the production of step-level public goods in structured...

18
©Copyright JASSS Francisco José León Medina, Francisco José Miguel Quesada and Vanessa Alcaide Lozano (2014) The Production of Step-Level Public Goods in Structured Social Networks: An Agent-Based Simulation Journal of Artificial Societies and Social Simulation 17 (1) 4 <http://jasss.soc.surrey.ac.uk/17/1/4.html> Received: 26-May-2012 Accepted: 03-Nov-2013 Published: 31-Jan-2014 Abstract This paper presents a multi-agent simulation of the production of step-level public goods in social networks. In previous public goods experimental research the design of the sequence ordering of decisions have been limited because of the necessity of simplicity taking priority over realism, which means they never accurately reproduce the social structure that constrains the available information. Multi-agent simulation can help us to overcome this limitation. In our model, agents are placed in 230 different networks and each networks’ success rates are analyzed. We find that some network attributes -density and global degree centrality and heterogeneity-, some initial parameters of the strategic situation -the provision point- and some agents’ attributes -beliefs about the probability that others will cooperate-, all have a significant impact on the success rate. Our paper is the first approach to an explanation for the scalar variant of production of public goods in a network using computational simulation methodology, and it outlines three main findings. (1) A less demanding collective effort level does not entail more success: the effort should neither be as high as to discourage others, nor so low as to be let to others. (2) More informed individuals do not always produce a better social outcome: a certain degree of ignorance about other agents’ previous decisions and their probability of cooperating are socially useful as long as it can lead to contributions that would not have occurred otherwise. (3) Dense horizontal groups are more likely to succeed in the production of step-level public goods: social ties provide information about the relevance of each agent’s individual contribution. This simulation demonstrates the explanatory power of the structural properties of a social system because agents with the same decision algorithm produce different outcomes depending on the properties of their social network. Keywords: Public Goods, Collective Behaviour, Decision Making, Social Networks Introduction 1.1 Since individual rationality can lead to a suboptimal outcome, a collective action problem emerges in the production process of public goods (Taylor 1987). Several characteristics of public goods explain why their production takes the form of a "social dilemma" (Hardin 1982; Olson 1965; Taylor 1987): a) the good is jointly produced by the contributions of the individual group members, but typically not every single member's contribution is required; b) contributions are costly even when the value that is obtained from the public good is higher than the individual cost of the contribution; c) once the good is produced, it will be available to all members of the group since excluding non-contributing members from its enjoyment is difficult or costly; d) a member's usage of the good does not diminish its availability to other members (non-rivalness). This last characteristic is very infrequent, and here we will consider that the good provides an equal benefit to all members of the group. 1.2 The production of public goods has been studied extensively through the following experimental design: a) a group of individuals receive an endowment of ( e) units and have to decide, anonymously, how much (x) they wish to contribute to a common pool; b) the share of the endowment that is not contributed to the common pool (e-x) retains its original value, while the share that is contributed is multiplied by a factorc; c) the common pool (∑x*c) is divided in equal parts among all the members of the group; d) factor c is determined such that it would be better for each player to keep rather than contributing a unit, but it would be worse for each player if no one contributes rather than if everybody contributes all their endowments. 1.3 This experimental design allows for countless variations: subjects can have equal or unequal endowments, decisions on how much to contribute to the common pool can be binary –to contribute all or not to contribute– or discrete –to contribute a fraction of e–, etc. The variant we are interested in specifies a different production function; production functions make the level of production of a good dependent on the contributions. The main distinction is between continuous and step-level functions (León 2010) [1] . In economic terms, we can stipulate that either the good has a price by unit (continuous good) or it has a unique price that has to be reached (step-level good). 1.4 In the experimental design presented above, the production function is continuous. Each contribution adds to the production of the good, and all contributions add the same (it is a linear function). However, in this article we focus on public goods with a step-level production function. These "step-level goods" (SLG) or "lumpy goods" are characterised by the existence of a provision point: a minimum of contributions is needed to produce the public good. If the provision point is reached, the good is produced; if the point is not reached, the good is not produced at all. The basic SLG model states that once the provision point is reached additional contributions do not generate a higher level of production. [2] 1.5 In the early 1980s, Marwell and Ames (1980), Hardin (1982), Taylor (1987) and Taylor and Ward (1982) highlighted the distinction between continuous and step-level goods. It is now widely accepted that public goods analysis is not possible without explicitly stating the kind of good we are referring to (Kollock 1998: 189; Komorita and Parks 1995; Ledyard 1995; Marwell and Oliver 1993: 24). 1.6 The variation of the production function we introduce in our MAB experimental design is one of the most interesting possible variations, mainly because the inclusion of a provision point completely changes the logic of the situation. In truth, the variation of other design elements, such as the endowment – equitable/inequitable–, the type of distribution –equitable/inequitable–, the type of decision –binary/discrete– can have a significant impact on the results, but do not make a public goods game any different to an n-person prisoner's dilemma. The introduction of a provision point does achieve the difference: since there is a provision point, defection is no longer a dominant strategy in a one-shot game. [3] http://jasss.soc.surrey.ac.uk/17/1/4.html 1 16/10/2015

Upload: others

Post on 25-Apr-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

©CopyrightJASSS

FranciscoJoséLeónMedina,FranciscoJoséMiguelQuesadaandVanessaAlcaideLozano(2014)

TheProductionofStep-LevelPublicGoodsinStructuredSocialNetworks:AnAgent-BasedSimulation

JournalofArtificialSocietiesandSocialSimulation 17(1)4<http://jasss.soc.surrey.ac.uk/17/1/4.html>

Received:26-May-2012Accepted:03-Nov-2013Published:31-Jan-2014

Abstract

Thispaperpresentsamulti-agentsimulationoftheproductionofstep-levelpublicgoodsinsocialnetworks.Inpreviouspublicgoodsexperimentalresearchthedesignofthesequenceorderingofdecisionshavebeenlimitedbecauseofthenecessityofsimplicitytakingpriorityoverrealism,whichmeanstheyneveraccuratelyreproducethesocialstructurethatconstrainstheavailableinformation.Multi-agentsimulationcanhelpustoovercomethislimitation.Inourmodel,agentsareplacedin230differentnetworksandeachnetworks’successratesareanalyzed.Wefindthatsomenetworkattributes-densityandglobaldegreecentralityandheterogeneity-,someinitialparametersofthestrategicsituation-theprovisionpoint-andsomeagents’attributes-beliefsabouttheprobabilitythatotherswillcooperate-,allhaveasignificantimpactonthesuccessrate.Ourpaperisthefirstapproachtoanexplanationforthescalarvariantofproductionofpublicgoodsinanetworkusingcomputationalsimulationmethodology,anditoutlinesthreemainfindings.(1)Alessdemandingcollectiveeffortleveldoesnotentailmoresuccess:theeffortshouldneitherbeashighastodiscourageothers,norsolowastobelettoothers.(2)Moreinformedindividualsdonotalwaysproduceabettersocialoutcome:acertaindegreeofignoranceaboutotheragents’previousdecisionsandtheirprobabilityofcooperatingaresociallyusefulaslongasitcanleadtocontributionsthatwouldnothaveoccurredotherwise.(3)Densehorizontalgroupsaremorelikelytosucceedintheproductionofstep-levelpublicgoods:socialtiesprovideinformationabouttherelevanceofeachagent’sindividualcontribution.Thissimulationdemonstratestheexplanatorypowerofthestructuralpropertiesofasocialsystembecauseagentswiththesamedecisionalgorithmproducedifferentoutcomesdependingonthepropertiesoftheirsocialnetwork.

Keywords:PublicGoods,CollectiveBehaviour,DecisionMaking,SocialNetworks

Introduction

1.1 Sinceindividualrationalitycanleadtoasuboptimaloutcome,acollectiveactionproblememergesintheproductionprocessofpublicgoods(Taylor1987).Severalcharacteristicsofpublicgoodsexplainwhytheirproductiontakestheformofa"socialdilemma"(Hardin1982;Olson1965;Taylor1987):a)thegoodisjointlyproducedbythecontributionsoftheindividualgroupmembers,buttypicallynoteverysinglemember'scontributionisrequired;b)contributionsarecostlyevenwhenthevaluethatisobtainedfromthepublicgoodishigherthantheindividualcostofthecontribution;c)oncethegoodisproduced,itwillbeavailabletoallmembersofthegroupsinceexcludingnon-contributingmembersfromitsenjoymentisdifficultorcostly;d)amember'susageofthegooddoesnotdiminishitsavailabilitytoothermembers(non-rivalness).Thislastcharacteristicisveryinfrequent,andherewewillconsiderthatthegoodprovidesanequalbenefittoallmembersofthegroup.

1.2 Theproductionofpublicgoodshasbeenstudiedextensivelythroughthefollowingexperimentaldesign:a)agroupofindividualsreceiveanendowmentof(e)unitsandhavetodecide,anonymously,howmuch(x)theywishtocontributetoacommonpool;b)theshareoftheendowmentthatisnotcontributedtothecommonpool(e-x)retainsitsoriginalvalue,whilethesharethatiscontributedismultipliedbyafactorc;c)thecommonpool(∑x*c)isdividedinequalpartsamongallthemembersofthegroup;d)factorcisdeterminedsuchthatitwouldbebetterforeachplayertokeepratherthancontributingaunit,butitwouldbeworseforeachplayerifnoonecontributesratherthanifeverybodycontributesalltheirendowments.

1.3 Thisexperimentaldesignallowsforcountlessvariations:subjectscanhaveequalorunequalendowments,decisionsonhowmuchtocontributetothecommonpoolcanbebinary–tocontributeallornottocontribute–ordiscrete–tocontributeafractionofe–,etc.Thevariantweareinterestedinspecifiesadifferentproductionfunction;productionfunctionsmakethelevelofproductionofagooddependentonthecontributions.Themaindistinctionisbetweencontinuousand

step-levelfunctions(León2010)[1].Ineconomicterms,wecanstipulatethateitherthegoodhasapricebyunit(continuousgood)orithasauniquepricethathastobereached(step-levelgood).

1.4 Intheexperimentaldesignpresentedabove,theproductionfunctioniscontinuous.Eachcontributionaddstotheproductionofthegood,andallcontributionsaddthesame(itisalinearfunction).However,inthisarticlewefocusonpublicgoodswithastep-levelproductionfunction.These"step-levelgoods"(SLG)or"lumpygoods"arecharacterisedbytheexistenceofaprovisionpoint:aminimumofcontributionsisneededtoproducethepublicgood.Iftheprovisionpointisreached,thegoodisproduced;ifthepointisnotreached,thegoodisnotproducedatall.ThebasicSLGmodelstatesthatoncetheprovisionpointisreached

additionalcontributionsdonotgenerateahigherlevelofproduction.[2]

1.5 Intheearly1980s,MarwellandAmes(1980),Hardin(1982),Taylor(1987)andTaylorandWard(1982)highlightedthedistinctionbetweencontinuousandstep-levelgoods.Itisnowwidelyacceptedthatpublicgoodsanalysisisnotpossiblewithoutexplicitlystatingthekindofgoodwearereferringto(Kollock1998:189;KomoritaandParks1995;Ledyard1995;MarwellandOliver1993:24).

1.6 ThevariationoftheproductionfunctionweintroduceinourMABexperimentaldesignisoneofthemostinterestingpossiblevariations,mainlybecausetheinclusionofaprovisionpointcompletelychangesthelogicofthesituation.Intruth,thevariationofotherdesignelements,suchastheendowment–equitable/inequitable–,thetypeofdistribution–equitable/inequitable–,thetypeofdecision–binary/discrete–canhaveasignificantimpactontheresults,butdonotmakeapublicgoodsgameanydifferenttoann-personprisoner'sdilemma.Theintroductionofaprovisionpointdoesachievethedifference:sincethereis

aprovisionpoint,defectionisnolongeradominantstrategyinaone-shotgame.[3]

http://jasss.soc.surrey.ac.uk/17/1/4.html 1 16/10/2015

Page 2: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

1.7 Collectiveactionintheformofproducingastep-levelpublicgoodisratherfrequent.Forexample,thecasepresentedbyTaylor(1987)regardinganelectionwithtwooptions,AandB,onacommitteeoravolunteerassistanceparliamentinwhichtherearetwoparties:amajorityandaminority.ThelargestpartyprefersoptionAtoB.Whichmajoritygroupmemberswillattendthesession?Totheextentthatattendanceisvoluntaryandcostly,eachmemberfromthemajoritygroupwouldpreferthatothermembersofhisgroupattendsandhelpwiththeirvotetoelectoptionA.Moreover,inthissituation,agreaternumberofmajoritymembersthanminoritymembersattendingisenough,butbeyondthatpointfurtherattendeesofthemajoritygrouparenotrequiredtoproducethe"good":winningtheelection.NotethatinthiscaseitisnotstrictlynecessaryforAtobeapublicgood,butthelogicofthesituationisexactlywhatwemodelinthispaper.Anotherrealexampleofscalarpublicgoodswouldbethefollowing:insomeuniversitiesandcolleges,thedepartment'sbudgetpartiallydependsontheproductivityoftheirmembers,measuredasthetotalnumberofjournalpapersproducedbythedepartment.Ifthereisaminimumthresholdofarticlestoobtaintheadditionalbudget,thismaybeconsideredascalarpublicgood.Apartfromanyotherpersonalbenefitofpublication,eachmemberofthedepartmenthastoestablishitscontributionatalevelofeffort,knowingthatonlyifthecumulativeeffortsreachtherequiredlevelthedepartmentwillgetanadditionalamountintheannualbudget.

1.8 AnexperimentaldesignforstudyingSLGproductioncouldstartfromthesimplemodeloutlinedabove,andthenmodifythethirdcharacteristicbyaddingaminimumamountor"provisionpoint"requiredforthegoodtobeproduced.ThefirstexperimentswiththistypeofdesignwereconductedbyvandeKragtetal.(1983)andRapoport(1985,1987,1988,1993).Inthispaperweaimtocomputationallyreplicateavariantoftheseexperiments.Thepaperisorganisedasfollows.First,wepresentdifferenttypesofsequentialrunningorderingoftheagents'decisionsandourproposalofanetwork-dependentsequenceformalisation.Second,webrieflyreviewtheliteratureonsocialnetworksandexperimentationinpublicgoods.Third,wepresentanagent-basedmodelthatreplicatestheexperimentalsetting(Miguel2011).Fourth,wepresenttheresultsofdifferentsimulationsandthediscussion.

ANetwork-DependentSequenceOrderingRulefortheProductionofPublicGoods

2.1 TheliteratureonthesubjectsuggeststhatthedecisionsofexperimentalsubjectsinSLGexperimentsareverysensitivetothesequenceofthedecisions(AbeleandEhrhart2005;ErevandRapoport1990)becausedifferentsequencingimpliesdifferentinformationavailableatthedecisionmoment.Fourdifferentrulesfordeterminingtheorderofagents'decisionsequencingcanbefoundinexperimentalresearch(Budescuetal.1997):(1)the"simultaneousprotocol",wheresubjectshavenoinformationabouttheoneanother'sdecisions;(2)the"sequentialprotocol",wheresubjectshavecompleteinformationaboutallpreviousdecisions;(3)the"positionalprotocol",whereasubjecthasinformationaboutherpositioninthesequencebutnotaboutpreviousdecisions;and(4)the"cumulativeprotocol",wherethesubjectknowsthequantityofthecontributionsmadebeforeher,butnotherownpositioninthesequence.

2.2 Inlaboratoryexperiments,allthesesequencing–orrunningorder–ruleshavebeenlimitedintheirdesignbythenecessityofsimplicitytakingpriorityoverrealism,andcanthereforeresultinunrealisticassumptionsabouttheavailableinformationwhenpeoplemaketheirchoices.Thissimplificationimpliesahighercapacitytocontrolforvariables,generatingcumulativeknowledgeaboutfactorsthatinfluencebehaviour.Thelackofrealisminherentinthesesequencingrulesdoesnotnecessarilyimplyalackofvalueorheuristicutility.Ourproposalaimstoelaborateandapplyanewruletocomplementtheexistingones.

2.3 Amorerealisticsequenceorderingruleshouldrepresentadecisionalsituationwhereindividualsareembeddedinarelationalstructurethatconstrainstheavailableinformation.Inanon-artificialsituationofpublicgoodsproduction,individualstypicallyhavecertainlocalknowledgeaboutthedimensionsoftheirgroup,buttheyarenotindirectcontactwithallmembers.Inotherwords,itisthenetworkstructurethatdeterminesavailableinformationintermsofwhatothersthink,sayordo.Thatiswhy,withtheexceptionofexperimentsbySuriandWatts(2001),itiscommonlyacceptedinexistingliteraturethatasocialnetwork'stopology,ormorespecifically,thestructuralpropertiesofasocialnetworkhaveaneffectonthelevelofcontributiontocollectiveaction.Co-operativebehaviourcanspreadinanetworkduetosuchmechanismsasmimicry(FowlerandChristakis2010)orsocialcomparison(Zschache2012),butinthispaperwefocusontheroleplayedbytheuseoflocalinformationonthedecisionsofadjacentnodes.

2.4 Anexperimentalset-upaimedatreproducingthissituationwouldbetoocomplextobeexecutedinalaboratory,soweinsteadtestthisnewsequenceorderingrulewithinacomputationalmulti-agentsimulationmodel.Thissolutionallowustoavoidthetrade-offbetweensimplicityandrealismtypicalforlaboratoryexperiments,andsubstituteitforatrade-offbetweenrealisminsomeaspects–agentswillbeembeddedinanetwork–andrealisminothers–decisionswillbemadebysoftwareagentsnotbyrealpeople.Inadditiontotheaimtocomputationallyreplicatesomeexperimentalset-upsofthereviewedliterature,otherconsiderationssuchastherelativeadvantageintermsofcosts,resourcesandtime,orthefactthatthesimulationcanpredictwhentherearedifferencesintheparametersthatdescribeagivenenvironment,playinfavouroftheuseofABMresearchmethodology.

TheFormalModelDescription

3.1 Inthissectionweoutlinetheformalmodeltobecomputationallyimplemented:

1. AgroupofagentsN={1,…,n}participatesinapublicgoodsgame.2. Agentsknowthesizeofthegroup–thenetwork–theybelongto.3. Eachplayeri(i∈N)receivesanendowmentofeiunits(ei>0).Thevalueofallendowmentsisequal.4. Eachagenti∈Ndecidesaboutthequantity(xi)fromthesetX={0,ei}thatshewishestocontributetotheproductionofthepublicgood.Agentimust

makeabinarydecision:toeithercontributeallherendowment(xi=ei)orcontributenothing(xi=0).[4]

5. Thepositionofeachagentinthesequenceofdecisionsisrandomlydetermined.6. Agentsareplacedinanetwork,g,wheretheyonlyinteractwithsomemembersofN.Networkgisconstitutedbysymmetricallinks:gij=gji.7. Eachagenti∈NfacesasetofadjacentnodesorneighboursNi⊂Nwithwhichsheisconnected.Ni={j∈N|gij=1}.Thenumberofi'sneighboursiski–i's

nodedegreecentrality–.8. Ifgij=1agentjwillknowxiifintherandomlydecidedsequence(seepoint5)xihastakenplacebeforexj,andviceversa.9. Agentsknowtheirki,butnotthatoftheirneighbours.TheyalsodonotknowtheirpositioninthesequenceofdecisionsofN,althoughtheyknowtheir

positioninthesequenceofdecisionsofNi.10. Atthedecisionmomentt,eachagenthasthreepiecesofinformationaboutotheragents'decisions:a)howmanyneighbourshavemadedecisions

previously–B,for"Behaviours"–;b)∑ j∈Nixj,t-1,thatis,howmanyadjacentnodeshavecontributedtheirendowmenttotheproductionofthepublicgood

int-1–C,for"Cooperators"–;c)howmanyagentsparticipateinthenetworkaboutwhomshehasnoinformation,n-ki.11. Thereexistsaprovisionpoint(m)knownbyallagents,0<m<∑i∈Nei.12. If∑i∈Nxi≥m,eachagentreceivesapayoffr.If∑i∈Nxi<m,thegoodisnotproducedandallthecontributionsarelost.13. Thepayoffireceivesifthegroupreachestheprovisionpointisr=mc/n,wherecisafactorknowntoallagents,suchthatei<r.14. Thefinalpayoff(πi)thatagentireceivesisdeterminedbythefollowingpayofffunction:

http://jasss.soc.surrey.ac.uk/17/1/4.html 2 16/10/2015

Page 3: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

(1)

3.2 Inourmodel,called"NetCommons",peoplearelocatedinasocio-matrixthatconstrainstheavailableinformation,conditioningthedecisionprocess.Letusillustratetwodifferentdecisionsequencingordersinthisnetwork–thesequenceofdecisionsinthematrixfollowsanalphabeticalordering–.

Diagram1.AtypicaldecisionorderinginaN=7network

3.3 AgentBdecidesknowingthedecisionofAandknowingthatGhasnotdecidedyet.SincesheknowsherpositioninthesequenceofdecisionsofNBbutshedoesnotknowherpositioninthesequenceofN–seepoint9–,agentBcannotknowwhetherC,D,E,orFhavealreadydecided.Similarly,agentCdecidesknowingthedecisionofAandknowingthatEandDhavenotdecidedyet.SincesheknowsherpositioninthesequenceofdecisionsofNCbutshedoesnotknowherpositioninthesequenceofN,agentCcannotknowwhetherB,F,orGhasalreadydecided.Eachindividualhasaccesstoinformationaboutpartofthedecisionsalreadytaken,butfacesuncertaintywithrespecttotherestofthepastandfuturedecisions.

NetworksandExperimentationonPublicGoods

4.1 Moreno'swork(1951)isoftenreferredtoasthestartingpointofsocialnetworkanalysis.Onlyinthelast20yearshasnetworkanalysisbeenappliedinexperimentaleconomics–forareview,seeKosfeld(2004).Thesestudieshavefocusedmostlyoncoordinationnetworks,buyer-sellernetworksandnetworkformation.

4.2 Concerningexperimentalresearchonpublicgoods,labexperimentsconsideringtheroleofnetworksarescarce–someexceptionsareBonacich1990;CárdenasandJaramillo2009;Sonnemansetal.2006;FowlerandChristakis2010–andavailableresearchusuallyconsidersnetworkstobetheoutcomeofagenerativeprocessratherthanapreconditionforinteraction.KniggeandBuskens(2010),forexample,presentedanexperimentinwhichsubjectsseektoestablishrelationshipswithothersthatshareaccesstothegoodproduced,sothatthenetworkemergesfromtheinteractions.Thispaucityofnetworkanalysisinthecontextoflaboratoryresearchonpublicgoodsissurprisingastheconsensusisthatovercomingthefree-riderproblemrequiresestablishingsomeformofinterdependencebetweendecisions(Marwelletal.1988).Therelevanceofsocialtiesindeterminingparticipationincollectiveactionwaspointedoutalongtimeago(Tilly1978;Oberschall1973).Theexplanationforthisismostlikelythetechnicaldifficultiesofreproducingcountlessspecificconfigurationsofnetworksinalaboratorycontext.

4.3 Consistentwiththis,themaincontributionstounderstandingtherelationbetweennetworksandpublicgoodshavebeenrealisedoutsidethelab,throughmathematicalmodelsandsimulations.NovakandMay(1992)werethefirsttostudythedynamicsoftheprisoner'sdilemmawhenagentsareplacedinatwodimensionalspacethatconstrainsthemtointeractwiththeirneighbours.Sincethen,thesociallocationoftheagentshasbeenconsideredakeyelementtobemodelledintheoriesofcollectiveaction,thusopeningthedoortothestudyoftherelationshipbetweennetworksandpublicgoods.BramoulléandKranton(Bramoullé2007)werethefirsttopresentanetworkmodelofpublicgoods,whileotherauthorsuseanevolutionaryapproachtopublicgoodswithoutpayingattentiontothesocialstructure(e.g.,Yeetal.2011).Inthefieldofmathematicalandcomputationalmodelling,asinexperimentalresearch,theanalysisoftherelationshipbetweennetworksandpublicgoodshasemphasisedtheanalysisofreticulardynamics,i.e.theprocessesofnetworksemergency.Thus,forexample,Takácsetal.(2008)andSkyrmsandPemantle(2000)presentedmodelsinwhichtheparticipantsincollectiveactioncanstrategicallyreviewtheirrelationships,thusgeneratingreticulardynamics,andBravoetal.(2012)claimedthattheendogenousformationofthenetworkisevenmoreimportantforcooperationthanthenetworktopology.Ingeneral,thesemodelsarepartofabroadertradition,studyingtheemergencyofcomplexnetworksfromlocalinteractions,suchasmodelsinSutcliffeetal.(2012)andPujoletal.(2005).

4.4 Table1summarisesthedifferencesbetweenoursimulationmodelandtheleadingmodelsonnetworksandcollectiveactioninliterature.Ascanbeseen,NetCommonsisthefirstsimulationmodelofSLGproduction.Totheextentthatthelogicofthesituationiscompletelydifferentwhenthegoodhasaprovisionpoint,weunderstandthattheoriginalcontributionofthisarticleisthatitwillbethefirsttoexploretheroleofreticulartopologyonthesuccessofcollectiveactionwhichaimstoproduceastep-levelpublicgood.

Table1:NetCommonscomparedwithleadingmodelsofnetworksandpublicgoods

Marwell(1988) Gould(1993) Chwe(1999) NetCommons

Method Simulationandmathematicalmodel

Mathematicalmodel Mathematicalmodel Simulation

Agents Motivationanddecisionalmechanism

Expectedutilitymaximisation

Rationalityandjusticenorms

Probabilityofreachingthethreshold

Expectedutilitymaximisation

(criticality)Decisioncost Considered Irrelevant Notconsidered ConsideredEndowment Differentforeachagent Notconsidered Notconsidered Equalforallagents

Reward Differentforeachagent Notconsidered Notconsidered EqualforallagentsUnconditionalcontribution

Impossible Possible Impossible Impossible

Threshold Nothreshold Nothreshold Endogenousand Exogenousand

http://jasss.soc.surrey.ac.uk/17/1/4.html 3 16/10/2015

Page 4: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

precondition objectiveIndividualefficacy Dependentonthe

decisionofthegroupDependenton

previousdecisionsNotDependentonpreviousdecisions

Dependentonpreviousdecisions

Networks Links Asymmetricalanddirect Symmetricalanddirect

Asymmetrical,directandindirect

Symmetricalanddirect

Useofthelinks Costly Notcostly Notcostly NotcostlyAgent'sknowledgeof

thenetworkAgentsknowthe

structureAgentsknowtheir

neighboursAgentsknowthe

structureAgentsknowtheirneighboursandn

Informationobtainedinthenetwork

Perfectinformation Behavioursandpredispositions

Others'personalthresholds

Others'behaviours

Output Mainresult Sumofcontributions Sumofcontributions Sumofcontributions Successrate

TheSimulationModelDescription

5.1 ThesimulationwasconductedthroughtheNetlogoplatform.Thecodeisavailableat"CoMSESComputationalModelLibrary",thepublicrepositoryofOpenABMConsortium,http://www.openabm.org/model/2522/version/1/view(Miguel2011).Thismodelimplementstheformalnetwork-dependentdecisionsequenceorderingoutlinedintheprevioussection.However,sincethemodelintendstocapturetheessentialtraitsofSLGproductions,wecompleteherethespecificationoftheformalmodelbyincludingfourelements:(1)theagents'decisionprocess;(2)theinitialparameters;(3)thenetworkproperties;and(4)thedynamicsoftheprocessanditsoutput.

Howagentsdecideinthemodel

5.2 IntheNetCommonsmodel,agentsareassumedtobeself-interestedutilitymaximisers.Experimentshaveshownthatsubjectswhorepeatedlyfaceapublicgoodsgametendtoactinaccordancewiththepredictionsofrationalchoicetheory(Ledyard1995).Theoreticalpredictionsfromoursimulationcanbeconsideredaswell-groundedhypothesesofwhatwereasonablyexpecttofindinrealcases(includinglaboratoryexperiments).Moreover,ourmodelcanoffersomepracticaladviceonthemosteffectivewayofproducingapublicgood.Wecouldconsiderprosocialagents,butconstructingamodelbasedonthemostunfavourableassumptionsregardingtheproductionofpublicgoods–withself-interestedutilitymaximiseragents–isthebeststrategy.Prosocialagentswouldmerelyraisesuccessrates–theirprobabilityofcooperationishigher–.

5.3 Tomodeldecisionsweshallusetheconceptofself-efficacyorcriticality,anditsformalisationbasedonprobabilitytheory.InthecontextoftheproductionofSLG,criticalityreferstotheconditionwherebyadecisionisnecessaryandsufficienttoreachtheprovisionpoint.InNetCommons,agentshavethecognitivecapacitytoestimatetheirprobabilityofbeingcritical.

5.4 Whenanagentdecideswhethertocontributetothecommonpool,shewillconsidertheprovisionpoint(m),therewardthatsheexpectstoobtainifthegoodisproduced(r),thecostofhercontribution(x),thenumberofneighboursthathavepreviouslydecided(B),andthenumberofneighboursthathavealreadycooperated(C).Moreover,thedecisionwilldependonherbeliefthatanexactnumberofm-1willcontribute,makinghercontributioncritical,sothedecisionultimatelydependsontheprobabilityofcooperationagentsattributetoeachother(p).

5.5 AnobviousimplicationfortheNetCommonsmodelisthatanagentcannotrationallyforeseethedecisionsoftheotheragentsatthemomentofherdecision.Shedoesnotknowtheinformationthatotherswillhaveatthemomentoftheirdecision,sinceshedoesnotknowwhichotheragentsareneighboursorwhatinformationtheyreceivefromtheiradjacentlinks.Inthissituationofuncertaintyregardingothers'decisions,therationaloptionwouldbetoattributetoothersacertainprobabilityofcooperation,0≤p≤1.Inourmodel,weintegrateahomogeneityassumption(Rapoport1985),wherebyallagentsattributethesame

probabilitytoallotheragents[5].

5.6 Takingallthesefactorsintoaccount,agentswillfindthemselvesinoneofthefollowingsixexhaustiveandmutuallyexclusivesituations:

Situation1:Thenumberofneighbourswhohavealreadycooperatedisequalorhigherthantheprovisionpoint.Inthiscasethepublicgoodwillbeproducedandtheagent'scontributionisnotstrictlyrequired.Arationalagentwillnotcooperateinthissituation,forcontributionsabovetheprovisionpointhavenoeffectandconstituteapurelossfortheagent.

Firstsituation:IfC≥m,thenDONOTcooperate.

Situation2:Thenumberofneighboursthathavealreadycooperatedisequaltotheprovisionpointminusone,andeachagentbelievesthatsheisnotthelastonetodecide.Aslongastherewardishigherthanthecostofcontributing,weexpecttheagent'scooperationtoensuretheprovisionpointisreached.However,theagentalsotakesintoaccountthepossibilitythatotherswillbearthecostofcontribution,inwhichcaseshecanretaintheendowmentandreceive

ashareoftherewardwithoutexpendingthecost.Inthissituation,anagent'scooperationdependsonherbeliefthatnobodyelsewillco-operate,thatis(1-p)n-B-

1,becauseonlyinthiscaseishercontributioncritical.

Secondsituation:IfC=m-1andn-B>1,then:cooperateifr[1-(1-p)n-B-1]>eDONOTco-operateifr[1-(1-p)n-B-1]≤e

Situation3:Thenumberofneighbourswhohavealreadyco-operatedislowerthantheprovisionpointminusone,andthenumberofpendingdecisionsisequalorhigherthanthenumberofcontributionsneededtoreachtheprovisionpoint.Inthiscase,theagenthastocalculateherprobabilityofbeingcriticalby

consideringbothhowmanyneighbourshavenotdecidedyetandhowmanystrangersareinthenetwork.[6]BuildingontheprobabilisticmodelofRapoport(1985),whichhasbeenwidelyusedasareferencemodelforcalculating"objective"criticality(Kerr1989),wecalculatetheprobabilityofbeingcritical,Pm-C-1,asfollows:

(2)

Thatis,theprobabilitythatineachcombinationofn-B-1players,exactlym-C-1willco-operateandtherest(n-B-m)willnot.Thiswouldmakehercontributioncritical.

Thirdsituation:IfC<m-1andm-C≤n-B,then:co-operateifrPm-C-1>e.DONOTco-operateifrPm-C-1≤e

Situation4:Theagent'scontributioniscriticalonlyifitisaddedtothefuturecooperationofalltheagentsaboutwhomshehasnoinformation–becausethey

http://jasss.soc.surrey.ac.uk/17/1/4.html 4 16/10/2015

Page 5: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

havenotdecidedorbecausetheyarenotplacedinadjacentnodes.Thisrequiresabeliefthatsheisnotthelastonedeciding.

Fourthsituation:If1<n-B=m-C,then:co-operateifrpn-B>eDONOTco-operateifrpn-B≤e

Situation5:Anagentfacesasituationinwhichthenumberofinstancesofcooperationneededtoreachmishigherthaneitherthenumberofagentsthathavenotyetdecidedorthoseaboutwhomshehasnoinformation.Inthiscase,arationalagentwillnotcooperate,sincethegoalisunattainableandhercontributionwouldbelost.

Fifthsituation:Ifm-C>n-B,thenDONOTcooperate.

Situation6:Anagentfindsherselfinasituationwhereshebelievesthatsheisthelastonetodecide,andonlyonemorecontributionisneededtoreachm.

Givenri>ei,arationalagentwillalwaysco-operate.[7]

Sixthsituation:If1=n-B=m-C,thencooperate.

Theinitialparametersofthesimulation

5.7 TheinitialparametersinNetCommonsarealmostidenticaltothoseusedinthepioneeringSLGexperimentsofvandeKragtetal.(1983):

m=3,n=7,p=0.5,e=5,r=15

InthevandeKragtdesign,therewasnopparameter,andourreward(r)isslightlyhigherthantheirsinresponsetosomeearlytestingofNetCommons.However,thequantityofrisnotarbitrary.Ithasbeenestablishinsidethemarginsthatleadstoafailureintheproductionofthegoodwhendecisionsaresimultaneous(Rapoport1985:151).Toillustrateit,inasimultaneousprotocolwiththeparametersm=3,n=7,p=0.5,e=5,rationalutilitymaximisersevaluatetheirprobabilityofbeingcriticalasPm-1=0,234.IftheirdecisionistocooperateonlywhenrPm-1>e,onlyr=22couldleadthemtocooperate.Thesameisalsotrueforacumulativeprotocol.Ifthenetwork-dependentdecisionsequenceorderingleadstosuccesswherethesimultaneousandthecumulativesequenceorderingsfail,orifatleastdoessoundercertainwell-specifiedconditions,wecouldconcludethatworkinginanetwork,oratleastincertainnetworks,facilitatestheachievementofthecommongoal.Underequalconditionsofincentivesfortheproductionofthegood,thenetwork-dependentsequenceordering

wouldnotonlybemorerealistic,butalsomoreeffectivethanthesimultaneousandthecumulativeorderingrules.[8]

Inputnetworksproperties

5.8 Multi-agentsimulationnotonlyallowsustoplaceagentsinanetworkstructureandobservetheresults,butalsoallowsustocomparetheresultsobtainedwhentheseagentsareplacedindifferentkindsofsocialnetworks.Forthispurpose,wehavebuilt–withUcinet6.0andinDLformat–acatalogueofsymmetricalnetworkscomposedby7nodesandwithdifferentdensitiesandglobaldegreecentralityandheterogeneity.

5.9 Densityisdefinedastheratioofthenumberoftiesoverthemaximumpossiblenumberofties(Scott2005).

(3)

wherelisthenumberoftiesinthenetwork.Thevalueofdensityoscillatesbetween0,whennonodeislinked,and1,whenallthenodesareconnectedwitheachother.Fornetworkswithn=7andonlyonecomponent,densityrangesbetween0.29–foranetworkwiththeshapeofaline–and1.Densityisoneofthestructuralpropertiesofnetworksthatexistingliteraturehasidentifiedasrelevantduetoitsimpactonthesuccessofcollectiveaction.Thisimpact,however,isfarfromclear(Bravo2008).AuthorslikeTakácsetal.(2008),forexample,havenotedthatdensenetworksinvolvemoreinformationandgreatercapacityforpunishment,whichfosterscooperation.However,atthesametime,highdensitycanprovidetoolstoresistpressureandevenformutualaffirmationofdefectingbehaviours.Punishmentfordefectionmaybereducedpreciselybecausesomeneighboursanticipatefutureinteractionsthattheydonotwanttocompromise(Gould2003).Also,ifthedecisiontocontributedependsonobservedbehaviourinthelocalneighbourhood,thenobservingnon-co-operativebehaviourcanstimulatedefection(Gould1993).

5.10 Givenacertainlevelofdensity,itispossibletobuildnetworkswithdifferentglobaldegreecentralityofheterogeneity.Forexample,anetworkwithadensityof0.52canbecomposedofasetofnodeswithdegrees5,5,4,3,2,2,1orwithdegrees5,4,3,3,3,3,1(amongmanyotherpossiblecombinations).WeusedBlau'sindextomeasureheterogeneity.Insymmetricalnetworkstheindexisdefinedasfollows:

(4)

5.11 Thenetworksthatweusehavejustonecomponent;allnodesaredirectlyorindirectlyconnectedformingasinglestructure.Thisisfortwomainreasons.First,asamatterofrealism:insmallgroups(n=7)thelikelihoodofmorethanonecomponentisquitelow,andinthecaseofpublicgoodsthepopulationcouldbeseenasbelongingtocommunitiesthat,generally,aremorecohesivethesmallertheyare.Second,becauseourultimateinterestisinanalysingtheimpactofreticularpropertiessuchasdensityandheterogeneityinthesuccessorfailureofcollectiveaction.Thenumberofnetworkcomponentsaffectstheseindicatorsandthereforemustbeexperimentallycontrolledasseparatefromthesetofanalysedvariables.

5.12 Giventhesecharacteristics,thecatalogueiscomposedof230differentnetworks.[9]Thissetisexhaustive:itincludesallthepossiblenetworkswithn=7,symmetricaltiesandonlyonecomponent.Thismeansthatthecataloguecontainsverydifferentnetworktopologies:small-worldnetworks,scale-freenetworks,randomnetworks,etc.However,theaimofthisarticleisnottocomparethedifferentialsuccessofthesespecialtypesofnetworksbuttoexploretheimpactofreticularpropertiesonthesuccessorfailureoftheproductionofSLGs.

Dynamicsoftheprocessandoutputs

5.13 Thepreviousformalmodeldescriptionsectionspecifiesthedynamicsoftheproductionofthepublicgood,andcorrespondstoasingle"turn"intheNetCommonssimulationmodel,whereallmembersofthegroupdecideoneafteranotherresultingeitherintheprovisionpointbeingreached(success)ornot(failure).

5.14 Thetop-levelalgorithmforthemodelis:

Create initial network (options: random network OR import-DLfile)Unless maximum number of time steps (5000) has been reached, do:

http://jasss.soc.surrey.ac.uk/17/1/4.html 5 16/10/2015

Page 6: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

Reset all agents' attributes Unless all agents have decided, eavh agent will: Scan neighborhood for counting deciders (B) and cooperators (C) Estimate the decision situation Make a decision about cooperation behaviour Record micro-decisional outcomes in a buffer-matrix Next agent Compute the turn macro-indicators from the buffer-matrix Update outputs and plotting Next turnStop simulation run

5.15 Thefinalresultofaturndependsonthesequenceofdecisions,which,asstatedbefore,israndomlydecidedineachturn.Forn=7,thenumberofallpossiblesequenceorderingis7!=5,040foreachdifferentnetwork.Thismeansanetworkhas5,040differentpossibleresults.Inouranalysiswefocusonthesuccessrates(SR)ofeachnetwork,definedasthenumberofsuccessfulturns–wheretheprovisionpointisreached–overthetotalnumberofrealizedturns.Inthenextsectionwepresenttheresultsafterasimulationof5,000turnsforeachnetwork.

ResultsandDiscussion

6.1 Inthissectionwediscusstwodifferenttypesofresults.Thefirsttypeconcernshowthechangeincertaininitialparametersimpactsonthesuccessrate(SR).Forthisanalysiswefocusonasinglenetwork("18_16.dl")comprising18tiesandaheterogeneityof16%.Thereasonforselectingthisparticularnetworkwasitsaveragesuccessrate(0.63)whichallowedustobetteranalysetheimpactofchangesintheinitialparameters.

6.2 Thesecondtypeofresultsrelatestothevariabilityinsuccessratesofdifferentnetworksasafunctionoftheirstructuralproperties.Forthisanalysisweusedthefullcatalogueof230networks.

Thestabilityofthesuccessrate

6.3 Ifthereexistsafinitenumberofpossiblesequenceorderingsforanetwork,andiftheresultsofaparticularturndonotaffectthesubsequentturns,asuccessratethatisupdatedaftereachturncannotshowanunstablepattern.Asthesimulationisrunningturnafterturn,thenetworkwouldasymptoticallyapproachtheaveragesuccessrate(SR)producedby–hypothetically–calculatingtheresultofthe5,040possiblesequencesofthenetwork.WecanseeinFigure1,withthisstabilisationevident,theupdatedSRaftereachturnconcludes.

http://jasss.soc.surrey.ac.uk/17/1/4.html 6 16/10/2015

Page 7: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

Figure1.Successrateofthenetwork"18_16"(5000rounds)

6.4 Tworemarksshouldbemadehere.First,thefactthateachnetworkor"socialunit"hasitsownsuccessrateisunsurprisinginsociologicalanalysis.Lesscommonistheabilitytoidentifythesituationalmechanisms–responsibleforthemacro-microlink–,theaction-formationmechanisms–responsibleforthemicro-microlink–andthetransformationalmechanisms–responsibleforthemicro-macrolink–thatexplainhowindividualactionsjointlyproducethatoutcome(HedstromandSwedberg1996,1998).Inourcase,thesimulationallowsustogenerateandanalysethe35,000individualdecisionsthatcausethisrate,andmoreoverspecifyhowthishappens.

6.5 Second,weestimatethatthe"18_16"networkneeds212roundstostabiliseataratewithina±3%marginoftherateobtainedafter5,000rounds.Ifweuseamarginof±5%,ittakesthesimulation100roundstoachievestability.Inbothcases,itisanexcessivenumberofroundstobereproducedinalaboratoryexperimentwithhumansubjects.Thisiswhytheanalysisofthenetwork-dependentsequenceorderingcanonlybeproducedthroughamulti-agentsimulationoramathematicalmodel.

Theeffectsofthechangeintheparameters

6.6 InthefollowingsectionswepresentsomeresultsofanalysingtheeffectsontheSRofdifferentrelevantparameters.Wewillexplorethreespacesofparametersinturn:theprovisionpointofthepublicgood(m),theprobabilityofcooperationattributedtootheragents(p),andtwostructuralpropertiesofthenetworks(density,andglobaldegreeheterogeneity).Inallthreecasestheanalyticalstrategywillbefirsttopresentsomeaggregateormacro-results,andthenanalysethedetailedrecordstoestablishthemicro-foundationsexplainingtheseresults.

Changesintheprovisionpoint(m)

Macroresults

6.7 After5000rounds,thesuccessrateofthenetwork"18_16",whichcomprisesdifferentprovisionpointswhilekeepingtherestoftheparametersconstant,presentsthefollowingdistribution.

Figure2.Successratesofthenetwork"18_16"withdifferentprovisionpoints

http://jasss.soc.surrey.ac.uk/17/1/4.html 7 16/10/2015

Page 8: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

6.8 Inthisanalysiswedonotconsiderthepossibilityofm=1:thisparticularsituationdoesnotconstituteacollectiveactionproblem.Inaddition,ourformalisation0<m<∑i∈Neiallowsustodismissm=7aswell,becausethissituationexcludesthepossibilityofreceivingabenefitwithoutcost.

6.9 AsshownbyFigure2,highprovisionpointsgenerateafailureof100%.Thelowerprovisionpointm=2ismoresuccessful,butthisoneinturnisoutperformedbytheintermediateprovisionpointm=3.Toacertaindegree(m=3inourcase),higherSRsareachievedthemorepeoplecollaborate.

Microfoundations

6.10 Table2allowsustoidentifytheexplicativemicro-foundationsoftheSRsobtainedwithdifferentprovisionpoints.

Table2:Numberofdecisionsdependingontheprovisionpoint,theorderinthesequence,thesituationthattheagentfaces,andthetypeofdecisionsmade(network"18_16",5000roundsforeachprovisionpoint)

Situation1C>=m

Situation2C=m-1&n-B>1

Situation3C<m-1&m-C=<n-B

Situation4C=m-n+B

Situation5m-C>n-B

m Order Def. Coop. Def. Coop. Def. Coop. Def. Coop. Def. Coop.

2 1 50002 50003 50004 4602 3985 236 3902 8626 21 838 3181 852 1087 166 1393 2328 414 699

3 1 50002 50003 4163 8374 3171 18295 2279 2476 2456 25 1155 3239 463 1187 65 4251 222 462

4 1 50002 50003 50004 4590 4105 4143 624 2336 3966 370 6647 3644 665 691

Note:Anynetworkconsistingof7membersrepresentsatotalof7decisionseachround,sothat5000roundsinvolve35000choices.Eachofthesedecisions(a)hasbeenproducedinasequenceorderwithintheround(fromfirsttoseventh,thelast),(b)hasoccurredinoneofthe5decisionsituations,and(c)hasbeenwhethertoco-operateornot.Thetableabovedisplaystheabsolutefrequencycross-distributionof35,000decisions,accountingforthethreevariablesandcontrollingby3levelsofprovisionpoint(m).

6.11 Table2showsthatfirstchoicesarealwaysproducedinSituation3andalwaysconsistofadefection,independentofthevalueofm.TheexplanationofSRrequiresclarificationofhow,whenandwhyagentsstarttocooperate.InSituation3,PmC1indicatesthat(whileC=0)withalowBnoagentcooperates,butfromacertainthresholdinBvaluesonwardsagentsstartcooperating.Toexplainwhythishappenswehavetoexplorethethreecomponentsofthecriticalitycalculus(Formulae2).Thefirstcomponent,thebinomialcoefficientC(n-B-1,m-C-1),islowerthehigherBis.Inthesefirstroundsinwhichnobodycooperates,

thesecondcomponent(pm-C-1)isconstantandcanthereforebeignored.Thelastcomponent((1-p)n-B-m)providesthekeytounderstandingtheunderlyingmechanisms,becausehereweobservethatthehigherBis,thehigherthepossibilitythatn-mwillnotcooperate,whichinturnincreasestheprobabilityoftheagent'scriticality.

6.12 Thisisasituationalmechanism(HedströmandSwedberg1998),sinceitexplainsatwhichmomentandwhyapioneeringcooperativebehaviour–cooperationthatisbelievedtobepioneeringbytheagent–istriggered.Wecallthismechanismthepioneeringcooperationtrigger.Oncethismechanismisactivated,twosituationscanoccur.Insomecases,pioneeringcontributionsestablishtheconditionsforotheragentstoestimatetheirprobabilityofbeingcriticalashigh,andthereforetheydecidetoco-operate.Wecallthismechanismthespeculativecooperationtrigger.Inothercases,havingrelevantinformationconcerningpreviouscooperationdirectlyaffectsanagent'sbeliefthathercontributionmeanstheprovisionpointisreached(althoughshecannotbesurethathercontributioniscritical,sinceotheragentscouldbearthecostofthecontribution).Wecallthismechanismtheeffectivecooperationtrigger.

6.13 Theonlytransformationalmechanismthatexplainswhyaturnconcludessuccessfullyisthemechanismoftheaggregationofindividualchoices.However,thismechanismcantakedifferentformsdependingonthewayinwhichsituationalmechanismsthattriggerpioneering,speculativeandeffectivecooperationareactivated.DifferentcumulativeprocessesthatleadtosuccessexplainwhyeachcasehasitsSR,andwhythatrateisdifferentineachcase.

6.14 Further,differentcumulativeprocessesexplainwhyanetworkwithm=2achievesaspecificSR,andwhythisSRislowerthantheSRproducedwhenm=3.ThekeytounderstandingthisdifferenceisthefactthatthethresholdofBvalueswhichactivatethepioneeringcooperationtriggerishigherinnetworkswithlowerprovisionpoints.KeepingC=0,m=2requiresB=3toco-operate,whilem=3onlyrequiresB=2.

http://jasss.soc.surrey.ac.uk/17/1/4.html 8 16/10/2015

Page 9: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

6.15 ThisoutcomeisexplainedastheresultofachangeinthefirstcomponentofPm-C-1.KeepingBandCconstant,goodswithahighermproduceahighervalueforthebinomialcoefficientC(n-B-1,m-C-1),whichinturnimpliesthatthecriticalityestimationfavourscooperation.Whenp=0.5,theprobabilityofasituationwith1cooperationand5defectionsisequaltotheprobabilityofsituationswith2cooperationsand4defections.However,thelatterismorelikelythantheformerbecausetherearemorecombinationsofthisoutcome.Inotherwords,anagentismorelikelytobecriticalwhenthenumberofcontributionsneededishigher.

6.16 Whenm=2,startingfromthefourthplayeritispossiblethatsomeonemayhaveknowledgeofB=3andthendecidetocooperate.Whenm=3,thesameprocessstartsearlier,butnowitonlytakesB=2fortheagenttoco-operate.Thismeansthatwhenm=3pioneeringcooperationistriggeredearlierthanwhenm=2,whichinturninducestheearliertriggeringofspeculativeandeffectivecooperation.Inthecaseofm=3,previouscooperationleadseithertospeculativecooperationortotheagentsfindingthemselvesinSituation4characterisedbynocooperation.However,sincethebalanceispositive,theincreaseinspeculativecooperationisresponsiblefortheincreasingsuccessrate.Thecaseofm=3impliesanearlytriggeringofpioneeringcooperation,whichallowsasubsequenttriggeringofspeculativecooperationsufficientlyhighfortheprovisionpointtobereached.Whenm=2,pioneeringcooperationistriggeredatalaterpoint,whichmeansspeculativeandeffectivecooperationcanemergetoolatefortheprovisionpointtobereached.

6.17 SRgrowsfromm=2tom=3,butfallsfromm=3tom=4.Thisisexplainedbecause,whilewiththeseparameters B≤3leadstoadefection,incontrasttowhenm=3thepioneeringcooperationtriggerisnotactivated.Inthissituation,agentswithki≤3nevercooperate,whilethosewithki≥3eitherfaceSituation3,withacriticalityestimationthatdiscouragescooperation,orelsefaceSituation4,wherethecriticalitycalculusalwaysleadstodefection.Theconditionsthatleadtocooperativebehaviourarenevermet.

6.18 Inshort,theexplanationofthedifferentsuccessratesforeachvalueofmresemblesaninstanceofpath-dependence,becauseofthewayinwhichthecumulativeprocessofcontributionstakesplace.Wheretheconditionsforanearlytriggeringofpioneeringcooperationaremet,thesubsequenttriggeringofspeculativeandeffectivecooperationmaytakeplaceearlyenoughfortheprovisionpointtobereached.High(m=3)provisionpointstriggerpioneeringcooperationearlier,thusgeneratinganearliercascadeofdecisionsthatleadtotheproductionofthepublicgood.However,provisionpointsthataretoohigh(m>3)havetheoppositeeffect,sincetheaccumulationofdefectionsreducesthepossibilityofreachingaverydemandinggoal.Ingeneral,agentsdonotexpecttobecriticalwhenthegoaliseasy,becausetheybelieveotheragentswilllikelybearthecost,norwhenthegoalisverydemanding,becausetheythinktherewillnotbeenoughotheragentscontributing.

Changesintheprobabilityofcooperationattributedtootheragents(p)

6.19 Aswehavediscussedandargued,westartfromtheassumptionofhomogeneityasraisedbyRapoport(1985),i.e.allagentsattributetoothersthesameprobabilityofcooperation(p).Whenwefocusonanalysingtheroleofthestructuralpropertiesofthenetwork,wemustsetthisparametertoaspecificvaluetorunsimulations.Thisposesaproblem,becauseforarationalandmaximising-utilityagent,thereisnoreasonwhatsoevertosetthisparametertoanyvalue.Therefore,ifallpossibilitiesareequallylikely–andtheagenthasnoreasontobelieveotherwise–,thePrincipleofInsufficientReasonsuggeststhatwesettheparametertop=0.5.Thereareotherpossibilities,suchasfixingthevalueofpasafunctionofobservedbehavioursbytheagentinhisneighbourhood,buttoattributetothosewhohavenotyetdecidedacooperativeprobabilitydependentonthebehaviourofthosewhohavealreadydecidedseemsanexercisemoreunreasonablethanattributingtoallagentsanequallikelihood.However,ABMallowsustoexplorethebreadthofpossibilitiesbytestingforpossiblevaluesofp,andthatispreciselywhatwedointhenextsection.

Macroresults

6.20 Figure3showsthatthebeliefinahighprobabilitythatotheragentswillcooperatehasthesameeffectasalowprobability(theybothgeneratealowSR),whileintermediatelevelsoftrustordistrustinotheragents'cooperationproduceahigherSR.

http://jasss.soc.surrey.ac.uk/17/1/4.html 9 16/10/2015

Page 10: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

Figure3.Successrateofnetwork"18_16"dependingonbeliefsabouttheprobabilitythatotheragentswillco-operate(5000roundsforeachvalueofp)

Microfoundations

6.21 Table3helpsustoidentifytheexplanatorymechanismsoftheseresults.

Table3:Numberofdecisionsdependingontheprobabilityofcooperationattributedtootheragents,theorderinthesequence,thesituationthattheagentfaces,andthetypeofdecisionsmade(network"18_16",5000roundforeachvalueofp)

Situation1C>=m

Situation2C=m-1&n-B>1

Situation3C<m-1&m-C=<n-B

Situation4C=m-n+B

Situation5m-C>n-B

P Order Def. Coop. Def. Coop. Def. Coop. Def. Coop. Def. Coop.

0.3 1 50002 50003 50004 50005 4765 2356 4275 610 1157 4274 726

0.4 1 50002 2835 21653 1579 34214 113 899 39885 15 575 302 41086 106 1343 35517 506 1875 2619

0.5 1 50002 50003 4131 8694 3287 17135 2307 2432 2616 29 1183 3194 480 1147 72 4210 247 471

0.6 1 50002 50003 4200 8004 3164 18365 2303 2508 1896 19 1182 3186 485 1287 69 4208 252 471

0.7 1 50002 50003 50004 4617 3835 4148 626 226

http://jasss.soc.surrey.ac.uk/17/1/4.html 10 16/10/2015

Page 11: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

6 3927 367 598 1087 3551 742 707

Note:Anynetworkconsistingof7membersrepresentsatotalof7decisionseachround,sothat5000roundsinvolve35000choices.Eachofthesedecisions(a)hasbeenproducedinasequenceorderwithintheround(fromfirsttolast,theseventh),(b)hasoccurredinoneofthe5decisionsituations,and(c)hasbeentoco-operateornotco-operate.Thetableabovedisplaystheabsolutefrequencycross-distributionof35000decisions,accountingforthethreevariablesandcontrollingby5levelsofcooperationprobabilityattributedtoothers(p).

6.22 Again,thekeytoexplainingthedifferencesintheSRliesintheconditionsthatactivatethetriggermechanismsofpioneeringcooperation,thenumberofpreviousdecisions(B)necessaryforanagenttoevaluateherownprobabilityofbeingcriticalassufficientlyhightocooperate.

6.23 Forthisexplanation,webuildonFigure4.ThisfigurepresentsthemathematicalrelationbetweenB(keepingC=0)andtheprobabilityofcontributingtothecollectiveaction.OrdinateaxisrepresentsthevalueofD(p)foreachofthepossiblevaluesofp,whereD(p)=rPm-C-1−edenotesthedifferencebetweentheexpectedbenefitandtheexpectedcost.WhenD(p)hasapositivevaluetheagentdecidestocooperate,whileshedecidestodefectwhenD(p)isnegative.

Figure4.Probabilityofbeingcriticaldependingontheprobabilityofcooperationattributedtootheragentsandthenumberofpreviousdecisions

6.24 Figure4showstwoimportantpoints.First,thecriticalitycalculusnecessarilyhasaninvertedU-shapebecause,whilethefirstcomponentofPm-C-1isconstant,

theremainingtwoarenegativelycorrelated,suchthattheincreaseofpm-C-1necessarilyimpliesadecreaseof(1-p)n-B-m.Second,Figure4showsthatthevaluesofBthatleadtocooperationaredependentonthevaluesofp.Inotherwords,Figure4holdsthekeytoidentifyingthetransformationalmechanismsthatexplaintheresultsofFigure3.

6.25 Withp=0.3,thereisnopossibilityforpioneeringcooperationtobetriggered.Withp=0.4,however,cooperationbecomespossibleevenwhenB=1andB=2.Oncethesecooperationsareproduced,speculativeandeffectivecooperationcanbetriggered.Withp=0.5andp=0.6thesameprocessstartsatalaterpoint,whensomeagentsknowthatB=2.Thisimpliesthatinsomecasestheroundfailsbecausethetriggeringofspeculativecooperationtakesplacetoolate(notenoughcontributionsareproducedtoreachtheprovisionpoint).

6.26 Withp=0.7theSRfallsto0.0becausehereaminimumofB=3isnecessaryforthetriggeringofpioneeringcooperation,soonlyafractionofthosewhodecideinthefourthorahigherplacehavetheoptiontocooperate.Moreover,whenthishappenssomeoftheagentsfaceSituation4,whichagainimpliesthattherewillnotbeenoughcontributionsfortheprovisionpointtobereached.

http://jasss.soc.surrey.ac.uk/17/1/4.html 11 16/10/2015

Page 12: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

6.27 Inshort,ontheonehand,excessiveconfidenceaboutotheragents'cooperationincreasesthebeliefthatitisimprobablethatenoughdefectionswillbeproducedfortheagent'scontributiontobecritical.Ontheotherhand,excessivelackofconfidenceaboutotheragents'cooperationincreasesthebeliefthatitisimprobablethatenoughcontributionswillbeproducedfortheagent'scontributiontobecritical.Withintermediatevaluesofp,bycontrast,thenumberofpreviousdefectionsfortriggeringpioneeringcooperationislow,andthecascadeofcooperationstartssufficientlyearlyinaroundforthenecessarycontributionstobeaccumulated.

Successratesandthestructuralpropertiesofthenetworks

6.28 Inthissectionwepresenttheresultsobtainedforthe230networksofourcatalogueandwetrytoidentifysomemicro-foundationsofthoseresults.

Macroresults

6.29 InTable4wepresentOLSregressionmodelstotesttheinfluenceoftwostructuralpropertiesofthenetworksontheSR.Sincetherelationsbetweentheindependentvariablesandthedependentvariablehaveanexponentialcharacter,alltheindependentvariablesaretransformedintotheirbase-10logarithm.

Figure5.Networkdensityandsuccessrate

http://jasss.soc.surrey.ac.uk/17/1/4.html 12 16/10/2015

Page 13: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

Figure6.Networkglobaldegreeheterogeneityandsuccessrate

6.30 Table4showsthatmodelB,whichusesthebase-10logarithmofthedensityandheterogeneityasindependentvariables,explains84%oftheSRvariability.DensitypositivelycorrelateswithSRanditisthevariablewiththehighestpredictivevalue.GlobaldegreeheterogeneitynegativelycorrelateswithSRand,althoughitspredictivepowerislower,itsinclusionresultsinamodelwithabetterfit.

Table4:Successrate.OLSregressionmodels(non-standardisedcoefficientsandstandarderrors)

(A) (B)Non-standardisedcoefficientsandstandarderrors

BetaNon-standardisedcoefficientsandstandarderrors

Beta

Constant 1.271(.022)***

3.781(.260)***

LOG_DENSITY 2.114(.076)***

.881 1.505(.090)***

.627

LOG_HETEROGENEITY -2.195(.227)***

-.361

N 227 227AdjustedR-square .775 .841

F 780.461*** 598.051***R-squarechange -- .066

Fchange -- 93.787***

***Significancelowerorequalto0.001

6.31 TsukamotoandShirayama(2010)presentedamodelontheevolutionofcooperationincomplexnetworks,andfoundthatintermediatevaluesofheterogeneitygeneratemaximumcooperation.Inthesamevein,Fuetal.(2007)presentedanevolutionaryprisoner'sdilemmaonanetwork,andalsofoundthatthefrequencyofcooperationishighestatintermediatelevelsofheterogeneity,andYangetal.(2012)reachedthesameconclusionanalysingscale-freenetworks.Itseems,therefore,thatmuchresearchpointstothesameconclusion:intermediatevaluesofheterogeneityoptimisethelevelsofcooperation.However,Fuetal.(2007)studytheeffectofheterogeneitystartingfromanetworkwithagivenhomogeneityandthenaddinglinkstosomeselectednodes.Thus,itisnotonlytheheterogeneitywhichincreases,butalsothedensity.RegressionmodelB,presentedinTable4,allowsustoassessthenetimpactforeachofthetwoindependentvariables.Inourcase,unlikeFuetal.(2007),TsukamotoandShirayama(2010),andYangetal.(2012),wefindthat,whilecontrollingforthedensity,heterogeneityhasanegativerelationshipwiththerateofsuccess.Thedifferencebetweenthecontinuousproductionofagood,suchasthatmodelledbyFuetal.,TsukamotoandShirayama,andYangetal.,andascalarvariant–likeours–impliesadifferenceintheimpactoftheindependentvariables.

Microfoundations

6.32 Todiscoverthemicro-foundationsthatexplainthemacro-resultsoutlinedinTable4,weanalyseagents'decisionsinrelationtotheirnodaldegree.Afterall,networksaredistinguishedintermsoftheirrelationalstructures,resultingfromthedifferenceinthenodaldegreeoftheagents.Moreover,agentswiththesamenodaldegreebehaveinthesameway,independentofthenetworkinwhichtheyaresituated.

6.33 Forsimplicity,weuseasubsetof8networksrepresentativesofdifferentcombinationsofdensityandheterogeneity.Figure7showsthepercentageofcooperationanddefectionsofagentsdependingontheirnodaldegree.

http://jasss.soc.surrey.ac.uk/17/1/4.html 13 16/10/2015

Page 14: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

Figure7.Agents'decisionsdependingontheirnodedegreefor8selectednetworks(5000roundsforeachnetwork,280000decisionsintotal)

6.34 TherelationbetweenthenodedegreeandtheprobabilityofcooperationhasaninvertedU-shape,althoughtheprobabilityofcooperationgenerallyincreasesforhighernodedegrees.Thehigherthedensityofanetwork,thehighertheprobabilitythatasufficientnumberofagentscontributetotheproductionofthepublicgood.

6.35 Atthesametime,thehighertheglobaldegreeofheterogeneity,thehigherthedispersioninthenodaldegreesoftheagents.Inthiscasewecanfindmoreagentswithextremelyhighorextremelylownodaldegreesandthosearepreciselythenodaldegreesthatcorrespondwithalowerprobabilityofcooperation.Theeffectofheterogeneityisnotasclearastheeffectofdensity:whileahighheterogeneityleadstofewercontributions,alowheterogeneityleadseithertofewerormorecontributionsdependingonthemeannodaldegreeofthenetwork.

6.36 TheseeffectsofdensityandheterogeneityarerepresentedinFigures5and6above.Giventhattheinfluenceofdensityoutweighsthatofheterogeneity,thereductioninSRthatisproducedbyanincreaseinheterogeneitywillbelowerwhendensityishigh;thisexplainsthedecelerativeshapeofthepointcloudinFigure5.

6.37 Theinfluenceofthestructuralpropertiesofanetworkcanbeexplainedbythedifferentprobabilitiesofcooperationofagentswithdifferentnodaldegrees.However,westillneedtomakeexplicitwhyagentswithdifferentnodaldegreeshavedifferentprobabilitiesofcooperation.

Table5:Numberofdecisionsdependingonagents'nodedegree(k)andthesituationatthemomentofdecision,for8selectednetworks.

k Situation Defect Cooperate Total%row %col. %row %col. %row %col.

1 Situation3 100% 100% 100% 100%Total 100% 100% 100% 100%

2 Situation2 100% 8.7% 100% 2.9%Situation3 68.7% 100% 31.3% 91.3% 100% 97.1%

Total 66.7% 100% 33.3% 100% 100% 100%

3 Situation1 100% 1.9% 100% 100%Situation2 100% 17.6% 100% 8.6%Situation3 55.3% 98.1% 44.7% 82.4% 100% 90.4%

Total 50.9% 100% 49.1% 100% 100% 100%

4 Situation1 100% 4.8% 100% 2.1%Situation2 100% 17.7% 100% 9.9%Situation3 46.5% 90.1% 53.5% 82.3% 100% 85.7%

http://jasss.soc.surrey.ac.uk/17/1/4.html 14 16/10/2015

Page 15: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

Situation4 100% 5.1% 100% 2.3%Total 44.3% 100% 55.7% 100% 100% 100%

5 Situation1 100% 15.5% 100% 7.2%Situation2 100% 25.6% 100% 13.7%Situation3 45.6% 71.6% 54.4% 74.4% 100% 73.1%Situation4 100% 100% 100% 4.7%Situation5 100% 2.8% 100% 1.3%

Total 46.6% 100% 53.4% 100% 100% 100%

6 Situation1 100% 6.6% 100% 3.9%Situation2 100% 10.3% 100% 4.2%Situation3 46.1% 48.8% 53.9% 82.3% 100% 62.5%Situation4 81.1% 21.9% 18.9% 7.4% 100% 16,0%Situation5 100% 22.7% 100% 13.4%

Total 59.1% 100% 40.9% 100% 100% 100%

Note:Asetof8selectednetworksconsistingof7memberswithonedecisioneachperroundand5000roundsinvolvesatotalof280000choices.Eachofthesedecisions(a)hasoccurredinoneofthe5decisionsituationsand(b)hasbeentoco-operateornottoco-operate.Thetabledisplaysboththerowandcolumnpercentagecross-distributionof280000decisions,accountingforthetwovariablesandcontrollingby6levelsofagents'nodedegree(k).

6.38 Table5presentsbehaviouraldifferencesasafunctionofagents'nodedegree(k).Inthecaseofagentswithki=1,theeffectivecooperationtriggerisnotactivatedbecausem=3.Pioneeringandspeculativecooperationwillnotbetriggeredeither,sincefortheseagentsBandConlyhavevalues0or1andnocombinationofthesevaluesleadstoacriticalitycalculusfavourabletocooperation.Table5showsthatagentswithki=1arealwaysinthesamesituationanddecideinfavourofdefection.Giventhelargenumberofagentsaboutwhomtheyhavenoinformation,agentsconsidertheprobabilityoftheircontributionbeingcriticaltobeverylow.

6.39 Foragentswithki>1therangeofpossiblesituationsisbroader,whichfavourscooperationinsomecases.Forexample,agentswithki=2whofaceSituation3donotalwaysdefect,asagentswithki=1do.Infact,thecombinationsB=2&C=1andB=2&C=0favourcooperation,whilecombinationsB=1&C=1,B=1&C=0andB=0&C=0leadtodefection.Thisexplainswhyagentswithki=2inSituation3co-operateonlyinsomecases(Table5).However,agentswiththisnodaldegreemayalsofaceSituation2,inwhichtheagentbelievesthatonlyonecontributionisneededtoreachtheprovisionpointandcooperationistherationaldecision.Inotherwords,agentswithki=2faceasituationwherebothspeculativecooperation–whenthecombinationsofBandChold–andeffectivecooperationcanbetriggered–whenthenumberofpreviouscooperationsis2,thatism-1.Anincreaseinthenodaldegreeproducesabroaderrangeofpossiblesituationsthattheagentfaces,someofwhichfavourcooperation.

6.40 However,thecaseofki=6isanexception.Figure7showsthattheprobabilityofcooperationdecreasesfortheseagents.Table5suggeststhatthisreductionisexplainedbytherelationbetweenahigherkiandanincreasedproportionofoccasionsinwhichagentsfaceSituation4(C=m-n+B).WithourparametersthissituationisproducedwhenC=B-4.Thisoccursinthreesituations:withC=0&B=4,withC=1&B=5,andwithC=2&B=6.Anagentwithki=4facesonlyoneofthesesituations(C=0&B=4).Agentswithki=5andki=6facesituationsC=0&B=4andC=1&B=5,butthelatterhasahigherprobabilityofencounteringacombinationC=1&B=5becauseshehasmoreneighbours.DefectionischoseninallthesecombinationsofCandB.AslongasthemajorityofpossiblecombinationsofCandBinSituation4leadtodefection,ahigherkiincreasestherelativeweightofthesecombinationsinthesesituations.

6.41 Inshort,agentsthatarelinkedinawaythatallowsthemtoobtainsignificantinformationaboutotheragents'behaviourareawareofthelowprobabilitythattheprovisionpointwillbereached,andthereforedecidenottocontribute.Ontheotherhand,thelackofinformationthatcharacterisesthelessconnectednetworksfavoursspeculativecontributions.Whenki=6isreachedthepossibilityfortriggeringspeculativecooperationdecreases.Paradoxically,inthiscasewefoundthatconstrainingthelevelofinformationgenerateslargersocialbenefits–asElsterclaims(1979,2000)–,andthesemechanismscouldbeexaminedatadetailedlevelbymeansofcomputationalsimulation.

ConclusionsandFurtherWork

7.1 Weconcludethispaperbyhighlightingseveralresultsthatweconsiderespeciallyrelevant.Oursimulationbearswitnesstotheexplanatorypowerofthestructuralpropertiesofasocialsystem.Rationalagentswithanidenticaldecisionalgorithmcangenerateverydifferentsocialoutcomesdependingontherelationalstructureinwhichtheyareembedded.

7.2 Thetopologyorstructureofasocialnetworkisnottheonlyrelevantfactor;thedecisionsequenceorderingisalsoessential.Animportantaspectofthedecisionisthequantityofinformationavailabletotheagentatthemomentofdecision,andinasocialnetworkthisvariabledependsonthenodaldegreeoftheagentandonthesequencepositionofherdecision.Theseissuescanbeaddressedthroughtheanalysisofdynamicnetworks,throughobservationorsimulation.

7.3 Wetendtothinkthatthelowertherequiredeffortlevel,thehighertheprobabilityforapublicgoodtobeproduced.However,ourresearchsuggeststhat,insomecases,institutionaldesignersshouldbalancetheeffortlevel;nottoohightodiscourage,nortoolowtostimulateadefectiongroundedontheconfidencethatotherswillbearthecost.

7.4 Theresultsofoursimulationrunsalsosuggestthatbothhighandlowconfidenceaboutotheragents'cooperationcanhaveanegativeeffectontheprobabilityofsuccessofacollectiveaction.Intermediatelevelsofconfidenceseemtostimulatemorecooperationbecausetheysupporttheagent'sbeliefthathercontributionisrelevant.

7.5 Socialgroupswithadensestructurehaveahigherprobabilityofsuccessintheproductionofstep-levelpublicgoodsbecauseneighboursprovideusefulinformationabouttherelevanceoftheagent'scontribution.However,ifthelinksarecostly,thehighestnumberoflinksdoesnotnecessarilymakeamorecohesivegroup:fromacertaindegreeofdensityonwards,theprobabilitiesofsuccessincollectiveactiondonotchangesignificantly.

7.6 Groupswithahorizontalstructurehaveahigherprobabilityofsuccessintheproductionofstep-levelpublicgoods.Poorlyconnectedagentshavelittleusefulinformationabouttherelevanceoftheirdecision,whilewell-connectedagentsmayhaveanexcessofinformationsuchthattheymaybeawareofdifficultiesthat

http://jasss.soc.surrey.ac.uk/17/1/4.html 15 16/10/2015

Page 16: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

cannotbeperceivedfromotherpositions.Alimitationofinformationstimulatesagentstoco-operatebecausetheyignorecertaindifficulties,thereforegeneratingasociallyoptimalaggregateeffect.Inotherwords,moreinformationforanindividualdoesnotalwaysgenerateabetterresultforthegroup.Thisconclusionisinlinewithrecentfindingsaccordingtowhichlimitedinformation,cognitiveorsocial,canfavourtheevolutionofcooperation;forexample,Horváthetal.(2012)haveshownhowthenumberofroundsrememberedbyaniteratedgameagentdoesnotmonotonicallyincreasethelikelihoodofreachingaco-operativestate.

7.7 Thesameagentsthatfailtoproducethepublicgoodinasimultaneousorcumulativedecisionsequenceorderingcouldsucceedwhentheyaresituatedincertainnetworks,withrelevantimplicationsforinstitutionaldesignandformulationofmoreeffectivesocialpolicyalternatives.

7.8 Oncewehaveproceededwiththecomputationalreplicationofpreviouslaboratoryexperiments,ournextgoalistoextendthemodelandthestudywithfurtherABMpossibilities.Forinstance,infutureversionsofNetCommonsweplantoincludeanewagentheterogeneityassumptionregardingexpectationsofothers'behaviour;toovercomethehypothesisofhomogeneitywefocusoncertainnetworktopologytypesandweintroducenetworksizeasaparameterunderscrutiny.

Acknowledgements

ThisworkhasbenefitedfromaMICINNR&Dprojectgrant(referencenumberCSO2009-09890),fromaMINECOR&Dprojectgrant(referencenumberCSO2012-31401)andfromaCONSOLIDER-INGENIO2010projectgrant(referencenumberCSD2010-00034).PartofthisresearchhasbeencarriedoutduringaresearchstayofF.León'sattheEqualityStudiesCentre,UniversityCollegeDublin.

Notes

1Amongthecontinuousfunctionstherearealsoimportantdifferencesbetweenaccelerating,deceleratingandlinearfunctions(Heckathorn1996;Linares2007).

2Taylor(1987)writesthatwhenweareinterestedintheconstructionofabridge,moremoneydoesnotimplymorebridges.However,extracontributionscanbededicatedtomakingabetterbridge(useofsuperiormaterialsorbetterdesigns)Inthiscase,wewouldhavea"mixed"formpublicgood,inwhichthefirstparttakestheformofaSLGfollowedbyacontinuousfunctiononcetheprovisionpointisreached.Inexperimentalterms,thishasbeenmadeconcreteindifferent"rebaterules"(MarksandCroson1998).

3Taylorhasargued(1987:46)thatlumpygoodsareoftenbestmodelledbytheChickenGame:dependingontheexpectationsregardingthecontributionsofothergamers,cooperationmaybethepreferredoptionforarationalmaximizingutilityagent.

4Somemodelshavedemonstratedthepositiveeffectoncooperationofheterogeneitywithregardtodecisionsabouthowmuchtoinvest(see,e.g.,Caoetal.2010;Santosetal.2008).Inourmodel,however,andforreasonsofoutcometractability,allagentshavethesameendowmentandfaceabinarydecision(fullycontributeitornottothecommonpool).

5Accordingtotheformalisationofourmodel,thesamedistributionofvaluesofpbetweenagentsofagivennetworkwouldentaildifferentoutcomesdependingonwhichvaluesofpareassociatedwithagentshavingdifferentnodaldegrees.Incorporatingtheassumptionofheterogeneitywillexponentiallyincreasethenumberofcombinationsnecessarytoobtainarealisticassessmentofitsimpactonthesuccessrate.Consideringalsothatourultimategoalistotesttherelevanceofthestructuralpropertiesofthenetwork,andthatasdiscussedbelowwewillworkwithacatalogueof230networks,itseemsreasonabletochoosethehypothesisofhomogeneitytoassuretheanalyticaltractabilityofthemodeloutputs.

6Regardingthelatter,thissituationisstrategicallyequivalenttoasimultaneousdecision.

7InNetCommonsthissixthsituationhasbeensubsumedinSituation4,becauseindecisionaltermstheresultisthesame.

8Comparingthe"sequentialprotocol"orderingandournetwork-dependentsequenceorderingisbeyondthescopeofthispaper.Insequentialprotocolorderingthereisnonecessityforapparameter,becauseagentsknowwhichinformationothershaveatthemomentoftheirdecision.

9Togainaccesstothefullcatalogueofnetworksusedasinput,pleasecontacttheauthorsdirectly.

References

ABELE,S.&Ehrhart,K.M.(2005).Thetimingeffectinpublicgoodsgames.JournalofExperimentalSocialPsychology,41,470–481.[doi:10.1016/j.jesp.2004.09.004]

BONACICH,P.(1990).CommunicationDilemmasinSocialNetworks:AnExperimentalStudy.AmericanSociologicalReview,55(3),448–459.[doi:10.2307/2095768]

BRAMOULLÉ,Y.&Kranton,R.(2007).Publicgoodsinnetworks.JournalofEconomicTheory,135,478–494.[doi:10.1016/j.jet.2006.06.006]

BRAVO,G.(2008).ImitationandcooperationinDifferentHelpingGames.JournalofArtificialSocietiesandSocialSimulation11(1),8http://jasss.soc.surrey.ac.uk/11/1/8.html.

BRAVO,G.,Squazzoni,F.andBoero,R.(2012).Trustandpartnerselectioninsocialnetworks:anexperimentallygroundedmodel.SocialNetworks,34,481–492.[doi:10.1016/j.socnet.2012.03.001]

BUDESCU,D.V.,WingTung,A.andChen,X.P.(1997).Effectsofprotocolofplayandsocialorientationonbehaviorinsequentialresourcedilemmas.Organisationalbehaviorandhumandecisionprocesses,69(3),179–193.[doi:10.1006/obhd.1997.2684]

CAO,X.,Du,W.andRong,Z.(2010).Theevolutionarypublicgoodsgameonscale-freenetworkswithheterogeneousinvestment,PhysicaA,389,1273–1280.[doi:10.1016/j.physa.2009.11.044]

CÁRDENAS,J.C.&Jaramillo,C.R.(2009).cooperationinlargenetworks:anexperimentalapproach.http://wwwprof.uniandes.edu.co/~chjarami/cardenas_jaramillo_cooperation_in_large_networks_CEDE200706_20070430.pdf.Archivedathttp://www.webcitation.org/67w5FhNia.

CHWE,M.S.Y.(1999).Structureandstrategyincollectiveaction.AmericanJournalofSociology,105(1),128–156.[doi:10.1086/210269]

http://jasss.soc.surrey.ac.uk/17/1/4.html 16 16/10/2015

Page 17: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

ELSTER,J.(1979).UlyssesandtheSirens:Studiesinrationalityandirrationality.Cambridge[Eng.];NewYork:CambridgeUniversityPress.

ELSTER,J.(2000).Ulyssesunbound:Studiesinrationality,precommitment,andconstraints.Cambridge;NewYork:CambridgeUniversityPress.[doi:10.1017/CBO9780511625008]

EREV,I.&Rapoport,A.(1990).Provisionofstep-levelpublicgoods.Thesequentialcontributionmechanism.JournalofConflictResolution,34(3),401–425.[doi:10.1177/0022002790034003002]

FOWLER,J.H.&Christakis,N.(2010).co-operativebehaviorcascadesinhumansocialnetworks,PNAS,107(12),5334–5338.[doi:10.1073/pnas.0913149107]

FU,F.,Liu,L.-H.andWang,L.(2007).EvolutionaryPrisoner'sDilemmaonheterogeneousNewman-Wattssmall-worldnetwork,Eur.Phys.J.B,56,367–372.[doi:10.1140/epjb/e2007-00124-5]

GOULD,R.V.(1993)CollectiveActionandNetworkStructure.AmericanSociologicalReview,58(2),182–196.[doi:10.2307/2095965]

GOULD,R.V.,(2003).Whydonetworksmatter?Rationalistandstructuralistinterpretations.InM.Diani&D.McAdam(Eds.),SocialMovementsandNetworks(pp.233-257).OxfordUniversityPress.[doi:10.1093/0199251789.003.0010]

HARDIN,R.(1982).Collectiveaction.Baltimore:JohnsHopkinsUniversityPress.

HECKATHORN,D.D.(1996).Thedynamicsanddilemmasofcollectiveaction.AmericanSociologicalReview,61(2),250–277.[doi:10.2307/2096334]

HEDSTRÖM,P.&Swedberg,R.(1996).SocialMechanisms,ActaSociologica,39,281–308.[doi:10.1177/000169939603900302]

HEDSTRÖM,P.&Swedberg,R.(eds.)(1998).SocialMechanisms:AnAnalyticalApproachtoSocialTheory.Cambridge,CambridgeUniversityPress.[doi:10.1017/cbo9780511663901]

HORVATH,G.,Kovárík,J.andMengel,F.(2012).Limitedmemorycanbebeneficialfortheevolutionofcooperation.JournalofTheoreticalBiology,300,93–205.

KERR,N.(1989).Illusionsofefficacy:theeffectsofgroupsizeonperceivedefficacyinsocialdilemmas.Journalofexperimentalsocialpsychology,25(4),287–313.[doi:10.1016/0022-1031(89)90024-3]

KNIGGE,A.&Buskens,V.(2010).Coordinationandcooperationproblemsinnetworkgoodproduction.Games,1,357–380.[doi:10.3390/g1040357]

KOLLOCK,P.(1998).Socialdilemmas:theanatomyofcooperation.Annualreviewofsociology,24,183–214.[doi:10.1146/annurev.soc.24.1.183]

KOMORITA,S.S.&Parks,C.D.(1995).Interpersonalrelations:mixed-motiveinteraction.Annualreviewofpsychology,46,183–207.[doi:10.1146/annurev.ps.46.020195.001151]

KOSFELDM.(2004).EconomicNetworksintheLaboratory:ASurvey.ReviewofNetworkEconomics,3(1),BerkeleyElectronicPress.[doi:10.2202/1446-9022.1041]

LEDYARD,J.O.(1995).Publicgoods:asurveyofexperimentalresearch.InJ.Kagel&A.E.Roth(Eds.).Thehandbookofexperimentaleconomics.Princeton:PrincetonUniversityPress.

LEÓN,F.J.(2010).Lalógicadelaaccióncolectivaenlaproduccióndebienespúblicosescalares.Papers.Revistadesociología,95(2),363–387.

LINARES,F.(2007).Elproblemadelaemergenciadenormassocialesenlaaccióncolectiva:Unaaproximaciónanalítica.RevistaInternacionaldeSociología,46,131–160.

MARKS,M.&Croson,R.(1998).Alternativerebaterulesintheprovisionofathresholdpublicgood:Anexperimentalinvestigation.JournalofPublicEconomics,67,195–220.[doi:10.1016/S0047-2727(97)00067-4]

MARWELL,G.&Ames,R.E.(1980).Experimentsontheprovisionofpublicgoods.II.Provisionpoints,stakes,experience,andthefree-riderproblem.Americanjournalofsociology,85(4),926–937.[doi:10.1086/227094]

MARWELL,G.&Oliver,P.(1993).Thecriticalmassincollectiveaction.Amicro-socialtheory.Cambridge:CambridgeUniversityPress.[doi:10.1017/CBO9780511663765]

MARWELL,G.,Pamela,E.andPrahl,R.(1988)Socialnetworksandcollectiveaction:atheoryofthecriticalmass.AmericanJournalofSociology,94,502–534.[doi:10.1086/229028]

MORENO,J.L.(1951).Sociometry,ExperimentalMethodandtheScienceofSociety.AnApproachtoaNewPoliticalOrientation.BeaconHouse,Beacon,NewYork.

MIGUEL,F.J.(2011).NetCommons(code).Archivedat:http://www.webcitation.org/67w5ftehr.

OBERSCHALL,A.(1973).Socialconflictandsocialmovements.EnglewoodCliffs,Prentice-Hall.

OLSON,M.(1965).Thelogicofcollectiveaction.Cambridge,MA:HarvardUniversityPress.

PUJOL,J.M.,Flache,A.,Delgado,J.andSangüesa,R.(2005).Howcansocialnetworkseverbecomecomplex?Modellingtheemergenceofcomplexnetworksfromlocalsocialexchanges.JournalofArtificialSocietiesandSocialSimulation,8(4),http://jasss.soc.surrey.ac.uk/8/4/12.html.

RAPOPORT,A.(1985).ProvisionofpublicgoodsandtheMCSexperimentalparadigm.AmericanPoliticalScienceReview,79(1),148–155.[doi:10.2307/1956124]

RAPOPORT,A.(1987).Researchparadigms,andexpectedutilitymodelsfortheprovisionofstep-levelpublicgoods.PsychologicalReview,94,74–83.[doi:10.1037/0033-295X.94.1.74]

RAPOPORT,A.(1988).Provisionofstep-levelpublicgoods:Effectsofinequalityinresources.JournalofPersonalityandSocialPsychology,45,432–440.[doi:10.1037/0022-3514.54.3.432]

RAPOPORT,A.(1993).Incrementalcontributioninstep-levelpublicgoodsgameswithasymmetricplayers.Organisationalbehaviorandhumandecisionprocesses,55,171–194.[doi:10.1006/obhd.1993.1029]

SANTOS,F.C.,Santos,M.andPacheco,J.M.(2008).Socialdiversitypromotestheemergenceofco-operativebehavior.Nature,454,213-216.[doi:10.1038/nature06940]

http://jasss.soc.surrey.ac.uk/17/1/4.html 17 16/10/2015

Page 18: The Production of Step-Level Public Goods in Structured ...simulation.su/uploads/files/default/2014-leon-miguel-alcaide.pdf · Francisco José León Medina, Francisco José Miguel

SCOTT,J.(2005).Socialnetworkanalysis:Ahandbook.London:Sage.

SKYRMS,B.&Pemantle,R.(2000).Adynamicmodelofsocialnetworkformation.PNAS,97(16),9340–9346.[doi:10.1073/pnas.97.16.9340]

SONNEMANS,J.,vanDijk,F.andvanWinden.F.(2006).Onthedynamicsofsocialtiesstructuresingroups.Journalofeconomicpsychology,27,187–204.[doi:10.1016/j.joep.2005.08.004]

SURI,S.&Watts,D.J.(2011).cooperationandcontagioninweb-based,networkedpublicgoodsexperiments,PLoSONE,6(3),1–18.[doi:10.1371/journal.pone.0016836]

SUTCLIFFE,A.,Wang,D.andDunbar,R.(2012).Socialrelationshipsandtheemergenceofsocialnetworks.JournalofArtificialSocietiesandSocialSimulation,15(4),3,http://jasss.soc.surrey.ac.uk/15/4/3.html.

TAKÁCS,K.,Janky,B.andFlache,A.(2008).Collectiveactionandnetworkchange,SocialNetworks,30,177–189.[doi:10.1016/j.socnet.2008.02.003]

TAYLOR,M.(1987).Thepossibilityofcooperation.Cambridge:CambridgeUniversityPress.

TAYLOR,M.&Ward,H.(1982).Chickens,whalesandlumpygoods:alternativemodelsofpublic-goodsprovision.Politicalstudies,30,350–370.[doi:10.1111/j.1467-9248.1982.tb00545.x]

TILLY,Ch.(1978).Frommobilisationtorevolution.Reading:Addison-Wesley.

TSUKAMOTO,E.&Shirayama,S.(2010).Influenceofthevarianceofdegreedistributionsontheevolutionofcooperationincomplexnetworks,PhysicaA,389,577–586.[doi:10.1016/j.physa.2009.10.002]

VANDEKRAGT,A.J.C.,Orbell,J.M.andDawes,R.M.(1983).TheMinimalContributingSetasasolutiontopublicgoodsproblems.TheAmericanpoliticalsciencereview,77(1),112–122.[doi:10.2307/1956014]

YANG,H.,Wu,Z.andDu,W.(2012).Evolutionarygamesonscale-freenetworkswithtunabledegreedistribution.EPL,99,10006.[doi:10.1209/0295-5075/99/10006]

YE,H.,Tan,F.,Ding,M.,Jia,Y.andChen,Y.(2011).SympathyandPunishment:EvolutionofcooperationinPublicGoodsGame.JournalofArtificialSocietiesandSocialSimulation,14(4),20<http://jasss.soc.surrey.ac.uk/14/4/20.html>.

ZSCHACHE,J.(2012).Producingpublicgoodsinnetworks:Someeffectsofsocialcomparisonandendogenousnetworkchange.SocialNetworks,34,539–548.[doi:10.1016/j.socnet.2012.05.003]

http://jasss.soc.surrey.ac.uk/17/1/4.html 18 16/10/2015