict; however, it is morecommon for player 1 to offer $5 than to offernothing, and offers of $1, $2, $3, and $4 areapproximately evenly distributed. Thus, re-moving the threat of punishment reduces shar-ing behavior, but not by as much as game the-ory predicts.

Recognizing that the prospect of punish-ment might create expectations that changeplayer l 's behavior, Hoffman, McCabe,Shachat and Smith [1994] (hereafter HMSS)consider experimental treatments explicitlydesigned to affect subject expectations aboutoperating norms of social exchange. The ex-perimental instructions that describe the dif-ferent treatments might be viewed as signalsto the subjects of the expected social normoperating in each experiment.

Brewer and Crano [1994], a recent socialpsychology textbook, lists three norms of so-cial exchange that may apply in ultimatum

games. From our perspective, norms are theproduct of culture interacting with mentalmodules in order to solve specific problemsof social exchange. Such norms can then in-form a theory of mind mechanism as toanother's volitional state. Equality impliesthat gains should be shared equally in the ab-sence of any objective differences between in-dividuals suggesting another sharing rule. Eq-uity implies that individuals who contributemore to a social exchange should gain a largershare of the returns. Reciprocity implies thatif one individual offers a share to another in-dividual, the second individual is expected toreciprocate within a reasonable time. We dis-tinguish negative reciprocity—the use of pun-ishment strategies to retaliate against behaviorthat is deemed inappropriate—and positivereciprocity—the use of strategies that initiateor reward appropriate behavior.

The designs of KKT and FHSS invoke theequality norm. No distinction is made be-tween tie two individuals "provisionally allo-cated" $10, and they are told to "divide" themoney. Hence, deviations from equal divisionare more likely to be punished as "cheating"on the social exchange. Using the same taskdescription, HMSS replicate the FHSS resultsin a "random/divide $10" treatment.

To invoke equity, HMSS explore two vari-ations on their random/divide $10 treatmentin a 2x2 experimental design. First (the ex-change treatment), without changing the re-duced form of the game, HMSS describe it asa market in which the "seller" (player 1)chooses a price (division of $10) and the"buyer" (player 2) indicates whether he or shewill buy or not buy (accept or not accept).From the perspective of social exchange, aseller might equitably earn a higher returnthan a buyer. Second (the contest treatment),they make each seller earn the property rightto be a seller by scoring higher on a generalknowledge quiz than buyers. Winners are thentold they have "earned the right" to be sellers.Going back to Homans [1967], equity theorypredicts that individuals who have earned theright to a higher return will be socially justi-fied in receiving that higher return.

Figure 1 reproduces HMSS's random/di-vide and contest/exchange experimental re-sults. Social exchange theory predicts that, ina situation in which it is equitable for player1 to receive a larger compensation than player

342 ECONOMIC INQUIRY

FIGURE laUltimatum; Random Entitlement,

FHSS Instructions, Divide $10, N=24

Percentage

% Offer % Rejection

Percentage

20 -

10 -

FIGURE lbUltimatum; Contest Entitlement,

FHSS Instructions, Divide $10, N=24

% Offer % Rejection

HOFFMAN, MCCABE & SMITH: FOUNDATIONS OF RECIPROCITY 343

Percentage

FIGURE lcUltimatum; Random Entitlement,

Exchange, N=24

% Offer % Rejection

70

60

50

40

30

20

10

Percentage

FIGURE IdUltimatum; Contest Entitlement,

Exchange, N=24

% Offer

$ Offer

| ' | % Rejection

344 ECONOMIC INQUIRY

2 (i.e., contest/exchange), (a) player 1 willoffer significantly less to player 2; while (b)player 2 will accept any given offer withhigher probability. The data in Figure 1 areconsistent with prediction (a) and not incon-sistent with prediction (b). Player Is offer sig-nificantly less to player 2s, while rejectionrates are statistically indistinguishable. Theseresults suggest that the change from ran-dom/divide to contest/exchange alters theshared expectations of the two players regard-ing the social exchange norm operating to de-termine an appropriate sharing rule. Finally,the difference between random/divide andcontest/exchange carries over to dictator ex-periments as well, indicating that the changein expectations takes place even when thereis no threat of punishment from player 2.

But why do these treatments reduce offerswithout causing an increase in the rejectionrate? One hypothesis is that both players inferone anothers' mental states—in this case ex-pectations—from relevant information in theexperiment. "Mind reading" implies the abil-ity to take the perspective of another personwho has common information. In this experi-ment, player 1 expects player 2 to find a loweroffer acceptable, while player 2 expects, andis prepared to accept, a lower offer. At mini-mum, this involves a shared attention mecha-nism.

Observability is potentially powerful in theenforcement of social norms. Thus, FHSS re-cruited Player Is and Player 2s in separaterooms, and the players were anonymous withrespect to one another. However, subject de-cisions were not anonymous with respect tothe experimenter. Someone was still "watch-ing"; hence player Is were still not entirelyremoved from a social exchange setting wherereciprocity norms might unconsciously apply.

This led HMSS to design a "double-blind"dictator experiment, with several features thatwere later changed one or two at a time, toinvestigate the role of social isolation in ex-tinguishing behavior reflecting social norms(Hoffman, McCabe and Smith [1996a]). In thedouble-blind treatment, 64% of the Player Istake all $10; about 90% take at least $8.6

These results are strikingly different fromthe dictator results in FHSS, and from theHMSS random/divide and contest/exchangedictator experiments in which subjects wereobserved by the experimenters. Next, in three

stages, HMS vary each of the elements of thedouble-blind dictator experiment in ways in-tended to reduce the "social distance" be-tween the subjects and anyone who might seetheir choices. The experimental results form apredicted ordered set of distributions. As thesocial distance between the subject and othersdecreases, the cumulative distribution of of-fers to Player 2s increases. These results dem-onstrate the power of isolation from impliedobservability in the enforcement of norms ofequality, equity and reciprocity.

Signaling, Trust, and Punishment inBargaining Experiments

In this section we review the results of two-person extensive form bargaining/trust exper-iments in which players move sequentially,and one player can choose to play—signal—cooperatively. Berg, Dickhaut and McCabe[1995] have adapted the double-blind proce-dure to study trust and reciprocity in a two-stage dictator game. In stage one player 2 de-cides how much of $10 to send to player 1,and how much to keep. The amount sent tri-ples to M before reaching player 1. In stagetwo player 1, acting as a dictator, decides howto split the M dollars. Since the amount to besplit is endogenous, the two players now sharea common history before the dictator game isplayed. If reciprocity plays a significant rolein promoting social exchange, then their com-mon history should reduce the "social dis-

6. Bolton, Katok and Zwick [1993], using a differentversion of the dictator game and using different double-blind procedures, find no difference between their double-blind and single-blind treatments. The results from suchtreatment variations are always of interest, but claims thatthe experiments show that the results of HMSS do notreplicate exceed what is demonstrated. When examiningtreatment variations on an earlier study, a second experi-menter must first show that he/she can duplicate the orig-inal results using the same treatment and procedures, es-tablishing that the results replicate with different subjectsand different experimenters. Only then can the results usingthe new treatment, if different, be attributed to these con-ditions and not to the subjects, experimenter, or proceduresused. Thus, HMSS replicated the procedures and results ofForsythe et al. [1994] before attempting to compare themwith the results from new treatments. Given the sensitivityof the dictator game to procedures and instructions, it isimportant that other researchers be able to replicate suchfindings before changing the treatment. Eckel and Gross-man [1996] replicated the HMSS double-blind experimentsbefore conducting their interesting new treatment in whichthe recipient was the American Red Cross instead of an-other subject. Terry Bumham also replicated the HMSSdouble-blind experiments in a study currently in process(private communication).

HOFFMAN, MCCABE & SMITH: FOUNDATIONS OF RECIPROCITY 345

tance" between subjects in a two-stage dicta-tor game. While Berg, Dickhaut and McCabefind significant use of trust and reciprocity,subjects in their experiments had no alterna-tive except to rely on trust for mutual gain.

McCabe, Rassenti and Smith [1996a] studyan extensive form game in which a player canchoose between two subgames, each of whichcan result in mutual gain. In one subgame mu-tual gain can be achieved using reciprocityincentives, while in the other subgame mutualgain is achieved using self-interested incen-tives. By choosing the reciprocity subgamethe subject signals a desire to cooperate, andeach subject can earn 50. By choosing theself-interested subgame the subject signals adesire to play noncooperatively, and each sub-ject earns 40. In some of these experiments,the signaling player, at a cost to himself orherself, can directly punish the other playerfor "cheating" on the implied social exchange.In the other "trust" experiments, there is nodirect opportunity to retaliate against defec-tion from a signal to cooperate.

The Constituent Games: Payoffs. Figure 2shows the extensive form bargaining tree forthese two constituent, or stage, games playedby two persons. Player 1 begins with a moveright or down at node x,. A move right termi-nates the play with payoffs (35, 70), in cents,in repeat play (multiplied by 20 in singleplay), respectively for Players 1 and 2. If themove is down, then Player 2 moves left orright at node x2, and so on. Play ends with anymove that terminates at a payoff box on theright or the left of the tree. Game 1 shows thebaseline payoff structure used; Game 2 is thesame except for the payoffs in the boxes cor-responding to plays left at nodes x3 and x5.McCabe, Rassenti and Smith [1996a] havestudied behavior in these games under a vari-ety of matching protocols and informationtreatments.

In both Games 1 and 2 the right side of thetree contains the subgame perfect (SP) non-cooperative outcome (40, 40), where Player 2moves right at x6. This outcome is achievedby simple dominance, once Player 2 movesright at x2; i.e., it is in Player l's interest toplay down at x4, and for Player 2 then to playright at x6.

In Game 1, cooperative actions by the play-ers can lead to the largest symmetric (LS) out-come (50, 50), achieved if Player 1 moves leftat x3. Under complete payoff information, amove left at x2 by Player 2 can be interpretedas a signal to Player 1 that Player 1 should goleft at x3. (This is because 50 at LS is clearlybetter than 40 at SP for Player 2, allowingPlayer 1 to infer Player 2's reason for playingleft atx2.) Player 1, however, can defect, movedown at x3, and force Player 2, in his or herown interest, to move left at x5 giving Player

1 a payoff of 60. In fact this is the game the-oretic prediction if play occurs on the left sideof the tree in Game 1. In a single play, Player2 should see this and the theoretical predictionbecomes Selten's SP outcome on the right.

But a move left at x2 in Game 1 is morethan a signal that Player 2 wants to achievethe LS outcome (50, 50). It can also be inter-preted as a potential threat to play down atx5, punishing Player 1 if Player 1 defects or"cheats" by playing down at x3. This action,however, is costly to Player 2, since eachplayer gets 20 if Player 1 moves left atx7. But,given the way subjects behave in ultimatumgames, it is not unreasonable to assume thatsome subjects will move left at x2 and thenpunish defections at x3.

Game 2 contrasts with Game 1 in that toachieve LS, by Player 2 moving left at x5,Player 1 must resist the temptation to moveleft at x3. In Game 2, Player 1 can "cheat" onthe invitation to cooperate by choosing (60,30) without the prospect that Player 2 canpunish Player 1. Thus, Game 2 allows signal-ing, but not punishment; it is a game of trust.

Experimental Design. Table I shows fourtreatments that vary the protocol for matchingpairs in each experiment. An experiment con-sists of groups of 8—16 subjects who are ran-domly assigned to pairs. In Repeat Single webegin the session with 16 subjects, and eachperson plays every other counterpart once,with their roles alternating between Player 1and 2. Under Contingent each player indicatesher choice at each node. Then the computerexecutes the play. Single means that all pairsplay the constituent game exactly once for amultiple of 20 times the payoffs shown in theboxes of Figure 2.

346 ECONOMIC INQUIRY

FIGURE 2Games One and Two in Extensive Form

Player Id's

[50, 50][60, 30]

[60, 30][50, 50]

[20, 20]

Payoff Box

[35, 70]

DecisionNode

[30, 60]

[40, 40]

[15,30]

[0,0] [0,0]

* Game 2 differs from Game 1by these two payoff boxes.

Summary of Results. Table II lists the condi-tional outcome frequencies for each payoffbox. Reading across data row 1 for Single 1we observe that 13 of 26 Player 2s moved leftat x2 indicating cooperation; 10 of the 13 leftplays ended with Player 1 choosing (50, 50);three Player Is defected by playing down atx3; two of these Player 2s accepted the defec-tion and responded with (60, 30), while oneplayed down at xs to punish Player 1 who thenchose (20, 20). In the right game, played by13 of 26 Player 2s, 12 of 13 ended at the SPoutcome (40, 40); one play was at (15, 30).

The column labeled E^ILeft) computes theexpected profit, 44.6 cents, to player 2 fromplaying left at x2, based on the relative frequen-cies of subsequent play by both players.E(7i||Down) is the expected profit, 46.7 cents,to Player 1 from defecting at node x3. Effi-ciency is the percentage of the cooperativetotal payoff at (50, 50) that is realized by allplayers. Thus in Single 1 85.5% of the coop-erative surplus is collected by all pairs. At SPefficiency is 80%, so any greater efficiencyimplies a net social benefit from cooperativeinitiatives.

HOFFMAN, MCCABE & SMITH: FOUNDATIONS OF RECIPROCITY 347

TABLE IExperimental Design

Treatments and Number of Pairs8

Designation Constituent Game Matching Protocol Number Pairsc

Single 1 1Single 2 1Repeat Single 1 1

Single 1 Contingent*1 1

Single PlaySingle Play

Repeat Single Playb

Single Play, All Nodes

2617

2624

"Payoff information is complete—both players know both payoffs in all treatments,

each subject plays each counterpart once with type alternating between player 1 and 2.cSessions consist of at least eight subjects, four pairs. In Repeat Single 1,16 subjects each played every other subject

exactly once.dContingent play means that each player makes a response at each of his/her decision nodes. Then the computer

executes the play once.

Result 1. Game theory predicts that in Single1 all plays will be in the right subgame. Infact half are in the left subgame. In RepeatSingle 1, we observe that experience does nothelp to achieve SP; now 58% play the leftsubgame. Contrary to the theory, we observeboth too much attempted cooperation and toofew defections on these attempts. Conditionalon right-branch play however, game theorydoes very well in predicting the SP outcomefor both Game 1 and Game 2 in all treatments.

Result 2. In all treatments it is (weakly) ad-vantageous in the expected payoff sense toplay in the left subgame. This is indicated bythe fact that the expected profit to Player 2 ofleft-branch play is at least 40.0 cents in alltreatments, and 40 is the payoff to Player 2 atSP. Thus, right subgame play by the minorityis not profitable in both Games 1 and 2.

Result 3. Defections by Player 1 at node or3 ofGame 1 are not profitable under the Single 1and Repeat Single 1 treatments: the expectedprofit of playing down is always less than 50,the payoff to Player 1 by not defecting. Thus,the "punish cheaters" mental module hypoth-esized by Cosmides [1985] is alive. Moreoverit is used only just enough to be effective, butnot so much that efficiency is badly compro-mised.

Result 4. Single 1 Contingent converts Game1 from the extensive to the normal form byrequiring each player's choices at all nodes ofthe tree to be made in advance for simulta-neous play. It is equivalent to expressing allpayoff path outcomes in matrix form for si-

multaneous choice by both players. Game the-ory hypothesizes that the normal and exten-sive forms are equivalent, but previous re-search has shown that this is not generally thecase (Schotter, Wiegelt and Wilson [1994]).Comparing Single 1 with Single 1 Contingentwe see that left play declines (right play in-creases) in the latter. Why? Our hypothesis isthat the extensive form, with sequential turn-taking moves, allows the players to engage ina move interpreting conversation. Thus, atnode x2, Player 2 has just received the mes-sage, "I moved down at x] because I want todo better than receive 35," from Player 1. IfPlayer 2 now moves left, the message is "I amplaying left because I want to forgo the (40,40) on the right in favor of (50, 50) which isbetter for both of us. Also, note that if yourespond by playing down at x3, then I have theoption of punishing you with (20, 20)." Thishypothetical dialogue is disrupted with simul-taneous play, although under strict rationalityit is irrelevant: Player 2's message is not cred-ibly self enforcing. But as we have seen(Baron-Cohen [1995]), mindreading allowsplayers to infer mental states from actionsand, as shown by these results, may lead themto play differently in the extensive form thanin the normal form.7

7. Additional tests of the reciprocity hypothesis basedon comparisons of the extensive form with matrix normalform are reported in McCabe, Smith and Lepore [1997].The reciprocity hypothesis also implies that SP outcomeswill predominate under private information. This predic-tion is strongly supported in McCabe, Rassenti and Smith[1996b].

Treatment

Left

50,50

60,30

20,20

Right

30,60

40,40

15,30

£(ji2l Left)b

E(KI| Down)0

Efficiency %d

Conditional

Single 1

13/26 = .50

10/13 = .769

2/3 = .667

1/1 = 1

13/26 = .50

0/13 = 0

12/13 = .92

1/1 = 1

44.6

46.7

85.5

TABLE IISummary Data; All Treatments

Outcome Probabilities

Single 2'

12/26 = .462

6/12 = .5

6/6= 1

0

14/26 = .538

0/14 = 0

14/14= 1

0

40.0

—

86.9

by Treatment; for all Trials

Repeat Single 1

204/352 •» .580

133/204= .652

33/71 •= .549

36/36= 1

148/352 = .420

9/148 - .061

138/139 = .993

0/1 - 0

41.5

42.0

85.1

Single 1 contingent

9/23 = .391

8/9 •= .889

1/9 = .111

0

14/23 = .609

3/14 = .214

11/11 = 1

0

47.8

60

88.7

'Note that in Game 2 the play order of the outcomes (60, 30) and (50, 50) are reversed relative to game 1.bExpected payoff to player 2 from playing left at X2, given me relative frequencies of subsequent play by player 1 andcExpected payoff to player 1 from defecting at node xj.dEfficiency is the percent of the cooperative (50, 50) total payoff that is realized by all pairs.

2.

8oooc

HOFFMAN, MCCABE & SMITH: FOUNDATIONS OF RECIPROCITY 349

Result 5. The failure of the SP predicted out-come (Result 1) motivated the study of Game2 in which the cooperative (50, 50) outcomecannot be supported by the prospect of pun-ishment. Comparing Single 2 with Single 1(rows 2 and 1 of Table II), we see a slightreduction in left moves by Player 2s in Game2. Play in left subgame 2 produces fewer (50,50) outcomes (50%) than in Game 1 (76.9%).This reduces the expected profit of left playfrom 44.6 cents in Game 1 to a break-even 40cents in Game 2. Clearly, the strategic differ-ence between the two games is making a dif-ference in the game theoretic predicted direc-tion. The more interesting observation is thatthe trust element in Game 2 is sufficient toyield cooperation for half of the pairs whoplay the left subgame. This is consistent withresults reported by Fehr, Kirchsteiger andRiedl [1993] in labor market experiments, andby Berg, Dickhaut and McCabe [1995] in in-vestment dictator games. In these studies firstmovers trusted second movers to reciprocatewith no possibility of punishment.

If you think of noncooperative game theoryas applying to "foes," in these extensive formexperiments the theory accurately predicts be-havior in up to half the observations. The rel-evance of traditional game theory for a largesegment of this population cannot be dis-missed. However, the other half, who persistin cooperation, need also to be explained andmodeled. Their behavior is not extinguishedwith experience: in Repeat Single 1, the per-cent of play in the left reciprocity branch in-creases to 58%. We conjecture that minimalelements for a complete theory of mental phe-nomena in games of strategy should include:(1) a friend-or-foe detection mechanism, and(2) an intentionality detector mechanism,where the latter requires extensive form playto achieve its full scope.

V. WHEN DO PEOPLE ABANDONRECIPROCITY IN FAVOR OF

NONCOOPERATIVE PLAY

The above examples illustrate a model ofa mixture of individuals, some of whose playreflects game theoretic principles, whileothers' play reflects learned or innate re-sponses involving signaling, trust, punishmentand other ingredients of reciprocity behavior.In the latter, the play objective serves the typ-

ical subject well: they exceed the performanceof strict game-theoretic players in that sur-plus-improving cooperative outcomes aremore often attained than theory would predict.

In this section we consider a contrary ex-ample to those above, one in which subjectsbegin with their intuitive automatic responses,discover that these responses cannot sustaingood performance, then adjust in the directionof the noncooperative rational expectationsoutcome predicted by theory. In this case sub-jects are given common information, but thisis not sufficient to induce common knowledgein the sense of expectations. (Also see Smith,Suchanek and Williams [1988] and Harrisonand McCabe [1992]). This, we argue, is be-cause common information leaves behavioralor strategic uncertainty unresolved. The latteris resolved over time as subjects, in successiveextensive form rounds, come to have commonexpectations that predicted equilibrium out-comes will prevail.

McCabe [1989] reports a six person, sixperiod, extensive form game experiment usingfiat money. In successive periods subjects usebuy, sell and null messages to trade, or nottrade, a unit of fiat money against cash divi-dend paying bonds. In the last period a bondholder should not sell since he or she is leftwith worthless fiat money. Similarly, themoney should not be accepted on the penulti-mate round, and by backward inductionshould not be accepted in the first period. Al-though subjects have complete information onthis payoff structure, trade in the first play ofthe sequence yields trade in each period untilthe last one. Repeating this constituent game10 times (common information) causes some,but not a complete, unravelling backwardfrom the final trial. When subjects return fora second 15 trial experiment, the slow unrav-elling process continues, but trade persists, es-pecially in the early trials. In a third sessionfor 20 trials, gradually, trade is further dimin-ished, and is virtually eliminated by the 15thtrial.

These results can be understood in termsof a model in which people have been stronglyconditioned by reciprocity experience to ac-cept fiat money in trade because they expectothers to accept money when they offer it intrade. This expectation is unconscious; theynever ask themselves why they and others ac-cept money. It is a conditional reciprocity re-

350 ECONOMIC INQUIRY

sponse, which serves them effectively in dailylife. They are recruited to the laboratorywhere the conditions for ongoing repeated ex-change are not satisfied; in the end-game in-trinsically worthless money is refused intrade. This failure experience induces them toreevaluate their unconscious, accustomed re-sponse to money. Very slowly, in the limit, asplay is repeated in the finite horizon environ-ment, trade converges to zero.8

VI. CONCLUSIONS

The experimental game results summarizedin this paper suggest that people invoke re-ward/punishment strategies in a wide varietyof group interactive contexts. These strategiesare generally inconsistent with, but more prof-itable than, the noncooperative strategies pre-dicted by game theory. However, in contrastto CT's emphasis on punishing cheaters, weobserve substantial use of positive as well asnegative reciprocity strategies, even in single-play games. Hence behavior is much richerand more trusting than CT's model would pre-dict.

A punish-cheaters mechanism has the ad-vantage, as in tit-for-tat, that it can sustaincooperation. But is a pure trust/trustworthymechanism sustainable? Recall that the "co-operate" strategy C in the PD game cannotresist invasion by defectors. This is still anopen question, but Carmichael and MacLeod[1997] offer a model which is encouraging.They analyze gift exchange showing that astable gift-giving custom, which does not de-pend upon the use of punishment strategies,may emerge.

Consider the following hypothetical modelof the mind for human decision making. Weinherit a circuitry which is modularized forsolving social exchange problems. But theswitches are not set; that occurs sometime inour maturation, requires no formal instruction,and is not a self-aware process. In this senseit is like the way we "learn" natural languagewithout being taught The switches are set dif-ferently in different cultures, but the resultsare functionally equivalent across cultures; inparticular there is a propensity to be pro-

8. Similarly, Camercr and Weigelt [1988] report veryslow convergence in a sequential equilibrium reputationmodel.

grammed to try cooperation in dealing withother people who are not detected as foes. Butthere is variation so that we can talk aboutpopulation distributions of P, the probabilitythat a person will initiate cooperation, of Q,the probability that a person will defect on anoffer to cooperate, of R, the probability a de-fection will be punished, of S, the probabilitythat a person will be trusting, of T, that a per-son will be trustworthy, and so on. These dis-tributions of player types are an adaptationcapable of changing slowly over time.

Formal education is hard because it is con-cerned with conscious learning, expression,and action, and does not come naturally, justas written language is unnatural and hard toleam. When people are exposed to economicprinciples, most find it extremely hard to leamabout comparative advantage, opportunitycost, gains from exchange, and Nash equilib-ria. Many give up, but it does not follow thatif they are in an economics experiment thatthey will perform poorly. This is because theymay be good at reading other minds and rely-ing on their unconscious natural mental mech-anisms. These mechanisms help to define rep-utations that are applied repeatedly across dif-ferent life, and laboratory, games. A one-shotgame in the laboratory is part of a life-longsequence, not an isolated experience that callsfor behavior that deviates sharply from one'sreputational norm. Thus, we should expectsubjects to rely upon reciprocity norms in ex-perimental settings, unless they discover inthe process of participating in a particular ex-periment that reciprocity is punished andother behaviors are rewarded. In such casesthey abandon their natural instincts, and at-tempt other strategies that better serve theirinterests.

REFERENCESAdolphs, R., D. Tanel, H. Damasio and A. Damasio. "Im-

paired Recognition of Emotion in Facial ExpressionsFol lowing Bilateral Damage to the HumanAmygdala." Nature, 15 December 1994, 669-72.

Allman, John and Leslie Brothers. "Faces, Fear and theAmygdala." Nature, 15 December 1994,613-14.

Axelrod, Robert. The Evolution of Cooperation. NewYork: Basic Books, 1984.

Axelrod, Robert and William D. Hamilton. "The Evolutionof Cooperation." Science, 211, 1981, 1390-96.

Baron-Cohen, Simon. Mindblindness An Essay on Autismand Theory of Mind. Cambridge, Mass.: MIT Press,1995.

HOFFMAN, MCCABE & SMITH: FOUNDATIONS OF RECIPROCITY 351

Berg, Joyce, John Dickhaut and Kevin McCabe. "Trust,Reciprocity and Social History." Games and Eco-nomic Behavior, 10(1), 1995, 122-42.

Bolton, Gary, Elena Kator and Rami Zwick. "DictatorGame Giving: Rules of Fairness versus Random Actsof Kindness." Working Paper, University of Pitts-burgh, 1993.

Brewer, Marilyn and William Crano. Social Psychology.St Paul, Minn.: West Publishing Co., 1994.

Camerer, Colin and Keith WeigelL "Experimental Tests ofa Sequential Equilibrium Reputation Model."Econometrica, January 1988, 1—36.

Carmichael, Lome and W. Bentley MacLeod. "Gift Givingand the Evolution of Cooperation." InternationalEconomic Review, 1997, forthcoming.

Cosmides, Leda. "The Logic of Social Exchange: HasNatural Selection Shaped How Humans Reason?Studies with the Wason Selection Task." Cognition,31(3), 1989,187-276.

Cosmides, Leda and John Tooby. "From Evolution to Be-havior: Evolutionary Psychology as the MissingLink," in The Latest and the Best: Essays on Evolu-tion and Optimality, edited by John Dupre. Cam-bridge, Mass.: MIT Press, 1987, 277-306.

. "Evolutionary Psychology and the Generation ofCulture, Part II." Ethology and Sociobiology, 10(1-3), 1989, 51-97.

, "Cognitive Adaptations for Social Exchange," inThe Adapted Mind, edited by Jerome Barkow, LedaCosmides and John Tooby. New York: Oxford Uni-versity Press, 1992.

Davis, Douglas D. and Charles A. Holt Experimental Eco-nomics. Princeton, N.J.: Princeton University Press,1993.

Eckel, Catherine and Philip Grossman. "Altruism in Anon-ymous Dictator Games." Games and Economic Be-havior, 16(2), 1996, 181-91.

Fehr, Ernst, George KirchsteigeT and Arno Riedl. "DoesFairness Prevent Market Clearing: An ExperimentalInvestigation." Quarterly Journal of Economics,May 1993,437-59.

Forsythe, Robert, Joel Horowitz, N. Eugene Savin andMartin Sefton. "Replicability, Fairness and Pay inExperiments with Simple Bargaining Games."Games and Economic Behavior, 6(3), 1994, 347-69.

Fudenberg, Drew and Jean Tirole. Game Theory. Cam-bridge, Mass.: MIT Press, 1993.

Halgren, Eric. "Emotional Neurophysiology of theAmygdala within the Context of Human Cognition,"in The Amygdala: Neurobiological Aspects of Emo-tion, Memory and Menial Dysfunction, edited by JohnAggleton. New York: Wiley-Liss, 1992.

Harrison, Glenn and Kevin McCabe. "Testing Non-cooperative Bargaining Theory in Experiments," inResearch in Experimental Economics, vol. 5., editedby R. Mark Isaac. Greenwich, Conn.: JAI Press,1992, 137-69.

Hoffman, Elizabeth, Kevin McCabe, Jason Shachat andVernon Smith. "Preferences, Property Rights and An-onymity in Bargaining Games." Games and Eco-nomic Behavior, 7(3), 1994, 346-80.

Hoffman, Elizabeth, Kevin McCabe and Vernon Smith."Social Distance and OtheT Regarding Behavior inDictator Games."/4meWca/j Economic Review, 86(3),1996a, 653-60.

, "Trust, Punishment, and Assurance: Experimentson the Evolution of Cooperation." Paper presented atthe Economic Science Association Annual Meeting,October, 1996b.

Homans, George C. The Nature of Social Sciences. NewYork: Harcourt, Brace and World, 1967.

Isaac, Glynn L. "The Food-sharing Behavior of Protohu-man Hominoids. Scientific American, 238, 1978, 9 0 -108.

Isaac, R. Mark, Kenneth F. McCue and Charles R. Plott"Public Goods Provision in an Experimental Envi-ronment." Journal of Public Economics, February1985,51-74.

Isaac, R. Mark, David Schmitz and James M. Walker. "TheAssurance Problem in a Laboratory Market." PublicChoice, September 1989, 217-36.

Isaac, R. Mark and James M. Walker. "Group Size Effectsin Public Goods Provision: The Voluntary Contribu-tions Mechanism." Quarterly Journal of Economics,February 1988a, 179-200.

. "Communication and Free-Riding Behavior TheVoluntary Contributions Mechanism." Economic In-quiry, October 1988b, 585-608.

. "Costly Communication: An Experiment in aNested Public Goods Problem," in ContemporaryLaboratory Research in Political Economy, edited byThomas Palfrey. Ann Arbor University of MichiganPress, 1991,269-86.

Isaac, R. Mark, James M. Walker and Susan H. Thomas."Divergent Evidence on Free Riding: An Experimen-tal Examination of Possible Explanations." PublicChoice, 43(2), 1984, 113-49.

Isaac, R. Mark, James M. Walker and Arlington Williams."Group Size and the Voluntary Provision of PublicGoods: Experimental Evidence Utilizing LargeGroups." Indiana University Working Paper, 1991.

Kahneman, Daniel, Jack Knetsch and Richard Thaler."Fairness and the Assumptions of Economics." Jour-nal of Business, October 1986, S285-S300.

Kaplin, Hillary and Kim Hill. "Food Sharing Among AcheForagers: Test of Explanatory Hypotheses." CurrentAnthropology, March 1985, 223-46.

Kreps, David, Paul Milgrom, John Roberts and RobertWilson. "Rational Cooperation in the Finitely Re-peated Prisoners' Dilemma." Journal of EconomicTheory, 27(2), 1982, 245-52.

Marr, David. Vision: A Computational Investigation intothe Human Representation and Processing of VisualInformation. San Francisco: Freeman, 1982.

McCabe, Kevin. "Fiat Money as a Store of Value in anExperimental Market." Journal of EconomicBehaviorts and Organization, October 1989, 215-31.

McCabe, Kevin, Stephen Rassenti and Vernon Smith."Game Theory and Reciprocity in Some ExtensiveForm Bargaining Games." Proceedings NationalAcademy of Science, November 1996a, 13421-28.

, "Reciprocity, Trust and Payoff Privacy in Exten-sive Form Bargaining." Manuscripts, Economic Sci-ence Laboratory, University of Arizona, November1996b.

McCabe, Kevin, Vernon Smith and Michael Lepore. "In-tentionality Signalling: Why Game Form Matters."Manuscripts, Economic Science Laboratory, Univer-sity of Arizona, November 1997.

352 ECONOMIC INQUIRY

Pinker, Steven. The Language Instinct. New York: WilliamMorrow, 1994.

Rapoport, A. "Prisoner's Dilemma," in The New Palgrave,vol. 3, edited by John Eatwell, Murray Milgate andPeter Newman. London: Macmillan, 1987, 973-76.

Rice, William R. "Sexually Antagonistic Male AdaptationTriggered by Experimental Arrest of Female Evolu-tion." Nature, 16 May 1996, 232-34.

Roth, Alvin, Vesna Prasniker, Masahiro Okuno-Fujimaraand Shmuel Zamir. "Bargaining and Market Behaviorin Jerusalem, Ljubligana, Pittsburgh and Tokyo: AnExperimental Study." American Economic Review,December 1991, 1068-95.

Schotter, Andrew, Keith Wiegelt and Charles Wilson. "ALaboratory Investigation of Multiperson Rationalityand Presentation Effects." Games and Economic Be-havior, May 1994, 445-68.

Selten, Reinhard. "Reexamination of the Perfectness Con-cept for Equilibrium Points in Extensive Games."International Journal of Game Theory, 4(1), 1975,25-55.

Selten, Reinhard and Rolf Stoecker. "End Behavior inSequences of Finite Prisoner's Dilemma Super-games." Journal of Economic Behavior and Organi-zation, March 1986,47-70.

Smith, Vemon L., Gerry L. Suchanek and Arlington W.Williams. "Bubbles, Crashes and Endogenous Expec-tations in Experimental Spot Asset Markets."Econometrica, September 1988, 1119-51.

Tooby, John and I. De Vore. "The Reconstruction of Hom-inoid Behavioral Evolution through Strategic Model-ling," in Primate Models of Human Behavior, editedby Waren G. Kinzey. New York; SUNY Press, 1987,183-237.

Trivers Robert L. "The Evolution of Reciprocal Altruism."Quarterly Review of Biology, 46(4), 1971, 35-57.

Wason, Peter. "Reasoning," in New Horizons in Psychol-ogy, edited by Brian M. Foss. Harmondsworth: Pen-guin, 1966, 135-51.