The Evolution of Reciprocal Altruism Author(s): Robert L. Trivers Reviewed work(s): Source: The Quarterly Review of Biology, Vol. 46, No. 1 (Mar., 1971), pp. 35-57 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/2822435 . Accessed: 20/10/2012 18:02

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

.

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to The Quarterly Review of Biology.

http://www.jstor.org THE EVOLUTION OF RECIPROCAL ALTRUISM

BY ROBERT L. TRIVERS

BiologicalLaboratories, Harvard University, Cambridge,Mass. 02138

ABSTRACT

A model is presented to account for the natural selection of what is termed recipro- cally altruistic behavior. The model shows how selection can operate -against the cheater (non-reciprocator)in the system. Three instances of altruistic behavior are discussed, the evolution of which the model can explain: (1) behavior involved in cleaning symbioses; (2) warning cries in birds: and (3) human reciprocal altruism. Regarding human reciprocal altruism, it is shown that the details of the psycho- logical system that regulates this altruism can be explained by the model. Spe- cifically,friendship, dislike, moralisticaggression, gratitude, sympathy, trust, suspicion, trustworthiness,aspects of guilt, and some forms of dishonestyand hypocrisycan be explained as important adaptations to regulate the altruistic system. Each individual human is seen as possessing altruistic and cheating tendencies, the expression of which is sensitive to developmental variables that were selected to set the tendencies at a balance appropriate to the local social and ecological environment.

INTRODUCTION own perpetuation,regardless of which individ- ual the genes appear in. The term "kin selec- LTRUISTIC behavior can be de- A tion" will be used in this paper to cover in- finedas behavior that benefitsan- stancesof this type-that is, of organismsbeing otherorganism, not closelyrelated, selected to help theirrelatively close kin. while being apparently detrimen- The model presented here is designed to tal to the organism performing show how certain classes of behavior con- the behavior, benefitand detrimentbeing de- venientlydenoted as "altruistic" (or "recipro- fined in termsof contributionto inclusive fit- cally altruistic")can be selected for even when ness. One human being leaping into water, at the recipient is so distantlyrelated to the or- some danger to himself, to save another dis- ganism performingthe altruisticact that kin tantly related human from drowning may be selection can be ruled out. The model will said to display altruisticbehavior. If he were apply, for example, to altruisticbehavior be- to leap in to save his own child, the behavior tween membersof differentspecies. It will be would not necessarilybe an instance of "al- argued that under certain conditions natural -truism";he may merelybe contributingto the selection favors these altruistic behaviors be- survivalof his own genes investedin the child. cause in the long run theybenefit the organism Models that attemptto explain altruisticbe- performingthem. havior in termsof natural selectionare models designed to take the altruismout of altruism. THE MODEL For example, Hamilton (1964) has demon- strated that degree of relationship is an im- One human being saving another,who is not portant parameterin predictinghow selection closely related and is about to drown, is an will operate, and behavior which appears instance of altruism. Assume that the chance altruistic may, on knowledge of the genetic of the drowningman dying is one-half if no relationships of the organisms involved, be one leaps in to save him, but that the chance explicable in termsof natural selection: those that his potentialrescuer will drown if he leaps genes being selectedfor that contributeto their in to save him is much smaller, say, one in 35 36 TH E QUARTERLY REVIEW OF BIOLOGY [VOLUME 46

twenty. Assume that the drowningman always This argumentcan be made precise. Assume drowns when his rescuer does and that he is there are both altruistsand non-altruistsin a always saved when the rescuersurvives the res- population of size N and that the altruistsare cue attempt. Also assume that the energycosts characterizedby the fact that each performs involvedin rescuingare trivialcompared to the altruisticacts when the cost to the altruist is survival probabilities. Were this an isolated well below the benefitto the recipient,where event, it is clear that the rescuer should not cost is defined as the degree to which the be- bother to save the drowningman. But if the havior retardsthe reproductionof the genes of drowningman reciprocatesat some futuretime, the altruistand benefitis the degree to which and if the survival chances are then exactly the behavior increasesthe rate of reproduction reversed,it will have been to the benefitof each of the genes of the recipient. Assume that the participantto have riskedhis life for the other. altruisticbehavior of an altruist is controlled Each participant will have traded a one-half by an allele (dominant or recessive),a2, at a chance of dying for about a one-tenthchance. given locus and that (for simplicity)there is If we assume that the entire population is only one alternativeallele, a1, at that locus and sooner or later exposed to the same risk of that it does not lead to altruistic behavior. drowning,the two individuals who risk their Consider three possibilities: (1) the altruists lives to save each other will be selected over dispense their altruism randomly throughout those who face drowningon their own. Note the population; (2) they dispense it nonran- that the benefitsof reciprocitydepend on the domly by regarding their degree of genetic unequal cost/benefitratio of the altruisticact, relationshipwith possible recipients;or (3) they that is, the benefitof the altruisticact to the dispense it nonrandomlyby regarding the al- recipient is greaterthan the cost of the act to truistictendencies of possible recipients. the performer,cost and benefitbeing defined here as the increase or decrease in chances of (1) Random dispensationof altruism the relevant alleles propagating themselvesin There are three possible genotypes: a1a1, the population. Note also that,as defined,the a2al, and a2a2' Each allele of the heterozygote benefitsand costs depend on the age of the will be affectedequally by whatevercosts and altruist and recipient (see Age-dependent benefitsare associatedwith the altruismof such changes below). (The odds assigned above may individuals (if a2 is dominant) and by whatever not be unrealistic if the drowning man is benefitsaccrue to such individuals from the drowningbecause of a cramp or if the rescue altruismof others,so they can be disregarded. can be executed by extending a branch from If altruisticacts are being dispensed randomly shore.) throughouta large population, then the typi- Why should the rescued individual bother to cal a1ai individual benefitsby (1/N)lb1, where reciprocate? Selection would seem to favor be- bi is the benefit of the ith altruisticact per- ing saved fromdrowning without endangering formedby the altruist. The typical a2a2 indi- oneself by reciprocating. Why not cheat? vidual has a net benefitof (l/N)Jbi - (I/N)Icj, this paper ("Cheating" is used throughout wherec; is the cost to the a2a2 altruistof his jth solely for convenience to denote failure to altruistic act. Since -(1 /N)Icj is always less reciprocate;no conscious intent or moral con- than zero, allele a1 will everywherereplace notationis implied.) Selectionwill discriminate allele a2. against the cheaterif cheatinghas later adverse affectson his life which outweighthe benefitof (2) Nonrandom dispensationby reference not reciprocating.This may happen if the al- to kin truistresponds to the cheatingby curtailingall future possible altruisticgestures to this indi- This case has been treated in detail by vidual. Assumingthat the benefitsof these lost Hamilton (1964), who concluded that if the altruistic acts outweigh the costs involved in tendency to dispense altruism to close kin is reciprocating, the cheater will be selected great enough, as a function of the disparity against relative to individuals who, because between the average cost and benefit of an neither cheats, exchange many altruistic acts. altruistic act, then a2 will replace a1. Tech- MARCH 1971] RECIPROCAL ALTRUISM 37

nically,all that is needed for Hamilton's form question sometimesor oftendirectly benefits its of selection to operate is that an individual duplicate in anotherorganism. with an "altruisticallele" be able to distinguish If an "altruisticsituation" is definedas any between individuals with and without this al- in which one individual can dispense a benefit lele and discriminateaccordingly. No formal to a second greaterthan the cost of the act to analysishas been attemptedof the possibilities himself,then the chances of selecting for al- for selection favoringindividuals who increase truisticbehavior, that is, of keeping jcj+?bm their chances of receivingaltruistic acts by ap- small,are greatest(1) when thereare manysuch pearing as if they were close kin of altruists, altruisticsituations in the lifetime of the al- although selection has clearly sometimes fa- truists,(2) when a given altruist repeatedly vored such parasitism (e.g., Drury and Smith, interactswith the same small set of individuals, 1968). and (3) when pairs of altruistsare exposed "sym- metrically"to altruisticsituations, that is, in (3) Nonrandom dispensation by referenceto such a way that the two are able to render the altruistic tendencies of the recipient roughly equivalent benefits to each other at What is required is that the net benefitac- roughly equivalent costs. These three condi- cruing to a typicala2a2 altruistexceed that ac- tions can be elaborated into a set of relevant cruing to an alai non-altruist,or that biological parameters affectingthe possibility that reciprocally altruistic behavior will be - (I /p2) (lb k Cj) > (I /q2)lbm, selected for. wherebk is the benefitto the a2a2 altruist (1) Length of lifetime.Long lifetimeof indi- of the kth altruistic act performed toward viduals of a species maximizesthe chance that him, where ci is the cost of the jth altruistic any two individuals will encounter many al- act by the a2a2 altruist,where bm is the benefit truisticsituations, and all other things being of the mth altruistic act to the alai nonal- equal one should searchfor instances of recipro- truist,and where p is the frequency in the cal altruismin long-livedspecies. population of the a2 allele and q that of the a, (2) Dispersal rate. Low dispersalrate during allele. This will tend to occur if Ibm is kept all or a significantportion of the lifetimeof small (which will simultaneouslyreduce Ici). individuals of a species increases the chance And this in turn will tend to occur if an al- that an individual will interactrepeatedly with truistresponds to a "nonaltruisticact" (that is, the same set of neighbors,and other thingsbe- a failure to act altruisticallytoward the altruist ing equal one should search for instances of in a situationin which so doing would cost the reciprocalaltruism in such species. Mayr (1963) actor less than it would benefitthe recipient) has discussedsome of the factorsthat may affect by curtailing future altruisticacts to the non- dispersalrates. altruist. (3) Degree of mutual dependence. Interde- Note that the above form of altruism does pendence of members of a species (to avoid not depend on all altruistic acts being con- predators,for example) will tend to keep indi- trolled by the same allele at the same locus. viduals near each other and thus increase the Each altruistcould be motivatedby a different chance theywill encounter altruisticsituations allele at a differentlocus. All altruisticalleles together. If the benefit of the mutual de- would tend to be favored as long as, for each pendence is greatestwhen only a small number allele, the net average benefitto the homozy- of individuals are together,this will greatly gous altruistexceeded the average benefitto the increase the chance that an individual will re- homozygousnonaltruist; this would tend to be peatedly interact with the same small set of true if altruistsrestrict their altruismto fellow individuals. Individuals in primate troops,for altruists,regardless of what allele motivatesthe example, are mutually dependent for protec- other individual's altruism. The argumentwill tion frompredation, yet the optimal troop size thereforeapply, unlike Hamilton's (1964), to for foragingis often small (Crook, 1969). Be- altruisticacts exchanged between members of cause they also meet the other conditions out- differentspecies. It is the exchange that favors lined here, primatesare almost ideal species in such altriusm,not the fact that the allele in which to search for reciprocalaltruism. Clean- 38 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME46

ing symbiosesprovide an instance of mutual to that of the modern chimpanzeethan to that dependence between membersof differentspe- of the modern baboon (see, for example, Rey- cies, and this mutual dependence appears to nolds, 1966). have set the stage for the evolution of several (6) Aid in combat. No matter how domi- altruisticbehaviors discussed below. nance-orienteda species is, a dominant indi- (4) Parental care. A special instance of mu- vidual can usually be aided in aggressiveen- tual dependence is that found between parents counterswith other individuals by help froma and offspringin species thatshow parental care. less dominant individual. Hall and DeVore The relationshipis usually so asymmetricalthat (1965) have described the tendencyfor baboon few or no situationsarise in which an offspring alliances to formwhich fightas a unit in ag- is capable of performingan altruisticact forthe gressive encounters (and in encounters with parents or even for another offspring,but this predators). Similarly,vervet monkeys in ag- is not entirelytrue for some species (such as gressive encounters solicit the aid of other, primates)in which the period of parental care often less dominant, individuals (Struhsaker, is unusuallylong. Parental care, of course,is to 1967). Aid in combat is then a special case in be explained by Hamilton's (1964) model, but which relativelysymmetrical relations are pos- there is no reason why selection for reciprocal sible between individuals who differin domi- altruismcannot operate between close kin, and nance. evidence is presentedbelow that such selection The above discussionis meant only to suggest has operatedin humans. the broad conditions that favor the evolu- (5) Dominance hierarchy.Linear dominance tion of reciprocal altruism. The most impor- hierarchiesconsist by definitionof asymmetri- tant parametersto specifyfor individuals of a cal relationships; a given individual is domi- species are how many altruisticsituations occur nant over another but not vice versa. Strong and how symmetricalthey are, and theseare the dominance hierarchies reduce the extent to most difficultto specify in advance. Of the which altruisticsituations occur in which the three instancesof reciprocal altruismdiscussed less dominant individual is capable of perform- in this paper only one, human altruism,would ing a benefitfor the more dominantwhich the have been predicted from the above broad more dominant individual could not simply conditions. take at will. Baboons (Papio cynocephalus)pro- The relationship between two individuals vide an illustrationof this. Hall and DeVore repeatedly exposed to symmetricalreciprocal (1965) have described the tendency for meat situations is exactly analogous to what game caught by an individual in the troop to end up theoristscall the Prisoner'sDilemma (Luce and by preemptionin the hands of the most domi- Raiffa, 1957; Rapoport and Chammah, 1965), nant males. This ability to preempt removes a game that can be characterizedby the payoff any selectiveadvantage that food-sharingmight matrix otherwisehave as a reciprocal gesture for the most dominantmales, and thereis no evidence A2 C2 in this species of any food-sharingtendencies. Al R, R S, T By contrast, Van Lawick-Goodall (1968) has C1 T, S P, P shown that in the less dominance-oriented chimpanzeesmore dominant individuals often where S < P < R < T and where A1 and A2 do not preempt food caught by the less domi- representthe altruisticchoices possible for the nant. Instead, they besiege the less dominant two individuals, and C1 and C2, the cheating individual with "begging gestures,"which re- choices (the firstletter in each box gives the sult in the handing over of small portions of payofffor the firstindividual, the second letter the catch. No strongevidence is available that the payoff for the second individual). The this is part of a reciprocallyaltruistic system, other symbolscan be given the followingmean- but the absence of a stronglinear dominance ings: R stands for the reward each individual hierarchyhas clearly facilitated such a possi- gets from an altruistic exchange if neither bility. It is very likely that early hominid cheats; T stands for the temptationto cheat; groups had a dominance systemmore similar S stands for the sucker'spayoff that an altruist MARCH 1971] RECIPROCAL ALTRUISAM 39

gets when cheated; and P is the punishment for it means that only under highlyspecialized that both individuals get when neither is al- circumstancescan the altruist be reasonably truistic (adapted from Rapoport and Cham- guaranteed that the causal chain he initiates mah, 1965). Iterated games played between the with his altruisticact will eventuallyreturn to same two individuals mimic real life in that him and confer,directly or indirectly,its bene- they permit each player to respond to the be- fit. Only under these conditions will the havior of the other. Rapoport and Chammah cheater be selected against and this type of (1965) and othershave conducted such experi- altruisticbehavior evolve. ments using human players,and some of their Althoughthe preconditionsfor the evolution results are reviewedbelow in the discussionof of reciprocal altruism are specialized, many human altruism. species probably meet them and display this W. D. Hamilton (pers. commun.) has shown type of altruism. This paper will limit itself, that the above treatmentof reciprocalaltruism however, to three instances. The first,be- can be reformulatedconcisely in termsof game havior involvedin cleaning symbioses,is chosen theoryas follows. Assuming two altruistsare because it permits a clear discriminationbe- symmetricallyexposed to a series of reciprocal tween this model and that based on kin selec- situations with identical costs and identical tion (Hamilton, 1964). The second, warning benefits, then after 2n reciprocal situations, calls in birds, has already been elaborately each has been "paid" nR. Were one of the two analyzed in terms of kin selection; it is dis- a nonaltruistand the second changed to a non- cussed here to show how the model presented altruisticpolicy after firstbeing cheated, then above leads to a very differentinterpretation the initial altruistwould be paid S + (n - I)P of these familiar behaviors. Finally, human (assuming he had the firstopportunity to be reciprocal altruism is discussed in detail be- altruistic) and the non-altruistwould receive cause it representsthe best documentedcase of T + (n - I)P. The important point here is reciprocal altruism known, because there has that unless T >> R, then even with small n, apparently been strong selection for a very nR should exceed T + (n - I)P. If this holds, complex systemregulating altruisticbehavior, the nonaltruistictype, when rare, cannot start and because the above model permitsthe func- to spread. But there is also a barrier to the tional interpretationof details of the system spread of altruismwhen altruistsare rare, for that otherwiseremain obscure. P > S implies nP > S + (n-1)P. As n in-

creases,these two total payoffstend to equality, ALTRUISTIC BEHAVIOR IN CLEANING SYMBIOSES so the barrierto the spread of altruismis weak if n is large. The barrierwill be overcome if The preconditionsfor the evolution of re- the advantages gained by exchanges between ciprocal altruism are similar to those for the altruists outweigh the initial losses to non- operation of kin selection: long lifetime,low altruistictypes. dispersal rate, and mutual dependence, for ex- Reciprocal altruismcan also be viewed as a ample, tend to increase the chance that one is symbiosis,each partnerhelping the other while interactingwith one's close kin. This makes it he helps himself. The symbiosishas a time lag, difficultto discriminatethe two alternativehy- however;one partnerhelps the other and must potheses. The case of cleaning symbiosis is then wait a period of time before he is helped importantto analyze in detail because altruistic in turn. The returnbenefit may come directly, behavior is displayed that cannot be explained as in human food-sharing,the partner directly by kin selection,since it is performedby mem- returningthe benefitafter a time lag. Or the bers of one species for the benefitof members returnmay come indirectly,as in warningcalls of another. It will be shown instead that the in birds (discussed below), where the initial behavior can be explained by the model pre- help to other birds (the warning call) sets up sented above. No elaborate explanation is a causal chain through the ecological system needed to understand the evolution of the (the predatorfails to learn useful information) mutually advantageous cleaning symbiosis it- which redounds aftera time lag to the benefit self; it is several additional behaviorsdisplayed of the caller. The time lag is the crucial factor, by the host fish to its cleaner that require a 40 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME 46

special explanation because they meet the Soonwe watchedour secondsurprise: the grouper criteriafor altruistic behavior outlined above- made a movementwhich in the precedingsix thatis, theybenefit the cleanerwhile apparently yearswe had neverseen him make: he spreadthe beingdetrimental to thehost. rightgill-covering so wide that the individual gill-plateswere separatedfrom Feder (1966) and Maynard (1968) have re- each other at greatdistances, wide enough to let the cleaner centlyreviewed the literatureon cleaning sym- through(translated from Hediger, 1968 p. 93). bioses in the ocean. Briefly,one organism(e.g., the wrasse, Labroides dimidiatus) cleans an- When Hediger added two additional L. di- other organism(e.g., the grouper,Epinephelus midiatus to the tank, all three cleaned the striatus) of ectoparasites (e.g., caligoid cope- grouper with the result that within several pods), sometimesentering into the gill cham- days the grouperappeared restlessand nervous, bers and mouthof the "host" in order to do so. searchedout places in the tank he had formerly Over forty-fivespecies of fishare known to be avoided, and shook himselfoften (as a signal cleaners,as well as six species of shrimp. In- that he did not wish to be cleaned any longer). numerablespecies of fishserve as hosts. Stom- Apparently three cleaners working over him ach analyses of cleaner fish demonstratethat constantlywas too much for him, yet he still they vary greatlyin the extent to which they failed to eat any of them. When Hediger re- depend on theircleaning habits for food,some moved two of the cleaners, the grouper re- apparentlysubsisting nearly entirelyon a diet turned to normal. There is no indication the of ectoparasites.Likewise, stomachanalyses of grouperever possessedany edible ectoparasites, host fishreveal that cleaners differin the rate and almost two yearslater (in December, 1968) at which theyend up in the stomachsof their the same cleaner continued to "clean" the hosts, some being apparently almost entirely grouper (pers. observ.) although the cleaner immuneto such a fate. It is a strikingfact that was, in fact, fed separatelyby its zoo-keepers. thereseems to be a strongcorrelation between Eibl-Eibesfeldt(1959) has described the mor- degree of dependence on the cleaning way of phology and behavior of two species (e.g., As- life and immunityto predation by hosts. pidontutstaeniatus) that mimic cleaners (e.g., Cleaning habits have apparentlyevolved in- L. dimidiatus) and that rely on the passive dependentlymany times(at least threetimes in behavior of fishwhich suppose they are about shrimps alone), yet some remarkable conver- to be cleaned to dart in and bite offa chunk gence has taken place. Cleaners, whether of their fins. I cite the evolution of these shrimp or fish, are distinctivelycolored and mimics, which resemble their models in ap- behave in distinctiveways (for example, the pearance and initial swimming behavior, as wrasse,L. dimidiatus,swims up to its host with evidence of strongselection for hosts with no a curious dipping and rising motion that re- intentionof harmingtheir cleaners. minds one of the way a finchflies). These dis- Of especial interestis evidence that therehas tinctivefeatures seem to serve the functionof been strongselection not to eat one's cleaner attractingfish to be cleaned and of inhibiting even afterthe cleaning is over. Eibl-Eibesfeldt any tendencyin themto feed on theircleaners. (1955) has made some strikingobservations on There has apparentlybeen strongselection to the goby,Elacitinus oceanops: avoid eating one's cleaner. This can be il- lustrated by several observations. Hediger I neversawv a groupersnap up a fishafter it had (1968) raised a grouper (Epinephelus) from cleaned it. On the contrary,it annoouncedits impendingdeparture by twodefinite signal move- infancyalone in a small tank for six years,by ments. Firstit closed its mouthvigorously, al- which time the fish was almost four feet in thoughnot completely,and immediatelyopened length and accustomed to snapping up any- it wide again. Upon this intentionmovement, thing dropped into its tank. Hediger then all the gobiesleft the mouthcavity. Then the dropped a small live cleaner (L. dimidiatus) groupershook its bodylaterally a fewtines, and into the grouper'stank. The groupernot onily all the cleaniersreturned to theircoral. If onie failed to snap up the cleaner but opened its frighteneda grouperit never neglectedthese mouthand permittedthe cleaner freeentry and forewarningmovements (translated from Eibl- exit. Eibesfeldt,1955, p. 208). MARCH 1971] RECIPROCAL ALTRUISM 41

Randall has made similar observationson a parasites may be sufficientto inhibit hunger. moray eel (Gymnothoraxjaponicus) that sig- Both possibilitiesare contradictedby Hediger's nalled with a "sharp lateral jerk of the eel's observation,cited above, and seem unlikelyon head," after which "the wrasse fairlyflew out functionalgrounds as well. of the mouth, and the awesome jaws snapped A fishto be cleaned seems to performseveral shut" (Randall, 1958, 1962). Likewise, Hedi- "altruistic" acts. It desists from eating the ger's Kasper Hauser grouper shook its body cleaner even when it easily could do so and when it had enough of being cleaned. when it must go to special pains (sometimes Why does a large fishnot signal the end to a at danger to itself)to avoid doing so. Further- cleaning episode by swallowing the cleaner? more, it may performtwo additional behaviors Natural selection would seem to favor the which seem of no direct benefitto itself (and double benefitof a good cleaning followed by which consume energyand take time), namely, a meal of the cleaner. Selection also operates, it signals its cleaner that it is about to depart of course,on the cleaner and presumablyfavors even when the fishis not in its mouth, and it mechanismsto avoid being eaten. The distinc- may chase offpossible dangers to the cleaner: tive behavior and appearance of cleaners has While diving with me in the Virgin Islands, been cited as evidence of such selection. One Robert Schroederwatched a Spanish hogfish can also cite the distinctivebehavior of the fish groominga bar jack in its bronze color state. being cleaned. Feder (1966) has pointed out When a second jack arrivedin the pale color that hostsapproaching a cleaner react by "stop- phase, the firstjack immediatelydrove it away. ping or slowing down, allowing themselvesto But later when anotherjack intrudedon the assume awkwardpositions, seemingly in a hyp- sceneand changedits pale color to dark bronze notic state." Fishes sometimesalter their color it was not chased. The bronzecolor would seem harm intended;I need service" dramaticallybefore and while being cleaned, to mean "no (Randall,1962 p. 44). and Feder (1966) has summarizedinstances of this. These forms of behavior suggest that The behavior of the host fishis interpreted natural selection has operated on cleaners to here to have resulted from natural selection avoid attemptingto clean fish without these and to be, in fact,beneficial to the host because behaviors,presumably to avoid wasting energy the cleaner is worthmore to it alive than dead. and to minimize the dangers of being eaten. This is because the fishthat is cleaned "plans" (Alternatively,the behaviors, including color to returnat later dates formore cleanings,and change, may aid the cleaners in findingecto- it will be benefitedby being able to deal with parasites. This is certainly possible but not, the same individual. If it eats the cleaner, it I believe, adequate to explain the phenomenon may have difficultyfinding a second when it completely. See, for example, Randall, 1962.) needs to be cleaned again. It may lose valuable Once the fishto be cleaned takes the proper energy and be exposed to unnecessarypreda- stance, however,the cleaner goes to work with tion in the search for a new cleaner. And it no apparent concernfor its safety: it makes no may in the end be "turned down" by a new effortto avoid the dangerous mouth and may cleaner or serviced very poorly. In short, the even swim inside,which as we have seen, seems host is abundantly repaid for the cost of its particularlyfoolhardy, since fishbeing cleaned altruism. may suddenly need to depart. The apparent To support the hypothesisthat the host is unconcern of the cleaner suggeststhat natural repaid its initial altruism,several pieces of evi- selection acting on the fishbeing cleaned does dence mustbe presented: that hostssuffer from not, in fact, favor eating one's cleaner. No ectoparasites;that findinga new cleaner may speculation has been advanced as to why this be difficultor dangerous; that if one does not may be so, although some speculation has ap- eat one's cleaner,the same cleaner can be found peared about the mechanismsinvolved. Feder and used a second time (e.g., that cleaners are advances two possibilities, that of Eibl- site-specific);that cleaners live long enough to Eibesfeldt (1955) that fishcome to be cleaned be used repeatedly by the same host; and if only aftertheir appetite has been satisfied,and possible, that individual hosts do, in fact,reuse one of his own, that the irritation of ecto- the same cleaner. 42 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME 46

(1) The cost of ectoparasites. It seems al- Cleaning fishesand cleaning shrimpshave regular most axiomatic that the evolution of cleaners stations to which fishes wanting to be cleaned entirely dependent on ectoparasites for food can come (p. 367). implies the selective disadvantage for the Limbaugh, Pederson, and Chase (1961) have cleaned of being ectoparasite-ridden.What is reviewed available data on the six species of perhaps surprisingis the effectthat removing cleaner shrimps, and say: all cleaners froma coral reef has on the local "hosts" (Limbaugh, 1961). A, 'animal . . . and is not tend repeatedlyto return to familiar cleaners. highly dependent upon its host for survival. So The only observationof fishbeing disappointed far as is known, it does not display itself to attractfishes (p. 238). in their search for cleaners comes from Eibl- Eibesfeldt (1955): "If the cleaners fail to ap- It is H. californica that is occasionally found in pear over one coral in about half a minute,the the stomachs of at least one of its hosts. The large fishes swim to another coral and wait striking correlation of territoriality and soli- there a while" (translatedfrom p. 210). It may tariness with cleaning habits is what theory be that fish have several alternative cleaning would predict. The same correlation can be stationsto go to, since any particular cleaning found in cleaner fish. Labroides, with four station may be occupied or unattended at a species, is the genus most completely dependent given moment. So manyfish tend to be cleaned on cleaning habits. No Labroides has ever been at coral reefs (Limbaugh, 1961, observed a found in the stomach of a host fish. All species cleaner service300 fishin a 6-hourperiod), that are highly site-specific and tend to be solitary. predatorsprobably frequent coral reefsin search Randall (1958) reports that an individual swim as much as of fishbeing cleaned. Limbaugh (1961) suggested L. dimidiatus may sometimes cleaning station, servicing fish that good human fishingsites are found near 60 feet from its the But he cleaning stations. One finalreason whycoming on way. notes, to be cleaned maybe dangerousis thatsome fish This was especially true in an area where the must leave their element to do so (Randall, highly territorialdamsel fish Pomacentris nigri- 1962): cans (Lepede) was common. As one damsel fish was being tended, another nearby would assume Most impressive were the visits of moray eels, a stationarypose with finserect and the Labroides which do not ordinarily leave their holes in the would move oni to the latter with little hesita- reef during daylight hours, and of the big jacks tion (p. 333). which swam up from deeper water to the reef's edge to be "serviced" before going on their way Clearly, what matters for the evolution of re- (p. 43). ciprocal altruism is that the same two indi- (3) Site specificityof cleaners. Feder (1966) viduals interact repeatedly. This will be facili- has reviewed the strikingevidence for the site tated by the site specificityof either individual. specificityof cleaners and concludes: Of temperate water cleaners, the species most MARCH 1971] RECIPROCAL ALTRUISM 43

specialized to cleaning is also apparently the of cleaning organismsperform several kinds of most solitary(Hobson, 1969). altruisticbehavior, including not eating their (4) Lifespan of cleaners. No good data exist cleaner after a cleaning, which can be ex- on how long cleaners live, but several observa- plained on the basis of the above model. A tions on both fishand shrimpsuggest that they review of the relevant evidence suggeststhat easily live long enough for effectiveselection the cleaner organismsand theirhosts meet the against cheaters. Randall (1958) repeatedly preconditionsfor the evolution of reciprocally checked several ledges and found that different altruisticbehavior. The host's altruismis to be feedingstations were occupied for"long periods explained as benefitinghim because of the of time," apparently by the same individuals. advantage of being able quickly and repeatedly One such feeding station supported two indi- to return to the same cleaner. viduals for over three years. Of one species of WARNING CALLS IN BIRDS cleaner shrimp,Stenopus hispidus, Limbaugh, Pederson, and Chase (1961) said that pairs of Marler (1955, 1957) has presented evidence individuals probably remain months, possibly that warningcalls in birds tend to have charac- years,within an area of a square meter. teristicsthat limit the informationa predator (5) Hosts using the same cleaner repeatedly. gets fromthe call. In particular,the call char- There is surprisinglygood evidence that hosts acteristicsdo not allow the predator easily to reuse the same cleaner repeatedly. Feder (1966) determinethe location of the call-giver.Thus, summarizesthe evidence: it seems that giving a warningcall must result, at least occasionally,in the otherwiseunneces- Many fishesspend as muchtime getting cleaned sary death of the call-giver,either at the hands as theydo foragingfor food. Some fishesreturn of the predator that inspired the call or at the again and again to the same station, and showa hands of a second predatorformerly unaware of definitetime pattern in their daily arrival. exact location. Otherspass fromstation to stationand return the caller'spresence or manytimes during the day; this is particularly Given the presumed selection against call- trueof an injuredor infectedfish (p. 368). giving, Williams (1966) has reviewed various models to explain selection for warning cries: Limbaugh, Pederson, and Chase (1961) have (1) Warning calls are functional during the presentedevidence that in at least one species breeding season in birds in that they protect of cleaner shrimp (Stenopus scutellus), the one's mate and offspring.They have no func- shrimp may reservice the same individuals: tion outside the breeding season, but they are One pair was observedin the same football-sized not deleted then because "in practice it is not coral boulder fromMay throughAugust 1956. worth burdening the germ plasm with the in- Duringthat period, we changedthe positionand formationnecessary to realize such an adjust- orientationof the boulderseveral times within a ment" (Williams, 1966, p. 206). radius of approximatelyseven meterswithout (2) Warning calls are selected for by the disturbingthe shrimp.Visiting fishes were mo- mechanismof group selection(Wynne-Edwards, mentarilydisturbed by the changes,but they 1962). soon relocatedthe shrimps(p. 254). (3) Warning calls are functionaloutside the Randall (1958) has repeatedly observed fish breeding season because thereis usually a good swimmingfrom out of sight directlyto clean- chance that a reasonably close kin is near ing stations,behavior suggestingto him that enough to be helped sufficiently(Hamilton, they had prior acquaintance with the stations. 1964; Maynard Smith, 1964). Maynard Smith During two months of observationsat several (1965) has analyzed in great detail how closely feeding stations,Eibl-Eibesfeldt (1955) became related the benefitedkin mustbe, at what bene- personally familiar with several individual fit to him the call must be, and at what cost groupers (Epinephelus striatus)and repeatedly to the caller, in order for selection to favor observed them seeking out and being cleaned call-giving. at the same feedingstations, presumably by the The firstis an explanation of last resort. same cleaners. While it must sometimesapply in evolutionary In summary,it seemsfair to say that the hosts arguments,it should probablyonly be invoked 44 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME 46

when no otherexplanation seemsplausible. The (1963) has presentedevidence thatkin selection second is not consistentwith the known work- in some cryptic saturnid moths has favored ings of natural selection. The thirdis feasible rapid, post-reproductivedeath to minimize and may explain the warning calls in some predation on the young. Blest's evidence thus species and perhaps even in many. But it does providesan instanceof a predatorgaining use- depend on the somewhatregular nearby pres- ful informationthrough the act of predation. ence of closelyrelated organisms,a matterthat It does not matterthat in giving a warning may oftenbe the case but that has been demon- call the caller is helping its non-callingneigh- stratedonly as a possibilityin a few species and bors more than it is helping itself. What counts that seems very unlikely in some. A fourth is that it outcompetes conspecificsfrom areas explanation is suggestedby the above model: in which no one is giving warning calls. The (4) Warning calls are selected for because non-calling neighbors of the caller (or their theyaid the bird givingthe call. It is disadvan- offspring)will soon find themselvesin an area tageous for a bird to have a predator eat a withoutany caller and will be selected against nearby conspecificbecause the predator may relative to birds in an area with callers. The then be more likelyto eat him. This may hap- caller, by definition,is always in an area with pen because the predator will at least one caller. If we assume that two callers are preferableto one, and so on, then selection (i) be sustained by the meal, will favor the spread of the warning-callgenes. (ii) be more likelyto forma specificsearch Note that this model depends on the concept image of the prey species, of open groups, whereas "group selection" (iii) be more likely to learn the habits of (Wynne-Edwards,1962) depends partly on the the prey species and perfecthis preda- concept of closed groups. torytechniques on it, It mightbe supposed that one could explain (iv) be more likely to frequentthe area in bird calls more directlyas altruisticbehavior which the birds live, or that will be repaid when the other birds re- (v) be more likelyto learn usefulinforma- ciprocate, but there are numerous objections tion about the area in which the birds to this. It is difficultto visualizehow one would live. discover and discriminateagainst the cheater, In short,in one way or another,giving a warn- and there is certainlyno evidence that birds ing call tends to preventpredators from special- refrainfrom giving calls because neighborsare izing on the caller's species and locality. not reciprocating.Furthermore, if the relevant bird There is abundant evidence for the impor- groupings are very fluid,with much emi- tance of learning in the lives of predatoryver- gration and immigration,as they often are, tebrates (see, for example, Tinbergen, 1960; then cheating would seem to be favored and Leyhausen, 1965; Brower and Brower, 1965). no selectionagainst it possible. Instead, accord- Rudebeck (1950, 1951) has presentedimportant ing to the model above, it is the mere fact that observationson the tendencyof avian predators the neighborsurvives that repays the call-giver to specialize individually on prey types and his altruism. hunting techniques. Owen (1963) and others It is almost impossible to gather the sort of have presented evidence that species of snails evidence that would discriminatebetween this and insects may evolve polymorphismsas a explanation and that of Hamilton (1964). It protectionagainst the tendencyof their avian is difficultto imagine how one would estimate predatorsto learn their appearance. Similarly, the immediatecost of giving a warning call or Kuyton (1962; cited in Wickler, 1968) has de- its benefitto those within earshot,and precise scribedthe adaptation of a moth that minimizes data on the geneticrelationships of bird group- the chance of its predators forminga specific ings throughoutthe year are not only lacking search image. Southern (1954), Murie (1944), but would be most difficultto gather. Several and numerous others have documented the lines of evidence suggest,however, that Hamil- tendencyof predators to specialize on certain ton's (1964) explanation should be assumed localities within their range. Finally, Blest with caution: MARCH 1971] RECIPROCAL ALTRUISM 45

(1) There exist no data showing a decrease zee (Van Lawick-Goodall, 1968) and not of the in warning tendencies with decrease in more rigidly linear form characteristicof the the genetic relationshipof those within baboon (Hall and DeVore, 1965). Aid in intra- earshot. Indeed, a striking feature of specificcombat, particularly by kin, almost cer- warningcalls is that theyare given in and tainlyreduced the stabilityand linearityof the out of the breeding season, both before dominance order in early humans. Lee (1969) and after migration or dispersal. has shown that in almost all Bushman fights (2) There do exist data suggestingthat close which are initially between two individuals, kin in a number of species migrate or othershave joined in. Mortality,for example, disperse great distances from each other often strikes the secondaries rather than the (Ashmole, 1962; Perdeck, 1958; Berndt principals. Tool use has also probably had an and Sternberg,1968; Dhont and Huble, equalizing effecton human dominance rela- 1968). tions, and the Bushmen have a saying that (3) One can advance the theoretical argu- illustratesthis nicely. As a dispute reaches the ment that kin selection under some cir- stage where deadly weapons may be employed, cumstancesshould favor kin dispersal in an individual will oftendeclare: "We are none order to avoid competition (Hamilton, of us big, and otherssmall; we are all men and 1964, 1969). This would lead one to we can fight;I'm going to get my arrows,"(Lee, expect fewerclosely related kin near any 1969). It is interestingthat Van Lawick-Goodall given bird, outside the breeding season. (1968) has recorded an instance of strongdom- inance reversalin chimpanzeesas a functionof The argumentsadvanced in this section may tool use. An individual moved from low in also apply,of course,to speciesother than birds. dominance to the top of the dominance hier- archy when he discoveredthe intimidatingef- HUMAN RECIPROCAL ALTRUISM fects of throwinga metal tin around. It is likely that a diversityof talentsis usually pres- Reciprocal altruism in the human species ent in a band of hunter-gathererssuch that the takes place in a number of contextsand in all best makerof a certaintype of tool is not often known cultures (see, for example, Gouldner, the best maker of a differentsort or the best 1960). Any complete list of human altruism user of the tool. This contributesto the sym- would contain the followingtypes of altruistic metaryof relationships,since altruisticacts can behavior: be traded with referenceto the special talents (1) helping in timesof danger (e.g. accidents, of the individuals involved. predation, intraspecificaggression; To analyze the details of the human recipro- (2) sharing food; cal-altruisticsystem, several distinctionsare im- (3) helping the sick, the wounded, or the portant and are discussed here. veryyoung and old; (1) Kin selection. The human species also (4) sharing implements; and met the preconditionsfor the operation of kin (5) sharing knowledge. selection. Early hominid hunter-gathererbands almost certainly(like today's hunter-gatherers) All these forms of behavior often meet the consisted of many close kin, and kin selection criterionof small cost to the giver and great mustoften have operated to favorthe evolution benefitto the taker. of some types of altruisticbehavior (Haldane, During the Pleistocene,and probablybefore, 1955; Hamilton, 1964, 1969). In general, in a hominid species would have met the precon- attemptingto discriminatebetween the effects ditions for the evolution of reciprocalaltruism: of kin selection and what might be called re- long lifespan; low dispersal rate; life in small, ciprocal-altruisticselection, one can analyze the mutuallydependent, stable, social groups (Lee formof the altruisticbehaviors themselves. For and DeVore, 1968; Campbell, 1966); and a long example, the existenceof discriminationagainst period of parental care. It is very likely that non-reciprocalindividuals cannot be explained dominance relations were of the relaxed, less on the basis of kin selection,in which the ad- linear formcharacteristic of the living chimpan- vantage accruingto close kin is what makes the 46 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME 46

altruisticbehavior selectivelyadvantageous, not should show interestingdifferences, but the its chance of being reciprocated. The strongest analysis is complicatedby the possibilityof re- argumentfor the operation of reciprocal-altru- ciprocityto the kin of a deceased altruist(see isticselection in man is the psychologicalsystem Multi-partyinteractions below). controlling some forms of human altruism. (4) Gross and subtle cheating. Two formsof Details of thissystem are reviewedbelow. cheating can be distinguished,here denoted as (2) Reciprocal altruismamong close kin. If gross and subtle. In gross cheating the cheater both forms of selection have operated, one fails to reciprocateat all, and the altruistsuffers would expect some interestinginteractions. One the costsof whateveraltruism he has dispensed might expect, for example, a lowered demand without any compensating benefits. More forreciprocity from kin than fromnonkin, and broadly, gross cheating may be defined as re- thereis evidence to support this (e.g., Marshall, ciprocatingso little, if at all, that the altruist 1961; Balikci, 1964). The demand that kin receivesless benefitfrom the grosscheater than show some reciprocity(e.g., Marshall, 1961; the cost of the altruist'sacts of altruismto the Balikci, 1964) suggests,however, that reciprocal- cheater. That is, Icai> Nbaj where cai is the altruisticselection has acted even on relations cost of the ith altruisticact performedby the between close kin. Although interactionsbe- altruistand where bai is the benefitto the al- tween the two formsof selectionhave probably truistof the jth altruisticact performedby the been importantin human evolution,this paper gross cheater; altruisticsituations are assumed will limit itself to a preliminarydescription of to have occurred symmetrically.Clearly, selec- the human reciprocallyaltruistic system, a sys- tion will stronglyfavor prompt discrimination tem whose attributesare seen to result only against the gross cheater. Subtle cheating,by from reciprocal-altruisticselection. contrast,involves reciprocating,but always at- (3) Age-dependentchanges. Cost and benefit temptingto give less than one was given, or were defined above without reference to the more precisely,to give less than the partner ages, and hence reproductive values (Fisher, would give if the situation were reversed. In 1958), of the individuals involved in an altru- this situation,the altruiststill benefitsfrom the istic exchange. Since the reproductivevalue of relationship but not as much as he would if a sexually mature organismdeclines with age, the relationship were completely equitable. the benefitto him of a typicalaltruistic act also The subtle cheaterbenefits more than he would decreases,as does the cost to him of a typical if the relationship were equitable. In other act he performs.If the intervalseparating the words, two acts in an altruisticexchange is shortrela- I - - - tive to the lifespans of the individuals, then (bqi cqj) > t (bqi Cai) > i. (ba cai) the erroris slight. For longerintervals, in order where the ith altruisticact performedby the to be repaid precisely,the initial altruistmust altruisthas a cost to him of cal and a benefit receive more in return than he himselfgave. to the subtle cheater of bqi and where the jth It would be interestingto see whetherhumans altruisticact performedby the subtle cheater in fact routinelyexpect "interest"to be added has a cost to him of cq, and a benefit to the to a long overdue altruisticdebt, interestcom- altruist of baj. Because human altruism may mensuratewith the interveningdecline in re- span huge periods of time,a lifetimeeven, and productivevalue. In humansreproductive value because thousandsof exchangesmay take place, declines most steeply shortlyafter sexual ma- involving many different"goods" and with turityis reached (Hamilton, 1966), and one many differentcost/benefit ratios, the problem would predict the interest rate on altruistic of computingthe relevant totals,detecting im- debts to be highest then. Selection might also balances, and deciding whether they are due favorkeeping the intervalbetween act and re- to chance or to small-scalecheating is an ex- ciprocation short,but this should also be fa- tremelydifficult one. Even then, the altruistis vored to protect against complete non-recipro- in an awkwardposition, symbolized by the folk cation. W. D. Hamilton (pers.commun.) has sug- saying,"half a loaf is better than none," for if gested that a detailed alialysisof age-dependent attempts to make the relationship equitable changes in kin altruismand reciprocalaltruism lead to the ruptureof the relationship,the al- MARCH 1971] RECIPROCAL ALTRUISM 47

truist,assuming other thingsto be equal, will of this mutual dependence are a positive func- sufferthe loss of the substandard altruism tion of group size and that altruisticbehaviors of the subtle cheater. It is the subtletyof the may be selected for because they permit addi- discriminationnecessary to detect this formof tional individuals to survive and therebycon- cheatingand the awkwardsituation that ensues fer additional indirect (non-altruistic)benefits. that permitsome subtlecheating to be adaptive. Such an argumentcan only be advanced seri- This sets up a dynamic tension in the system ously for slowlyreproducing species with little that has importantrepercussions, as discussed dispersal. Saving an individual's life in a below. hunter-gatherergroup, for example, may permit (5) Number of reciprocalrelationships. It has non-altruisticactions such as cooperativehunt- so far been assumed that it is to the advantage ing to continue with more individuals. But if of each individual to formthe maximumnum- there is an optimum group size, one would ber of reciprocal relationships and that the expect adaptations to stay near that size, with individual suffersa decrease in fitnessupon the individualsjoining groups when the groups are rupture of any relationshipin which the cost below this size, and groups splittingup when to him of acts dispensed to the partner is less they are above this size. One would only be than the benefitof acts dispensed toward him selected to keep an individual alive when the by the partner. But it is possible that relation- group is below optimum and not when the ships are partly exclusive, in the sense that group is above optimum. Although an abun- expanding the number of reciprocal exchanges dant literature on hunter-gatherers(and also with one of the partners may necessarilyde- nonhuman primates) suggeststhat adaptations crease the number of exchanges with another. exist to regulate group size near an optimum, For example, if a group of organismswere to there is no evidence that altruisticgestures are split into subgroupsfor much of the day (such curtailed when groups are above the optimum as breakingup into huntingpairs), then altru- in size. Instead, the benefitsof human altruism istic exchanges will be more likely between are to be seen as coming directlyfrom reciproc- membersof each subgroup than between mem- ity- not indirectly through non-altruistic bers of differentsubgroups. In that sense, re- group benefits. This distinctionis important lationships may be partly exclusive, member- because social scientistsand philosophershave ship in a given subgroup necessarilydecreasing tended to deal with human altruism in terms exchanges with others in the group. The im- of the benefitsof living in a group, without portance of this factor is that it adds further differentiatingbetween non-altruisticbenefits complexityto the problem of dealing with the and reciprocal benefits (e.g., Rousseau, 1954; cheater and it increases competitionwithin a Baier, 1958). group to be membersof a favorable subgroup. An individual in a subgroup who feels that THE PSYCHOLOGICAL SYSTEM UNDERLYING HUMAN another member is subtly cheating on their RECIPROCAL ALTRUISM relationship has the option of attemptingto restorethe relationshipto a completelyrecipro- Anthropologistshave recognized the impor- cal one or of attemptingto join another sub- tance of reciprocityin human behavior, but group, therebydecreasing to a minimum the when they have ascribed functionsto such be- possible exchanges between himself and the havior they have done so in terms of group subtle cheater and replacing these with ex- benefits,reciprocity cementing group relations changesbetween a new partneror partners. In and encouraginggroup survival. The individ- short,he can switchfriends. There is evidence ual sacrificesso that the group may benefit. in hunter-gatherersthat much movement of Recently psychologistshave studied altruistic individuals from one band to another occurs behavior in order to show what factorsinduce in response to such social factorsas have just or inhibit such behavior. No attempthas been been outlined (Lee and DeVore, 1968). made to show what functionsuch behavior may (6) Indirect benefitsor reciprocal altruism? serve, nor to describe and interrelatethe com- Given mutual dependence in a group it is possi- ponents of the psychologicalsystem affecting ble to argue that the benefits (non-altruistic) altruisticbehavior. The purpose of this section 48 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME 46 is to show that the above model for the natural organizes it differently,much of the material selection of reciprocallyaltruistic behavior can supportingthe assertionsbelow is taken from readilyexplain the functionof human altruistic his paper. All referencesto Krebs below are to behavior and the details of the psychological this review. Also, Hartshorneand May (1928- systemunderlying such behavior. The psycho- 1930) have shownthat childrenin experimental logical data can be organized into functional situationsdo not divide bimodallyinto altruists categories,and it can be shown that the compo- and "cheaters" but are distributednormally; nents of the systemcomplement each other in almostall the childrencheated, but theydiffered regulatingthe expressionof altruisticand cheat- in how much and under what circumstances. ing impulses to the selectiveadvantage of indi- ("Cheating" was defined in their work in a viduals. No concept of group advantage is slightlydifferent but analogous way). necessary to explain the function of human (2) Friendshipand the emotionsof likingand altruisticbehavior. disliking. The tendency to like others, not There is no direct evidence regarding the necessarilyclosely related, to form friendships degree of reciprocal altruism practiced during and to act altruisticallytoward friendsand to- human evolutionnor its geneticbasis today,but ward those one likes will be selected for as the given the universaland nearlydaily practiceof immediateemotional rewardsmotivating altru- reciprocal altruism among humans today, it is istic behavior and the formationof altruistic reasonable to assume that it has been an impor- partnerships.(Selection may also favorhelping tant factorin recenthuman evolution and that strangersor disliked individualswhen they are the underlyingemotional dispositionsaffecting in particularly dire circumstances). Selection altruisticbehavior have importantgenetic com- will favora systemwhereby these tendenciesare ponents. To assume as much allows a number sensitive to such parameters as the altruistic of predictions. tendencies of the liked individual. In other (1) A complex, regulatingsystem. The hu- words,selection will favorliking those who are man altruistic systemis a sensitive,unstable themselvesaltruistic. one. Often it will pay to cheat: namely,when Sawyer(1966) has shown thatall groupsin all the partnerwill not findout, when he will not experimental situations tested showed more discontinue his altruism even if he does find altruisticbehavior toward friendsthan toward out, or when he is unlikely to survive long neutral indivduals. Likewise, Friedrichs (1960) enough to reciprocate adequately. And the has shown that attractivenessas a friend was perception of subtle cheating may be very most highly correlated among undergraduates difficult.Given this unstable characterof the with altruistic behavior. Krebs has reviewed system,where a degree of cheating is adaptive, other studies that suggestthat the relationship natural selection will rapidly favor a complex betweenaltruism and likingis a two-waystreet: psychologicalsystem in each individual regulat- one is more altruistictoward those one likes and ing both his own altruistic and cheating one tends to like those who are most altruistic tendenciesand his responsesto thesetendencies (e.g., Berkowitz and Friedman, 1967; Lerner in others. As selection favorssubtler formsof and Lichtman, 1968). cheating, it will favor more acute abilities to Others (Darwin, 1871; Williams, 1966; and detect cheating. The systemthat resultsshould Hamilton, 1969) have recognized the role simultaneouslyallow the individual to reap friendshipmight play in engenderingaltruistic the benefitsof altruisticexchanges, to protect behavior, but all have viewed friendship(and himselffrom gross and subtle formsof cheat- intelligence)as prerequisitesfor the appearance ing, and to practicethose forms of cheatingthat of such altruism. Williams (1966), who cites local conditions make adaptive. Individuals Darwin (1871) on the matter,speaks of this be- will differnot in being altruists or cheaters havior as evolving but in the degree of altruismthey show and in in aniimalsthat live in stable social groupsand the conditionsunder which theywill cheat. have the intelligenceand othermental qualities The best evidence supportingthese assertions necessaryto forrna systemof personalfriend- can be found in Kreb's (1970) review of the ships and animositiesthat transcendthe limits relevant psychologicalliterature. Although he of familyrelationships (p. 93). MARCH 19711 RECIPROCAL ALTRUISMW 49

This emphasison friendshipand intelligenceas gen, 1968; Gilula and Daniels, 1969). The prerequisitesleads Williams to limit his search grounds for expecting,on functionalgrounds, for altruismto the Mammalia and to a "min- a highlyplastic developmentalsystem affecting orityof thisgroup." But accordingto the model moralisticaggression is discussedbelow. presented above, emotions of friendship(and (4) Gratitude,sympathy, and the cost/benefit hatred) are not prerequisites for reciprocal ratio of an altruistic act. If the cost/benefit altruism but may evolve after a system of ratio is an importantparameter in determining mutual altruism has appeared, as important the adaptiveness of reciprocal altruism, then waysof regulatingthe system. humans should be selectedto be sensitiveto the (4) Moralistic aggression. Once strong posi- cost and benefit of an altruisticact, both in tiveemotions have evolved to motivatealtruistic deciding whetherto performone and in decid- behavior,the altruistis in a vulnerableposition ing whether,or how much, to reciprocate. I because cheaterswill be selected to take advan- suggestthat the emotion of gratitudehas been tage of the altruist'spositive emotions. This selectedto regulatehuman responseto altruistic in turn sets up a selection pressurefor a pro- acts and that the emotion is sensitive to the tective mechanism. Moralistic aggression and cost/benefitratio of such acts. I suggestfurther indignationin humanswas selectedfor in order that the emotion of sympathyhas been selected (a) to counteractthe tendencyof the altru- to motivatealtruistic behavior as a functionof ist, in the absence of any reciprocity,to con- the plight of the recipient of such behavior; tinue to performaltruistic acts for his own crudelyput, the greaterthe potential benefitto emotionalrewards; the recipient,the greaterthe sympathyand the (b) to educate the unreciprocatingindi- more likely the altruistic gesture, even to vidual by frighteninghim with immediate strange or disliked individuals. If the recipi- harm or with the future harm of no more ent's gratitudeis indeed a functionof the cost/ aid; and benefit ratio, then a sympatheticresponse to (c) in extreme cases, perhaps, to select the plight of a disliked individual may result directlyagainst the unreciprocatingindivid- in considerablereciprocity. ual him. by injuring,killing, or exiling There is good evidence supporting the Much of human aggressionhas moral over- psychological importance of the cost/benefit tones. Injustice, unfairness,and lack of reci- ratio of altruisticacts. Gouldner (1960) has re- procityoften motivate human aggressionand viewed the sociologicalliterature suggesting that indignation. Lee (1969) has shown that verbal the greaterthe need state of the recipientof an disputes in Bushmen usually revolve around altruisticact, the greater his tendency to re- problemsof gift-giving,stinginess, and laziness. ciprocate; and the scarcer the resourcesof the DeVore (pers. commun.) reports that a great donor of the act, the greater the tendencyof deal of aggressionin hunter-gatherersrevolves the recipient to reciprocate. Heider (1958) has around real or imagined injustices-inequities, analyzed lay attitudes on altruism and finds for example, in food-sharing(see, for example, that gratitudeis greatestwhen the altruisticact Thomas, 1958; Balikci, 1964; Marshall, 1961). does good. Tesser, Gatewood, and Driver (1968) A common featureof this aggressionis that it have shown that American undergraduates oftenseems out of all proportionto the offenses thoughtthey would feel more gratitudewhen committed. Friends are even killed over ap- cost bene- parently trivial disputes. But since small in- the altruisticact was valuable and the equities repeated many times over a lifetime factora great deal. Pruitt (1968) has provided may exact a heavy toll in relative fitness,selec- evidence that humans reciprocate more when tion may favor a strong show of aggression the original act was expensive for the bene- when the cheating tendency is discovered. factor. He shows that under experimentalcon- Recent discussionsof human and animal aggres- ditions more altruism is induced by a gift of sion have failed to distinguishbetween moral- 80 per cent of $1.00 than 20 per cent of $4.00. istic and other formsof aggression(e.g., Scott, Aronfreed(1968) has reviewedthe considerable 1958; Lorenz, 1966; Montague, 1968; Tinber- evidence that sympathymotivates altruistic be- 50 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME 46 havior as a functionof the plight of the indi- is based on the fact that many transgressions vidual arousingthe sympathy. performed in private are likely to become (5) Guilt and reparativealtruism. If an orga- public knowledge. It should often be advan- nism has cheated on a reciprocal relationship tageous to confess sins that are likely to be and this facthas been found out, or has a good discoveredbefore theyactually are, as evidence chance of being found out, by the partnerand of sincerity(see below on detectionof mimics). if the partnerresponds by cuttingoff all future (6) Subtle cheating: the evolution of mimics. acts of aid, then the cheater will have paid Once friendship,moralistic aggression, guilt, dearlyfor his misdeed. It will be to the cheater's sympathy,and gratitudehave evolved to regu- advantage to avoid this,and, providingthat the late the altruistic system,selection will favor cheatermakes up forhis misdeed and does not mimickingthese traitsin order to influencethe cheat in the future,it will be to his partner's behavior of others to one's own advantage. benefitto avoid this,since in cuttingoff future Apparent acts of generosityand friendshipmay acts of aid he sacrificesthe benefitsof future induce genuine friendshipand altruismin re- reciprocalhelp. The cheatershould be selected turn. Sham moralisticaggression when no real to make up for his misdeed and to show con- cheating has occurred may neverthelessinduce vincing evidence that he does not plan to reparativealtruism. Sham guilt may convince a continue his cheating sometimein the future. wronged friend that one has reformedone's In short, he should be selected to make a ways even when the cheating is about to be reparativegesture. It seems plausible, further- resumed. Likewise, selection will favor the more, that the emotion of guilt has been se- hypocrisyof pretendingone is in dire circum- lected forin humans partlyin order to motivate stances in order to induce sympathy-motivated the cheater to compensate his misdeed and to altruistic behavior. Finally, mimicking sym- behave reciprocallyin the future,and thus to pathy may give the appearance of helping in preventthe ruptureof reciprocalrelationships. order to induce reciprocity,and mimicking Krebs has reviewed the evidence that harm- gratitude may mislead an individual into ex- ing another individual publicly leads to altru- pecting he will be reciprocated. It is worth isticbehavior and concludes: emphasizingthat a mimic need not necessarily Many studies have supportedthe notion that be conscious of the deception; selection may public transgressionwhether intentional or un- favorfeeling genuine moralisticaggression even intentional,whether immoral or onlysituationally when one has not been wronged if so doing unfortunate,leads to reparativealtruism (p. 267). leads another to reparative altruism. Instances of the above formsof subtle cheat- Wallace and Sadalla (1966), for example, ing are not difficult find. For in- showed experimentallythat individuals who to typical stances fromthe literatureon broke an expensivemachine were more likelyto hunter-gatherers see Rasmussen (1931), Balikci (1964), and Lee volunteer for a painful experimentthan those and DeVore (1968). The importanceof these who did not, but only if theirtransgression had formsof cheating can partly be been discovered. Investigatorsdisagree on the inferredfrom the adaptations to detect such cheating dis- extent to which guilt feelingsare the motiva- cussed below and from the importance and tion behind reparative altruism. Epstein and prevalence of moralistic aggressiononce such Hornstein (1969) supply some evidence that cheatingis detected. guilt is involved, but on the assumption that one feels guilt even when one behaves badly (7) Detection of the subtle cheater: trust- in private,Wallace and Sadalla's (1966) result worthiness, trust, and suspicion. Selection contradicts the view that guilt is the only should favor the ability to detect and dis- motivatingfactor. That private transgressions criminate against subtle cheaters. Selection are not as likely as public ones to lead to will clearlyfavor detecting and counteringsham reparativealtruism is preciselywhat the model moralistic aggression. The argument for the would predict,and it is possible that the com- others is more complex. Selection may favor mon psychologicalassumption that one feels distrustingthose who perform altruistic acts guilt even when one behaves badly in private without the emotional basis of generosityor MARCH 1971] RECIPROCAL ALTRUISM 51 guilt because the altruistictendencies of such 1966; Schopler and Thompson, 1968). Krebs individuals may be less reliable in the future. concludes that, "When the legitimacyof ap- One can imagine, for example, compensating parent altruismis questioned,reciprocity is less for a misdeed withoutany emotional basis but likelyto prevail." Lerner and Lichtman (1968) with a calculating,self-serving motive. Such an have shown experimentallythat those who act individual should be distrusted because the altruisticallyfor ulterior benefit are rated as calculating spirit that leads this subtle cheater unattractiveand are treated selfishly,whereas now to compensate may in the future lead those who apparently are genuinely altruistic him to cheat when circumstancesseem more are rated as attractiveand are treated altru- advantageous(because of unlikelihoodof detec- istically. Berscheid and Walster (1967) have tion, for example, or because the cheated indi- shown thatchurch women tend to make repara- vidual is unlikely to survive). Guilty motiva- tionsfor harm theyhave committedby choosing tion, in so far as it evidences a more enduring the reparationthat approximates the harm (that commitmentto altruism,either because guilt is, is neither too slightnor too great), presum- teaches or because the cheater is unlikely not ably to avoid the appearance of inappropriate- to feel the same guilt in the future,seems more ness. reliable. A similar argument can be made Rapoport and Dale (1967) have shown that about the trustworthinessof individuals who when two strangers play iterated games of initiate altruistic acts out of a calculating Prisoner'sDilemma in which the matrix deter- rather than a generous-hearteddisposition or mines profitsfrom the games played there is a who show either false sympathyor false grati- significanttendency for the level of cooperation tude. Detection on the basis of the underlying to drop at the end of the series,reflecting the psychological dynamics is only one form of fact that the partnerwill not be able to punish detection. In many cases, unreliability may for "cheating" responses when the series is more easily be detected through experiencing over. If a long seriesis brokenup into subseries the cheater's inconsistent behavior. And in with a pause between subseriesfor totalingup some cases, thirdparty interactions (as discussed gains and losses,then the tendencyto cheat on below) may make an individual's behavior pre- each otherincreases at the end of each subseries. dictable despite underlying cheating motiva- These results,as well as some others reported tions. by Rapoport and Chammah (1965), are sug- The anthropologicalliterature also abounds gestive of the instabilitythat exists when two withinstances of the detectionof subtle cheaters strangersare consciously trying to maximize (see above referencesfor hunter-gatherers).Al- gain by tradingaltruistic gestures, an instability though I know of no psychologicalstudies on that is presumablyless markedwhen the under- the detectionof sham moralisticaggression and lying motivation involves the emotions of sham guilt, there is ample evidence to support friendship,of liking others,and of feelingguilt the notion that humans respond to altruistic over harminga friend. Deutsch (1958), for ex- acts accordingto theirperception of the motives ample, has shown that two individuals playing of the altruist. They tend to respond more iterated games of Prisoner's Dilemma will be altruisticallywhen they perceive 'the other as more cooperative if a third individual, dis- acting "genuinely"altruistic, that is, voluntarily liked by both,is present. The perceivedmutual dispatchingan altruisticact as an end in itself, dislike is presumed to create a bond between without being directed toward gain (Leeds, the twoplayers. 1963; Heider, 1958). Krebs (1970) has reviewed It is worthmentioning that a classic problem the literatureon this point and notes that help in social science and philosophy has been is more likely to be reciprocated when it is whetherto define altruismin termsof motives perceived as voluntary and intentional (e.g., (e.g., real vs. "calculated" altruism)or in terms Goranson and Berkowitz, 1966; Lerner and of behavior,regardless of motive (Krebs, 1970). Lichtman, 1968) and when the help is appro- This problem reflectsthe fact that, wherever priate, that is, when the intentions of the studied, humans seem to make distinctions altruistare not in doubt (e.g., Brehmand Cole, about altruism partly on the basis of motive, 52 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME 46

and this tendency is consistentwith the hy- favorlearning about the altruisticand cheating pothesis that such discriminationis relevant to tendencies of others indirectly,both through protectingoneself fromcheaters. observing interactions of others and, once (8) Setting up altruisticpartnerships. Selec- linguistic abilities have evolved, by hearing tion will favor a mechanism for establishing about such interactionsor hearing characteriza- reciprocalrelationships. Since humans respond tions of individuals (e.g., "dirty, hypocritical, to acts of altruismwith feelingsof friendship dishonest,untrustworthy, cheating louse"). One that lead to reciprocity,one such mechanism important result of this learning is that an might be the performingof altruistic acts individual may be as concernedabout the atti- toward strangers,or even enemies, in order to tude of onlookers in an altruisticsituation as induce friendship.In short,do unto others as about the attitude of the individual being you would have them do unto you. dealt with. The mechanismhypothesized above leads to (ii) Help in dealing with cheaters. In dealing results inconsistentwith the assumption that with cheaters selection may favor individuals humans always act more altruisticallytoward helping others,kin or non-kin,by direct coer- friendsthan towardothers. Particularlytoward cion against the cheateror by everyonerefusing strangers,humans may initiallyact more altru- him reciprocal altruism. One effectof this is isticallythan towardfriends. Wright (1942) has that an individual, throughhis close kin, may shown, for example, that third grade children be compensatedfor an altruisticact even after are more likelyto give a more valuable toy to a his death. An individual who dies saving a strangerthan to a friend. Later, some of these friend, for example, may have altruistic acts childrenverbally acknowledged that theywere performedby the friend to the benefitof his tryingto make friends.Floyd (1964) has shown offspring.Selection will discriminate against that, after receiving many trinkets from a the cheater in this situation, if kin of the friend,humans tend to decrease their giftsin martyr,or others,are willing to punish lack of return,but afterreceiving many trinketsfrom a reciprocity. neutral or disliked individual, they tend to (iii) Generalized altruism. Given learning increase their gifts in return. Likewise, after from others and multiparty action against receiving few trinketsfrom a friend,humans cheaters,selection may favora multipartyaltru- tend to increase their gifts in return,whereas isticsystem in whichaltruistic acts are dispensed receiving few trinketsfrom a neutral or dis- freely among more than two individuals, an liked individual resultsin a decrease in giving. individual being perceived to cheat if in an This was interpretedto mean that generous altruisticsituation he dispenses less benefitfor friendsare taken forgranted (as are stingynon- the same cost than would the others,punish- friends). Generosityfrom a non-friendis taken ment coming not only fromthe other individ- to be an overtureto friendship,and stinginess ual in that particular exchange but from the froma friendas evidence of a deterioratingre- othersin the system. lationship in need of repair. (Epstein and (iv) Rules of exchange. Multipartyaltruistic Hornstein, 1969, provide new data supporting systemsincrease by several-foldthe cognitive this interpretationof Floyd, 1964.) difficultiesin detectingimbalances and deciding (9) Multipartyinteractions. In the close-knit whetherthey are due to cheatingor to random social groups that humans usually live in, se- factors. One simplifyingpossibility that lan- lection should favormore complex interactions guage facilitatesis the formulationof rules of than the two-partyinteractions so far discussed. conduct, cheating being detected as infraction Specifically,selection may favor learning from of such a rule. In short,selection may favor the altruisticand cheatingexperiences of others, the elaboration of normsof reciprocalconduct. helping others coerce cheaters,forming multi- There is abundant evidence for all of the party exchange systems,and formulatingrules above multipartyinteractions (see the above for regulated exchanges in such multiparty referenceson hunter-gatherers).Thomas (1958), systems. forexample, has shown thatdebts of reciprocity (i) Learning from others. Selection should do not disappear with the death of the MARCH 19711 RECIPROCAL ALTRUISM 53

"creditor"but are extended to his kin. Krebs nism's sense of guilt to be educated, perhaps has reviewed the psychological literature on partly by kin, so as to permit those formsof generalized altruism. Several studies (e.g., cheating that local conditions make adaptive Darlington and Macker, 1966) have shown that and to discourage those with more dangerous humans may directtheir altruism to individuals consequences. One would not expect any sim- other than those who were hurt and may ple systemregulating the developmentof altru- respond to an altruisticact that benefitsthem- istic behavior. To be adaptive, altruisticbe- selves by acting altruisticallytoward a third havior must be dispensed with regard to many individual uninvolvedin the initial interaction. characteristicsof the recipient (including his Berkowitzand Daniels (1964) have shown ex- degree of relationship,emotional makeup, past perimentally,for example, that help from a behavior, friendships,and kin relations), of confederate leads the subject to direct more other membersof the group, of the situation help to a third individual, a highlydependent in which the altruisticbehavior takesplace, and supervisor. Freedman, Wallington, and Bless of many other parameters,and no simple de- (1967) have demonstratedthe surprisingresult velopmental system is likely to meet these that, in two differentexperimental situations, requirements. humans engaged in reparativealtruism only if Kohlberg(1963), Bandura and Walters (1963), it could be directedto someone other than the and Krebs have reviewed the developmental individual harmed,or to the original individual literatureon human altruism. All of themcon- only if they did not expect to meet again. In clude that none of the proposed developmental a systemof strongmultiparty interactions it is theories (all of which rely on simple mecha- possible that in some situationsindividuals are nisms) can account for the known diverse de- selected to demonstrategeneralized altruistic velopmental data. Whiting and Whiting (in tendenciesand that their main concern when prep.) have studied altruisticbehavior directed theyhave harmed another is to show that they towardskin by childrenin six differentcultures are genuinelyaltruistic, which theybest do by and find consistentdifferences among the cul- acting altruisticwithout any apparent ulterior tures that correlate with differencesin child- motive, e.g., in the experiments,by acting rearing and other facetsof the cultures. They altruistic toward an uninvolved third party. argue that the differencesadapt the childrento Alternatively,A. Rapoport (pers. commun.)has differentadult roles available in the cultures. suggestedthat the reluctance to direct repara- Although the behavior analyzed takes place tive altruismtoward the harmedindividual may between kin and hence Hamilton's model be due to unwillingness to show thereby a (1964) may apply rather than this model, the recognitionof the harm done him. The re- Whitings' data provide an instance of the direction serves to allay guilt feelingswithout adaptive value of developmental plasticityin triggeringthe greater reparation that recogni- altruisticbehavior. No careful work has been tion of the harmmight lead to. done analyzing the influenceof environmental (10) Developmentalplasticity. The conditions factors on the development of altruistic be- under which detection of cheating is possible, havior,but some data exist. Krebs has reviewed the range of available altruistic trades, the the evidence that altruistictendencies can be cost/benefitratios of these trades, the relative increased by the effectsof warm, nurturant stabilityof social groupings,and other relevant models, but little is known on how long such parameters should differfrom one ecological effectsendure. Rosenhan (1967) and Rettig and social situation to another and should (1956) have shown a correlationbetween altru- differthrough time in the same small human ism in parentsand altruismin theircollege-age population. Under these conditionsone would children, but these studies do not separate expect selectionto favordevelopmental plastic- genetic and environmental influences. Class ity of those traits regulating altruistic and differencesin altruistic behavior (e.g., Berko- cheating tendencies and responses to these witz, 1968; Ugurel-Semin,1952; Almond and tendencies in others. For example, develop- Verba, 1963) may primarily reflect environ- mental plasticitymay allow the growingorga- mental influences. Finally Lutzker (1960) and 54 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME 46

Deutsch (1958) have shown thatone can predict altruisticsystem. The inherentinstability of the the degree of altruisticbehavior displayed in Prisoner's Dilemma, combined with its impor- iterated games of Prisoner's Dilemma from tance in human evolution,has led to the evolu- personality typing based on a questionnaire. tion of a very complex system. For example, Such personalitydifferences are probablypartly once moralistic aggression has been selected environmentalin origin. for to protect against cheating, selection fa- It is worth emphasizing that some of the vors sham moralisticaggression as a new form psychologicaltraits analyzed above have applica- of cheating. This should lead to selection for tions outside the particularreciprocal altruistic the abilityto discriminatethe two and to guard systembeing discussed. One may be suspicious, against the latter. The guarding can, in turn, for example, not only of individuals likely to be used to counter real moralisticaggression: cheat on the altruisticsystem, but of any indi- one can, in effect,impute cheating motives to vidual likely to harm oneself; one may be sus- another person in order to protect one's own picious of the knowntendencies toward adultery cheating. And so on. Given the psychological of another male or even of these tendenciesin and cognitive complexity the system rapidly one's own mate. Likewise, a guilt-motivated acquires, one may wonder to what extent the show of reparation may avert the revenge of importanceof altruismin human evolution set someone one has harmed,whether that individ- up a selection pressure for psychologicaland ual was harmed by cheating on the altruistic cognitive powers which partly contributed to systemor in some otherway. And the systemof the large increase in hominid brain size during reciprocal altruismmay be employed to avert the Pleistocene. possible revenge. The Bushmen of the Kala- hari, for example, have a saying (Marshall, ACKNOWLEDGMENTS 1959) to the effectthat, if you wish to sleep with I thankW. H. Drury,E. Mayr,I. Nisbet,E. E. someone else's wife,you get him to sleep with Williamsand E. 0. Wilson for usefulcomments yours,then neitherof you goes afterthe other on earlierdrafts of this paper, and I thankespe- with poisoned arrows. Likewise,there is a large cially I. DeVore and W. D. Hamiltonfor detailed literatureon the use of reciprocityto cement commentand discussion. I thank A. Rapoport friendshipsbetween neighboringgroups, now and D. Krebs for access to unpublishedmaterial. engaged in a common enterprise(e.g., Lee and This workwas completedunder a NationalScience DeVore, 1968). Foundationpre-doctoral fellowship and partially The above review of the evidence has only supported by grant number NIMH 13156 to begun to outline the complexitiesof the human I. DeVore.

LIST OF LITERATURE

ALMOND, G. A., and S. VERBA. 1963. The Civic and social distance in help-giving: an experi- Culture. Princeton University Press, Prince- mental investigation of English social class ton,N. J. differences. J. Exp. Soc. Psychol., 4: 66-669. ARONFREED, J. 1968. Conduct and Conscience. BERKOWITZ, L., and L. DANIELS. 1964. Affecting Academic Press,N. Y. the salience of the social responsibilitynorm: ASHMOLE, M. 1962. Migration of European effectsof past help on the response to depen- thrushes: a comparativestudy based on ringing dency relationships. J. Abnorm. Soc. Psychol., recoveries. Ibis, 104: 522-559. 68: 275-281. BAIER, K. 1958. The Moral Point of View. Cor- BERKOWITZ, L., and P. FRIEDMAN. 1967. Some nell UniversityPress, Ithaca, N. Y. social class differencesin helping behavior. J. BALIKCI, A. 1964. Development of basic socio- Personal. Soc. Psychol.,5: 217-225. economic units in two Eskimo communities. BERNDT, R., and H. STERNBERG. 1968. Terms, stud- National Museum of Canada Bulletin No. 202, ies and experiments on the problems of bird Ottawa. dispersion. Ibis, 110: 256-269. BANDURA, A., and R. H. WALTERS. 1963. Social BERSCHEID, E., and E. WALSTER. 1967. When does Learning and Personality Development. Holt, a harm-doercompensate a victim? J. Personal. Rhinehart and Winston,N. Y. Soc. Psychol.,6: 435-441. BERKOWITZ, L. 1968. Responsibility, reciprocity BLEST, A. D. 1963. Longevity, palatability and MARCH 1971] RECIPROCAL ALTRUISM 55

natural selection in five species of New World and man's struggle to adapt. Science, 164: Saturuiid moth. Nature, 197: 1183-1186. 395-405. BREHM,J. W., and A. H. COLE. 1966. Effect of GORANSON,R., and BERKOWITZ.1966. Reciproc- a favor which reduces freedom. J. Personal. ity and responsibilityreactions to prior help. Soc. Psychol.,3: 420-426. J. Personal. Soc. Psychol.,3: 227-232. BROWER, J. V. Z., and L. P. BROWER. 1965. Ex- GOULDNER,A. 1960. The norm of reciprocity: a perimentalstudies of mimicry. 8. Am. Natur., preliminary statement. Am. Sociol. Rev., 47: 49: 173-188. 73-80. CAMPBELL, B. 1966. Human Evolution. Aldine, HALDANE, J. B. S. 1955. Population genetics. Chicago. New Biology, 18: 34-51. CROOK, J.H. 1969. The socio-ecologyof primates. HALL, K. R. L., and I. DEVORE. 1965. Baboon In J.H. Crooke (ed.), Social Behavior in Birds social behavior. In I. DeVore (ed.), Primate and Mammals, p. 103-166. Academic Press, Behavior: Field Studies of Monkeys and Apes, London. p. 53-110. Holt, Rhinehart and Winston,N. Y. DARLINGTON,R. B., and C. E. MACKER. 1966. Dis- HAMILTON, W. D. 1964. The genetical evolution placement of guilt-producedaltruistic behavior. of social behavior. J. Theoret. Biol., 7: 1-52. J. Personal. Soc. Psychol.,4: 442-443. *. 1966. The moulding of senescence by natu- DARWIN, C. 1871. The Descent of Man and ral selection. J. Theoret. Biol., 12: 12-45. Selection in Relation to Sex. Random House, -. 1969. Selection of selfish and altruistic be- N. Y. havior in some extreme models. Paper pre- DEUTSCH, M. 1958. Trust and suspicion. J. Con- sented at "Man and Beast Symposium" (in flictResolution, 2: 267-279. press,Smithsonian Institution). DHONT, A. A., and J. HUBLE'. 1968. Fledging-date HARTSHORNE, H., and M. A. MAY. 1928-1930. and sex in relation to dispersal in young Great Studies in the Nature of Character. Vol. 1, Tits. Bird Study,15: 127-134. Studies in Deceit; Vol. 2, Studies in Self-Con- DRURY,W. H., and W. J. SMITH. 1968. Defense trol; Vol. 3, Studies in the Organization of of feeding areas by adult herring gulls and Character. Macmillan, N. Y. intrusionby young. Evolution, 22: 193-201. HEDIGER, H. 1968. Putzer-fische im aquarium. EIBL-EIBEsFELDT, I. 1955. Qber Symbiosen, Para- Natur und Museum, 98:89-96. sitismusund andere besondere zwischenartliche HIEIDER, F. 1958. The Psychology of Interper- Beziehungen tropischer Meeresfische. Z. f. sonal Relations. Wiley, N. Y. 7Tierpsychol.,12: 203-219. HOBSON, E. S. 1969. Comments on certain recent 1959. Der Fisch Aspidontus taeniatus als generalizationsregarding cleaning symbiosesin Nachahmer des Putzers Labroides dimidiatus. fishes. Pacific Science,23: 35-39. Z. f. Tierpsychol.,16: 19-25. KOHLBERG, L. 1963. Moral development and EPSTEIN, Y. M., and H. A. HORSTEIN. 1969. Penalty identification. In H. W. Stevenson (ed.), Year- and attractionas factorsinfluenc- interpersonal book of the National Society for the Study of ing the decision to help another person. J. Education. Part 1. Child Psychology,p. 277- Exp. Soc. Psychol.,5: 272-282. 332. Universityof Chicago Press,Chicago. FEDER, H. M. 1966. Cleaning symbioses in the KREBS, D. 1970. Altruism-an examination of the marine environment. In S. M. Henry (ed.), concept and a review of the literature. Psy- Symbiosis,Vol. 1, p. 327-380. Academic Press, N. Y. chol. Bull., 73: 258-302. an FISHER, R. A. 1958. The Genetical Theory of KUYTON, P. 1962. Verhalten-beobachtungen Natural Selection. Dover, N. Y. der Raupe des Kaiseratlas. Z. d. Entomol., FLOYD, J. 1964. Effectsof amount of award and 72: 203-207. friendshipstatus of the other on the frequency LEE, R. 1969. !Kung Bushman violence. Paper of sharing in children. Unpublished doctoral presented at meeting of American Anthropo- dissertation. University of Minnesota. (Re- logical Association,Nov. 1969. viewed in Krebs, 1970). LEE, R., and I. DEVORE. 1968. Man the Hunter. FREEDMAN, J. L., S. A. WALLINGTON, and E. BLESS. Aldine, Chicago. 1967. Compliance without pressure: the effect LEEDS,R. 1963. Altruismand the norm of giving. of guilt. J. Personal. Soc. Psychol., 7: 117-124. Merrill-PalmerQuart., 9: 229-240. FRIEDRICHS, R. W. 1960. Alter versus ego: an LERNER,M. J.,and R. R. LICHTMAN. 1968. Effects exploratory assessment of altruism. Am. of perceived norms on attitudes and altruistic Sociol. Rev., 25: 496-508. behavior toward a dependent other. J. Per- GILULA, M. F., and D. N. DANIELS. 1969. Violence sonal. Soc. Psychol.,9: 226-232. 56 THE QUARTERLY REVIEW OF BIOLOGY [VOLUME 46

LEYHAUSEN, P. 1965. Ober die Funktion der rela- . 1962. Fish service stations. Sea Frontiers, tiven Stimmungshierarchie (dasgestellt am 8: 40-47. Beispiel der phylogenetischenund ontogene- RAPOPORT,A., and A. CHAMMAH. 1965. Prisoner's tischen Entwicklungdes Beutefangsvon Raub- Dilemma. Universityof Michigan Press, Ann tieren. Z. f. Tierpsychol.,22: 412-494. Arbor. LIMBAUGH, C. 1961. Cleaning symbioses. Scient. RAPOPORT,A., and P. DALE. 1967. The "end" and Am., 205: 42-49. "start" effectsin iterated Prisoner's Dilemma. LIMBAUGH, C., H. PEDERSON, and F. CHASE. 1961. J. ConflictResolution, 10: 363-366. Shrimps that clean fishes. Bull. Mar. Sci. Gulf RASMUSSEN,K. 1931. The Netsilik Eskimos: social Caribb., 11: 237-257. life and spiritual culture. Report of the Fifth LORENZ, K. 1966. On Aggression. Harcourt, Thule Expedition 1921-1924, Vol. 8(1,2). Brace and World, N. Y. GyldendalskeBoghandel, Copenhagen. LUCE, R. D., and H. RAIFFA. 1957. Games and RETTIG, S. 1956. An exploratorystudy of altruism. Decisions. Wiley, N. Y. Dissert.Abstr., 16: 2220-2230. LUTZKER,D. 1960. Internationalismas a predictor REYNOLDS, L. 1966. Open groups in hominid of cooperative game behavior. J. Conflict evolution. Man, 1: 441-452. Resolution, 4: 426-435. ROSENHAN, D. 1967. The Origins of Altruistic MARLER,P. 1955. The characteristicsof certain Social Autonomy. Educational Testing Ser- animal calls. Nature, 176: 6-7. vice, Princeton,N.J. . 1957. Specific distinctiveness in the com- RoussEAu,J. J. 1954. The Social Contract. Henry munication signals of birds. Behavior, 11: RegneryCo., Chicago. 13-39. RUDEBECK, G. 1950. The choice of prey and MARSHALL, L. K. 1959. Marriage among !Kung modes of hunting of predatorybirds with spe- Bushmen. Africa,29: 335-365. cial reference to their selective effort. Oikos, . 1961. Sharing, talking and giving: relief of 2: 65-88. social tension among !Kung Bushmen. Africa, * 1951. The choice of prey and modes of 31: 231-249. hunting of predatorybirds, with special refer- MAYNARD,E. C. L. 1968. Cleaning symbiosisand ence to their selective effort(cont.) Oikos, 3: oral grooming on the coral reef. In P. Person 200-231. (ed.), Biology of the Month, p. 79-88. Philip SAWYER, J. 1966. The altruism scale: a measure Person, American Association for the Advance- of cooperative, individualistic,and competitive ment of Science,Wash., D. C. interpersonal orientation. Am. J. Social. 71: MAYNARD SMITH, J. 1964. Kin selection and group 407-416. selection. Nature, 201: 1145-1147. SCHOPLER,J., and V. T. THOMPSON. 1968. The role . 1965. The evolution of alarm calls. Amer. of attributionprocess in mediating amount of Natur., 99: 59-63. reciprocityfor a favor. J. Personal. Soc. Psy- MAYR, E. 1963. Animial Species and Evolution. chol., 10: 243-250. Belknap Press,Cambridge. SCOTT, J. P. 1958. Aggression.University of Chi- MONTAGU, F. M. A. 1968. Man and Aggression. cago Press,Chicago. Oxford UniversityPress, N. Y. SOUTHERN, H. N. 1954. Tawny owls and their MURIE, A. 1944. The Wolves of Mount McKinley. prey. Ibis, 96: 384-410. Fauna of National Parks, Faunal Series # 5; STRUHSAKER, T. 1967. Social structure among Wash., D.C. vervet monkeys (Cercopithecus aethiops). Be- OWEN,D. F. 1963. Similar polymorphismasin an havior,29: 83-121. insect and a land snail. Nature, 198: 201-203. TESSER, A., R. GATEWOOD, and M. DRIVER. 1968. PERDECK,A. 1958. Two types of orientation in Some determinants of gratitude. J. Personal. migrating starlings, Starnus vulgaris L., and Soc. Psychol.,9:\232-236. chaffishes,Fringilla coeleus L., as revealed by THOMAS, E. M. 1958. The Harmless People. displacement experiments. Ardea, 46: 1-35. Random House, N. Y. PRUITT,D. G. 1968. Reciprocityand credit build- TINBERGEN, L. 1960. The dynamics of insect and ing in a laboratory dyad. J. Personal. Soc. bird populations in pine woods. 1. Factors Psychol., 8: 143-147. influencingthe intensityof predation by song RANDALL,J. E. 1958. A review of the Labrid fish birds,Arch. Neerl. de Zool., 13: 265-343. genus Labroides with descriptionsof two new TINBERGEN, N. 1968. On war and peace in ani- species and notes on ecology. Pacific Science, mals and man. Science, 160: 1411-1418. 12: 327-347. UGUREL-SEMIN, R. 1952. Moral behavior and MARCH 1971] RECIPROCAL ALTRUISM 57

moral judgment of children. J. Abnorm. Soc. WICKLER,W. 1968. Mimicry in Plants and Ani- Psychol.,47: 463-474. mals. McGraw-Hill, N. Y. VANLAWICK-GoODALL, J. 1968. A preliminary WILIIAMS, G. C. 1966. Adaptation and Natural report on expressivemovements and communi- Selection. Princeton University Press, Prince- in Gombe Stream chimpanzees. In cation the ton, N. J. P. Jay (ed.), Primates, p. 313-374 Holt, Rhine- WRIGHT,B. 1942. Altruism in children and the hart and Winston, N.Y. perceived conduct of others. J. Abnorm. Soc. WALLACE, J., and E. SADALLA. 1966. Behavioral consequences of transgression: the effects of Psychol.,37: 218-233. social recognition. J. Exp. Res. Personal., 1: WYNNE-EDWARDS, V. C. 1962. Animal Dispersion 187-194. in Relation to Social Behavior. Hafner, N. Y.