k1Pj20icjwherethelasttermissummedover0ithenetworkneighborhoodACMSIGecomExchangesVol10No2June2011Pages38 61SSuriandDWattsorintheaveragecontributionsofgroupsde12nedbythenetwo ID: 822180
Download Pdf The PPT/PDF document "4S.SuriandD.Wattscomesofinterest,suc..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
4S.SuriandD.Wattscomesofinterest,suc
4S.SuriandD.Wattscomesofinterest,suchasaggregatelevelsofcooperation,plausiblydependonthestructureofthenetworkaswellasonthestrategiesoftheindividualsinthepopulation[Nowaketal.2010].Therearetwomainreasonstosuspectthatcooperationshoulddependonnet-workstructure.Therstreasonisthatmanytheoreticalmodelsofsocialdilemmasassumethatcooperationisconditional,inthesensethatanindividualwillonlycooperateontheconditionthatitspartnersarealsocooperating.ArguablytheclearestexampleoftheprincipleofconditionalcooperationisthecelebratedTit-For-Tatstrategy,whichhasconsistentlybeenshowntooutperformmoreexploita-tivestrategiesinarangeofsimulationstudies,inlargepartbecauseitperformswellwheninteractingwithothercooperativestrategies[Axelrod1984].Inaddition,relatedstrategieshavealsobeenproposedthatgeneralizetheideaofconditionalcooperationtomulti-playersettings[Watts1999;GlanceandHuberman1993],usuallybyspecifyingsomeformofthresholdrequirement|i.e.\IwillcooperateifatleastXofmyneighborscooperatedlastround,elseIwilldefect."Regardlessofthespecicsoftherule,theimplicationoftheseresultsfornetworksisthatnetworkscharacterizedbyhighlevelsoflocalclustering[WattsandStrogatz1998],meaningthatanindividual'sneighborsarealsolikelytobeneighborsofeach
other,oughttosustainhigheraggregatelevel
other,oughttosustainhigheraggregatelevelsofcooperationthanpopulationsinwhichindividualsarerandomlymixed[AxelrodandHamilton1981].Putanotherway,localreinforcementwouldimplythatwhenanindividual'sneighborsalsointeractwitheachother,theyareinabetterpositiontoreinforceoneanother'spro-socialbehavior,andsomaybeexpectedtoresist\invasion"bydefectingstrategiesbetterthanwheneachneighborinteractswithadierentsetofothers.Thesecondreasontosuspectthatnetworkstructureshouldimpactcooperationisthatcooperationinnetworksmightbe\contagious".Specically,ifAisacon-ditionalcooperatorsurroundedmostlybycooperatingneighbors,Awillcooperatemore;butthenA'sincreasedcooperationmaycauseitsremainingneighborstocooperatemoreaswell.Theseneighborsmayinturncausetheirneighborstoco-operatemoreaswell,andsoon,leadingtoacascadeofcooperationthatsustainsitselfovermultiplesteps.Infact,recentlyithasbeenclaimedthatcooperationischaracterizedbya\threedegreesofin uence"rule[FowlerandChristakis2010],meaningthatanindividualwhoincreaseshisorherlevelofcooperationcanpos-itivelyimpactthecontributionofanindividualwhoisthreestepsremovedfromtheminthenetwork.Becausethenumberofindividualswhocanbereachedwithinthreedegreesofacooperatingindividualwillingeneraldependontheno
n-localstructureofthenetwork[WattsandStr
n-localstructureofthenetwork[WattsandStrogatz1998],thepresenceofsocialcontagionwouldimplythatnetworkfeaturesotherthanlocalclusteringshouldalsoimpactaggregatecooperationlevels.2.EXPERIMENTALDESIGNIncontrastwithstandardpublicgoodsgames,inwhichparticipants'contributionsaresharedamongmembersofthesamegroup,hereparticipantsarearrangedinanetworkandtheirpayoisonlyaectedbytheactionsoftheirneighbors.Tore ectthischange,players'payosaredenedbythefunctioni=eici+ak+1Pj2(i)cj,wherethelasttermissummedover(i),thenetworkneighborhoodACMSIGecomExchanges,Vol.10,No.2,June2011,Pages3{86S.SuriandD.Wattsorintheaveragecontributionsofgroupsdenedbythenetworktopologiesovereachround.Thusweconcludethattopologydoesnotexertanoticeableimpactoncontributionsatanylevel:individual,group,oraggregate.Theabsenceoftopologicaldependencyofcontributionssuggeststhatoneorbothofthehypothesesoutlinedabove(reinforcementandcontagion)mustbewrong.Wethereforeconductedtwofurtherseriesofexperimentsdesignedtotestthereinforce-mentandcontagionhypothesesrespectively.Intherstseriesof30experimentswefollowedthesamedesignasabove,butwiththekeydierencethatineachex-perimentfournodeswereselectedandtheircontributionswerealla
rticiallyxedeitherat10(the\coo
rticiallyxedeitherat10(the\cooperative"condition)or0(the\defection"condition)forallrounds.Thus,wewereabletestthereinforcementhypothesisbydirectlymeasur-ingthepositive/negativein uenceofunconditionalcooperators/defectorsontheirimmediateneighbors.Theseedplayerswerearrangedinordertocoverthenet-work,meaningthateachhumanplayerwasadjacenttopreciselyoneseedplayer(intherandomregularcase,aperfectcoverarrangementdidnotexistforthese-lectednetwork;thusacloseapproximationwasusedinstead).Anadvantageofthisarrangement,isthatallhumanplayersweresubjectedtothesameexperimentallymanipulatedin uence.Thepresenceofcooperatingseedsstimulatedconsistentlyhigheraggregatecon-tributionsfromtheremaining20players,whilethepresenceofdefectingseedshadtheoppositeeect.Possessingahigh(orlow)contributingneighborthereforedidincrease(ordecrease)theaveragecontributionlevels;thusoursubjectswereindeedbehavingasconditionalcooperators.Neverthelesstheeectoftheseedplayerswasnotconsistentlybiggerinthegraphswiththehighestclustering.Forexampletheeectoftheseednodesinanetworkofdisjointcliques,whichhadthemaximumnumberoftrianglesincidentoneachnode,wasverysimilartotheeectoftheseedsnodesintherandomregularnetwork,whichhadfewerthan1/10thasman
ytrian-gles.Thisresultimpliesthattwonode
ytrian-gles.Thisresultimpliesthattwonodesthatformatrianglewithacooperating(ordefecting)seeddonothaveanappreciablylarger(orsmaller)averagecontributionlevelthentwodisconnectednodeswithacooperating(ordefecting)seedneighborincommon.Mutualreinforcementofthecontributionsamongtheneighborsofaseednodeislargelyabsent,whetherornotthereisanedgebetweentheneighbors.Next,inaseriesof20experimentsover2weeks,wetestedthecontagionhypoth-esisbykeepingthenumberofunconditionallycooperatingseedsconstantatfourpernetwork(wedidnotintroduceunconditionaldefectorsintheseexperiments),butconcentratingthemtogetherintotwoadjacentpairs.Thisarrangementofseedsexposedsomehumanplayerstotwounconditionalcooperatorsasimmediateneighbors,whileotherswerenotexposedtoanyseedsdirectly,butwereconnectedindirectlytotheseedsviaahumanintermediary.Ifpositivecontagionwerepresentinthenetwork,wewouldexpecttoseenodesatdistancetwofromtheseedsin-creasetheircontributionsrelativetotheall-human(i.e.noseeds)condition.Quitetothecontrary,infact,thetwo-stepneighborsofthecooperatingseedscontributedslightlylessthanthenodesinthecorrespondingnetworkpositionscontributedintheall-humanexperiments.ACMSIGecomExchanges,Vol.10,No.2,June2011,Pages3{88S.SuriandD.WattsFowler,J.H.andChrist
akis,N.A.2010.Cooperativebehaviorcascade
akis,N.A.2010.Cooperativebehaviorcascadesinhumansocialnetworks.ProceedingsoftheNationalAcademyofSciences107,9,5334{5338.Glance,N.S.andHuberman,B.A.1993.Theoutbreakofcooperation.JournalofMathemat-icalSociology17,4,281{302.Hamilton,W.D.1964.Thegeneticalevolutionofsocialbehaviour.JournalofTheoreticalBiology7,1,17{52.Kearns,M.,Judd,S.,Tan,J.,andWortman,J.2009.Behavioralexperimentsonbiasedvotinginnetworks.ProceedingsoftheNationalAcademyofSciences106,5,1347.Kearns,M.,Suri,S.,andMontfort,N.2006.Anexperimentalstudyofthecoloringproblemonhumansubjectnetworks.Science313,5788,824{827.Newman,M.E.J.2003.Thestructureandfunctionofcomplexnetworks.SiamReview45,2,167{256.Nowak,M.A.,Tarnita,C.E.,andAntal,T.2010.Evolutionarydynamicsinstructuredpopulations.PhilosophicalTransactionsoftheRoyalSocietyB:BiologicalSciences365,1537,19.Strogatz,S.H.2001.Exploringcomplexnetworks.Nature410,6825,268{276.Watts,D.J.1999.SmallWorlds:TheDynamicsofNetworksBetweenOrderandRandomness.PrincetonUniversityPress,Princeton.Watts,D.J.,Dodds,P.S.,andNewman,M.E.J.2002.Identityandsearchinsocialnetworks.Science296,1302{1305.Watts,D.J.andStrogatz,S.H.1998.Collectivedynamicsof'small-world'networks.Na-ture393,6684,440{442.ACMSIGecomExchanges,Vol.10,No.2,June2011,Pag