Group Competition, Reproductive Leveling, and the Evolution of Human Altruism Samuel Bowles, et al. Science 314, 1569 (2006); DOI: 10.1126/science.1134829 The following resources related to this article are available online at www.sciencemag.org (this information is current as of December 8, 2006 ): Updated information and services, including high-resolution figures, can be found in the online version of this article at: http://www.sciencemag.org/cgi/content/full/314/5805/1569 Supporting Online Material can be found at: 6 0 http://www.sciencemag.org/cgi/content/full/314/5805/1569/DC1 0 2 This article cites 11 articles, 3 of which can be accessed for free: r 8, http://www.sciencemag.org/cgi/content/full/314/5805/1569#otherarticles e b m e This article appears in the following subject collections: c e Evolution D http://www.sciencemag.org/cgi/collection/evolution n o g Information about obtaining reprints of this article or about obtaining permission to reproduce r o this article in whole or in part can be found at: g. http://www.sciencemag.org/help/about/permissions.dtl ma e c n e ci s w. w w m o r d f e d a o nl w o D Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. Copyright c 2005 by the American Association for the Advancement of Science; all rights reserved. The title SCIENCE is a registered trademark of AAAS. RESEARCH ARTICLES Group Competition, Reproductive variance.(E{}indicatesaweightedaverageover demes.)Thecoefficientb istheeffectofvariation G inp ontheaveragefitnessofmembersofdemej Leveling, and the Evolution of j (w); b is the effect of variation in p (namely, j i ij switchingfromanNtoanA)onthefitnessofan Human Altruism individualindemej(w).Abehaviorisaltruisticif ij adopting it lowers one's expected fitness while increasingtheaveragefitnessofone'sdeme(18). Samuel Bowles Giventhisdefinition,weareinterestedinthecase whereb isnegativeandb ispositive. i G Humans behavealtruistically innatural settingsandexperiments. Apossibleexplanation—that Using Eq. 1, we see that whether altruism groups withmorealtruistssurvive whengroups compete—haslongbeenjudgeduntenableon evolves(Dp>0)dependsontheoutcomeofa empiricalgroundsformostspecies.Buttherehavebeennoempiricaltestsofthisexplanationfor raceinwhichthebetween-selectionprocesspro- humans.My empiricalestimatesshowthatgeneticdifferences between earlyhumangroupsare motingitsspread[var(p)b ]competeswiththe j G likelytohavebeen greatenoughsothatlethalintergroup competition couldaccountforthe within-selection processtending toeliminateit evolutionofaltruism.Crucialtothisprocessweredistinctivehumanpracticessuchassharingfood (E{var(p )}b).Forthebetween-demeeffect to ij i beyond theimmediate family,monogamy,andotherformsof reproductiveleveling.These exceedthewithin-demeeffect(rearrangingEq.1), culturallytransmitted practicespresupposeadvanced cognitiveand linguisticcapacities,possibly itmustbethat 6 accountingforthe distinctiveformsofaltruism foundinourspecies. 0 0 varðpÞ 2 Darwinthoughtthatthe“moralfaculties” Models(14),computersimulations(15),and Efvarðpj Þg>−bi=bG (2) 8, had proliferated among early humans empiricalstudies(16)haveconfirmedthatinter- ij r e because a tribe of “courageous, sym- groupcompetitioncouldinfluencetheevolution b m pathetic and faithful members who were al- of culturally transmitted behavior. This study Theleft-handsideofthisconditionisameasure e ways ready to…aid and defend each other… investigates whether, as an empirical matter, ofpositiveassortmentarisingfromthefactthat ec D would spread and be victorious over other intergroup competition and reproductive level- ifthefractionofA'sindemesdiffer[thatis,var(p) tribes” (1, p. 134). Recent experiments have ing might have allowed the proliferation of a ispositive],thenA'saremorelikelythanN'stoj on extensivelydocumentedaltruisticbehaviorsnot geneticallytransmittedpredispositiontobehave interactwithA's. g r only in laboratories but also among hunter- altruistically. To determine the facts necessary Becausethewithin-demebenefitsofaltruism o gathererpopulations(2–4).Butinorderforthe for this inquiry, a model was developed that are randomly distributed, between-deme differ- ag. m survivalofmorealtruisticgroupsincompetition captures the main aspects of ancestral human encesintheprevalenceofA's[i.e.,var(p)>0]is j e withothergroupstoaccountfortheevolution geneticdifferentiation,between-groupcompeti- theonlyreasonwhyA'saremorelikelythanN'sto c n ofapredispositiontoactaltruistically,thegroup tion,andgroupsocialstructure. interactwithA'sandthustobenefitmutually.But e extinction process would have to be strong Framework for the empirical analysis. ifA'sarelikelytobenefitforthisreason,theyare sci enough to offset the lower fitness of altruists Consideralargemetapopulationofindividuals also more likely to compete over deme-specific w. w comparedtoothermembersoftheirgroup.For livinginpartiallyisolatedsubpopulations(called resources(19,20).Iassumethemoststringentform w thistobethecase,therewouldhavetobesub- demes).Altruists(A's)takeanactioncostingc of local density-dependent constraints on re- m stantialdifferencesin thefractionofaltruists that confers a benefit b on an individual ran- productive output: Sites are saturated so that ter- o r in groups,which isthought tobeunlikely be- domlyselectedfromthenmembersofthedeme. ritorial expansion is required for deme growth. d f cause migration among groups tends to limit (PayoffsaregiveninTable1,andthemodeland Thus, altruism can proliferate only by helping a e d between-group differences in group composi- notation are summarized in Table 2.) A's are demetoacquiremoreterritory,notbyanyofthe a tion.Thus,manyhaveconcludedthatbetween- bearersofahypothetical“altruisticallele”;those otherwaysthatmembersofpredominantlyaltru- nlo group genetic differences are too small for withouttheallele(N's)donotbehavealtruisti- isticdemesmightproducemoresurvivingoffspring. w o selective group extinction to offset the within- cally.Reproductionisasexual.Intheabsenceof Selective group extinction. Selective ex- D group selective pressures that oppose the evo- reproductiveleveling,individualfitnessisiden- tinction may allow the evolution of altruism if lution of a genetic predisposition to behave ticaltothepayoffsinTable1.Forexample,anA predominantly altruistic demes are more likely altruistically[(5),butseealso(6)]. who interacts solely with A's will expect a than other demes to survive between-deme However, early humans lived under con- number of offspring surviving to reproductive contestsandtocolonizeandrepopulatethesites ditionssuchthatselectivegroupextinctionmight agethatisb−cgreaterthanthefitnessofanN vacatedbydemesthatfail(21).Thisprocessis have been a powerful evolutionary force. Cul- whointeractsonlywithN's. captured by the term b , the size of which is G turally transmitted norms supporting resource Letp =1ifindividualiindemejisanA determined by the frequency of contests, the ij and information sharing, consensus decision- withp =0otherwise.Letp bethefractionof fitness effects of surviving a contest, and the ij j making, collective restraints on would-be ag- demej’smembershipthatareA’s;pandp'bethe contributionofaltruiststosurviving. grandizers, monogamy, and other reproductive A-fractionofthemetapopulationinagivenand In every generation with probability k, each leveling practices that reduced within-group subsequentgeneration,respectively;andDp≡ demeengagesina“contest.”(Acontestmaybea differences in fitness may have attenuated the p'−p.Then,followingPrice(17)andassuming selectivepressurestowhichaltruistsaresubject themetapopulationsizedoesnotchange,wecan (7–11).Theimpactofintergroupcompetitionis express the possible evolution of altruism as Table 1. Payoffs to within-deme interactions. heightened by the fact that although group ag- summarized by Dp as a between-deme effect Entries are the payoffs of the row individual wheninteractingwithanindividualwhosetype gression is not uniquely human (12), among plusawithin-demeeffect: isgivenbythe column head. humansitisextraordinarilylethal(13). Dp=var(p)b +E{var(p )}b (1) j G ij i Altruist Not Santa Fe Institute, 1399 Hyde Park Road, Santa Fe, NM Altruist(A) 1+b−c 1−c Thetermsvar(p)andE{var(p)},are,respective- 87501, USA, and Universitá di Siena, 17 Piazza San j ij Not(N) 1+b 1 Francesco,Siena,Italy.E-mail:[email protected] ly, the between-deme and within-deme genetic www.sciencemag.org SCIENCE VOL314 8DECEMBER 2006 1569 RESEARCH ARTICLES hostile encounter or an environmental chal- To see how this works, suppose an N were Ifnislarge,thisisapproximatedby lengewithoutdirectdemeinteraction.)Demes instead an A. In the absence of reproductive that fail are eliminated, and surviving demes leveling,itsfitnesswouldbelessbyanamountc. F ð1(cid:1)tÞc ST > (6) repopulatethevacatedsites.Earlyhumandemes Buttheindividualwouldalsohavea1/nchanceof ð1(cid:1)F Þ k2l ST A probably faced frequent intergroup, environ- garneringthebenefitb,whichisdistributedran- mental, and other challenges resulting in oc- domlytomembersofthegroup.Additionally,by Like Hamilton's rule for the evolution of al- casional fatalities or territorial losses or gains increasingthechanceofsurvivalofthedeme(in truism by inclusive fitness, this model thus [more closely resembling boundary skirmishes which case, like every member of the surviving yields a condition indicating the minimum de- amongchimpanzees(22)thanthisall-or-nothing deme, it will be doubled), it also contributes gree of positive assortment necessary to allow deme-extinction scenario]. I show (13) that es- directly to its own fitness an amount equal to altruismtoproliferate.Theleft-handterm, like timatesoflong-termfitnesseffectsofcontinuous 1/n(i.e.,theeffectoftheswitchfromNtoAon Hamilton’s degreeof relatedness(r), is a mea- low-level losses or gains are equivalent to a p)timesb (theeffectofvariationsinp onthe sureofpositiveassortment;buthereassortment j G j complete extinction-repopulation scenario oc- averagefitnessofthedeme).Thus arisessolelyfrombetween-demedifferencesin curringinfrequently. theprevalenceofA's.Theright-handterminEq. Demesarethesamesize(normalizedto1), bi≡dwij∕dpij=−c+b∕n+k2lA∕n (3) 6istheratioofindividualcoststogroup-level exceptthatdemesthathaveoccupiedthesiteof benefits. We now ask if ancestral humans are an eliminated deme are momentarily of size 2 Reproductive leveling can now be intro- likely to havelivedunder conditions such that (and eliminated demes are of size zero). The duced by representing it as a convention, Eqs. 5 or 6 would be satisfied. Table 3 is a 6 surviving deme divides, forming two daughter conformity to which is in the interest of each summary of the main parameters and the 0 0 demesofequalsize.Lettheprobabilitythatthe deme member (25). Let some portion of the estimatedrangeofempiricallyplausiblevalues. 2 deme survives be l. The size of deme j in the payoffs initially acquired by an individual be Empirical estimates of FST. Wright [(27), 8, next generation is thus 1, 2, or 0 with distributed equally among all deme members. p.203]speculatedthatanequilibriumF among r ST e probabilities (1 − k), kl, and k(1 − l), re- Reproductivelevelingthentakestheformofa human groups—namely, that which would b m spectively,sotheexpectedsizeiswj=1−k+ proportional deduction at rate t of each mem- balance the offsetting effects of migration and e 2kl.TheeffectoftheprevalenceofA'sonthe ber’s payoffs, the proceeds of which are drift—mightbeabout0.02,avaluethatwould ec D expectedsizeofthedemeinthenextgeneration distributedequallytoallmembersofthedeme. precludeinterdemiccompetitionasanimportant (that is, bG ≡ dwj∕dpj) is the likelihood of a Theeffectistoreducewithin-demefitnessdif- evolutionaryforce.Butmostempiricalestimates on contest (k), times the effect on deme size of ferencesbetweenA'sandN'sfrom−cto−(1−t)c, areconsiderablylarger.Themeasuresofgenetic g r surviving or not (2), times the effect of the sob =−(1−t)c+b∕n+k2l ∕n. differentiation in Table 4 are from recent for- o i A g. prevalenceofA'sontheprobabilityofademe Positive assortment and the evolution of aging populations whose population structure, a survivingshouldacontestoccur(thatis,dl/dp ≡ altruism. Substituting these values for b and geographical and linguistic proximity, and m j i e l );sob =k2l .Thereisnowaytoestimate b inEq.1,wehave livelihoodmayresemblethoseofforagingbands c A G A G n l empirically, so I explore two alternative ofthelatePleistoceneandearlyHolocene(about e vAalues(13):lA=1isderivedfromamodelin Dp=var(pj)k2lA 150,000 to 10,000 years before the present). sci which all-A and all-N demes (respectively) These estimates are based on genetic material, w. surviveandfailwithcertaintyshouldacontest −E{var(pij)}{(1−t)c− (b+k2lA)∕n} most of which was collected before the mid- w w occur; whereas if lA = ½, an all-A deme (4) 1970s, and in most cases are averages over a m surviveswithprobability¾andanall-Ndeme large number of genetic systems and over F- o r surviveswithprobability¼. We will assess this condition with genetic data statistics among a large number of subpopula- d f Reproductive leveling. Distinctive human fromrecenthunter-gathererpopulations,usinga tions.Anestedthree-levelhierarchyofmeasures e d practiceswithingroupsalsocreatedafavorable commonly measured statistic from population ofgeneticdifferentiationisestimated,depending a o nichefortheevolutionofaltruism.Individual genetics,thefractionofthetotalgeneticvariance onthesizeofthesubpopulationunits(13):FDG nl differencesinsize,health,behavior,andother atalocusthatisbetweengroups,alsoknown measures genetic differentiation among demes w influences on access to scarce resources are asWright'sinbreedingcoefficient(26):FST≡ (D) in the same ethno-linguistic group (G), Do typically reflected in differences in reproduc- var(p)∕[var(p)+E{var(p)}].Usingthisdefini- whereas F and F , respectively, measure j j ij GT DT tive success. Among some primates (23, 24), tion,werewriteEqs.2and4andfindthattheA's differentiation among groups and demes in a and especially among humans, reproductive shareofthemetapopulationwillincreaseif metapopulation (T). If most competition is levelingattenuatesthisrelation.Becausealtru- betweendemesacrossethno-linguisticbounda- ists receive lower payoffs than other deme FST >(cid:1)bi ¼ð1(cid:1)tÞc(cid:1)b=n(cid:1)1 ries,thenFDTistherelevantstatistic. members (by the definition of altruism), they ð1(cid:1)F Þ b k2l n I think it is unlikely that Table 4 over- ST G A benefitfromreproductiveleveling,resultingin estimates the relevant degree of genetic differ- areductionofthetermb. (5) entiation among early humans. First, extreme i Table 2. Summary of model and notation. b and c: benefits and costs to deme members; p: percent of deme k that are A’s; and p: percent of k metapopulation that are A’s. Notation Eq.no. Equation Comment GenericPriceequation(PE) 1 Dp=var(p)b +E{var(p )}b Dp=betweendeme+withindeme j G ij i GenericPEconditionforAtoincrease 2 var(p)∕E{var(p )}≡F ∕(1−F )>−b∕b F ≡between-demevar∕totalvar j ij ST ST i G ST EffectofAondeme-averagefitness(b ) b ≡dw∕dp =k(dw∕dl)(dl∕dp)=k2l k=probabilityofinterdemiccontest G G j j j j A EffectofAonindividualfitness(b) b ≡dw ∕dp =−(1−t)c+b∕n+k2l ∕n t=extentofreproductiveleveling i i ij ij A ConditionforA’stoincrease(Priceequation) 4 Dp=var(p)k2l −E{var(p )}{(1−t)c−(b+k2l )∕n} Dp=between–deme+within-demeeffect j A ij A ConditionforA’stoincrease 5 F ∕(1−F )>−b∕b ={(1−t)c−b/n}∕k2l −1/n LargerF favorsA’s. ST ST i G A ST ConditionforA’stoincrease(ifn=V) 6 F ∕(1−F )>(1−t)c∕k2l >individualcost∕demebenefit ST ST A 1570 8DECEMBER 2006 VOL 314 SCIENCE www.sciencemag.org RESEARCH ARTICLES climate variability during the late Pleistocene an N can mimic the A's so as to evade their sus)sizeinthemostcomprehensivesurvey(13) (fig. S3) probably induced frequent deme ex- choosiness is to adopt the altruistic behavior is19.Individualbandsmayhavecompetedfor tinctions, population crashes, and subsequent itself and thus to bear its costs. Thus, the in- survival,but itislikelythat bands in coalition growth,resultinginthecolonizationofnewsites stability arising in the case of assortation by alsoengagedincontests.Aplausiblebenchmark bysmallpropagules.Naturalexperiments[e.g., “greenbeards”(34)doesnotarise. isthatademeisfivebands,givingn=32;Iwill withtheplantSilenedioica(28)]suggestthatthe But there is nonetheless an impediment to alsoconsiderverylarge(strictly,infinite)demes. effect may be a considerable elevation of selective assortment that is sometimes over- Plausiblevalues of c and b will depend on between-groupgeneticvariance.Second,genet- looked:ExclusionofN'sislikelytobecostlyfor theparticularaltruisticbehaviorinquestion.For ic differentiation among a subspecies of chim- the A's, whereas the associated benefits are example,awarningcallwouldhaveadifferentb panzees(Pantroglodytesschweinfurthii)whose sharedbyalldememembers.However,itisnot and c than defending the community against spatial distribution and demographic history implausiblethataltruistswouldundertakesome hostileneighbors.Tofacilitatetheexplorationof may resemble those of early humans (29) is moderatelevelofN-exclusionasacontribution avarietyofaltruisticbehaviors,Ipresentresults substantially higher than the median of the es- tothepublicgood.Thereisampleethnographic foragivenb=0.05andcvaryingfrom0to0.08. timatesinTable4(F =0.102). evidence(11)thatforagerspracticeselectiveas- (Eqs. 5 and 6 make it clear that for sizable ST However,geneticdifferentiationatthelocus sortmentwhentheyostracizeorshunindividu- demes,bisoflittleimportance.) ofanallelethatisexpressedinanaltruisticbe- als who violate behavioral norms. Models and Deme extinction. The extent of hostile haviormaydifferfromthatestimatedforneutral simulations(35)confirmthatthesepracticescan group interactions during the late Pleistocene loci(thosenotunderselection)suchasthosein proliferatewhenrareandpersistindefinitelyina and early Holocene may be suggested by cli- 6 Table4. First, an altruistic allele would be(by plausibleevolutionarydynamic.Moreover,itis matic data, hunter-gatherer demographics, ar- 0 0 definition) under directional selection. This readily shown (13) that a modest amount of chaeological evidence, and recent histories of 2 wouldbeexpectedtoreduceinterdemicgenetic selectiveassortmentgeneratessubstantiallevels foragingpeoples,andisamatterofsomedebate 8, differentiationatleastintheverylongrun,be- ofbetween-demedifferences. [theevidenceisreviewedin(13)]. r e cause in the absence of offsetting effects, the Within-demeselection.Althoughtheeffects We know from ice and deep-sea cores that b m frequencyoftheA’sinthepopulationwilleven- ofmostformsofreproductivelevelingcannotbe average temperature during the late Pleistocene e c tually go tozero. However, thistendencymay estimated, the degree of within-deme resource variedbyasmuchas8°Coverperiodsoflessthan e not work over time scales relevant to human sharing is known from empirical studies of the twocenturies—thedifferenceinaveragecontem- D n demes.Simulations(13)showthatevenforvery acquisition and consumption of nutrition among poraryannualtemperaturesbetweenCapeTown o strong selection against the A’s and for plau- foragers(13).Onthisbasis,Itaket=⅔asaplau- and Mombasa, 4000 km to the north (fig. S3). g r sible initial distributions of A’s in demes, the siblebenchmarkwith⅓analternativevalue(13). Mortalchallengesresultingfromclimaticadversi- o g. FSTrisesfortensofgenerations.Formoderate Theappropriatevalueofnisthenumberof tymusthavebeenfrequent,aswellasfromhostile a selectionagainsttheA’s,theF mayrisefor dememembersofabreedinggeneration(abouta interactions among groups migrating over un- m ST e morethanahundredgenerationsbeforefalling. thirdofthecensussize).Themedianband(cen- familiarterrainwithoutestablishedarrangements c n Becausefissionandextinctioneventsthaten- e hofanmceaginntieturddeemmicovrearfiarenqcueeanrtetlhikaenlythtois,beitaanpopredaersr Table4.Geneticdifferentiationamong13hunter-gatherersubpopulations(13).Themedianand w.sci thathighlevelsofF couldpersistindefinitely. mean values (respectively) are 0.076and 0.081. The median and meanfor the FDT estimates are w ST 0.081and 0.093. w Even with random fission (and relatively small m demes), additional simulations (25) show that Population Index F o r exceptionally strong directional selection against IndigenouscircumpolarEurasianpopulations F 0.076 d f theA's(c=0.1)iscompatiblewiththeindefinite DT e NativeSiberianpopulations F 0.170 d maintenanceofhighlevelsofFST. NativeSiberianpopulations FDT 0.114 oa Second,altruistswillsometimesbeableex- !Kungdemes(SouthernAfrica) FDG 0.007 nl cludenonaltruistsfromtheirdemes,resultingin DG w whatEshelandCavalli-Sforzacalled“selective SouthernAfricangroups FGT 0.075 Do assortment” (30, 31). This is particularly com- SouthernAfricandemes(from18groups) FDT 0.081 AboriginalAustralians F 0.042 mon when demes fission, a process Hamilton GT (32)called“associativetribesplitting.”Directed Kaiadilt-Lardiilgroups(Australia) FDT 0.081 Asmat-Mappi(LowlandWesternNewGuinea): F 0.056 migration(33)willalsoenhancebetween-deme DT Mbuti(CentralAfrica)–San(SouthernAfrica) F 0.149 variance and reduce within-deme variance. GT Aka(CentralAfricabetween“villages”inthesamegroup) F 0.042 Here,selectiveassortmentiscontingentonpast DG Aka (betweengroups) F 0.057 behaviorthatisitselfanobservableexpression GT Aka(between“villages”inallgroups) F 0.097 ofthealtruisticallele.Asaresult,theonlyway DT Table 3.Parameter estimates.Benchmark values are inbold.Entries not inbold are alternative values(d= 0.4 notused). Determinant Rangeexplored Comment/methodofestimation(13) Interdemicgeneticdifferentiation F 0.007–0.170;0.076 Geneticmarkers(recentforagers) ST Reproductiveleveling t 0.66,0.33 Foodsharing(recentforagers) Gains−lossesfromcontestspergeneration d 0.30,(0.40) Archaeologicalandethnographicevidence Per-generationprobabilityofadecisive(2,0)contest k =d∕2 Basedonestimatesofmortalityinongoingconflict Effectofpercentaltruistsondemesurvival lA 1/2,1 Arbitrary(seeFig.1) Effectivedemesize(one-thirdofcensussize) n 32,V Coalitionof5median-sizedbands Costtoaltruist c 0.0to0.08 Dependsonbehaviorunderconsideration Benefitstodememembers(withoutacontest) b 0.05 Asimmediatelyabove www.sciencemag.org SCIENCE VOL314 8DECEMBER 2006 1571 RESEARCH ARTICLES estimateincludesextinctionsinduceddirectlyby 5. G.C.Williams,AdaptationandNaturalSelection: climatechangeorothereventsunrelatedtowar. ACritiqueofSomeCurrentEvolutionaryThought (PrincetonUniv.Press,Princeton,NJ,1966). I use the smaller estimate of the frequency of 6. D.S.Wilson,Ed.,Am.Nat.150(suppl.)(1997). conflicts(k=1=7). 7. C.Boehm,J.Soc.Biol.Struct.5,413(1982). Discussion.Theaboveestimatesaresubject 8. I.Eibl-Eibesfeldt,J.Comp.Ethol.60,177(1982). tosubstantialerrorgiventhattheyareinferences 9. R.D.Alexander,TheBiologyofMoralSystems(Adinede Gruyter,NewYork,1987). aboutconditionsoccurringtensofthousandsof 10. H.Kaplan,M.Gurven,inMoralSentimentsandMaterial yearsagoforwhichverylittledirectevidenceis Interests:TheFoundationsofCooperationinEconomic available. With this caveat in mind, suppose Life,H.Gintis,S.Bowles,R.Boyd,E.Fehr,Eds.(MIT earlyhumans'demographicsandsocialpractices Press,Cambridge,MA,2005). resultedingeneticdifferentiationatthelocusof 11. C.Boehm,HierarchyintheForest(HarvardUniv.Press, Cambridge,MA,2000). analtruisticallelethatwasthemagnitudeofthe 12. J.H.Manson,R.W.Wrangham,Curr.Anthropol.32,369 median in Table 4 (F = 0.076). For the (1991). benchmarkvaluesoft,n,andl thesolidlines 13. SupportingmaterialsareavailableonScienceOnline. A, inFig.1givethecombinationsofcandksuch 14. R.Boyd,P.J.Richerson,J.Theor.Biol.145,331 (1990). that Eq. 5 is satisfied as an equality. More fre- 15. R.Boyd,H.Gintis,S.Bowles,P.Richerson,Proc.Natl. quent contests or less costly forms of altruism Acad.Sci.U.S.A.20,123(2003). (pointsabovetheline)allowaltruismtoprolifer- 16. J.Soltis,R.Boyd,P.J.Richerson,Curr.Anthropol.36, 6 ate. Dashed lines do the same for more strin- 473(1995). 0 17. G.R.Price,Ann.Hum.Genet.35,485(1972). 0 gentalternativeparametervalues.Forexample, 2 for the estimated k, if c = 0.05, altruism pro- 18. B1.9K,e1r3r,5P(.2G0o0d4f)r.ey-Smith,M.Feldman,TrendsEcol.Evol. 8, liferates (for both values of lA) under the 19. D.S.Wilson,G.B.Pollock,L.A.Dugatkin,Evol.Ecol.6, er Fig. 1. Conditions for the evolution of altruism benchmarkassumptions,butnotforverylarge 331(1992). b m byselectiveextinctionandreproductivelevelingif demeswithlimitedreproductiveleveling.Sim- 20. P.Taylor,Evol.Ecol.6,352(1992). e Fval=ues0.e0s7t6im.aTtheed isnoltidhelitneexst (anre=th3e2,bten=ch0m.6a6r)k. iplraerseanntaeldysiins(f1o3r).all of the data in Table 4 is 2221.. DAKm..AW.oJa.ktitP,sr,iEmMvoa.lt.MolIun6ltl.e8r,J,.1SO6.r1Agm.(2Es0lve0or6l,.)G.3.6M,b8a3b2az(i1,9J.82C)..Mitani, Dec Line1:n=V,t=0.33;line2:n=32,t=0.33; FormanyofthepopulationsinTable4and 23. S.Pandit,C.vanShaik,Behav.Ecol.Sociobiol.55,161 on line3:n=V,t =0.66.Pointsaboveeachline forplausibleparametervalues,then,geneticdif- (2003). g givecombinationsofcandksuchthataltruism ferentiation is such that even very infrequent 24. R.Noe,A.A.Sluijter,Int.J.Primatol.16,77(1995). or wouldproliferateaccordingtoEqs.5and6.(A) 25. S.Bowles,J.-K.Choi,A.Hopfensitz,J.Theor.Biol.223, g. lA=1/2;and(B)lA=1.Forbothpanels,b=0.05. cqounitteesctosswtlyoufoldrmhsavoefabletreunissmu.ffBiceiecnautsteothspereinaid- 26. 1S.35Wr(i2g0h0t,3)A.m.Nat.56,330(1922). ma tial spread of altruism among humans could 27. S.Wright,J.Cell.Comp.Physiol.235,187(1950). ce 28. B.Giles,J.Goudet,Am.Nat.149,507(1997). n for peaceful coexistence. Frequent catastrophic have been propelled by just a few of the vast e mortality is the most plausible way to reconcile number of late Pleistocene demes, the above 29. T(1.9G9o7ld).berg,L.M.Ruvolo,Mol.Biol.Evol.14,976 sci two facts about hunter-gatherer demography— dataandreasoningsuggestthatselectivedeme 30. I.Eshel,L.L.Cavalli-Sforza,Proc.Natl.Acad.Sci.U.S.A. w. namely,thathumanpopulationgrewextraordinar- extinction may be part of the account of the 79,1331(1982). w 31. D.S.Wilson,L.A.Dugatkin,Am.Nat.149,336(1997). w ilyslowlyornotatallforthe100,000yearsprior evolutionofaltruism.Thisislikelyinthepres- 32. W.D.Hamilton,inBiosocialAnthropology,R.Fox,Ed. m to20,000yearsbeforethepresent(36),yetmodels enceofappreciablelevelsofreproductivelevel- (Wiley,NewYork,1975),pp.133–155. o r and data on hunter-gatherer demographics show ing (and not in its absence), suggesting an 33. A.Rogers,L.B.Jorde,Ann.Hum.Genet.51,169 d f that they are capable of growth rates exceeding important role for culturally transmitted prac- (1987). e 34. M.Ridley,A.Grafen,Anim.Behav.29,954(1981). d 2%perannum(37). ticesincreatinganicheinwhichageneticpre- a 35. S.Bowles,H.Gintis,Theor.Popul.Biol.65,17(2004). o A few archaeological sites from the late disposition to behave altruistically might have 36. M.N.Cohen,inBiosocialMechanismsofPopulation nl Pleistocene suggest that exceptionally lethal evolved,andperhapsaccountingforthedistinc- Regulation,M.N.Cohen,RoyS.Malpass,H.G.Klein,Eds. w warfaretookplaceandthatviolenceintensified tiveaspectsofhumanaltruismnotfoundinother (YaleUniv.Press,NewHaven,CT,1980)pp.275–303. Do duringperiodsofclimaticadversityandresource species.Whetherrelatedprocessesofinterdemic 37. F.A.Hassan,inBiosocialMechanismsofPopulation Regulation,M.N.Cohen,RoyS.Malpass, stress (13). Deaths due to warfare constitute a competitionmightsupporttheevolutionofco- H.G.Klein,Eds.(YaleUniv.Press,NewHaven,CT,1980). substantialfractionofalldeathsamongforagers, operativebehaviorsintheabsenceofhighlyde- 38. E.O.Wilson,B.Holldobler,Proc.Natl.Acad.Sci.U.S.A. averaging 13% on the basis of archaeological veloped cultural transmission and cognitive 102,13367(2005). data(violentdeaths,tableS3)and15%onthe capacities [as has recently been suggested for 39. ThisresearchwassupportedbytheBehavioralSciences ProgramoftheSantaFeInstituteandtheRussellSage basisofethnographicstudies.Thisismuchmore euosocialinsects(38)]isanempiricalquestion Foundation.IthankR.Alexander,M.Alexander, thanforEuropeandtheUnitedStatesinthe20th thatremainstobeaddressed. C.Bergstrom,C.Boehm,R.Boyd,L.L.Cavalli-Sforza, century(lessthan1%ofmaledeaths).Territorial Nothing hereimplies that ageneticdisposi- J.-K.Choi,T.Clutton-Brock,A.Dreber,L.Dugatkin, lossorgainsduetowarfareamongasmallsam- tionfavoringhumanaltruismexists,orthatcul- H.Gintis,J.Goudet,H.Harpending,A.Hopfensitz,C.Huff, K.Hunley,H.Kaplan,L.Keller,L.Lehmann,J.Peck, pleofforaginggroupsaveraged16%pergener- tural or other possible explanations of human P.Richerson,A.Rogers,P.Roscoe,E.AldenSmith, ation.Basedonaveragesofthreelargesamples altruismareoflesserimportance.Theevidence T.Taylor,E.Szathmary,M.Tommaseo-Ponzetta, from the ethnographic record (table S4), war doessuggestthatifsuchadispositionexists,it M.vanVeelen,J.Wilkins,D.SloanWilson,andE.Woodfor was“rare”inonlyafifthofthehunter-gatherer may be the result of a gene-culture coevolu- theircontributionstothisresearch. societiesand“continuous”inathird. tionaryprocessinwhich,asDarwinwrote,group Ishow(13)thatthelevelofongoinghostility conflictplayedakeyrole. SupportingOnlineMaterial www.sciencemag.org/cgi/content/full/314/5805/1569/DC1 indicated by these data would produce fitness SOMText effectsequivalenttotheextinction-repopulation ReferencesandNotes Figs.S1toS5 scenariomodeledaboveoccurringeveryfiveto 1. C.Darwin,TheDescentofMan(PrometheusBooks, TablesS1toS4 sevengenerations,thelatterfigureignoringwar Amherst,NY,1998). ReferencesandNotes 2. D.deQuervainetal.,Science305,1254(2004). casualtiesandconsideringonlythedemographic 3. E.Fehr,S.Gaechter,Nature415,137(2002). 7September2006;accepted13November2006 effects of territorial losses and gains. Neither 4. J.Henrichetal.,Behav.BrainSci.28,795(2005). 10.1126/science.1134829 1572 8DECEMBER 2006 VOL 314 SCIENCE www.sciencemag.org Supporting online materials for Group competition, reproductive leveling and the evolution of human altruism Samuel Bowles, Santa Fe Institute and Universitá di Siena Contents: 1. Background 2. Associative tribe-splitting 3. Inter-demic genetic differentiation 4. Deme extinction and survival 5. Selection within demes 6. Dynamics of an F at a locus under directional selection st 7. Relaxing the carrying capacity constraint 8. Conditions for the evolution of altruism for forager-based estimates of F st 9. Figures S1. Hierarchical measures of genetic differentiation S2. Survival probability for deme j if half of the paired deme are A's S3. Pleistocene temperature variations S4. Dynamics of an F at a locus under directional selection. st S5. Conditions for the evolution of altruism 10. Tables S1. Associative tribe-splitting S2. Sources of estimates in Table 4 S3. Fraction of deaths due to violence: archeological evidence S4. Warfare in hunter-gatherer societies 11. Works cited 1. Background The main features of this model – reproductive leveling and deme extinction – have received considerable attention among biologists interested in social behavior. Haldane (1):210- 214 suggested that in a population of small endogamous “tribes,” an altruistic trait might evolve because the “tribe splitting” that occurs when successful demes reach a certain size would by chance create a few successor demes with a very high frequency of altruists. This would enhance the force of selective extinction by increasing between-deme genetic differences. Wright (2):114 similarly held that ‘isolation by distance’ would support “statistical differences among local populations...[that] provide a possible basis for intergroup selection of genetic systems, a process that provides a more effective mechanism for adaptive advance of the species as a whole than does the mass selection that is all that can occur under panmixia.” Like Wright and Haldane, Hamilton (8) remained skeptical that inter-demic competition would be a powerful evolutionary force; but he noted that if the assignment of members to successor demes following tribe-splitting was “associative” (p.137) rather than random, its importance would be enhanced. From classic early works to recent contributions, phenotype-based models abstracting from the details of diploid genetic transmission have played a central role in the literature on inter-demic selection (3-5). Closest to the model here is the selective extinction model of Aoki (6). Though based on an entirely different derivation (with an explicit analysis of the dynamics of the between- and within-deme variances), the condition for altruism to evolve is the same in the two models. His equation (6) is (S1) f > s(cid:39)(s+ k) 2 where f = F , s = the selection coefficient measuring the cost of the altruistic behavior (that is, c ST in the notation of this paper) and k is a measure of “intensity of group selection” which (because the average probability of survival is ½) is equal (cid:54)2(cid:56) . Using this equivalence and rearranging the A equation S1 we have (S2) F (cid:39)(1-F ) > s/k = c(cid:39)(cid:54)2(cid:56) ST ST A which is my equation (6) in the absence of reproductive leveling. By contrast to the selective extinction model used here, the selective emigration model (7) works because the average fitness of members of predominantly altruistic demes is above the meta-population average and thus they contribute disproportionately to the next generation. (This is possible because sites are assumed not to be saturated and demes can either accommodate or export increments to their population.) For simplicity of comparison with the selective extinction model assume that n is sufficiently large so that we can ignore terms in 1/n and that (cid:74) = 0.. Then (cid:36) = - c while the expected average fitness of members of deme j is w = 1 + p(b - c) so (cid:36) (cid:47) i j j G dw(cid:39)dp = b - c. Using these values and equation (1) the condition for the A trait to increase j j (analogous to equations 2 and 6) is (S3) var(p)(cid:39)E{var(p )} > - (cid:36)(cid:39)(cid:36) = c(cid:39)(b - c) j ij i G or, upon rearranging, (S4) var(p)(cid:39){E{var(p )} + var(p)} (cid:47) F > c(cid:39)b j ij j ST which reproduces Hamilton's rule for the evolution of altruism by inclusive fitness with r = F . ST Inspection of the estimates in table 4 shows that for this process to proliferate altruism, the benefits of altruism relative to the costs would need to be over ten for most groups. 2. Associative tribe-splitting. 3 An examination of fission among Amazonian tribal peoples ((8):198) reports that “fissioning ... keeps close kin together but separates them from more distant kin ... [T]he potential line of cleavage is furnished by the division in patrilineages.” Among two pairs of Yanomamo daughter villages formed by a fission resulting from hostile relations within the parent deme “the net effect of lineal fission is to reduce the effective size of the village at the time of fission by a factor of four, relative to expectation from random fission.” (9):179 (A reduction in effective size of this magnitude increases the level of equilibrium genetic differentiation by a factor of almost 4.) The authors conclude: “The process of village fission is strongly nonrandom socially, and results in pronounced genetic cohesion within, and great genetic differences between daughter villages.” 194. About two-thirds of the genetic differentiation among South American tribes has been estimated to be the effect of associative fission (10) rather than isolation and drift stressed in the equilibrium F formulations due to Wright (11). In these cases, genetic differentiation of ST demes is an unintended byproduct of lineal fission, as the alleles in question are not expressed in observable ways. For these alleles the between-deme genetic differentiation created by associative tribe splitting is already captured by the data in Table 4. But in the case of a behavioral trait such as altruism, one would expect deliberate associative fission and other forms of selective assortment, as altruists seek to exclude non-altruists from their demes. Because excluding an N may be costly to an A, and because the benefits of the exclusion are shared with all members of the deme, selective assortment will frequently be a form of public goods provision (that is, an n-person prisoners' dilemma). Gintis and I modeled the evolution of a form of individually costly selective assortment behavior (12), in which substantial numbers of individuals who behave altruistically and ostracize those who violate a norm of altruistic 4 behavior are sustained indefinitely, even when competing with cheaters who conform to the norm so as to avoid ostracism but do not contribute to the public good by ostracizing norm violators. The ethnographic literature suggests that selective assortment is common among some groups of foragers (see the works cited in (13) as well as (14)) To assess the effects on between-deme genetic differentiation of non-random deme fission we compare the expected between-deme variance resulting from associative fission and infinite deme size with that resulting from random fission and small deme size. Suppose an infinitely large “parent” deme splits into two “daughter” demes. We assume the secession is organized by a single head who seeks to compose a deme made up of individuals like himself, thereby setting aside the public goods aspect of selective assortment mentioned above and in the text. Because information about the past behavior of the members (the only basis of his selection of deme members) is noisy, with probability (1-r) he selects randomly from the parent deme and with probability r he selects one of his own type. Thus r is the assortment coefficient for associative fission. If the departing head is an A, and p is the fraction of A's in the parent deme, the daughter deme will be composed of a fraction of A’s approximately equal to r + (1-r)p. The fraction of A's among those remaining after the departure of the first daughter deme will be 2p - r- (1-r)p. If the departing head is an N, the fraction of A's in his deme will be p - rp, and the fraction of those remaining who are A's is 2p - p + rp. This process is summarized in Table S1 [Table S1 here] If the departing head is drawn at random, he will be an A with probability p and the variance between the daughter deme (1) and the remaining deme (2), denoted var(p; r, (cid:52)), j (meaning the variance of the deme means p (j (cid:48) [1,2] ) given r and assuming infinite deme j 5