Reciprocal altruism, rather than kin selection, maintains nepotistic food transfers on an Ache ...

14 Pages·2008·0.47 MB·English

EvolutionandHumanBehavior29(2008)305–318
Reciprocal altruism, rather than kin selection, maintains nepotistic food
☆
transfers on an Ache reservation
⁎
Wesley Allen-Aravea, , Michael Gurvenb, Kim Hillc
aDepartmentofAnthropology,UniversityofNewMexico,Albuquerque,NM87131,USA
bDepartmentofAnthropology,UniversityofCaliforniaSantaBarbara,SantaBarbara,CA93106,USA
cSchoolofHumanEvolutionandSocialChange,ArizonaStateUniversity,Tempe,AZ85287,USA
Initialreceipt9December2006;finalrevisionreceived24March2008
Abstract
Cooperationamongrelativesisoftenregardedasevidenceofkinselection.Yetaltruismnotrequiringsharedgenescanalsoevolveamong
relatives. If characteristics of relatives (such as proximity, familiarity, or trust) make kin preferred social partners, the primary causes of
nepotisticbiases mayresideprincipallyindirectfitnesspayoffsfromcooperation ratherthan indirectfitnesspayoffsacquiredfromaiding
collateralkin.WeconsidertherolesofkinselectionandreciprocalaltruisminmaintainingnepotisticfoodtransfersonanAchereservationin
northeasternParaguay.Householdsdonotprimarilydirectaidtorelatedhouseholdsthatreceivelargercomparativemarginalgainsfromfood
intakeaswewouldpredictunderkinselectiontheory.Instead,(1)foodtransfersfavorhouseholdscharacterizedbylowerrelativenetenergy
productionvaluesirrespectiveofkinshipties,(2)householdsdisplaysignificantpositivecorrelationsinamountsexchangedwitheachother,
suggestingcontingencyinfoodtransfers,and(3)kinshipinteractswiththesepositivecorrelationsinamountshouseholdsexchangewitheach
other,indicatingevenstrongercontingencyinsharingamongrelatedhouseholdsthanamongunrelatedhouseholds.Whilekinarepreferred
recipientsoffoodaid,fooddistributionsfavorkinthathavegivenmoretothedistributinghouseholdinthepastratherthankinthatwould
benefitmore from the aid.Such discrimination amongkin accordsbetter with reciprocal altruism theory than with kin selection theory.
©2008Elsevier Inc. All rightsreserved.
Keywords: Nepotism;Kinselection;Reciprocalaltruism;Resourcesharing;Humanbehavioralecology;Ache
1. Introduction to inclusive fitness benefits without a careful consideration
ofalternatives.However,pathwaystoaltruismnotrequiring
Behavioralstudiesdemonstratethatindividualsinsmall- shared genes can lead to increased levels of cooperation
scalesocieties preferentiallyaidclose kinovermoredistant among relatives over nonrelatives if kin possess character-
kin and nonkin (e.g., Betzig, 1988; Betzig & Turke, 1986; istics that are preferredin social partners.
Chagnon, 1981; Chagnon & Bugos, 1979; Flinn, 1988; We examine the roles of indirect fitness impacts and
Gurven, Hill, Kaplan, Hurtado, & Lyles, 2000b; Hames, reciprocalexchangesinmaintainingnepotisticfoodtransfers
1987; Hawkes, 1983; Patton, 2005). Such nepotistic biases among reservation-living Ache forager-horticulturists of
are often cited as evidence that indirect fitness payoffs northeastern Paraguay. We previously reported that Ache
(Hamilton,1964;MaynardSmith,1964)haveshapedhuman householdsgivepreferenceinfooddistributionstorecipient
social interactions. Kin selection theory is so elegant and households that contain at least one close relative (Gurven,
appealingthattheoristsoftenattributeinstancesofnepotism Allen-Arave,Hill,&Hurtado,2001).Thisnepotisticbiasin
foodtransfersfollowslinesofgenealogicalrelatednessrather
☆ The National Science Foundation (grant no. 9617692), L. S. B. thanlinesofAchesocialkinshipterminology(Allen-Arave,
LeakeyFoundation,UniversityofNewMexicoStudentResearchAlloca- Gurven,Hill,&Hurtado,1999).Theoristshaveusedsimilar
tionsCommittee,andUniversityofNewMexicoOfficeofGraduateStudies resultsfromotherpopulationstoarguefortheimportanceof
ResearchProjectandTravelgrantsprovidedsupportforthisresearchandits
indirect fitness payoffs in patterning human social interac-
dissemination.
⁎ tions.Yet,ourpreviousreportalso reveals thatevenamong
Correspondingauthor.
E-mailaddress:[email protected](W.Allen-Arave). householdslinkedbyaclosekinshiptie,theamountoffood
1090-5138/$–seefrontmatter©2008ElsevierInc.Allrightsreserved.
doi:10.1016/j.evolhumbehav.2008.03.002
306 W.Allen-Araveetal./EvolutionandHumanBehavior29(2008)305–318
any household D (donor) transfers to any household R The ages of household members matter as much as the
(recipient) is correlated with the amount household D numberofresidentsahouseholdcontainsfordeterminingthe
receivesfromhouseholdR(Gurvenetal.,2001).Weexpect fitness impact a transferred unit of food may have for a
such a result if returns from reciprocation provide the household because energy requirements and reproductive
adaptive payoffs of the transfers but not if nepotistic values peak in young adulthood. Resting metabolic energy
investments in indirect fitness benefits provide the adaptive expenditure rates indicate that individuals aged from their
payoffsofthetransfers.The presenceofboth nepotism and late teens to fifties require more energy than younger and
correlated amounts of food transferred between related older individuals do (National Research Council, 1989a;
households challenges us to disaggregate the relative World Health Organization, 1985). Young adults also
contributions of indirect fitness impacts and reciprocal possess a larger potential to translate food energy into
benefits in maintaining nepotistic Ache food transfers. The inclusive fitness gains than other age classes, owing to the
present paper presents new analyses to (1) examine the greater number of childbearing years likely to await young
directionofimbalancesinfoodtransfersbetweenhouseholds sexually mature and about-to-mature individuals in the
and (2) consider the difference in net caloric production future. Thus,individuals inthemiddleofthelifecourse can
betweenhouseholds. returngreaterindirectfitnessbenefitstodonorkinfromlarge
amounts offood than younger and older individualscan.
1.1. Kin selection theory
Despitestraightforwardtheoreticalexpectationsthatfood
flows should favor individuals of high reproductive value,
Researcherscommonlypredictfrom kinselectiontheory
application of this logic to human populations presents
thataltruisticaidwillpositivelycorrelatewiththedegreeof
complications. Several theorists (e.g., Charlesworth &
relatedness between interactants. Yet, kin selection theory
Charnov, 1981; Hamilton, 1964; Rogers, 1993; Taylor &
does not presume that individuals should always act
Frank,1996;Trivers,1971)havenotedthatthereproductive
altruisticallytowardallrelatives,norshouldtheynecessarily
value of donors and recipients should alter the costs and
share mainly with close relatives. Mathematical models
benefits of giving and receiving aid. However, measures of
illuminate that natural selection can favor nepotistic acts
reproductive value do not provide an adequate estimate of
when the benefit to the recipient, B, discounted by the
the expected inclusive fitness contribution made by indivi-
coefficientofgeneticrelatedness,r,isgreaterthanthecostto
duals in species, such as ours, with child altriciality and
the provider, C: BrNC (Hamilton 1964). Whenever a
common allocare. While prereproductive and postreproduc-
household can obtain higher inclusive fitness payoffs by
tive individuals cannot produce copies of their genes in the
hoarding resources rather than providing them to relatives,
formofoffspring,theyregularlyassistcopiesoftheirgenes
kin selection theory suggests that no transfer will occur.
located in other relatives through activities such as baby-
Likewise, when distant relatives obtain a much larger
sitting (Bock, 1995, Fig. 57; Ivey, 2000; Turke, 1988;
positive fitness impact than close kin from assistance, kin
Weisner&Gallimore,1977),passingonimportantskillsand
selectiontheorypredictshigherratesoftransfertodistantkin
knowledge (Biesele & Howell, 1981; Liederman & Lieder-
thantoclosekin.Thus,anevaluationofkinselectiontheory
man, 1977), provisioning during times of need (Hawkes,
must consider not only relatedness, but also the costs and
O'Connell, & Blurton Jones, 1997), or offering protection
benefitsof aid.
andsupport(Chagnon&Bugos,1979).Theexpectedfitness
1.2. Direction and magnitude of imbalances contributionmadebyindividuals—especiallypostreproduc-
tive individuals—is therefore underestimated by reproduc-
If nepotistic transfers constitute investment in indirect tivevaluemeasuresalonebecausedirectreproductionisnot
fitness, the direction and magnitude of imbalances within the only way to increase inclusive fitness. Still, food
dyads of related households should attend to (1) the requirements and fertility measures alike indicate that food
capability of household members to produce food calories, transfers, which enhance inclusive fitness, should predomi-
(2)thenumberofhungrymouthsahouseholdcontains,and nately favor households containing young reproductive-
(3)theagesofhouseholdresidents.Allofthesefactorsaffect agedresidentsoverhouseholdscontainingotherageclasses,
the marginal gains of food intake on household summed when we control for the number of residents and their
reproductive value (Fisher, 1958). Given the reasonable production abilities.
assumption that the curve relating food intake to fitness is
1.3. Reciprocal altruism
negatively accelerated, kin selection theory implies that
imbalances in food transfers between related households Any valid evolutionary explanation accounting for
should favor households that produce less food over exchanges between nonkin may also apply to economic
households that produce more food, when we hold other interactions between kin. Thus, we should never a priori
factors constant. Holding all else constant, kin selection assume that cooperation among kin results from inclusive
theory also implies that imbalances between related house- fitness benefits to the exclusion of other pathways to
holds should favor households with more mouths to feed cooperation. We now consider the role reciprocal altruism
over households withfewer mouthsto feed. mayplayinfoodexchangesamongrelatives.
W.Allen-Araveetal./EvolutionandHumanBehavior29(2008)305–318 307
Reciprocal altruism (Trivers 1971)canevolveaslongas Pygmies (Gurven, 2004), Dolgan/Nganasan (Ziker, 2005),
the cost of aiding another individual is outweighed by the Hiwi (Gurven et al., 2000b), Mikea (Tucker, 2004), Pilaga
benefit of receiving aid from that individual later, devalued (Gurven, 2004), Yanomamo (Hames, 2000), and
bythe probabilitythat aidwill be returned(Boyd, 1990).If Ye'kwana (Hames & McCabe, 2007). However, amounts
thereciprocalexchangeisprofitable,individualaltruistscan of food given to all others does not correlate with
expectpaybackinthefuturefromself-interestedactorswho amounts of food received from all others for Hadza large
wish to continue obtaining the benefits that accrue from game (Hawkes, O'Connell, & Blurton Jones, 2001) nor
long-termcooperation.Suchcooperationisevenmorelikely Meriam turtle meat exchanges (Bliege Bird, Bird, Smith,
toappearwhenpunishmentofdefectorsispossible(Fehr& & Kushnick, 2002).
Gächter, 2002; Ostrom, Walker, & Gardner, 1992; Yama-
gishi1986).Nepotismmayemergeindependentofinclusive
1.5. Reciprocity among kin
fitness benefits if individuals find their relatives more
desirable as reciprocal exchange partners than nonrelatives. Individualsmaypreferclosekintodistantkinandnonkin
as partners for reciprocity. Relatives can make ideal
candidates for reciprocal exchanges due to factors such as
1.4. Contingent reciprocity
familiarity, trust, proximity, a high probability of future
Sincereciprocalaltruismprovidesexchangepartnerswith interaction, or an expectation that relatives will cooperate.
the temptation to defect by accepting the benefits of their When choosing among potential reciprocity partners,
partner'saltruism,withoutlaterpayinganycostsofaltruistic individualsshouldgenerallypreferpartnerswhowillprovide
actsthemselves,reciprocalaltruistsmustidentifyandpunish thehighestexpected return benefit. Due toadditiveindirect
oravoidfreeriders.Forthisreason,HillandKaplan(1993) fitnessbenefitsontopofthedirectbenefitsthatcollaborators
argue that reciprocal altruism makes a central prediction of gain from cooperation, reciprocal exchanges with relatives
“contingency” in food exchanges. Hill & Kaplan define will often yield larger expected return benefits than
contingency as giving that is conditional upon expectations reciprocal exchanges with nonkin.
of future receiving, where individuals infer expectations of Familiarity and emotional bonds fostered over time may
future receiving from prior sharing patterns. In modern makeclosekineasierto“read”andtrustthandistantkinand
societies, despite legal enforcement of reciprocity, contin- nonkin.Would-betransgressorslikelyexperiencemoreguilt
gency is implemented through practices such as credit from cheating victims with whom they have emotional ties
checks.Onlythosewhohavemetobligationstorepayinthe (Frank, 1988). Furthermore, individuals may have good
past are provided current goods and services with the reason to trust close kin over other potential exchange
expectation ofrepaymentin thefuture. partnersbecauseindirectfitnesscostsmakecheatingaclose
Contingent reciprocity may prove difficult to detect relativelessprofitablethancheatingnonkin.Ifanexchange
because reciprocal altruism does not imply perfectly partner doesfailtoreciprocateduetodeliberatecheatingor
balanced exchanges between individuals. For an exchange an inability to repay (as can occur with a move, injury, or
to occur, reciprocal altruism theory predicts only that the death), the loss is not complete for a slighted relative who
average utility of the expected return outweighs the utility still receives an indirect fitness benefit from their non-
of the resource given up today. For example, a satiated reciprocating relative's gain. Therefore, individuals assume
household may pay little cost in providing (say) 2000 lessriskininitiatingreciprocalexchangeswithrelativesthan
calories now, while benefiting greatly from (say) 500 with nonrelatives.
calories at a future date when household members are ill Finally, the close proximity that kin often maintain can
or hungry. This may be analogous to the logic of create more opportunities for exchange and increase the
insurance coverage in which one may pay premiums at probability of future interaction, which promotes coopera-
a low utility cost for years in order to cover any high tion(Andreoni&Miller,1993;Axelrod&Hamilton,1981).
utility needs in the event of a future catastrophic shortfall. At our study site, the homes of households joined by a
Additionally, if households engage in reciprocal exchanges close kinship tie tend to be nearer to each other than the
that include several goods and services, an evaluation of homes of households not joined by a close kinship tie
food exchanges alone may underestimate the true (Gurven etal., 2001).
contingent reciprocity occurring in the society. Despite The factors discussed above (familiarity, emotional
these complications, we expect to find that the amount of bonds, trust, proximity, and indirect fitness costs and
food provided by any household D to any household R benefits) can promote an expectation among kin that a
will correlate with the amount of food provided by relativewillcooperate,andexperimentalresearchhasshown
household R to household D if reciprocal altruism plays a that expectations of cooperation promote and stabilize
role in food transfers over the time scale of observation. altruistic behavior (Dawes, 1980; Messick & Brewer,
Researchers studying other forager-horticulturalists popu- 1983).Thus,we might expect close kin toprovide frequent
lations have found dyadic correlations in food shares goods and services in a stable arrangement of reciprocal
among the Achuar/Quichua/Zapara (Patton, 2005), Aka altruism ratherthan simple kin-directed charity.
308 W.Allen-Araveetal./EvolutionandHumanBehavior29(2008)305–318
2. Study case: the Northern Ache 3. Research methods
The Northern Ache are indigenous peoples of lowland
3.1. Measurements
northeastern Paraguay. At peaceful contact in 1971, they
subsisted as nomadic hunter-gatherers. Beginning in the Allen-Arave and Gurven sampled food production and
mid 1970s, theNorthern Ache began residing inpermanent distribution from February to May 1998 through random
horticultural settlements. Yet, they continue to utilize the 3-hour time block observations of two to three mutually
forests around their settlements and spend up to half of visible neighboring households. We observed every house-
their time on extended foraging trips (McMillian, 2001). hold in Arroyo Bandera for a total of 51–60 h using this
Characteristics of Northern Ache forest life, including method. We made weight measurements of resources with
resource acquisition, time allocation, food sharing, life spring scales whenever possible. When weighing an item
history theory, and group composition have been exten- provedunfeasible,wemadeanumericcountoftheitemand
sivelystudiedoverthepast25years(Hill&Hawkes,1983; converted the measurement into kilograms using weight
Hill, Hawkes, Hurtado, & Kaplan, 1984; Hill & Hurtado, measurements we obtained from large trials of counted
1996; Hill, Kaplan, Hawkes, & Hurtado, 1985; Hurtado, resources. We subsequently converted kilogram measure-
Hill, Hawkes, & Kaplan, 1985; Kaplan & Hill, 1985; ments into calorie equivalents using conversion estimates
McMillian, 2001). Aspects of Northern Ache reservation obtainedfromtheFoodCompositionTableforUseinLatin
life have been the focus of a few recent studies (Gurven et America (The Institute of Central America & Panama and
al., 2001; Gurven, Hill, & Kaplan, 2002; Hawkes, Kaplan, The Interdepartmental Institutes of Health, 1961) and from
Hill, & Hurtado, 1987). lab analyses on food samples conducted at Hill's request in
Thefoodtransfersobservedinthisstudyoccurredonthe the early 1980s. The sample consists of 380 complete food
Arroyo Bandera reservation where 117 permanent residents distributions,forwhichweobservedtheconsumptionofthe
arranged into 23 households resided during the 1998 study entirefoodpackage, and635incompletefooddistributions,
period. Small wooden dwellings (typically about 4×4 forwhichsomedistributionoftheresourceoccurredoutside
meters) allow Ache families to store goods, but caches of ofthe observation block.
food are rare. The Ache lack technology for refrigeration, We define households as married adults and their
freezing, drying, or canning and customarily consume dependents and treat households as the unit of observation.
resources shortly after harvesting them. We only observed Withinahousehold,residentsofteneatfromacommonplate
occasional caching of purchased goods (sugar, hard bread and freely pass food items back and forth. The intensity of
rolls,rice,andnoodles)obtainedfrominfrequentwagelabor food sharing within a household likely means that when
opportunities and large bundles of harvested peanuts that donorssendfoodsharestoaspecifichouseholdtheycannot
familymembersandvisitorswouldsnackonthroughoutthe targetonedesiredrecipientbut,instead,expectallmembers
day. These infrequent caches are likely to be known of the recipient household to consume portions (Hames,
throughout the community given the close proximity of 1987; Kurland, 1979). There are 23 households in this
dwellingssituatedanaverageof21metersapartandarrayed, sample constituting 253 householddyads.
in a circular pattern, with entries that are visible from most We calculate the coefficients of relatedness between
other households inthe community. all individuals residing at Arroyo Bandera during the
The Ache prize food sharing and remark with pride on study period using genealogical data obtained by Hill &
their culture's ethic of generosity. Traditional food sharing Hurtado (1996) over the past 29 years through observing
norms exhort group members to share with all present, to pregnancies and co-habitation patterns, frequent censuses,
give to those in need, and to refrain from excessive and from retrospective interviews covering reproductive
personal consumption of large package resources that one histories and family genealogies. We code kinship
produces. Nearly all food preparation and cooking is done between any two households, D and R, as the average
in plain view of other community members around open coefficient of relatedness of each member of household D
fires located in front of the doorway to each family's to each member of household R. This measure takes into
respective house. From any given house, one can see account the interests of all members of a household. Ache
nearly half of the other open fires in the community. An men, women, and children alike routinely distribute shares
ample depth (measured as the percentage of production to recipient households and it is rarely clear from simple
transferred to other households) but restricted breadth observation as to who within the donor household
(measured as the number of other households that receive ultimately initiates a transfer. Adults likely have more
a portion) characterizes food transfers at Arroyo Bandera. power than children do in determining which households
A typical household at Arroyo Bandera distributes just receive shares of their own household's production. Yet,
over eight food items per day, keeps only 20–30% of even children as young as 5 years were often observed to
the food it acquires, and shares the remaining 70–80% share “leftovers” and “snacks” from their homes with
with two to three other households, on average (Gurven members of other households when their parents were not
et al., 2001). home to sway the transfer. Because we are unable to
W.Allen-Araveetal./EvolutionandHumanBehavior29(2008)305–318 309
assert that any particular class of individuals is powerless families in our sample (Pearson r=.88, pb.0001). Age-
in influencing which households receive shares of their specific food consumption values, however, have the
own household's production, we consider the average advantage of not discounting postreproductive individuals
relatedness of all members of household D to all members as rapidly as age-specific reproductive values and thus
of household R. provideamorebiologicallyrealisticmeasureofthepositive
With the relatedness measure employed here, the house- influencesanindividualcanhaveoninclusivefitnessateach
holds of two full brothers who each live with an unrelated age (see Section1.2).
mate attain a relatedness value of 0.125. This measure In order to assess “need,” we calculate net caloric
remainsunaffectedbytheadditionofchildrentoahousehold production for each household by subtracting the house-
as long as the man of the household fathered the children. hold'sstandardizeddailyconsumptionrequirementfromthe
However,thepresenceofstepchildrenmaylowerorincrease household'sstandardizedobservedfoodproductionoverthe
relatedness between any two given households. The mean sample period. We divided consumption and production
relatedness between dyads of households in our sample is figures by their mean responses to obtain two standardized
.02±.04. Four dyads have an average relatedness of .25 estimates in order to contrast these two figures of different
between them, which is the highest average relatedness time scales in a single measure. As the observed food
betweenanydyadsinthissample;152householddyadshave productionofahouseholdincreases,sodoesthehousehold's
no genealogical kinship ties joining them. For graphic netcaloricproductionvalue.Asthenumberofconsumers—
displays, we divide dyads into four groups to create the andparticularlyconsumersagedfromtheirlateteenstoearly
ordinal categories of “close” kin (rN.05, n=34), “near” kin 50s—in a household increases, the household's net caloric
(.018brb.047, n=34), “distant” kin (0brb.018, n=33), and production value decreases.
nonkin (r=0, n=152).
3.2. Data analysis
To measure disparity in amounts of food exchanged
between any two households, we follow Hames (1987) and Table 1 provides summary statistics and a description of
calculate “specific imbalance” as the number of calories the variables we use in the present study. To account for
transferred from household D to household R minus the biases in variances, degrees of freedom, and significance
number of calories transferred from household R to tests that would result from ignoring the nonindependence
household D over the sample period. As the inequality in inherentindyadicdata(Kenny,1995;Kenny&Judd,1986),
exchange between two households increases, the specific we perform statisticalanalysesatthelevelof thedyad.The
imbalance measuring their exchanges attains values further information supplied by each household in a dyad is often
from 0, achieving positive values if the imbalance favors redundant with the information supplied by the other
household D and negative values if the imbalance favors household in the dyad since each household is paired twice
household R. witheachoftheotherhouseholdsinthecommunity,onceas
We sum age-specific daily food consumption estimates adonorandonceasarecipient.Foranalysesinwhichthere
across all members of a household to obtain a measure of is a theoretically meaningful ordering of household roles
each household's daily consumption requirement. Kaplan (e.g.,testingwhetherimbalancesfavorthehouseholdwithin
(1994) calculated age-specific consumption estimates from each dyad characterized by the lower net caloric value), we
restingmetabolicenergyexpenditureratesbyageandsexfor assigneachhouseholdwithinadyadtotheroleofdonoror
the Ache following a method used by the World Health recipient and consider each household dyad once in
Organization (1985) and the National Research Council theanalysis.
(1989b).Whensummedacrossallmembersofahousehold, For analyses in which dyad members lack theoretically
thismeasurecapturesboththenumberofindividualsandthe distinguishable roles (e.g., testing whether the number of
age of each individual within a household. The mean calories transferred from household D to household R
summed consumption requirement for a household in our covaries with the number of calories transferred from
Achesampleis7706cal/daywitharangefrom3502cal/day household R to household D), we employ multilevel
(an elderly couple with no dependents) to 14,643 cal/day regression modeling to recognize the hierarchical structure
(a middle-aged couple supporting their five young children of our data with households nested within dyads. The first
andtwounrelated teenagers). step in multilevel data analysis entails fitting a simple two-
Age-specific food consumption estimates are highly levelmodelwithoutanypredictors,whichisoftencalledan
correlatedwithage-specificreproductivevalues1fornuclear “unrestricted”model.Thismodelisanalogoustoaone-way
randomanalysisofvariancemodel.Next,weaddpredictors
to subsequent multilevel models to evaluate how well
the predictors model caloric transfers. Reductions in the
Pl 1 Reproductive value, VA, for individual A was calculated as: VA¼ modeled variance of subsequent models over that in the
lyme(cid:1)rðy(cid:1)Aþ1Þ;wherel issurvivorshipfromage0toA,m isfertility
ay¼tAaglAeAy,andristheinstantAaneousgrowthrateofthepopulationA.Estimates unrestricted model indicate how well the added predictors
explain variation in caloric transfers. We estimate the
ofl m ,andrforpost-contactAchewereobtainedfromHill&Hurtado
A, A
(1996). proportion of variance explained by a within-dyad
310 W.Allen-Araveetal./EvolutionandHumanBehavior29(2008)305–318
Table1
Summarystatisticsanddescriptionofvariables
Valuea Typeb Description
Classificationvariables
Household 23 c,w HouseholdID
Dyad 253 c,b Multipledummiestorepresentdyadmembershippairingeachhouseholdwitheveryotherhousehold
onceasthedonor(D)householdandonceastherecipient(R)household
Dependentvariables
Specificimbalance 1636.74±(6033.3)c i,b (Calories transferred from household D to household R)−(calories transferred from household R to
householdD)c
Caloriestransferred 2298.85±(4874.84) i,w CaloriestransferredfromhouseholdDtohouseholdR
Independentvariables
Relatedness 0.02±(0.04) i,b AveragecoefficientofrelatednessofeachmemberofhouseholdDtoeachmemberofhouseholdR
Need 0.75±(0.68) i,w DifferenceinnetcaloricproductionofhouseholdDandhouseholdR,wherenetcaloricproduction=
(food production−the age-specific food consumption estimate summed across all members of the
household)
Kin-directedaltruism 0.01±(0.04)c i,w Relatedness(asmeasuredabove)×need(asmeasuredabove)interactionterm
Kin-favoredreciprocity 253 c,b Dyad(asmeasuredabove)×relatedness(asmeasuredabove)interactionterm
a Summaryvalueisthesamplesizefornominalclassificationvariablesandthemean±S.D.forintervalvariables.
b Variabletypeis:nominalclassification(c),interval(i),within-dyad(w),between-dyad(b).
c Forthismeasure,householdsarearrayedindyadssothatthehouseholdwiththelargernetcaloricproductionvalueisassignedtheroleofdonor(D)andthe
householdwiththesmallernetcaloricproductionvalueisassignedtheroleofrecipient(R).
explanatory variable w and a between-dyad explanatory 4. Results
variable b, respectively, as:
4.1. Kin selection
(cid:1) (cid:2) (cid:1) (cid:2)
r2(cid:1)r2 s (cid:1)s AveragerelatednessbetweenallmembersofhouseholdD
Pseudo−R2 ¼ u c and Pseudo−R2 ¼ 00ju 00jc to all members of household R significantly predicts the
1 r2 2 s
u 00ju amountofcaloriesprovidedbyhouseholdDtohouseholdR
ð1:1 and 1:2Þ overthesampleperiod(t =3.52,p=.0005).Thisnepotistic
251
bias is consistent with kin selection theory, but it does not
where σ2 is the within-dyad variance for the unrestricted offer sufficient evidence to conclude that the adaptive
u
model, σ2 is the within-dyad variance for the conditional function of these food transfers result from indirect fitness
c
model with predictor w, τ is the between-dyad variance benefits. Rigorous tests of kin selection theory require that
for the unrestricted mode0l0,|uand τ is the between-dyad we look beyond a simple statistical tendency to share with
00|c kinandattendtothecosts,benefits,anddirectionalityofaid.
variance for the conditional model with predictor b
Toinvestigatewhetherthelikelyfitnesscostsandbenefits
(Raudenbush & Bryk 2002). Information-theoretic methods,
of aid determine the direction of imbalances, we array
such as Akaike's information criterion (AIC), allow us to
households within dyads so that specific imbalance values
compare models with different predictors to determine
reflectthenetcalorictransfersurplus(ordeficit)affectingthe
which potential predictors best describe the data. Lower
householdinthedyadwiththelowernetcaloricproduction
AIC values indicate a model that better optimizes the
value.Withthisassignmentofhouseholdroles,theslopeof
tradeoff between underfitting and overfitting the data
theregressionofspecificimbalancebynetcaloricproduction
(Burnham & Anderson 2002). Multilevel modeling also
should be positive if food transfers favor households
allows computation of the intraclass correlation, which
according to need. Table 2 shows the results of a multiple
provides a measure of correspondence in the number of
calories households transfer to each other. We compute the
intraclass correlation as:
Table2
s
r ¼ 00 ð2Þ Multipleregressionanalysisofspecificimbalanceincalorictransfersfrom
I s00þr2 householdDtohouseholdR
Parameterestimatea
where τ is the between-dyad variance in caloric transfers
00 Independentvariable
and σ2 is the within-dyad variance in caloric transfers. To Relatedness −0.1298
the extent that there are no other sources of nonindepen- Need 0.2882⁎
dence among household dyads in the amount of calories Kin-directedaltruism 0.1310
they transfer to one another, the intraclass correlation a Parameterestimatesarepartialstandardizedestimates.
provides a measure of dyadic contingency. ⁎ Significantatthe.01level.
W.Allen-Araveetal./EvolutionandHumanBehavior29(2008)305–318 311
Fig.1.Linearregressionoftherelationshipbetweenthedifferenceinnetcaloricproductionbetweendyadsofhouseholdsandspecificimbalanceintheirfood
transfers(arrayedsothatpositivespecificimbalancevaluesareattainedwhenanimbalancefavorsthehouseholdwiththelowernetcaloricvalue).Plotsfor(a)
closekin(rN.05),(b)nearkin(.018brb.047),(c)distantkin(0brb.018),and(d)unrelateddyads(r=0).
regression model of the association between relatedness, holds (t =1.38, p=.1688). Even kinship alone is not
252
need(Dnetcaloricproduction−Rnetcaloricproduction),and associated with imbalances in food transfers on the Ache
specific imbalance in caloric transfers. The model also reservation (t =−1.41, p=.1594).2 The only significant
252
includes a term for the interaction of relatedness and predictor in the model is the difference in net caloric
difference in net caloric production (kin altruism). This production between the households (t =4.41, pb.0001).
252
interaction term should have a strong positive association Imbalances in food transfers tend to favor the household
withimbalancesinfoodtransfersiffooddistributionsserve within each dyad characterized by the lower net caloric
the function of increasing indirect fitness contributions. production value regardless of relatedness. This indicates
Under kin selection theory, neither relatedness alone nor that something other than indirect fitness impacts leads
difference in net caloric value alone should significantly “richer” households to provide more of their production to
predict specific imbalances in food transfers when the their “poorer” exchange partners than the “richer” house-
interaction term is included in the model. We would expect holdsreceiveinkindfromtheir“poorer”exchangepartners.
no food transfers among even very close relatives (high r)
whenthedisparityinutilitytheyeachgainfromthesameunit
offoodislimited(lowdifferenceinnetcaloricproduction).
2 Gurven (2006) reported finding a positive correlation between
Likewise, kin selection does not favor transfers among
imbalance in food transfers among any two households and the closest
nonkin(lowr)evenwhenaverylargedisparityinfoodvalue kinshiptiejoining the twohouseholds.The computations Gurven(2006)
ispresent(largedifferenceinnetcaloricproduction). reportedweremadewithoutregardtothedirectionoftransferimbalancesor
Table 2 reveals that contrary to the prediction from kin byonlyconsideringpositiveimbalances.In contrast,thepresentanalysis
considers relative “need” so that positive specific imbalance values are
selection theory, the interaction term of relatedness by
attainedwhenanimbalancefavorsthe“poorer”householdinthedyadand
difference in net caloric production does not significantly
negativespecificimbalancevaluesareattainedwhenanimbalancefavors
predict specific imbalance in food transfers between house- the“richer”householdinthedyad.
312 W.Allen-Araveetal./EvolutionandHumanBehavior29(2008)305–318
Table3
MultilevelmodelsofcalorictransfersfromhouseholdDtohouseholdR
a)Unrestricted b)Modelwith c)Modelwith d)Modelwith e)Modelwith f)Fullmodel
model relatednessonly needonly kin-directedaltruism kin-favoredreciprocity
Parameterestimatea Parameterestimatea Parameterestimatea Parameterestimatea Parameterestimatea Parameterestimatea
Fixedeffects
⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎
Intercept 2298.16 1910.41 2298.16 1910.41 20,057 20,057
Relatedness – 19,568⁎⁎ – 19,568⁎⁎ 561,348 561,348
Need – – 1255.28⁎⁎ 1222.05⁎⁎ – 1222.05⁎⁎
Kin-directedaltruism – – – 1931.71 – 1931.71
Kin-favored – – – – 284,097.26⁎⁎ 284,097.26⁎⁎
reciprocityb
CovarianceComponents
⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎
Within-dyadvariation 19,467,855 19,467,855 16,328,912 16,381,350 19,467,855 16,381,350
⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎ ⁎
Between-dyad 5,291,654 4,642,678 6,861,125 6,185,930 2,098,587 3,641,839
variation
⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎
Intraclasscorrelation 0.21 0.19 0.30 0.27 .10 0.18
Fitstatistics
AICc 10,043.6 10,033.5 1000.2 9991.8 9859.6 9815.9
a Parameterestimatesarerestrictedmaximumlikelihoodestimates.FixedeffectsrepresenttheestimatedincreaseincaloriestransferredfromhouseholdDto
householdRduetoaone-unitincreaseintheindependentvariable.
b Pooledparameterestimaterepresentingtheaverageacrossalldyadsforthisclassificationvariable.
c AICvaluesforcomparingmodelswereobtainedwithfullmaximumlikelihoodestimates.
⁎
Significantatthe.05level.
⁎⁎
Significantatthe.01level.
Fig. 1 provides a graphic display of these results. Under transfers to any household R varies from dyad to dyad
kinselectiontheoryalone,wewouldhaveexpectedFig.1to (z=3.32, p=.0009). We find a statistically significant and
revealasteeppositiveslopeamongthemostcloselyrelated positive intraclass correlation, suggesting that across all
dyads (the top left panel), no slope among unrelated dyads levelsofrelatedness,householdsdisplaydyadiccontingency
(the bottom right panel), and intermediate slopes with most [r=.21, F(252,253)=1.54, two-tailed p=.0006].
I
givingoccurringamongdyadpairsintherightportionofthe If investments in inclusive fitness gains provide the
x-axisfordyadsinthemiddlerangesofrelatedness.Instead, adaptive function of food transfers, contingency should be
imbalances in food transfers between households tend to relatively unimportant among kin. To test this prediction,
favor the household with the lower net caloric production we next fit a conditional multilevel model that includes a
value at all relatedness levels (y=3483.6x−1603.4, R2=.18 fixed effect for relatedness. The between-dyad covariance
for “close kin”; y=4262.2x–1094.3, R2=.34 for “near” kin; component in Table 3, Column b, reveals that when we
y=2351.4x+325.2, R2=.05 for “distant” kin; y=2527.1x− control for relatedness, the average number of calories
317.72, R2=.08 for nonkin). transferred from household D to household R still differs
from dyad to dyad (z=3.00, p=.0027). The residual
4.2. Reciprocal altruism intraclass correlation in this model estimates the degree of
correspondence in amounts households transfer to each
While we have focused thus far on determinants of other after parceling out the effect of relatedness. If close
imbalances in calories of food exchanged between house- kin tolerate larger imbalances in exchanges than distant kin
holds, transfers between any two households at Arroyo and nonkin as we might predict under the precepts of kin
Bandera are usually not unidirectional over time. If selection theory, we would obtain a larger intraclass
households display contingency in food transfers, over correlation from the conditional model that parcels out the
time, the amounts two households transfer to each other effect of relatedness (Table 3, Column b) than from the
will display correspondence. To examine dyadic contin- unrestricted model (Table 3, Column a). Surprisingly, we
gency in food transfers, we pair each household with each find just the opposite effect. When we control for
of the other households on the reservation once as the relatedness, we still find a correlation in the amounts
donor (D) household and once as the recipient (R) households exchange to each other [r=.19, F(251,252)=
I
household in a multilevel model that hierarchically nests 1.47, two-tailed p=.0023], but this correlation is smaller
households within dyads. than that obtained from the model that did not consider
Results from our baseline unrestricted model are relatedness [r=.21, F(252,253)=1.54, two-tailed p=.0006].
I
presented in Table 3, Column a. The between-dyad This suggests that dyadic contingency is stronger, not
covariance component in the unrestricted model suggests weaker, among close kin than among more distant kin
that the average number of calories any household D and nonkin.
W.Allen-Araveetal./EvolutionandHumanBehavior29(2008)305–318 313
Fig.2.Linearregressionofdyadicfoodtransfers(arrayedsothelargeramountgivenineachhouseholdpairisplottedonthex-axis).Plotsfor(a)closekin
(rN.05),(b)nearkin(.018brb.047),(c)distantkin(0brb.018),and(d)unrelateddyads(r=0).
For a graphic display of these results, Fig. 2 plots the Toeliminateconfoundingeffectsandunderstandhowall
numberofcaloriesoffoodproductionhouseholdDprovided ofthepredictorssimultaneouslyinfluencefoodtransfers,we
to household R against the number of calories of food discuss the results from fitting the multilevel model that
productionhouseholdRprovidedtohouseholdDbykinship includes all of the predictors (Table 3, Column f). Mean-
level. As exchanges become more equitable between while, we compare each of the single-explanatory mechan-
households in a dyad, the slope of the linear fit in Fig. 2 ism conditional models (Table 3, Column b–e) to the
increases. Interestingly, the slope for “close” kin is steeper unrestricted model (Table 3, Column a) to estimate the
(y=.5648x+240.12, R2=.54) than the slopes for “near” kin explained variance of each of the individual predictors. In
(y=.1511x−2.2553, R2=.38), “distant” kin (y=.0423x+ thefullmodel(Table3,Columnf),thefixed-effecttermfor
353.43,R2=.02), and nonkin (y=.0911x+275.62, R2=.10). the intercept estimates that over an observation period of
102–120h,ahouseholdtransfersameanof2000.57calories
4.3. Relatedness, need, and contingency to an unrelated household when the donor household and
recipient household do not differ in net caloric production.
We have established that relatedness, need, and con- IncreasesintheamountoffoodanyhouseholdDtransfersto
tingency all significantly predict the number of calories any household R are associated with: (1) the degree of
household D transfers to household R. We next fit greater net energy production of household D relative to
conditional multilevel models to include need, the interac- householdR(“need”),(2)theinteractionofkinshipwiththe
tion of relatedness with difference in net caloric value amount of food household D transferred to household R
(kin-directed altruism), and the interaction of relatedness (“kin-favored reciprocity”), and (3) dyadic contingency
with contingency (kin-favored reciprocity) as fixed effects (indicated bythe intraclasscorrelation).
(Table3,Columnsc–e).Wealsofitaconditionalmultilevel The statistical significance of the interaction term of
model incorporating all of these potential explanatory relatedness with dyadic contingency in the full model
variables offood transfers (Table 3, Column f). (Table 3, Column f) indicates that households give
314 W.Allen-Araveetal./EvolutionandHumanBehavior29(2008)305–318
preference in their food distributions to other households ment in model fit is attributable to the combined effect of
that contain kin and provide ample food in return relatedness and need together, not their interaction. The
[F(100,151)=1.54, p=.0082], even when we control for interactiontermofrelatednessandneedistheonlyfactorin
the other explanatory variables. The model that considers the model that truly tests the cost–benefit conditions
only the “kin-favored reciprocity” predictors (Table 3, necessary for inclusive fitnessgains from altruism.
Columne)yieldsalowerAICthanthemodelsthatconsider
only relatedness measures, need measures, or kin-directed
altruism measures (Table 3, Column b–d). This suggests 5. Discussion
that the “kin-favored reciprocity” model provides the best
fit of any single-explanatory mechanism model. The Household pairs at Arroyo Bandera display correspon-
pseudo-R2 [Eq. (1.2)] estimates that the interaction of denceintheamountsofcaloriestheytransfertooneanother.
2
relatedness with dyadic contingency models 60% of the While nepotistic biases are present in the data, households
explainedvarianceamongdyadaverages(compareTable3, give preference to kin who are likely to reciprocate, rather
Column e to Column a). The predictive power of this “kin- thankinwhowillbenefitmorefromtheaid.Closelyrelated
favored reciprocity” term is consistent with a view that households display even higher correspondence in the
households mainly engage in reciprocal exchanges and numberofcaloriestheyprovideforeachotherthandistantly
prefer kin as their exchange partners. relatedandunrelatedhouseholds,suggestingthatcontingent
The residual intraclass correlation in the full model reciprocity is the norm among close kin. To the degree that
(Table 3, Column f) indicates that after controlling for the exchangesbetweenhouseholdsareuneven,imbalancestend
effects of the other predictors, 18% of the variation in the tofavorhouseholdsthatrequiremorecalorieseitherbecause
amount of calories transferred from household D to they are low producers, contain many mouths to feed, or
household R is attributable to the dyad [F(248,249)=1.44, both. These food flows favoring “poorer” households over
two-tailed p=.0042]. This measure of dyadic contingency “richer”households occur irrespective ofkinship distance.
conforms to the predictions of reciprocal altruism and
5.1. The selective forces of kin selection and
indicates that while kin are preferred reciprocal food
reciprocal altruism
exchange partners, households also commonly engage in
reciprocal exchanges with nonkin. TheresultssuggestthatAchehouseholdsmainlytransfer
The final significant predictor in the full model is the food on the reservation because of fitness gains from
difference in net caloric production (“need”) between expectedfoodrepaymentratherthanbecauseofpayoffsfrom
household D and household R [F(1,251)=0.19, pb.0001]. nepotistic investment in indirect fitness benefits. This may
Thepseudo-R12[Eq.(1.1)]estimatesthatthedifferenceinnet notbesurprisinggiven that,unlikemodelsofkinselection,
caloric production (“need”) between household D and theratiooftherecipient'sbenefittotheactor'scostneednot
household R models 16% of the explained variance in food beverygreatforreciprocalaltruismtofavorafoodtransfer
transferamountswithindyads(compareTable3Columncto (Kaplan & Hill, 1985). In order for kin selection to favor a
Columna).Thisresultisconsistentwithaviewthatbeyond food transfer among even full siblings, the average fitness
their reciprocal exchanges, households also help “needy” impact of the transferred food should be twice as great for
families that produce insufficient food relative to their recipientasforthedonor.Yet,forreciprocalaltruismtofavor
consumption requirement. the same transfer, the fitness cost of giving food now need
Relatedness retains no significant predictive power only be less than the benefit of food received in the future
[F(1,151)=0.11, p=.7430] in any model that includes the devalued by the probability of future repayment. Consider
interaction of relatedness and dyadic contingency (kin- Hamilton'srule (1964) for assisting kin:
favoredreciprocity).Thisimpliesthatwhilekinarefavored
withfoodshares,suchfavoritismonlypersistsiffoodaidis BrNC ð3:1Þ
returned. Such discrimination among kin is not consistent
with models of kin selection, because the benefits of in comparison to the rule Axelrod and Hamilton (1981)
nepotism in these models accrue from indirect fitness derived for tit-for-tat based cooperation:
benefits regardless of how the related recipient responds.
Thetermfortheinteractionofrelatednesswithdifferencein BpNC ð3:2Þ
net caloric value (kin-directed altruism) is also not
significant [F(1,251)=0.19, p=.6605] and loses predictive where B is the benefit to the recipient, C is the cost to
power in any model that also includes relatedness and the provider, r is the coefficient of genetic relatedness, and
difference in net caloric value (need) separately. Although, p is the probability of future interaction. This comparison
the model that only considers “kin-directed altruism” as an suggests that aid will be more strongly favored by the
explanatory mechanism (Table 3, Column d) yields a lower forcesofreciprocalaltruism thantheforcesofkinselection
AIC than the model that only considers “need” as an whenever pNr (Gurven et al., 2001). If siblings trust
explanatory mechanism (Table 3, Column c), this improve- each other, interact frequently, and rarely defect on