Page 1
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
ExploringtheImpactsofNetworksonInformationalandEconomicInterventionsinSolarPVAdoption
StudentTeam:AmaraUyannaandMatthewHaley
InternalTechnicalLead:D.CaleReeves
AdvisorandProjectDirector:Dr.VarunRai
Page 2
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
Abstract
FornoveltechnologiessuchasdistributedPV,effectiveapproachestoacceleratingadoptionmustfacilitateacomplexhousehold-leveldecision-makingprocesscharacterizedbybotheconomicandinformationalbarriers.Usinganempiricallygroundedandvalidatedagent-basedmodel(ABM)ofsingle-familyhouseholdPVadoption,theprojectteamsimulatedarangeofinformation-basedstrategiesincludingrecruitinginterestedindividualsto“champion”installingsolarandconnectingcurrentsolarownerstopotentialadopters.Thesestrategiesaimtoincreaseadoptionbyexplicitlyleveraginginformationexchangeatthelocallevel,whichisknowntobeanimportantdriverofPVadoption.Resultsoftheinformation-basedstrategiesarecomparedtheresultstoasimplesimulatedeconomic-basedstrategy(subsidizedadoption).
Threekeyfindingsemergedfromtheanalysis.First,information-basedstrategiesthatbreakfreeofexistingsocialnetworksandcreatenew“weakties”amongpreviouslyunconnectedindividualsappeartobemoreeffectivethaninterventionsthatrelysolelyonexistingpersonalconnectionstospreadinformation.Second,whenpotentialadoptersarealreadydenselyconnectedandinformationflowsfreelybetweenindividualsbasedonpre-existingrelationships,providingadditionalnewinformationisnecessarytoincreaseadoption.Third,simulatedinformation-basedstrategiescanincreaseadoptionwithareasonableestimatedreturn-on-investmentcomparedtoasimulatedeconomic-basedstrategy.
Thesefindingsyieldtwoconclusions.First,information-basedstrategieshavethepotentialtoplayanimportantroleinPVadoptionwherepotentialadoptersfaceinformationalbarrierstoadoption.ThoseinterestedinPVadoptionmayusefullycontinuedesigning,implementing,andevaluatingstrategiesthatexplicitlyleverageinformationexchangeamongpeers.Second,information-basedstrategiesshouldaimtoformnewtiesandinformationexchangebetweenindividualsratherthansolelyleveragingpre-existingrelationships.Whenencouragingnewconnectionsisnotpossible,newinformation—asintraining—mightinsteadbeprovided.Futureresearchcouldexploretheeffectivenessofcombinationsofinformation-basedandeconomic-basedstrategieswhenappliedsimultaneously.
Page 3
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
IntroductionRelationshipbetweenEconomicsandsolarPVadoptiondecision-makingOverthetwopastdecades,therehasbeenaglobalincreaseineffortstosupportsolaradoption.GlobalinvestmentinsolarPVismotivatedbyadesiretocurbglobalwarmingtransitiontoemissionfreeenergygeneration.EconomicincentiveshavebeenasignificantdrivingforcebehindresidentialsolarPVadoption.Federalinvestmenttaxcredits(FITC),feed-in-tariffs(FIT)aswellasvariousformsofrebatesandsubsidieshavebeenintroducedbygovernmentsworldwide.InBelgium,acombinationofsubsidies,taxcreditsandloanswereintroducedbetween2002and2015toincreasesolarPVadoption.By2012,BelgiumhadasolarPVadoptionrateof8.5%(DeGrooteetal.2016).In2002,Germanyintroducedfeed-in-tariffstoencouragetheadoptionofrenewableenergytechnologiesincludingsolarPV.By2011,theFITpolicieswereintheirsecondphaseofimplementation,andtherehadbeenasignificantincreaseinsolaradoptioninthecountry.Bymid2012,5.1%oftheGermannationalelectricityproductionwasfromsolarPVinstallations(Fulton,Capalino,&Auer2012).TheUnitedStateshasalsoadoptedpoliciestopromotesolaradoptionamongcitizens.In2011,thefederalgovernmentdisbursed$1.1billioninformoffinancialassistancetoincreasesolaradoption.Thisamountwasa500%increasefromtheamountallocatedin2007.Inthesameyear,legislatureacrosstenstatesincludedprovisionsforfinancialincentivestoincreasesolaradoption(Chernyakhovskiy,Ilya,2012).Figure1belowpresentstheobservedandpredictedgrowthofinstalledU.S.solarPVcapacityfrom2010-2023(WoodMackenzie2018).In2012,Austin'selectricutility–AustinEnergy-implementedthevalueofsolar(VOS)policy.TheVOSisaversionofnetenergymetering(NEM)policiesthatenablesthesecustomerstoreceivecreditforexcessrenewableenergygenerationexportedtothegrid.Typically,underNEMthesecustomersarecreditedatthesameretailrateforwhichthecustomerspurchaseelectricity.VOSdiscountsthecreditcustomersreceivebasedonthevalueofthesolartotheelectricitygrid,whichAustinEnergyassessesannually.InAustin,VOSreplacedNEMandisonlyapplicableinresidenceswithsolarPVsystemssmallerthan20kW(USDOE,2018).FollowingtheimplementationofNEMinTexas,statelegislatorsbegantointroduceNEMlegislationtotheir2013legislativesession.InQ1of2013,therewasa53%increaseintherateofsolaradoptionacrosstheUS.TocontinuegrowthinPVadoptionitisimportantforstateandlocalgovernmentstoprovideinformationandfinancialresourcestosustaintheincreaseinsolarPVadoption(Nolletal.2014).
Page 4
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
Figure1EmpiricalandpredictedgrowthininstalledcapacityofSolarPVintheUnitedStatesfrom2010-2023
(WoodMackenzie2018)
RelationshipbetweenInformationandsolarPVadoptiondecision-making
Informationistheadhesivethatbindsdifferentchoices,businessstructuresanditisanintegralpartofaproduct’svaluechain(Evans&Wurster,1997).Wordofmouth(WOM)especially,isapowerfulmechanismforconveyinginformation(Jalilvandetal.,2011).Neighborhoodpeer-effectshavebeenparticularlyeffectiveindrivingtheinnovationofnew,riskyproductsandhavebeenidentifiedasverybeneficialwheninfluentialindividualsinsocialnetworksareconnectedinwaysthateaseinformationdistribution(Kumaretal.,2007).Potentialsolaradoptersfaceseveralbarrierstoadoptionbeyondeconomics.Uncertaintiesabouttechnologyperformance,adoptionmodels(buyorlease),localpolicycontextsandalackofindividuallyrelevantinformationcomplicatethedecisiontoadoptsolarPV(Rai&Robinson2013).Informationinterventions–suchassolarmarketingandutilityseminars-areapowerfulresourceforovercomingtheseuncertainties.PeereffectsarerecognizedtobeparticularlypowerfulinpromotingthediffusionofsolarPV.BollingerandGillingham(2012)findthateachinstallationofsolarPVintheCaliforniamarketincreasestheprobabilityofanotheradoptioninthesamezipcodeby0.78.Informationinterventionsthatareabletoleveragepeereffectsandwordofmouthhavepotentialtoreducetheuncertaintyandnon-monetarycostsassociatedwithPVadoption.InformationinterventionsareabletoidentifytherelativeadvantagesofsolarPV,demonstratethecompatibilityofsolarPVwithacustomer'sbeliefsystem,reduceperceptionsoftechnologyandpolicycomplexityandshowtheresultsofotherinstallations(Nolletal2014.)ThisresearchdevelopsanempiricallyjustifiedmethodformodellinginformationandeconomicinterventionsdesignedtoforincreasesolarPVadoption.WeusetheSECADmodel–anempiricallygeneratedagent-basedmodelofresidentialsolarPVadoptioninAustin,TX–totestandevaluatefour“real-world”policyinterventionsdesignedtoovercomenon-monetarybarrierstosolarPVadoption.WeassesstherateofPVdiffusionbyestablishingbaselineeffectivenessmetricsforvariousinterventionsinanempiricallyandtheoreticallygroundedsetting.
Page 5
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
BackgroundOurresearchattemptstoimprovepolicyandbridgeexistinggapsbetweeninterventiontargetingbasedonnetworkstructuralcharacteristicsandreal-worldsolarPVinterventions.Wereviewtheavailableliteratureonthesetopicsandhighlightthefindingsthatinformourwork.ReviewofSolarInformationInterventions
TounderstandthecurrentrangeofinformationprogramsrelatedtosolarPVwesurveyedtheliterature.InparticularwefocusedonprogramspromotedbySolarCommunityOrganizations(SCOs.)SCOsareformalorinformalorganizationsthatattempttopromoteadoptionofsolarPV.SCOsprovideaccesstocredibleandtransparentinformationaboutthebenefitsofsolarPVandactivelycampaigntopromoteadoptionofsolarPVwithintheiroperationalboundaries(Nolletal2014.)TheresultsofthisreviewaresummarizedinTable1.Solarinformationprogramsarediverseandvarysignificantlyfrommarkettomarket.Thisislikelyduetothegeographicalvariationinsolarmarketsandsolarpolicyacrosstheworld.Assuchmostinformationcampaignsarelocalandtargetedtospecificcommunities.Mosttargetedprograms–suchastheSolarChampionsprogramortheHappyHourprograms-relyonindividualvolunteerstospreadinformationwithintheircommunities.Otherprograms,includingtheOregonSolarAmbassadorprogram,createdatabasestoconnectpotentialsolaradopterswithpeoplewhohavealreadyadoptedsolar.Criticaltothesuccessofsolarinformationinterventionsisestablishingtrustbetweentheorganizationsthatdistributeinformationandtherecipientsofinformation(Nolletal2014.)Manyinterventionsprimarilyfocusresourcesontrainingandeducatingprogramvolunteerswhohaveestablishedtrustedrelationshipswiththeircommunities.
ProgramType Description
UtilityMailers/Pamphlets/CallCenters/Massmarketadvertising
Themostextensiveinformationalappealstosolarconsumershavetraditionallybeenthroughmassmarketmechanisms(Costanzoetal.1986).Acursoryinternetsearchrevealsthatalmosteverysolarstakeholderreliesheavilyonbroadlydistributedmass-marketinformationsuchasmailers,electronicadvertisingandpamphletsasaprimarymarketingmechanism.Massadvertising,howeverisahighlypassiveformofinformationdiffusion.SurveysshowthatmassmediaisineffectiveatinfluencingmembersofthepublictoovercomeadoptionbarrierstoPV(Palm2016).Thehighattitudinalbarrierstosolaradoption(Raietal.2016)limitstheabilityofpassiveinformationtoleverageactivepeereffectstospreadinformation.
OpenHouseorSolarGuidedTours
SolartoursandopenhousesareanemergingPVadvocacymechanism.TheAmericanSolarEnergySocietycoordinatestheNationalSolarTour,wheremembersofthepublicareinvitedto
Page 6
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
visitthehomesofpreviousPVadopters.In201290,000participatedintheASESNationalTour.Thetourengagesactivepeereffectsbygivingpotentialsolaradopterstheopportunitytointeractwithexistingadopters.SolartourshavealsobeeninstrumentalinthehighrateofPVdiffusiontheSwedishcommunityofBerg(Palm2016).
Education/SchoolEvents
SolarOregon,PGEandmanyotherutilitiesinstallPVonPublicSchoolsandhavedesignedsolarbasedcurriculumtrainingpackagesforteachersthatfitstateengineeringanddesignandsciencebenchmarks.Between2004and2009PGEsolarcurriculumwastaughttoover200,000students(CleanEnergyGroup2009).
SolarChampions
Manyutilitiesandsolaradvocacyorganizationsoffer"SolarUniversities."Theutilityrecruitsmembersofthepublicfromneighborhoodsinitsserviceareaandprovidesthemtechnicaleducationandtraining.Thesesolarchampionsarethenempoweredtocoordinate,trainandleadsolarprogramsintheirowncommunities.
PorchTalks
PermembersoftheMuellerMegawattprojectinAustin,themosteffectiveapproachtospreadingsolarinformationwaswhattheycalled“porchtalks.”ThesetalksinvolvedsolaradvocateswhowerecommunitymemberssittingontheirporchafterdinnerandaskinganypasserbyiftheyknewaboutsolarandwereinterestedinputtingPVontheirroof(Nolletal.2014)
CelebrityEndorsement
Celebrityendorsementsaredesignedtoincreasecustomerknowledgeandawarenessofaproduct(Kowalska-Pyzalska2017).CaliforniasolarprogramadministratorsattributepartofsolartakeoffinCaliforniatotheeffectofcelebrityendorsementfromtheGovernorArnoldSchwarzenegger(CleanEnergyGroup2009).
Local
Seminars,HappyHoursandWorkshops
Attendingseminars,workshopsandhappyhourshostedbySCOs,utilitiesorsolarretailersisinfluentialintheadoptionprocess(Palm2016).HappyhourswereidentifiedascrucialtothesuccessoftheMuellerMegawattproject(Nolletal.2014).SCOsalsocoordinateworkshopsandseminarstoeducatemembersaboutthebenefitsofsolar(Nolletal.2014).seminars,workshopsandhappyhourshostedbySCOs,utilitiesorsolarretailersisinfluentialintheadoptionprocess(Palm2016).HappyhourswereidentifiedascrucialtothesuccessoftheMuellerMegawattproject(Nolletal.2014).SCOsalsocoordinateworkshopsandseminarstoeducatemembersaboutthebenefitsofsolar(Nolletal.2014).
Page 7
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
GroupCompetitions
SmartPowermanagedaprogramthatorganizedfriendlycompetitionsbetweenmunicipalitiestoencouragesign-upstogreenpowerprograms(CleanEnergyGroup2009).Similarprogramsareenvisionedforneighborhoodsolarinstallations.EnergyTrustofOregonapproachedcompaniesthathadpubliclyadvocatedforsolartoruninter-companychallengestoenlistemployeesasnewsolarcustomers(CleanEnergyGroup2009)
OregonSolarAmbassadorProgram
Oregoncreatedadatabasewhereprospectivesolaradoptersareputincontactwithexistingsolaradopterstodiscusswhytheychosetoadoptsolar.Adoptersareencouragedtosharekeyinformationthatenabledthemtoovercomeattitudinalbarrierstosolaradoption(Nolletal.2014).
Table1Descriptionofreal-worldsolarinformationinterventions
ReviewofNetworkInformationDiffusion
Theliteraturesuggeststhatasmallsubsetofinfluentialindividualscanshifttheattitudesofthemajority(Kovács&Barabási2015).Theseinfluentialindividualsarelocatedatcentralnetworknodes,thataremostefficientatdisseminatinginformationthroughouttheentirenetwork.Toinvestigatethestrengthofdifferentstrategies,weincorporatednetworkcentralityinouranalysis.Specifically,weexploredtwotargetingmethodsforseedinginformation:
o Randomlyseedingagentsasabaselinetoevaluatealternativestrategies(Araletal.2013).
o Targetinghighlyconnectedagents(Rai&Robinson2015;Centola&Macy2007).
ReviewofSECADABM
Agent-basedmodellingenablesustostudyinteractiveaspectsofanagent’slife.Theabilitytomanipulatevariousinteractionsenablesustoexplorethedifferentwaysthroughwhichindividualdecisionmakingcanmanifestasobservablemacroscopicprocesses(Rai&Robinson2015).UsingABM,wecanbridgethegapbetweenindividualbehavioralresponsesandmoreaggregatelong-termpossibilities(Hoganetal.2004).TheSECADABMfirstpublishedbyRobinson&Rai(2015)isatheoreticallybasedmodelofsolaradoptioninAustin,Texas.Themodelcontains~177,000agentsthatrepresenteachoftheGISmappedhouseholdsinAustin.Themodelassignseconomicandattitudinalattributesforeachagentbasedonsurveyandpropertyvaluedata.Themodelisempiricallyvalidatedandcanpredictadoptiontrendsonatestdata-setwithheldduringmodelfitting.
InterventionExperimentsWeusetheSECADmodeltounderstandhowinformationprogramscandriveadoption.Inthispaper,wejustifyhowinformationwillupdateattitudewithinthemodelandestablisha
Page 8
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
methodofdeterminingthecostofinformationprograms.Inthisresearch,weexplorethenatureofsocialnetworksandtheirimpactsonincreasingsolarPVadoption.Thisworkwillattempttoidentifythecharacteristicsofeffectivesolarinformationprograms.Interventionsareassessedontheirabilitytoincreaseadoptionswhileminimizingcost.Thepurposeoftheseexperimentswillnottobetodesignanoptimalprogramthatmaximizesadoptionforminimalcost,buttotestseveralinterventionstrategiesandrationalizetheirresults.Weareaimingtoidentifysalientprogramfeaturesthatimproveadoption.
MethodsABMOverviewWeassesstheimpactofseveralpracticalsolarinformationprogramsthatcurrentlyarebeingimplemented.Fourrealworldinformationinterventions–chosenfromtheninepresentedabovewillbereplicatedwithintheSECADmodel.UsingtheSECADmodelwecanevaluatethestrengthofinformationinterventions,potentialreturnoninvestmentandtheintervention'sabilitytoleveragewordofmouthattitudediffusion.Wecanalsoevaluatethedifferentinformationstrategiesagainstoneanother.
TheSECADmodelcanbeusedavirtuallaboratorytotesttheefficacyofvarioussolarinterventions.TheSECADmodelallowstestingofpurelyinformationalinterventionswithinthemodel.TheseinterventionsareperformedwithinSECADbyestablishingconditionsthataltertheevolutionofagent'ssociallyinformedattitudeswithintheSECADmodel.WithinSECADthesociallyinformedattitudeofagentsisrepresentedbythenumericalparametersia.Onceanagentssiaexceedsacertainthreshold(siathreshold)theyareclassifiedasattitudinallyactivated.TheSECADmodelreliesonadualthresholdmodelforanagenttobeconsideredasolaradoptertheymustbebothattitudinallyandeconomicallyactivated(seeFigure2).Tobeeconomicallyactivatedanagent’sperceivedallowablepaybacktimeisgreaterthantheirempiricalpaybackperiod.TheseconditionsmayalsobealteredtostudytheoutcomeofeconomicinterventionswithinSECAD,howevereconomicinterventionsareoutsidethescopeofthiswork.ThedesiredoutcomeofinformationinterventionswithinSECADwillbetoincreaseindividualagent’ssiaanddrivesolaradoptionwithinthemodel.
Figure2FlowsheetforSECADadoptiondecisionprocess(Rai&Robinson2015)
Page 9
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
TheSECADmodelallowsthemodelertounderstandtheimpactofinformationinterventionsovertime.Ateachtimestep(t=0,1,2,3,4....)agents'attitudesaboutsolarPV(sia)andtheuncertaintiesaroundthoseattitudes(U)areupdatedthroughinteractionswithotheragents.ThisallowsustodynamicallymonitorhowinformationinterventionscanmanifestasattitudechangesovertimewithintheSECADmodel.Thelayerofinteractionspossibleintheagent-basedmodelallowsustorecreatethehumanrealityinwhichvariousinformationdiffusionprocessareinterconnected.Interactionbetweenagentsismodeledusingarelativeagreement(RA)algorithm(Deffuantetal.,2002,2000;HegselmannandKrause,2002;MeadowsandCliff,2012).Ateachtimesteppairsofagent’sexchangeattitudeswithagentswithintheirsocialnetworkthatsharesimilarattitudestowardsolar(sia)anduncertaintyaroundthoseattitudes(U).AdetailedformulationoftheRAalgorithmispresentedinRai&Robinson2015andRobinson&Rai2015.Fortheinformationinterventionsmodelledinthispaper,theSECADmodelranfor24quarters(2008-2014),withfourtimestepsforagentupdatesperquarter.
Twoprimarytypesofnetworks–geographicalandsocial-aredefinedwithintheSECADmodel.Geographicneighborhoodsaroundanagentiaredefinedwithinthemodelasthecollectionofagentsthatarewithinacertainradius(2000ft)ofagenti.WithinSECADthemeannumberofagentswithineachgeographicalneighborhoodis498.
Socialnetworksarederivedfromgeographicalneighborhoods;however,theyarefurtherconstrainedbytheeconomicsimilarity(calculatedusinghomevalueasaproxyforwealth)ofagents.Therefore,onlyeconomicallysimilaragentswereallowedineachsocialnetwork.Finally,10%ofeachsocialnetwork’sconnectionwererandomlyrewiredwithagentsanywherewithintheSECADmodel,toaddadegreeofrandomizationtothesocialnetworks.ItisthesocialnetworksforeachagentwithinSECADthatdictatewhoeachagentupdateswithduringrelativeagreement.Agentswillonlyupdatesiawithotheragentswithintheirsocialnetwork.WithinthispaperweexploreonlythemanipulationswemaketoSECADtoperformourinterventionexperiments.ForacomprehensiveexplanationofthefullyvalidatedSECADmodelrefertoRai&Robinson2015andRobinson&Rai2015.
SelectionofRealWorldSolarInformationProgramsFromtheninesolarinformationinterventionswereviewedweselectfourtoreplicatewithintheSECADmodel.Thefourchoseninterventionsare:
• SolarChampions• PorchTalks• PhoneAmbassadors• CelebrityInterventions
Twoprimaryfactorsdictatedthechoiceoftheseprograms,theirsimilaritywithotherprogramsreviewedandtheirabilitytoleveragethesocialnetwork/wordofmouthinteractionsthatgoverntheSECADmodel.Forexample,pamphletsandmailerswereignoredfrommodelingsinceitisunlikelythattheywouldgeneratesignificantfollowonwordofmouthinteractions
Page 10
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
SolarChampionswaschosensinceitaprogramthatleveragesthesocialnetworksofthepeoplewhoparticipateintheprogramtospreadinformation.Inthissense,itisliketheHappyHourandWorkshopsprograms.Inourmodeling,weusetheSolarChampionprogramasanumbrellaforsimilarprogramsthatrelyonparticipant’ssocialnetworkstospreadinformation.PorchtalkswaschosenforexperimentationinSECADsinceitaprogramthatutilizesthegeographicnetworksofparticipantsintheprogramtospreadinformation.Inthissense,itislikesolaropenhousesandguidedtours,inthattheinterventionisconfinedwithinalocalneighborhood.Inourmodeling,weusetheporchtalkprogramasanumbrellaforsimilarprogramsthatrelyonparticipant’sgeographicalneighborhoodstospreadinformation.Phoneambassadorsandcelebrityinterventionswerechosenbecauseoftheirabilitytospreadinformationbeyondtheexistingnetworkofindividuals.Theseprogramscanspreadinformationbetweenpeoplewhohavenopre-existingconnectiontooneanother.InformationalSeedingExperiments
SeedsandFollow-OnsSeedsaretheagentswhowegiveinformationtoowithintheexperimentalinterventions.Thesearetheinfluenceragentswhoareabletopassoninformationwithinthemodelthatupdatesotheragent’sattitudes.Forexample,inthecaseofthesolarchampionsprogram,theseedistheagentwhobecomesachampionbyreceivingonsolartrainingfromhisutility.Existingagentsarechosenasseedswithinthemodel(exceptforthecelebrityintervention)basedoneithertheirattitude(siai>siathresh)oriftheyareanadopter,inwhichcasetheywillstillhaveattitudegreaterthantheadoptionthreshold.Seedsbydefinitionhavehighsia.Withineachoftheinterventionexperimentsweassignbetween1and500agentsasseeds.
Follow-onsaretheagentsthattheinitialseedagentsareabletoinfluencewiththeinformationfromtheintervention.Follow-onsaretheagentswhoareinfluencedandupdatetheirattitudesonsolar.Dependingontheinterventionbeingmodeledpotentialfollow-onsareidentifiedasagentswithinaseedssocialnetwork,agentswithinaseedsgeographicalnetwork,agentsthatareinterestedinsolar(approachingattitudinalactivation)orcanhavenopre-existingconnectiontotheseedagent.Thefinalgroupoffollow-onsisdeterminedbyrandomlyselectingapercentageofagentsfromthepoolofpotentialfollow-onagentsidentifiedineachintervention.Thepercentageofagentsselectedasfollows-onsisexpressedasaratefrom0-1.Thisisknownasthefollow-onrate.Thisraterepresentshoweffectiveaseedagentisininfluencingtheirfollowers.
TheimpactofinformationwithinSECADTheSECADmodelusesadualthresholdconcepttodeterminewhetheranagentwilladoptsolarPV.Anagentmustbebothattitudinallyandeconomicallyactivatedifitistoadopt.Informationalandeducationalsolarincentiveprogramsaredesignedtoinfluenceaperson’sattitudetowardssolar.Tomodeltheeffectsofinformationalinterventionswithin
Page 11
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
theSECADmodelwehavejustifiedtwoalternativesforhowanagent’sattitudesiawillbemodifiedwhenexposedtoinformationresources.
- MaximalEffectUpdating
Weassumethateachfollow-onagentthatencountersinformationdisseminatedfromourmodeledinterventionswillbecomeattitudinallyactivated(siai>siathresh).Theamountofsolaradoptiondrivenbythese“perfectinformationprograms”isatheoreticalupperlimittowhatevercanbeachievedforeachinterventionwithinthemodel.However,this“idealizedattitude”adjustmentallowsustoeasilyevaluatecompetinginformationalinterventionstrategiesagainstoneanotherbyevaluatingtheupperboundforeachstrategy.
- RelativeAgreementUpdating
RelativeagreementupdatingbuildsuponthetheorydevelopedwithintheoriginalSECADmodel.UnliketherelativeagreementprocesswhichtakesplaceintheSECADmodel4timesperquarter,interventionupdatesonlyoccurintimeperiodstheinterventionisactive(Fortheseexperimentsthisisalwaysatinitializationt=0).Ateachtime-step,seeds–whichhavehighattitude-willupdateusingrelativeagreementwithallthefollow-onstheyareconnectedto.Theeffectoftheseupdateswillbetoincreasefollow-onagentattitudesthroughoutthemodel.Howseedsareselected
Twotargetingmechanismsareusedforidentifyingseedagentswithintheinterventions:- Randomselection
Fromthepoolofpotentialseedagents,werandomlyselectseeds.Thistargetingstrategyrepresentsaminimumcostapproachtoselectinganinformationprogramparticipant.
- HighKselection
Fromthepoolofpotentialseedagents,weselectseedsthathavethehighestnumberofconnections(K)tootheragents.Essentially,weidentifytheagentswiththemostpotentialfollow-ons.Thistargetingstrategyrepresentsamaximumcostapproachtoselectinganinformationprogramparticipant,whopotentiallycaninfluencethemostfollow-onagents.Otheradvancedseedtargetingtechniqueswereconsidered,howeverimplementationofthesemethodsisbeyondthescopeofthisreport.
InterventionTimingWhenmodelingeachinterventionwithinSECADwegenerateseedandfollow-onagentsonlyonce–duringinitializationatt=0.While“inreality”informationprogramsarelikelytobecarriedoutoveradurationoftime,forthepurposeoftheseexperimentsweseedinformationonlyoncesotheeffectofeachinterventioncanbeobservedwithinSECADovertime.Seedingatmultiplytimestepswouldcomplicateseparatingtheadoptionsgeneratedbyeachindividualseedingevent.
InterventionDesign
Page 12
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
ThedesignofeachindividualinterventionissummarizedintheTable2below.Interventionsaredesignedtoapproximatetherealityofeachofthefourselectedinterventions.Thesuccessofeachadoptionismeasuredbyitsabilitytogenerateadditionaladoptionsoverthebaseline(no-intervention)model.
Program Seeds/Follow-OnRate Mechanismwithinmodel
SolarChampions:
Seeds:AgentsthatareseededwithinformationareSolarChampions.SolarChampionsaretrainedbytheutility/Solaradvocacyorganizationanddisseminateinformationtotheirimmediatesocialnetwork.
FollowOn:Follow-onsaretheagentswithinthesolarchampionseedssocialnetwork.
InitialTargeting:Tobeachampion(seed)anagentmusthavesiagreaterthansiathreshold.Atinitializationagentswithhighestattitude(sia>siathreshold)areidentifiedaspotentialchampions.Fromthepoolofpotentialchampions,weselectagentsaschampions(seeds)randomlyorbasedonhighK.
Follow-onsareagentsthatarewiththechampionsimmediatesocialnetwork(Alters).
Operation:Theinformationeffectofagentsispassedontofollowonagentsusingrelativeagreementalgorithmor‘maximaleffect’.
SolarAmbassadors
Seeds:AgentsthatareseededwithinformationareSolarAmbassadors.Solarambassadorsareaselectgroup(phonebook)ofagentswhohaveadoptedsolar.
FollowOn:Follow-onsareagentswhoareinterestedininstallingsolarbutrequiremoreinformation.Theyreachouttoambassadorstogetmoreinformation.
InitialTargeting:Atinitializationagentswithwhohaveadoptedsolarareidentifiedaspotentialambassadors.Fromthepoolofpotentialambassadors,weselectagentsasambassadors(seeds)randomlyorbasedonhighK.
Potentialfollow-onsareagentswithinthemodelthatareclosetobeingattitudinallyactivated(Thisiscalculatedusingathresholdof0.6siathreshold.)Theseagentsdonotneedtohaveanypriorsocialorgeographicconnectionwithambassadors.
Operation:Anetworkedgethatconnectseachagentwithinourfollow-onpoolandourchosenambassadorsiscreated.Thisisactionedbyaddingtheambassadorstothetargetsalternetwork.
PorchTalk
Seeds:Agentsareseededwithinformationtohostporchtalks.Porchtalkersareagentswhoareadopters.Porchtalksdisseminateinformationtotheirimmediategeographicalnetwork.
FollowOn:Follow-onsaretheagentswithinthesolarchampionsgeographicalnetwork.
InitialTargeting:Atinitialization,fromapoolofalltheagentsinthemodelwhoareadopterswe-randomlyorusingHighK–selectasubsetofadopterstohostporchtalks.Thesearetheseeds.
Wethendetermineasubsetoftheseedsimmediategeographicalneighbors.Thispoolisrefinedusingaparticipationrate.Thesearethefollow-ons.
Operation:Theinformationeffectofseedagentsispassedontofollow-onagentsusingtherelativeagreementalgorithmor‘maximaleffect’.
CelebrityAdvocates
Seeds:Agentsthatareseededwithinformationarecelebrities.Celebritiesabletotranscendsocialandgeographicalnetworkswithinthemodelanddisseminateinformationtoanyagentwithinthemodel.Inthismodelonly20agentsareseeded.
InitialTargeting:ThecelebrityagentsweseedwithinformationdonotexistwithintheSECADmodel.Wegenerateattitudinallyactivatedcelebrityseedagentswithinthemodelwithsiabetween0.6and1(Unif[1,0.6,1])andU=1/sia.
Thepoolofpotentialfollow-onsforthisinterventionarealltheagentswithintheSECADmodel.
Page 13
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
Table2:Describesthefourinterventionsmodelledinthisresearchandexplainstheiroperationwithinthemodel
SECADSetupEachinterventionissimulatedwithinSECADovera6-yearperiod.TogenerateadoptioninformationeachiterationoftheSECADmodelruns5batchesof24quarterindividualruns.
SeedsvsFollow-OnPhaseDiagramsMostourresultsarepresentedasphasediagramsthattrackadditionaladoptionsgeneratedbyeachinterventionasafunctionofseedquantityandfollowonrate.Phasediagramsallowvisualizationoftheeffectivenessofeachinterventionasafunctionofthenumberofseedsandtheseedsabilitytogeneratefollow-ons.Thisisausefulvisualtoolforattemptingtoviewchangesinadoptionbetweencausedbytheinterventionexperiments.Phasediagramsaregeneratedby1920separateparameter-blastiterationsofeachinterventiontoexploretheentireparameterspace.Parameter-blastsiterationsareconductedbyrandomlygeneratingtheseedquantity(1-500)andfollowonrate(0-1)priortoeachrun.ThesolarPVadoptionsplottedwithinthephasediagramresultsaretheadditionaladoptionsgeneratedbytheinterventioncomparedtotheempiricallyobservedadoptionsin2014.
Figure3Examplephasediagram.
FollowOn:Allagentsarepotentialfollow-ons.
Operation:Theinformationeffectofseedagentsispassedontofollow-onagentsusingtherelativeagreementalgorithmor‘maximaleffect’.
Page 14
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
MeasuringtheValueofInformationandEconomicInterventionsWeusetheSECADvirtuallaboratorytotesttheefficacyofvarioussolarinformationinterventions.Asinformationalresourcesaredispersedwithinthemodeltheywillalteranagent’ssiaanddrivesolaradoptionwithinthemodel.ToassesstheefficiencyofinformationalexperimentswithinSECADitisnecessarytoestablishaframeworkformeasuringthevalueofinformationalinterventionswemodel.ThisallowsexperimentalinterventionswithintheSECADmodeltobecomparedtotheeffectivenessofreal-wordinformationalandeconomicinterventions.Tofacilitatethesecomparisons,weusethefollowingmethodtomeasurethecostofinformationalinterventionsinSECAD.Themethodweutilizeisacomparisonofthereturnoninvestment–measuredinadditionaladoptions-betweenourexperimentalinformationalinterventionandarandomlydispersedeconomicinterventionwithinthemodel.ReturnoninvestmentoradditionaladoptionsinSECADismeasuredbytotalincreasedsolaradoptionsovertheobservedempiricaladoptionsinAustinduring2008-2013.
𝐴𝐴 = 𝐴$ −𝐴& (1)
where:
AAisadditionaladoptionsAiistheamountofadoptionsforaninformationalintervention
Aeistheamountofobservedempiricaladoptions.TheeconomicinterventionwemodelinSECADsimplymeasurestheadditionaladoptionscreatedbyrandomlyselectingagentswithinthemodelandseedingthemwithsolarPV.WithinSECADthissimplyinvolveschangingtheseedagent’sadoptionstatusfromnon-adoptertoanadopteratinitialization.Tomodelthesameparameterspaceasourinformationinterventionswerandomlyseedbetween1and500agentsandrecordtheresultingadditionaladoptions.Tomeasuretheeffectivenessoftheeconomicinterventionweevaluatetheregression:
𝐴𝐴& = 𝛽& × 𝑆& + 𝑐& (2)where:
AAeisadditionaladoptionsfromtheeconomicintervention
beistheamountofadditionaladoptionsforonemoreeconomicseed
Seistheamountofeconomicseeds
ceisthemeanadditionaladoptionswith0economicseeds.
Thisresultofthisregressioncanthenbecomparedtothefollowingregressiononeachofthephasediagramsgeneratedbyourinformationinterventionexperiments:
𝐴𝐴$ = (𝛽$,. ×𝑆$) +(𝛽$,0 × 𝐹𝑂𝑅)+𝑐$ (3)
Page 15
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
where:
AAiisadditionaladoptionsfromtheinformationalintervention
bi,sistheamountofadditionaladoptionsforonemoreinformationalseed
bi,fistheamountofadditionaladoptionsforincreasingfollow-onrate
Siistheamountofinformationalseeds
FORisthefollow-onrateceisthemeanadditionaladoptionswith0informationalseeds.
Theresultsofeachoftheseregressionscanbecomparedtoprovideanestimateofthedifferentialreturnoninvestmentprovidedbyaneconomicseedcomparedtoaninformationalseed.Weachievethisbycomparingtheregressioncoefficientsbeandbi,s,assumingbothregressionsarestatisticallysignificant.Ifweassumethecostofaneconomicseedissimplythecostofasolarsystemforthatseed,wetheratioofbi,s/beprovidesanestimateofthehowmanysolarsystemseachinformationalseedisworth.Thisresultgivesasimpleindicationofhowmucheachinformationseedisworth,andhowmuchareal-worldsolarinformationinterventionprogramshouldbewillingtopayeachseed.
ResultsInthissection,wepresenttheresultsoftheparameterblastsimulationsdescribedinthemethodssectionabove.Foreachintervention,resultsarepresentedonphasediagramsthatevaluateadditionaladoptionsgeneratedovertheseedandfollowonparameterspace.Foreachintervention,wepresenttheresultsforbothrandomandHighKseeding.Withinthephase-diagramswearelookingfordiscernabletrendsandgrowthinsolarPVadoptionresultingfromthemodeledinformationinterventions.SolarChampions
ThephasediagramsforbothHighKandrandom(Figure4)seedingofthesolarchampionsinterventionareshownbelow.Inbothsolarchampionmodels,weobservethattheinformationinterventionhasnotresultedinsignificantenoughadditionaladoptionstoovercometherandomstochasticityofadoptionsintheSECADmodel.Weseenodiscernableadoptiontrendacrosstheseedandfollowonparameterspace,andincreasesineitherparameterfailtogenerateadoptionpatterns.ThisresultisfurtherillustratedinFigure5whichpresentstheobservedempiricaladoptions(yellowcurve)inAustinoverthe2008-2014timespanagainsttheadoptionresultsgeneratedbythesolarchampionintervention(reddots,blackcurveismeanadoptions).Weseethattheadoptionsgeneratedbytheinterventionarespreadrelativelyevenlyaroundtheempiricaladoptionscurve,sometimesimprovinguponandsometimesunderperformingtheempiricaladoptions.Weseethatadditionaladoptionstracksempiricaladoptionsrelativelycloselyandourinterventionhasnotcausedasignificantchangeinsolaradoption.
Page 16
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
Figure5Observedadoptionsofthesolarchampionsinterventionovertime(individualobservationsinredand
meaninblack)againstempiricallyobservedadoptionsbetween2008and2014.
PorchTalksThephasediagramsforbothHighKandrandom(Figure6)seedingoftheporchtalkinterventionareshownbelow.Bothphasediagramsgiveverydifferentresults,withrandomseedingsignificantlyoutperformingHighKseeding.ThisfindingprovidesaninterestinginsightintothenetworklocationsoftheHighKagentswithintheSECADmodel,andthepowerofthistargetingstrategy.ItappearsthatthefailureofHighKtargetinginthismodelisafunctionofthelocationofagentswithHighKcharacteristics.ItisapparentthatallHighKagentsexistwithinasmallnumberofneighborhoods.Thisissensiblewhenconsideringhowsocialnetworksaredefined(seeABMoverview),usinggeographical
Figure 4 Phase diagram of solar champion intervention: Random (on left) and HighK (on right)
Page 17
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
neighborhoodsandanadditionalhomophilyconstraint.Therefore,smallanddenselypopulatedareasarelikelytocontainthemajorityofHighKagents.Asweaddnewseedswecontinuetore-engagethesamefollow-onnetworksthathavepreviouslybeentargetedandsolarinformationiscontinuouslyre-distributedwithinthesamefewnetworks.Thisstrategyfailstodriveanynoticeableadoptiontrendswithinthemodel,anddoesnotovercometherandomstochasticityofadoptionsintheSECADmodel.Alternatively,randomlyselectingagentsasseedsproducesanoticeableadoptiontrendwithinthephasediagram,andhasresultedinsignificantadditionaladoptions.Asthenumberofseedsandadoptionrateincreaseanobvioustrendofadditionaladoptionsemerges.Thisresultisobserveddespiterandomseedshavinglessfollow-ons(bydefinition)thenHighKseeds.Itappearstherelativesuccessofrandomseedingiscausedbythelargercumulativefollow-onnetworktheinterventioncanreachcomparedtoHighKseeding.Whileeachrandomseedmayhavelessindividualfollow-ons,itislikelythateachofthesefollow-ongroupsisunique.WhileontheotherhandHighKseedshavefollow-ongroups,itislikelythesegroupsarehighlysimilarandthecompleteinterventionfailstospreadsolarinformationandcausesolaradoption.
SolarAmbassadorThephasediagramsforbothHighKandrandom(Figure7)seedingofthesolarambassadorinterventionareshownbelow.Inbothphasediagrams,weobservethatthesolarambassadorinterventionhasresultedinsignificantadditionaladoptionsabovetheempiricallyobservedadoptionsin2014.Theadoptiontrendswithinthephasediagramsclearlyindicatethatthesolarambassadorprogramhassuccessfullydrivenadoptionwithin
Figure 6 Phase diagram of porch talk intervention: Random (on left) and HighK (on right)
Page 18
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
themodel.Evenwithrelativelyfewseedsandlowfollow-onratesanobservableadditionaladoptiontrendemerges.TherandomandHighKmodelsforsolarambassadorbothgenerateverysimilarphasediagrams.Thisisexpectedasthesolarambassadorinterventiondoesnotrelyontheseedsexistingnetworkstospreadinformation,ratherinformationisspreadtofollow-onagentsbyconnectingthemtoaseedagentthattheywerenotpreviouslyconnectedto.Therefore,itisunlikelythattargetingseedagentsbasedontheirexistingnetworks(asisdoneinHighK)willgeneratesignificantlydifferentresultsthanrandomlyseedingagents.
Additionaladoptionsinthephasediagramincreasesignificantlyacrosstheverticalaxisofthephasediagram.Asthenumberoffollow-onsconnectedtoeachseedincreases,sodoesadoptionwithinthemodel.However,thephasediagramsappeartoshowaweakercorrelationbetweenthenumberofseedsandadditionaladoptionwithinthemodel.Thisresultisconfirmedbymultivariatelinearregressionoffollowonrateandnumberofseedsonadditionaladoption.Theregressionconfirmsthattheseedsparameterisnotrelatedtotheresponse.Thisresultisproblematicasitindicatesthatthenumberofseedsisunrelatedtotheresponse.Itislikelythatthisisanartefactofapoorlyspecifiedseedvariablewithinthemodel,sinceeachseedupdateswiththesamegroupoffollow-onswithinthismodelthemarginaleffectofadditionalseedsisrapidlyreducedafteronlyafewseedsareincludedinthemodel.
CelebrityAdvocatesThephasediagramsforbothHighKandrandom(Figure8)seedingofthecelebrityadvocateinterventionareshownbelow.Thecelebrityinterventiononlyexplorestheseedparameterbetween1and20.Inbothphasediagrams,weobservethattheintervention
Figure 7. Phase diagram of phone ambassador intervention: Random (on left) and HighK (on right)
Page 19
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
hasresultedinsignificantadditionaladoptionsabovetheempiricallyobservedadoptionsin2014.Theclearadoptiontrendswithinthephasediagramsclearlyindicatethatthecelebrityadvocateshaveverysuccessfullydrivenadoptionwithinthemodel.Asthenumberofseedsandadoptionrateincreaseanobvioustrendofadditionaladoptionsemerges.Theeffectivenessofthecelebrityinterventionissomewhatexpected,ascelebritieshaveaccesstothelargestfollow-ongroupwithinthemodelanddonotrequireaconnectiontothefollow-ons.ThisresultisfurtherillustratedinFigure9,whichagaincomparestheobservedempiricaladoptions(yellowcurve)inAustinoverthe2008-2014timespanagainsttheadoptionresultsgeneratedbythecelebrityintervention(reddots,blackcurveismeanadoptions).Weseethatataveryearlytimetheadoptionsgeneratedbytheinterventionseparatesfromtheempiricaladoptionscurve.AninterestingobservationwithinthecelebrityphasediagramsistheextenttowhichrandomseedingoutperformedtargetedHighKseeding.HighKcangenerallybethoughtofasamore‘administrativelyexpensive’seedingtechniqueasitrequiresidentifyingnetworkcharacteristicsofeachagent.
Figure 8. Phase diagram of celebrity advocate intervention: Random (on left) and HighK (on right)
Page 20
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
Figure9.Observedadoptionsofthecelebrityadvocatesinterventionovertime(individualobservationsinredand
meaninblack)againstempiricallyobservedadoptionsbetween2008and2014
DeterminingtheValueofInterventionsToevaluatetheeconomicsoftheinterventionsmodeledinthisworkweusethemethoddescribedpreviously.WefirstevaluatetheresultsofasimpleeconomicinterventionwherebywegiveeachseedagentsolarPVatinitialization(t=0).Thisinterventionhasaneasilyevaluatedeconomiccostofandvalueof:
𝐶𝑜𝑠𝑡($) = 𝑁 × 𝑃Value($)=𝑁 × 𝑃 ×𝛽&
(4)(5)
where:
Nisthenumberofseeds Pisthepriceofeachsolarsystem.
beistheamountofadditionaladoptionsforonemoreeconomicseed
Wethencomparetheperformanceofeachinformationinterventionagainsttheperformanceofaneconomicinterventiontodeterminethevalueofeachinformationalseedcomparedtothecostofaneconomicseed.
Page 21
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
Figure10.Regressiononeconomicintervention
TheresultsoftheeconomicinterventionarepresentedinFigure10above.Weseefromsimplelinearregressionusingequation(2)ontheseresultsthateachadditionaleconomicseed(Se)isworthapproximately1.69(be)additionaladoptions.Wethenperformmulti-variatelinearregressionlinearusingequation(3)todeterminebi,sforeachoftheinformationalinterventions.TheseresultsaresummarizedinTable3below.
Program Targeting bi,s RelativeValue(bi,s/be)
SolarChampions Random 0.02 0.012
HighK -0.02 -0.012
PorchTalk Random 2.54** 1.5**
HighK -0.10** -0.06**
SolarAmbassador Random 0.03 0.018
HighK 0.07** 0.04**
Celebrity Random 168.1** 99**
HighK 78.7** 47**
**statisticallysignificant
Table3Assessingthevalueofinformationinterventions.
Theresultsofeachoftheseregressionsarecomparedintheright-handcolumnofTable3andprovideanestimateofthedifferentialreturnoninvestmentprovidedbyaneconomicseedcomparedtoaninformationalseed.Ifweassumethecostofaneconomicseedis
AA»(1.69xSe)+70
Page 22
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
simplythecostofasolarsystemforthatseed,theratioofbi,s/beprovidesanestimateofthehowmanysolarsystems(oreconomicseedsSe)eachinformationalseedisworth.
DiscussionDesigningInterventionstoSucceedAkeygoalofthisresearchwastoidentifyfactorsthatcausesolarinformationinterventiontosucceed.Whilewehavenotattemptedtooptimizeinterventionsthemselves,ourresultsprovideusefulinsightsintothecharacteristicsthatdefinethesuccessofourmodeledinterventions.Interventionsthatrelyonexistingnetworkconnections(socialnetworkinSECAD)tospreadinformationarelesseffectiveatdrivingsolaradoptionthaninterventionsthatforceagentstocreatenewconnections.ThisismadeclearbyacomparisonoftheSolarChampionsprogramtotheotherinterventionswemodel.TheSolarChampionsinterventionreliesonagentsspreadinginformationaboutsolartotheagentswithintheirexistingsocialnetworks.Unliketheotherinterventionswemodel,seedsandfollowonagentsinthismodelarealreadyconnected,thustheeffectivenessofthisprogramessentiallyreliesonseedagentsknowingasmanyuniquefollow-onagentsaspossible.However,asourresultsdemonstrateevenwithupto500seedagentsthisinterventionisnotpowerfulenoughtogenerateobservableadditionaladoptions.Theseresultsindicatethatspreadinginformationusingtheexistingsocialnetworkofagentsisapoorstrategyforgeneratingsolaradoptions.Incontrasttothisresult,allinterventions(withtheexceptionofPorchTalks-HighKwhichisdiscussedlater)thatforgednewnetworkconnectionsbetweenseedandfollow-onagentsdemonstratedclearlyobservableincreasesinadoption.Theseinterventionscreateconnectionsbetweenseedagentsandfollow-onsoutsideofeachagentsexistingsocialnetwork.Informationaboutsolarisabletoescapebeyondtheseedsocialnetworkclustersthataretargetedinsolarchampionsandreachfollow-onagentsthatareconnectedwithinothernewsocialnetworks.Thisisafindingthatremainssensiblewhentranslatedtothe“real-world”,interventionsthatforcepeopletointerconnectandexchangeinformationarelikelytobemoresuccessful.AnotherkeyobservationwithinourmodelingistheextenttowhichrandomtargetingofseedagentsoutperformsHighKtargeting.Thiswouldappeartobecontradictoryfinding,asHighKisthoughtofasamoreexpensiveformoftargetingasitrequiresassessingthe“popularity”(K)ofeachagent.However,theresultinfacthighlightsthelimitationsofHighKtargeting.
ThelimitationsofHighKtargetingisbestevidencedintheresultsfromthePorchTalkprogram.LookingatFigure6weseethatwithinthePorchTalkprogramHighKtargetingsignificantlyunderperformscomparedtorandomtargeting.TheproblemwhentargetingseedsusingHighKisthatagentswithHighKhaveahighprobabilityofbeingconnectedtooneanother.InthecaseofthePorchTalkprogramitisclearthatamajorityoftheagentswithseededduetoHighKresidedwithinthesamenetwork.Therefore,werepeatedlyprovidedinformationtothesamegroupoffollow-onagentsbutwereunabletoreachany
Page 23
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
agentsoutsidetheHighKneighborhood.Incontrastrandomseeding-whileinexact–managedbychancetospreadinformationtoasignificantlylargerfollow-onnetworkthanHighK’ssaturationofanindividualnetwork.
Theseobservationsinformthecentralfindingofthiswork,thatinformationprogramsthatareabletoreachthelargestuniquenetworkoffollow-onagentswillbethemostsuccessfulinpromotingdiffusingofsolarPV.Interventiondesignthatforcessolaradvocates(seeds)tointeractwithpeople(follow-ons)outsidetheirexistingsocialnetworksandwhomaynotsharesimilarcharacteristicsasthem(K)willleadtobetteroutcomes.
EvaluatingtheEconomicsofInterventionsFromTable3weseethatasexpectedcelebrityinterventionshavethehighestvalueofallseedsandareworthover50timesthecostofaneconomicseed.Therefore,asacoarseapproximationifthecosttohireacelebrityadvocatewaslessthan50solarunitsitwouldbeaneconomicaldecision.Thesameanalysiscanbeappliedtoeachoftheotherprogramstoprovideanestimateofathresholdvalueforeachinterventionseed.Finally,whileitmayseemobviousthatcelebritiesprovidethebestvalueofalltheinformationprograms,itisalsotruethatacelebritywouldbethemostexpensivepersontoemployasasolaradvocate.Whilethevalueofaporchtalksandsolarambassadorseedmayappearcomparativelysmall,theyarealsolikelytobeordersofmagnitudelessexpensivethanacelebrity.Therefore,itisimportanttonotethatifinterventionsgeneratestatisticallysignificantadditionaladoptionstheyareworthevaluatingagainstthecostofseeds.
ConclusionandPolicyImplicationsInthisresearchweidentifythatinformationinterventionsthatspreadbeyondexistingsocialnetworksaremoreeffectivethanthosethatrelysolelyonsocialnetworks.Putanotherway,wefindthatthecreationofnew“weakties”facilitatestheeffectivediffusionofinformation(Granovetter,1978).Wealsofindthatseedinghighlyconnectedagentsisinferiortorandomseeding.Thesefindingshaveseveralinterestingpolicyimplications.Basedontheseresultswerecommenddesigningsolarinformationinterventionstomaximizetheamountofinteractionsbetweenpeopleindifferentsocialgroups.Thisallowsinformationtotravelbeyondafinitenumberofexistingsocialgroupsandinsteadconnectsalargernumberofsocialgroups.Informationspreadswithintheselargerinterconnectedsocialgroupsovertimeviawordofmouthandpeereffects.Interventionsshouldthereforebedesignedsuchthateach“seed”orinfluencerisabletointeractwithpeoplebeyondtheirownsocialnetwork.Wealsofindthatrepeatedlyseedinghighlyconnectedagentsisanineffectivestrategy.Programsthatrepeatedlyidentifyseedsbasedonsimplecharacteristicssuchaspopularity,locationorwealthmayfindsimilarlimitations.Seedingbythesecharacteristicswillincreasethelikelihoodthateachnewseedisbelongstoasimilarsocialgroupasapreviousseed.Thereforeevenwithlargenumbersofseeds,theprogramreachesasubstantiallysmallernumberofsocialnetworks.Forsuccessfulinterventionsseedsshouldbeselectedthatareabletoreachthebroadestnumberofnewsocialgroups.
Page 24
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
TheSECADmodelusedinthisresearchisvalidatedforAustin,Texas.Necessarymodificationsshouldbemadewhenapplyingthefindingspresentedinthispapertocitieswithdifferenteconomicandpolicycontexts.
FutureWorkThisresearchpresentsseveralopportunitiesforfurtherrefinementandstudy.Improvedmodellingofthecostofeachinformationinterventionwillfacilitatebettercomparisonsofthereturnoninvestmentofeachprogram.Onepotentialmethodwouldbetousethenumberoffollow-onsaseedagenthasatinitialization(t=0)asaproxyfortheircost.Forexampleitwouldbeexpectedthatacelebrityseedwith100,000follow-onsis10,000timesmoreexpensivethanaprogramvolunteerseedwith10follow-ons.Furthermore,increasinglygranularmodelingofthequalityofinformationhasthepotentialtoimprovethefindingofthiswork.Forexample,Rai,Reeves&Margolis(2015)findthattheinfluenceofneighborsandinstallerstobestrongmotivatorstowardPVadoption.Anabilitytomodelthe“qualityofinformation”ortrustworthinessofeachindividualseedwouldimprovethereliabilityofthiswork.Inthisresearchwetestonlytwoseedtargetingstrategies,randomandHighK.However,severalmorepowerfulnetworkcentralitytheoriesexist.Forexample,Kitsaketal.(2010),offeranoveltheoryfordeterminingthemost“influential”nodesinanetwork.FurtherworkcouldtestthesetheorieswithintheSECADmodeltoseewhethertheyareabletooutperformrandomandHighKseeding.Finally,allinterventionsmodeledinthisresearchwerepurelyinformational.Noconsiderationwasgiventomodelingeconomicinterventions–forexampleachangeinthePVrebate–incombinationwithinformationalinterventions.Furtherworkshouldexplorecombinationsofeconomicandinformationinterventionstounderstandiftheyofferbetterreturnoninvestment.
Page 25
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
ReferencesAral,S.,Muchnik,L.&Sundararajan,A.,2013.Engineeringsocialcontagions:Optimal
networkseedinginthepresenceofhomophily.NetworkScience,1(2),pp.125–153.Availableat:http://www.journals.cambridge.org/abstract_S2050124213000064.
Bollinger, B., Gillingham, K., 2012. Peer Effects in the Diffusion of Solar Photovoltaic Panels. Mark. Sci. 31, 900–912. doi:10.1287/mksc.1120.0727
Carley,S.,2009.Distributedgeneration:Anempiricalanalysisofprimarymotivators.EnergyPolicy371648–1659
Centola,D.&Macy,M.,2007.ComplexContagionandtheWeaknessofLongTies.AmericanJournalofSociology,113(3),pp.702–734.
CleanEnergyGroup,2009.SmartSolarMarketingStrategies,Chernyakhovskiy,I.,2012.SolarPVAdoptionintheUnitedStates:AnEmpirical
InvestigationofStatePolicyEffectiveness.MastersThesis.Availableat:https://scholarworks.umass.edu/cgi/viewcontent.cgi?referer=&httpsredir=1&article=1097&context=masters_theses_2
Costanzo,M.Archer,D.,Aronson,E.,&Pettigrew,T.F.,1986.Energyconservationbehavior:Thedifficultpathfrominformationtoaction.AmericanPsychologist,41(5),pp.521–528.Availableat:http://doi.apa.org/getdoi.cfm?doi=10.1037/0003-066X.41.5.521.
DeGroote,O.,Pepermans,G.&Verboven,F.,2016.HeterogeneityintheadoptionofphotovoltaicsystemsinFlanders.EnergyEconomics,59,pp.45–57.Availableat:http://dx.doi.org/10.1016/j.eneco.2016.07.008.
Deffuant, G., Neau, D., Amblard, F., Weisbuch, G., 2000. Mixing beliefs among interacting agents. Adv. Complex Syst. 3, 87–98.
Deffuant,G.,Amblard,F.,Weisbuch,G.,&Faure,T.,2002.OpinionDynamicsandBoundedConfidenceModels,AnalysisandSimulation.JournalofArtificalSocietiesandSocialSimulation,5(3).
Evans,P.,&Wurster,T.S.,1997.StrategyandtheNewEconomicsofInformation.HarvardBuinessReview.Availableat:https://hbr.org/1997/09/strategy-and-the-new-economics-of-information
Fischlein,M.,&Smith,T.,2013.Revisitingrenewableportfoliostandardeffectiveness:policydesignandoutcomespecificationmatter.PolicyScience,6:277–310.DOI10.1007/s11077-013-9175-0
Fulton,M.,Capalino,&Josef,A.,2012.TheGermanFeed-inTariff:RecentPolicyChangesGranovetter, M.S., 1978. Threshold Models of Collective Behavior. Am. J. Sociol. 83, 1420–
1443.
Hegselmann,R.&Krause,U2002.Howcanextremismprevail?Astudybasedontherelativeagreementinteractionmodel.JournalofArtificalSocietiesandSocialSImulation,5(4).
Page 26
Policy Research Project (PRP), LBJ School of Public Affairs, The University of Texas at Austin, May 2018.
Project Director: Dr. Varun Rai
Hogan,J.E.,Lemon,K.N.&Barak,L.,2004.QuantifyingtheRipple:Word-of-MouthandAdvertisingEffectiveness.JournalOfAdvertisingResearch,44(September),pp.271–280.
Jalilvand,M.R.,Esfahani,S.S.,&Samiei,N.,2011.Electronicword-of-mouth:Challengesandopportunities.ProcediaComputerScience,3,pp42-46.
Kitsak,M.etal.,2010.Identificationofinfluentialspreadersincomplexnetworks.NaturePhysics,6(11),pp.888–893.Availableat:http://dx.doi.org/10.1038/nphys1746.
Kovács,I.A.&Barabási,A.L.,2015.Destructionperfected.Nature,524(7563),pp.38–39.
Kowalska-Pyzalska,A.,2017.Whatmakesconsumersadopttoinnovativeenergyservicesintheenergymarket?Areviewofincentivesandbarriers.RenewableandSustainableEnergyReviews,(October2016),pp.1–12.Availableat:http://linkinghub.elsevier.com/retrieve/pii/S1364032117314697.
Kumar,V.,J.Petersen,A.,&Leone,R.P.,2007.HowValuableisWordofMouth.Marketing.HarvardBusinessReviewhttps://hbr.org/2007/10/how-valuable-is-word-of-mouth
Meadows,M.&Cliff,D.,2022.ReexaminingtheRelativeAgreementModelofOpinionDynamics.JournalofArtificalSocietiesandSocialSimulation,15(4).
Morone,F.&Makse,H.A.,2015.Influencemaximizationincomplexnetworksthroughoptimalpercolation.Nature,524.Availableat:http://arxiv.org/abs/1506.08326%0Ahttp://dx.doi.org/10.1038/nature14604.
Noll,D.,Dawes,C.&Rai,V.,2014.SolarcommunityorganizationsandactivepeereffectsintheadoptionofresidentialPV.EnergyPolicy,67,pp.330–343.Availableat:http://dx.doi.org/10.1016/j.enpol.2013.12.050.
Palm,A.,2016.LocalfactorsdrivingthediffusionofsolarphotovoltaicsinSweden:Acasestudyoffivemunicipalitiesinanearlymarket.EnergyResearchandSocialScience,14,pp.1–12.Availableat:http://dx.doi.org/10.1016/j.erss.2015.12.027.
Rai,V.,Reeves,D.C.&Margolis,R.,2016.OvercomingbarriersanduncertaintiesintheadoptionofresidentialsolarPV.RenewableEnergy,89,pp.498–505.Availableat:http://dx.doi.org/10.1016/j.renene.2015.11.080.
Rai,V.&Robinson,S.A.,2015.Agent-basedmodelingofenergytechnologyadoption:Empiricalintegrationofsocial,behavioral,economic,andenvironmentalfactors.EnvironmentalModellingandSoftware,70.
Robinson,S.A.&Rai,V.,2015.Determinantsofspatio-temporalpatternsofenergytechnologyadoption:Anagent-basedmodelingapproach.AppliedEnergy,151C
USDOE,2018.AustinEnergyValueofSolarRate.Energy.gov.Availableathttps://www.energy.gov/savings/austin-energy-value-solar-residential-rate
WoodMackenzie,2018.USSolarMarketInsight:2017YearinReview.