Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
©CopyrightJASSS
MaxwellBrown(2013)
CatchingthePHEVer:SimulatingElectricVehicleDiffusionwithanAgent-BasedMixedLogitModelofVehicleChoice
JournalofArtificialSocietiesandSocialSimulation 16(2)5<http://jasss.soc.surrey.ac.uk/16/2/5.html>
Received:06-Aug-2012Accepted:02-Nov-2012Published:31-Mar-2013
Abstract
ThisresearchdevelopsthenmergestwoseparatemodelstosimulateelectricvehiclediffusionthroughrecreationoftheBostonmetropolitanstatisticalareavehiclemarketplace.Thefirstmodelisamixed(randomparameters)logisticregressionappliedtodatafromtheUSDepartmentofTransportation's2009NationalHouseholdTravelSurvey.Thesecond,agent-basedmodelsimulatessocialnetworkinteractionsthroughwhichagents'vehiclechoicesetsareendogenouslydetermined.Parametersfromthefirstmodelareappliedtothechoicesetsdeterminedinthesecond.Resultsindicatethatelectricvehiclesasapercentagesofvehiclestockrangefrom1%to22%intheBostonmetropolitanstatisticalareaintheyear2030,percentagesbeinghighlydependentonscenariospecifications.Alowerpriceisthemainsourceofcompetitiveadvantageforvehiclesbutothercharacteristics,suchasvehicleclassificationandrange,aredemonstratedtoinfluenceconsumerchoice.Governmentfinancialincentiveavailabilityleadstogreatermarketsharesinthebeginningyearsandhelpstospreaddiffusioninlateryearsduetoanincreasedbaseofinitialadopters.AlthoughseenasapotentialhindrancetoEVdiffusion,batterycostscenarioshaverelativelysmallimpactsonEVdiffusionincomparisontopolicy,range,milespergallon(MPG),andvehiclemilestravelled(VMT)asapercentageofrangeassumptions.PessimisticrangeassumptionsdecreaseoverallPHEVandBEVpercentagesofvehiclestockby50%and30%,respectively,relativetotheEPA-estimatedrangescenarios.FuelcostscenariosdonotconsiderablyalterestimatedBEVandPHEVstockbutincreasetheratioofcarstocktolighttruckstockintheinternalcombustionengine(ICE)vehiclespectrum.Specifically,carsareestimatedat55%ofICEvehiclestockinthedefaultfuelpricescenariobutincreaseto62%ofICEvehiclestockinthehighworldoilpricescenario,withLTscoveringtheappropriatedifferences.
Keywords:ElectricVehicle,Diffusion,MixedLogit,VehicleChoice,NetworkEffects
Introduction
1.1 BeforetheModelT,engineswereprimarilydesignedtorunonsteamandthenethanolorbiodiesel(Hamlyn1967).HenryFord'swife,ClaraFord,droveanelectric-poweredvehicle;atthetime,electricvehiclesheldasmallbutsubstantial-enoughnichemarketwithwell-to-dourbanhousewives(HenryFordEstate2007;Romero2009).Thesevehicletypeswerequicklyphasedoutbythemass-producedinternalcombustionengine(ICE)vehicles.Overthepastcentury,automobileenthusiastshaveintroducedalternativefuelvehicles(AFVs)butnonehaveheldasubstantialshareintheUStransportationmarket.Untilrecently,electricvehicles(EVs)wereprimarilyputtogetherbytheend-usersfromkitssoldbythemanufacturer.Now,in2012,thelargestautomobilemanufacturersintheUShaveintroducedelectricvehiclesintotheirmodellineupandarepushingforwardwithresearchanddevelopment.
1.2 Proponentsofalternativefuelsarechallengingpetroleum'sdominanceinthetransportationsector.Realizingclimatechangeandenergysecuritybenefits,USgovernmentsatthestateandfederallevelhaveincentivizedAFVpurchasesinanattempttoincreasedeployment.Overthepastdecade,morethan20USstateshaveofferedfinancialincentivesofvarioustypesandtheUSfederalgovernmenthasofferedtaxwaiverstowardsthepurchaseofhybridelectricvehicles(HEVs),batteryelectricvehicles(BEVs),andplug-inhybridelectricvehicles(PHEVs).Additionally,energysecurityconcernsandtheeconomiceffectsofnationalpetroleumrelianceduringperiodsofsubstantialpetroleumpricefluctuationhavebecomemoresalientfollowingsignificantgasolinepriceincreasesin2008.
1.3 Thetwolargest,broadly-groupedbarrierscommonlycitedinEVdeploymentaretechnologicaldevelopmentandsocialacceptance.Theprimaryconcernwiththeformeristherangeofthevehicle,typicallyreferencedas"rangeanxiety,"aswellasprice.Thebasisforthelatterconcernisthatoverthispastcenturyvehicleshaveprimarilyrunongasolineand,eveniftheywerecost-competitive,EVsstillfaceanuphillbattleinbeingacceptedasaprimarymeansofpersonaltransportation.Thisstudymodelsthespreadingofindividuals'willingnesstoconsider(WtC)electricvehiclesEVsthroughanagent-basedmodelofvehicleinnovationdiffusion.Agents'socialinteractionsareusedtomodelthespreadofWtCandifanagenthasexceededitsWtCthreshold(i.e.caughtthe"PHEVer")itwillintroduceEVsintotheirvehiclechoicesets.
LiteratureReview
SystemandAgent-BasedModelsofVehicleChoiceandAFVDiffusion
2.1 Themodeldevelopedinthispaperdrawsfromkeystrengthsincludedinotherrecentlycreated,agent-basedvehiclechoicemodels(VCMs).Thedecision-makingcomponentofthismodelutilizessurveydatafromthe2009NationalHouseholdTravelSurvey(NHTS;USDOT2009)todeterminevehicleattributevaluation,similartootherpaststudies(deHaanandMueller2009;deHaan,Mueller,andScholz2009;Zhangetal.2011;Cuietal.2011).Inpreviousagent-basedVCMs,researchershaveutilizedavariantoflogisticregression(logit)decisionmakingmethods(deHaanandMueller2009;deHaan,Mueller,andScholz2009;Cuietal2011;Zhangetal2011),whileotherresearchersdevelopedmathematicalandsystems-basedestimatesofconsumerbehavior(StrubenandSterman2008,Sullivanetal2009;Eppsteinetal2011).Priorresearchhasprimarilyutilizedtwomethodstoincorporateagentattributesintotheirvehiclechoicemodels.Thefirstmethodcreatessyntheticagentsfrompopulationcharacteristics(Eppsteinetal.2011;Cuietal.2011).Thesecondmethodreplicates
http://jasss.soc.surrey.ac.uk/16/2/5.html 1 14/10/2015
surveyrespondentsasagents(deHaanandMueller2009;deHaan,Mueller,andScholz2009;Zhangetal.2011).Governmentinterventionandmanufacturerresponsefunctionshavealsobeenincorporatedintoanagent-basedVCMframework(Zhangetal.2011).
2.2 Theprior-citedagent-basedVCMsgenerallyincludethevariablesoffuelsavingsandannualincomebut,unlikemostotherVCMs,alsoincludemediaand/orsocialinteractionsasinfluentialvariableswhichaffectconsumers'willingnesstoconsider(WtC)orpurchaseAFVs.Prioragent-basedVCMsmodelsocialinteractionssimulatedinaspatiallyexplicitenvironment(Eppsteinetal.2011;Cuietal.2011).ResearchershavesimulatedthresholdrangesforconsiderationofanAFV(Eppsteinetal.2011)whileothersendogenouslyincorporatedtheagents'socialnetworkconditionsintotheagents'utilityfunctions,derivedfromprimaryresearchandBayesianmixedlogitmethodologies(Zhangetal.2011).Inthemajorityofcitedstudies,aswellasothermodels,diffusionaspectsaremodeledusingthetheoryofhomophilywhich,initssimplestform,theorizesthatpeople'sbeliefstendtodrifttowardsthosethataresimilartothem(Schelling1978).
2.3 Eppsteinetal.(2011)createdauniqueattributeintheirsimulationswhichinvolvedathresholdtoconsiderAFVs;simulatedinteractionsamongagentsimpactedtheirconsiderationthresholdforEVs.ThethresholdisnormallydistributedaroundameanderivedfromsurveydataanalyzedbyCurtinetal.(2009).Theauthorsinitiatedtheenvironmentsuchthatinteractionsamongagentsaredonewithinasocialnetworkanddevelopaheuristicvaluationofmediaandsocialinfluences,determinedbyanagent'ssusceptibilitytoaswellastheassumedamountofmediacoverage.ThemodelpresentedheredrawsheavilyonEppsteinetal.'sconsumerinteractionframeworkwhileincorporatingmixedlogitparameterestimates.
2.4 StrubenandSterman(2008)developedaframeworkandmodelforAFVdiffusionbasedonconsumers'WtCAFVs.Intheirsimulations,WtCisaffectedbytotalexposuretotheplatformwhichisprimarilyinfluencedthroughthreemainchannels:marketing,socialexposurefromdriversofvehicles,andwordofmouthonvehiclesfromnon-drivers.Additionally,consumersgaugetheattractivenessofanAFVbasedonprice,performance,operatingcosts,safety,range,andecologicalimpactwhichmaintainadynamicrelationshipwithautomobileproducersandfossilfuelprices.ThemodelinthispaperseekstobeadaptableforseveraldifferentinstancesofinfluentialfactorsindicatedbyStrubenandSterman(2008)throughflexibilityinmodelassumptionsanddatasourcesrelatedtofuelcostsandvehiclecharacteristics.
SocialInfluenceasaFactorAffectingAFVConsumers
2.5 Drivingamoreenvironmentally-friendlyvehicleisgreatlyinfluencedbyindividuals'socialnetworks(Heffneretal.2005;KahnandVaughn2008;SextonandSexton2010).Theroleofpeerinfluenceonconsumerdecisionshasbeenobservedinmultiplepaststudies.BeardenandEtzel(1982)originatedthequantitativeassessmentofpeerinfluenceonconsumerdecisionsusinganestedrepeatedmeasuresdesign.Theirresultssuggestthatpeerinfluencehasastrongroleinconsumerdecision-makingprocesses,mostlyforthoseproductswhichareportrayedinpublicsettings.TheresultsofBeardenandEtzel'sanalysishavebeenrepeatedelsewhere(ChildersandRao1992;MakgosaandMohube2007).Similarly,evidencehasbeenfoundforover100differentstudiesinwhichhomophilywaspresent(McPhersonetal.2001).Blendingtheprimaryaspectsoftheaforementionedstudies,researchhasindicatedthatmediaandsocialnetworkscansignificantlyimpactconsumerdecisions(AralandWalker2011a;2011b)aswellastravelmodelchoice(Dugundjietal.2008;Dugundjietal.2011).
2.6 Consideringthatdrivingavehicleisapublicbehaviorandpeopletendtoidentifythemselvesandmakeastatementwiththeirvehicles(ChooandMokhtarain2004;SextonandSexton2010),peerinfluenceandsocialnetworkscanbededucedandhavebeenrecognizedasaninfluentialdeterminantofvehiclechoice.Realizingtheseinfluencesonvehiclechoice,studieshaveintroducedsocialandpsychological(Bolducetal.2008)aswellasneighboreffects(Mauetal.2008)intotheirVCMs.Thisstudymergeshomophilywithsocialnetworkeffectstoendogenouslydeterminethevehiclesavailableinconsumers'choicesets,theinclusionbeingspurredbyexceedingagents'WtCthresholdthroughsimulatedideadiffusion.
LogitModelsandSurveysofVehicleChoice
2.7 Researchershavedevelopedlogitmodelsforavarietyofpurposestoassessvehiclechoiceandconsumervaluationofvehiclefeatures.WhetheritisdeterminingthereasonbehindfallingmarketsharesofUSautomobilemanufacturers(TrainandWinston2007),examiningtheeffectsoftimeseriesvariationingasolinepriceexpectationsonthepricesandsharesofvehiclewithdifferentfueleconomyratings(AllcottandWozny2010),orgaugingwhetheraconsumerwillleaseorbuyavehicle(Dasgupta,Siddarth,andSilva-Risso2007),vehiclechoicemodelshavebeenusedforavarietyofdifferentapplicationsandpurposes.Thecoreideaisconsistentacrossallstudies:estimating,andincertaincasessimulating,consumerbehaviorandvaluationofvariousvehicleattributes.Theresearchpresentedheredevelopsasimilardiscretechoicestatisticalframeworkforvehiclechoicebutdeterminesthepowertypeofvehiclesavailableineachagent'schoicesetthroughsimulatedsocialnetworkinteractions.
2.8 SeveralVCMsarecapableofestimatingconsumervaluationoffueleconomyimprovements.Greene(2010)reviewedseveralpastconsumervaluationstudiesofvehiclesandfueleconomywithvariousmethodologies:mixedlogit,nestedlogit,statedandrevealedpreferencedata,hedonicprice,andpriceregression.Fueleconomyvaluationestimateswerefoundtoberatherinconsistentbetweenstudies;afindingsimilartotheUnitedStatesEnvironmentalProtectionAgency'sreviewoffueleconomyvaluation(USEPA2010).Greene'sanalysisandreviewindicatedthatconsumerswereestimatedtobewillingtopayfrom<1%upto400%ofdiscountedpresentvalueforfueleconomyimprovements.additionally,therewasnoevidentexplanationforthewidedifferencesamongtheestimates.
MethodsandModelOutline
MixedLogitModel
3.1 Foranoverviewofmixedlogitmethodology,seeTrain(2009).Mixedlogithastwobenefitstothepurposesoutlinedinthismodelandagent-basedmodelsingeneral.First,itpermitsaheterogeneoussetofconsumervaluationsofvehicleattributesbyallowingagentstoplaceuniqueemphases,drawnfromspecificdistributions,ontheobservablecharacteristicsinthechoiceset.Essentially,mixedlogitmodelsestimateboththemeanandspreadofanexplanatoryvariable'sinfluenceontheprobablyofthedependentvariable'soccurence.Second,afterestimationofparameters,thesimulationsarestraightforwardtoconductwithinanagent-basedframework.
3.2 Logitmodelsareoftenreferredtoasrandomutilitymodels(RUMs);themainpurposeofaRUMistoestimateandevaluateutilityfunctionsbasedondifferentproductattributes.WithinaRUMcontext,thismixedlogitmodel'sRUMrepresentationis:
(1)
http://jasss.soc.surrey.ac.uk/16/2/5.html 2 14/10/2015
Where:Ui,j:UtilityofobservationiforvehiclejCj:Classofvehiclej(e.g.CUV,SUV,sedan,etc.)HPj:HorsepowerofvehiclejFPj:Foot-poundoftorqueofvehiclejSafetyj:Safetyratinginstarsofvehiclej
K:Incomecategory(low,middle,high)[1]Dk:DummyvariableforincomecategorykMSRPj:Manufacturer'sSuggestedRetailPriceofvehiclejDPGi:Dollarspergallonofgasolineforobservationiduringmonthofpurchase,bystateMPGj:EPAmilespergallonestimateofvehiclejVMTi:Averagedailyvehiclemilestravelledofobservationi
3.3 Explanatoryvariablesinthispaper'smixedlogitmodelarerepresentativebutnon-exhaustive(duetodatalimitations)ofthosefoundinthestudiesreviewedbyGreene(2010).Thedailygascost(DGC)variableistheproductofstate-levelpricepergallonofgasolineatthemonthofpurchasedividedbytherespective
vehicle'smilespergallonofgasoline(MPG)andtheaveragedailyvehiclemilestravelled.Manufacturersuggestedretailprice[2](MSRP)andgasolinepricesareadjustedtoJanuary2011dollarsusingtheconsumerpriceindex.Differentclassesofvehiclesareimplementedasdummyvariables;allclassdummyparameterestimatesareinreferencetotheconvertibleclassbaselineanddefinedbyWard's(2011).Thegeometricmeanoftorque,infoot-pounds,andhorsepowerparametersattheWards-measuredenginerotationsperminute(RPM)isusedasameasureofvehicleperformance;forsimplificationthiswillbereferredtoasthe'power'variable.Thisvariable'sformprovedbestafterthreeprevailingconditionswereobserved:
1. Includingbothparametersseparatelyinthesamemodelintroducedcollinearityissues.2. Theformprovidesasignificantlyhigheradditiontothelog-likelihoodestimateforeachmarketsectorthaneitherparameterindividually.3. Thegeometricmeanprovidesasignificantlyhigheradditiontothelog-likelihoodestimateforeachmarketsectorthanonlytheproductofthetwoterms
Safetyofvehiclesismeasuredastheaverageofoverallimpact-testsofstarsbasedontheUSDepartmentofTransportation(USDOT)safetyratings(USDOT
2012)[3].ParameterestimatesaredisplayedinTable1anddatasourcesareexplainedinfurtherdetailinthe'Data'section.Severalfunctionalformswereassessed;theanalysiswhichprovidedthegreatestaveragelog-likelihoodvaluesacrossgroupswaschosenasthefinalform.
Table1:MixedLogitRegressionParameterResults
ParameterEstimates[4]
Grouping Single,NoChildren Single,WithChildren Married,NoChildren Married,WithChildren RetiredSedan 2.433**** 0.981** 1.735** 1.653*** 1.618***CUV 3.069** 1.272*** 1.704**** 1.954** 1.447***Pickup 2.040**** -0.271*** 1.538** 1.308** 1.294*SUV 2.067** 0.554** 1.184**** 0.8509**** 0.467**Coupe 2.841*** 0.323* 0.769*** 1.223**** 0.537**Hatchback 2.136**** -0.483* -0.021* 0.465**** 0.101*Wagon 0.415*** 0.746** -0.125** 0.978*** -0.464***Van 1.725*** 1.261**** 0.834** 1.732*** 1.380****ln(Safety) 2.105** 2.958*** 2.583*** 3.25**** 2.852***ln(Power) 2.41*** 1.950** 2.521*** 1.99** 2.64***ln(MSRP_Low) -4.267*** -5.067** -4.233**** -4.853*** -3.991****ln(MSRP_Mid) -3.426** -4.396*** -3.509**** -4.299*** -3.525***ln(MSRP_High) -3.044*** -3.904** -3.474*** -4.274*** -3.177***ln(DGC_Low) -2.937** -2.126*** -2.592**** -2.769*** -2.678****ln(DGC_Mid) -2.556**** -2.942*** -2.825**** -2.726**** -2.359***ln(DGC_High) -3.29*** -1.993** -2.254*** -1.985*** -2.654****
StandardDeviationln(Safety) 0.985** 0.568* 0.795*** 0.689*** 0.897**ln(Power) 0.567*** 0.298*** 0.354*** 0.453** 0.378***ln(MSRP_Low) 0.614** 0.490** 0.693**** 0.772**** 0.595****ln(MSRP_Mid) 0.524** 0.596** 0.594**** 0.658*** 0.595***ln(MSRP_High) 0.673** 0.880* 0.792** 0.635** 0.839**ln(DGC_Low) 1.252** 1.013** 1.375*** 1.235*** 1.2943***ln(DGC_Mid) 1.341*** 1.198** 1.125*** 1.123*** 0.986***ln(DGC_High) 1.120*** 1.001** 1.135** 1.356*** 0.978****
NumberofObservations 2136 527 17978 2162 10432Log-Likelihood -8.5E+6 -1.7E+6 -5.4E+07 -3.3E+6 -1.4E+07
SocialActorsandBehaviors
3.4 EachmodelagentisparameterizedforeveryNHTSsurveyrespondentfromMetropolitanStatisticalArea(MSA)groupings[5](USDOT2009).Theagentsarelocatedintheirenvironmentbyregionandthentothefurthestextentpossible,downtotheMSAandthenfurtherplacedaccordinglybytherespondent-indicated
http://jasss.soc.surrey.ac.uk/16/2/5.html 3 14/10/2015
populationdensityatthetract-levelcountysubdivisions(USCensusBureau2012).Locationattributesareimplementedtocontrolforpossiblecorrelationsamongagents'demographicsandtheirlocation.Theagentsarerandomlylocatedinanorbitalfashionaroundthecitycenterwithdistancedependentonthepopulationdensityrecordsinthe2009NHTS,theindividualswithhigherreporteddensitiesbeingclosertothecitycenter.
3.5 Theagentsareoriginallylimitedtoexploreoptionswithinthefueltypeclassoftheircurrently-ownedmodel[6]butwillexploreotheroptionsaftertheirindividualconsiderationthresholdhasbeenmetthroughsimulatedsocialinteractions.Interactionswithfourneighborsareimplementedtoincorporatethesimilaritieswithlatticeneighborhoodsmodels,bywhichtheideadiffusionequationisheavilyinfluenced.
3.6 Correspondingwiththebeliefsofpeerinfluenceonconsumerdecisionmakingandhomophily,aswellasEppsteinetal.'s(2011)vehicleconsiderationfunctions,interactionsmodeledbyMolofskyetal(1999),StrubenandSterman's(2008)implementationoffractionaldecay,andconsiderableinspirationfromHiebeler(forthcoming),consumers'willingnesstoexploreothervehicletypesintheirpurchasedecisionismodeledusingthefollowingequation:
(2)
Where:Ev,i,t:WillingnessparameterforagentiduringtimettoconsidervehicletypevF:Annualfractionaldecayratej:Otheragentswithintheagenti'ssocialnetworkJ:Totalnumberofagentswithwhichagentiinteractsv:Vehiclepowertype(Hybrid,Plug-inElectric)
(3)
W(j_i),t:Referencedasasimilarityindex[7]p:Agenttraitindex(age,income,highesteducationalattainment)Cp:Characteristicoftypepofagentiorj
3.7 WhenEv,i,texceedstherandomly-assignedheterogeneousthresholdforvehicleconsiderationtheconsumerwillconsider(throughmixedlogitderivedparameters)vehicleoptionsfromvehiclepowertype,v.Withinthemodel,theoptionexiststoallowforagentsonlytoconsiderBEVsifanothermemberofthehouseholdhasavehiclewhichisnotaBEV.Theheterogeneousthresholdisacrucialcomponentoftheanalysisandthusatargetofsensitivityanalysis.Last,iftheagentdoesnotpurchaseanAFVinthatyear,theirpresentWtCdecreasesatafractionaldecayrate,similartoStrubenandSterman(2008).
3.8 Figure1providesanoverviewandpseudo-codeofthedecisionmakingsequenceforeachagent.Orderisdeterminedbywhichagenthastheleastamountoftimeremaininginvehicleownership;tiesaresettledbycoinflips.ThetablefollowingFigure1providesalegendandinformationregardingthevariablespresentedinFigure1.
http://jasss.soc.surrey.ac.uk/16/2/5.html 4 14/10/2015
Figure1.FlowChartofAgentDecision-Making
Variable/Index Agent(A)orGlobal(G)
Description
Neighbors A OtheragentsinsocialnetworkAttributes A Attributesofagentsi A Focalagentj A OtheragentTraitRatio A W(j|i)ininteractionequationSimIndex A NumeratorinsecondportionofinteractionequationVEH G ICE,Hybrid,orElectricT G YearDecayRate G FractionaldecayrateofWtCfrom(T-1)toTMonthsRemaining A MonthsremaininginvehicleownershipYearsToOwn G YearsofownershipofnewvehiclepurchasersHybridsSold G HybridssoldinyearTAllVehSold G AllvehiclessoldinyearTMaximumMarketShare G MaximumvehiclepenetrationratefromNEMSAllHybridsSold G AllhybridssolduptoyearTMAN G ManufacturerindexPopRatio G P'inincentiveavailabityequationNationalVehicleSales G Total,nationalvehiclessoldfrom2012toyearTIncentiveLimit G IncentivelimitonnumberofvehiclessoldMSRP G Manufacturers'suggestedretailpriceHybModel G IndexofhybridmodelsiH A Otheragentsini'shouseholdDailyVMT A SurveyreportedaveragedailyVMTRange G EstimatedrangeofBEVModel,givenassumptionKBEVsSold G NumberofBEVssoldinyearTAllBEVsSold G AllBEVssoldupuptoyearTBEVModel G IndexofBEVmodelsBatCost G BatterycostfromUKCCC'sbatterycostreport(2012)Class G CarorLightTruckChargerCost G CostofchargingequipmentPresentValueBatDep G PresentvalueofbatterydepreciationDiscRate G Discountrate,assumedat7%AllPHEVsSold G AllPHEVssolduptoyearTPHEVModel G IndexofPHEVModelsDGC A DailyfuelcostinyearTDPGGE G DollarspergallonofgasolineequivalentMPG G MilespergallonofgasolineequivalentMPGROC G Milespergallonrateofchange(ANL2012)DailyVMT A DailyvehiclemilestravelledVehChoice A Vehicleselectedwhichmaximizesi'sutilityUtility A UtilityfromvehiclemodelsChoiceSet A VehiclechoicesetofagentML_Parameters A/G Agent'smixedlogitparametersfromglobaldistributionsR G Numberofrepetitionsforutilitycalculations,Train(2009)
Data
4.1 Figure2summarizesthevariousdatasourcesandtheirrespectivepurposesineachmodel.
http://jasss.soc.surrey.ac.uk/16/2/5.html 5 14/10/2015
http://jasss.soc.surrey.ac.uk/16/2/5.html 6 14/10/2015
Figure2.MapofDataFlow
4.2 Thelargestsourceofdatadetailingagentattributesisthe2009NHTS(USDOT2009).The2009NHTSisanationallyrepresentativesurveyoftravelbehaviorconductedfromApril2008throughApril2009.The2009NHTSsamplesizewas150,147households,includinganationalsampleof25,000householdsandseparatesamplesfromtwentystateDepartmentsofTransportationthattogetheradded125,147completedhouseholds.Thesetwosampleunitsbroughtthe2009NHTSsamplesizetoapproximately150,000householdsand300,000people.
4.3 Thesecondmainsourceofdata,relatedtovehiclecharacteristics,isWard'sAutomotiveInformationProducts(2011).Thesedataentaildetailedcharacteristicsofvehiclesproducedfrom2004-2011whichwereavailableforpurchaseintheUnitedStates.ThesedatawerepurchasedinJuly2011.TheNHTS2009doesnotsupplydetailedautomobileinformationintheirpubliclyavailabledatafilesbutNationalAutomobileSamplingSystem(NASS)codesareprovidedwhichareusedtomatcheachobservation'svehicletoitsrespectiveattributes.NASSvehiclecodesusedinthisanalysiswereextractedfromtheDepartmentofTransportation's(DOT)manual(NHTSA2011).Anotherrequirementtoproducethefinaldatasetwasthecollectionofsafetydatameasuredintheaveragestars.ThesedatawereretrievedthroughtheUSDOT(viaSafeCar.gov;USDOT2011).FuturevehiclespecificationsareexplainedinAPPENDIX2.
4.4 Althoughmultipledatasourcesareutilizedtoallowforassumptionsonvehicleattributes,itisnotentirelyrealistictoassumethatspecificmodeltypesusedtocalibratethemodel(e.g.a2012FordFocus;seeAPPENDIX2)willstillbearoundby2030.Insteadthemodel'sresultsshouldbeunderstoodandinterpretedthatthemarketsharesofamid-classvehiclewithattributessimilartosaidvehiclesislikelytogroworshrinkby2030comparedtoothervehiclescharacterizedthroughdataassumptionsrelatedtofuturevehicleattributes.
4.5 Thepriceofgasolineattheindividualstatelevelduringthemonthofpurchase(EnergyInformationAdministration2011),basedonthedifferencebetweenmonthwhenthesurveywastakenandstatedmonthsofownership,wasmatchedtoeachvehicleandhouseholdobservation.Costspermileanddailygascostswerethencomputedforeachvehiclechoiceavailabletotheconsumer.Surveyresearchindicatesthatvehicleconsumersnormallydonotorerroneouslycalculateannualfuelcosts(KuraniandTurrentine2004),thusthefuelcostsforthetimeatpurchaseareincludedanddailyVMTisusedtoscaleobservationsbasedontheiraverageamountofdriving.Averagegasolinepricesthroughout2011atthestatelevelareusedasthereferencepointsduringtheinitialtimeperiodandthenthegasolinepriceisadjustedaccordingtothe2012AnnualEnergyOutlook's(AEO;USEIA2012)marginalannualrateofchange.Thesevaluesremaininconstant2011dollarsthroughoutprojectedperiods.IntheUSDepartmentofEnergy's(USDOE)NationalEnergyModelingSystem(NEMS;USDOE2001),themarketsharesofanyvehicletypearecalculatedusingacombinationofvehicleandpopulationcharacteristicsaswellastechnologycostsandregulatorycostsimplementedonmanufacturerstoimprovefueleconomy.WithinNEMS,themaximumvaluesforalternativefuelcarandlighttruckmarketsharesaredependentonthepasttimeperiod'smarketshare,carvehiclemarketsbeingmoreresponsiveandflexiblethanthelighttruckmarketshares.Withinthispaper'smodel,thesedatadeterminethemaximumamountofAFVsavailablebyclassgivenextensivedemand;thisistoreflectproducers'responsesandabilitytoadjustmanufacturingschedules.AfutureextensionofthismodelwouldbetheinclusionofadynamicmanufacturerprofitmaximizationfunctionsimilartoZhangetal(2011).Thelimitsformaximummarketsharesarecomputedfollowingindividualagents'choices.
4.6 Survivabilityratesbyageschedulesareneededtoaccuratelyestimatevehicleexpirationrates.VehiclesurvivabilityratesaretakenfromtheUSDOTNationalHighwayTrafficSafetyAdministration's(NHTSA)2006technicalreport(NHTSA2006).
4.7 Tocomputethegramsofcarbon-dioxideequivalent(gCO2e)emissionsandsubsequentlyAverageFuelCarbonIntensity,thegCO2e/MJofelectricityaretakenfromtheGreenhouseGases,RegulatedEmissions,andEnergyUseinTransportation(GREET)Model(ArgonneNationalLaboratory2011).ThecarbonintensitiesforblendedmotorgasolinearecomputedusingvaluesfromtheTransportationRegulationandCreditTrading(TRACT)model(RubinandLeiby2012).
VehiclePurchaseIncentivePolicies
4.8 AstheresultssectionofthisanalysisfocusesontheStateofMassachusettsandmorespecificallytheMSAofBoston,onlythefederalincentivewillbeofferedonvehicles.ThefederalincentivesfromthePlug-InElectricVehicleCreditintheamountof$7,500areavailableuntilalimitof200,000vehiclesnationallysoldpermanufacturerisreached(IRC30andIRC30D;InternalRevenueService2011).Theincentiveisappliedinthemodeluntiltheproportionofvehiclesthathavebeenpurchasedofthatvehicletypeisgreaterthanorequaltothesameproportionrelativetotheestimatednationalvehiclefleetfromthe2012AEO(USEIA2012).Forexample,ifaspecificEVweretoreach20,000totalsalesandtheMSAhas200,000vehiclessoldalready,thentheamountassumedtohavebeensoldonanationallevelis10%oftheprojectednationalvehiclesalesuptothatyear.Essentially,if
(4)
thenincentivesarenotincludedonvehiclemanufacturerm'smodel(s).Where:
P:Populationadjustmentfactor[8]v:Vehiclessoldfrommanufacturermofmodelim:Indexofvehiclemanufacturersi:IndexofvehiclemodelsV:AllvehiclessoldinsimulatedmarketinyearyY:CurrentyearUSVy:AmountofvehiclessoldintheUSinyeary
4.9 SincethehybridandEVfederaltaxcreditsareavailableonaper-manufacturerbasis,thisrateisappliedonlytovehiclemanufacturersthathavenotexceededthelimitasofMay2012,asindicatedbyAdvancedFuelDataCenter(AFDC,2012).Additionally,thenationalamountofvehiclessold,asindicatedbyWards(2011),ofeachmanufacturer'sincentive-eligiblevehiclesisdeductedfromthefederalincentive'sthresholds.ThesameruleregardingproportionsofEVsispresentindeterminingavailabilityofhybridvehicletaxincentives.Althoughaphase-downperiodexistsfollowingtheexpirationofincentiveavailability,thescopeoftimeisarbitraryrelativemodeltimescale.
Results
5.1 Themodeliscapableofimportingagentcharacteristicsfromseveraldifferentmetropolitanstatisticalareas(MSAs).Thissetofsensitivityanalysisandresults
http://jasss.soc.surrey.ac.uk/16/2/5.html 7 14/10/2015
focusesontheBostonMassachusettsMSAwhichhasagoodbalanceof:
1. Highsurveyresponserate2. Representativepopulationproportionstothosepresentinthemixedlogitparameterestimationdata3. Adequatenumberofagentstoreliablydemonstratecoreconceptsandmodeldynamicswithoutconstrainingresultproductionandanalysisthrough
limitsoncomputerresources4. ThelimitedamountofpubliclyavailablehigherpercentagebiofuelsandCNGstationsavailable(AFDC,2012)minimizespotentialconcurrenteffects
fromotheralternativetransportationfuels
5.2 Therangeofforecastyearsissetat2009-2030throughoutthisresultssection.Themodelcanforecastoutfurther(to2100)whilekeepingseveralvariables
constantatthe2030level[9],buttheadditionaltimerequiredtorunthesimulationsoutto2100andthedecliningreliabilityofsimulationsdeterredtheauthorfromextended-rangeforecasting.
SensitivityAnalysisofCoreModelParametersandAssumptions
5.3 Resultspresentedwithinthisandothersectionsaretheaverageoveraburn-inof250runs.Recognizingthatinitialconditionsimpactoutcomesofnetworkeddiscretechoicemodels(DugundjiandGulyás2013),agentlocationandWtCthresholdsareresetatthebeginningofeachsimulation.Asopposedtovariousscenarioresults,thissectionexploresthedifferentimplicationsthatunderlyingfundamentalmodelassumptionshaveonestimatedresults.
Table2:PotentialScenariosfortheVehicleMarketplaceModel
IteminSecondaryLegend
Translation PotentialValues
MeanWtC MeanWtCParameter EagerNeutralReluctant
WtCFDRate AnnualFractionalDecayRateofWtCValues 0.50.751
EconScen EconomicScenario Business-As-Usual(BAU)LowEconomicGrowth(LEG)HighEconomicGrowth(HEG)
PopRatio Populationratioforpolicyanalysis 0.512
BatScen Batterycostsscenario Business-As-Usual(BAU)ConservativeOptimistic
VMTasRangeLimit PercentageofdailyVMTwhichtheBEVsrangehastobeinordertoconsider
50%75%100%
MPG Miles-per-Gallon/GGE EPAMFR
Range RangeofBEVandPHEV C&DEPAMFR
FuelPrice[10] Gasolineandelectricitypricescenarios Business-As-Usual(BAU)HighWorldOilPrice(HWOP)
5.4 Adrivingassumptionbehindthevehicleconsiderationcomponentoftheagent-basedmodelisthedistributionofWtCthresholds.Inthefirstgraph,thepercentageoftheagentswhichhavesurpassedtheirpersonalthresholdisplottedovertimeforvariousthresholdmeansandaconstantfractionaldecayrateof0.75,indicatingthateachagent'sWtCis75%ofthepriorperiods.Thelowerlimitfractionaldecayrateissetat0.5,approximatelythedecayratewithwhichdecaybecomestoosevereandnooneconsidersEVs.Thus,adecayrateof1indicatesthatagents'WtCdoesnotdeclinebetweentimeperiods.Intheanalyses,thedecayrateisassumedat0.75andthustheresultingeager,neutral,andreluctantscenarios'percentagesofpopulationexceedingWtCthresholdsaredisplayedinFigure3.
http://jasss.soc.surrey.ac.uk/16/2/5.html 8 14/10/2015
Figure3.ComparisonofMeanWtCwithDecayHeldConstant
5.5 Arecurring,andespeciallynoticeablewiththeeagerscenario,resultisthateventhoughtheWtCforEVsbeginsataslightlylowerpoint(sincehybridownersalreadyexist)itincreasesatanincreasingratefollowingthespreadinducedbyinitialadopters.Note:thebeginningpercentagesofpopulationexceedingthethresholdisgreaterthanzerosincetheleft-handtailofthenormaldistributionpassesovertheY-axiswhileWtCvaluesareassumedtorangebetween0and100.
5.6 AninterestingandimportantaspectregardingtheassumptionofmeanWtCistheamountofdiffusion,measuredinthisanalysisasthepercentageofvehiclestockofeachvehicletype.AsshowninFigure4,themeanWTCassumptionhasvaryingyetprofoundeffectsonthediffusionofdifferentvehicletypes.Inconjunctionwithaforementionedparameterassumptions,thevaluesdemonstratetheeffectsofthedifferentmeanWtCthresholds.Notably,thediffusionratesofchangeroughlyresembletheexponentialgrowthfunction,representativeofproductdiffusiontheory(Rogers1963).
Figure4.AFVStockPercentageswithDifferentMeanWtCThresholds
IncentivePolicyPopulationRatio
5.7 The2012AnnualEnergyOutlook(AEO;USDOE2012)estimatesgreatvariabilityinregionalusageofalternativefuels.GiventhatotherregionsareestimatedtohavedifferentusageratesofFFVs,CNGvehicles,andEVs,assumingaone-to-oneratioofnationalvehiclesalestoBostonMSAvehiclesalescouldhinderreliabilitybyskewingnationalEVconsumptionestimates.AratiogreaterthanoneindicatesthatalesseramountofthenationalEVincentivesareavailabletotheBostonMSAresidentsthanotherareas.BecauseofthemultipleconcurrentpoliciesrelatedtoAFVdeploymentandalternativefuelusage(e.g.theCaliforniaLowCarbonFuelStandardandtheRenewableFuelStandardrevisedundertheEnergyIndependenceandSecurityAct),theratioofBostonMSAvehiclestocktonationalvehiclestockbecomesafocusofincentivepolicystructuresensitivityanalysis.
5.8 ThemostpertinentresultfromtheincentivepolicystructureanalysisisthatthegreateravailabilityofincentivesinthebeginningperiodsaddstothebaseofinitialadopterswhichinturnincreasestheoverallspreadofWtCoverthemodeledperiods.Anincentive-dependentpopulationratiothatisbeneficialfortheBostonMSA(i.e.PopulationRatio<1)resultsinahigheroverallstockofevsbutpercentagesofvehiclestockbymodelhavevariedresponses.bothvehiclesestimatedtohavethehigheststocklevels,thefocusandvolt,incurrelativelylesserincreasesinestimatedvehiclestockwithincreasedincentives.themiev,givenitsunfavorablecategorizationofahatchback(basedonmixedlogitparameterestimates)butlowermsrp,incursthelargestrelativeincreasewithgreateravailabilityofincentivesasthevehicle'slowermsrpbeginstooutweighthegenerallyrelativedisutilityofbeingahatchback.fromobservingthecharacteristicsofbuyersbymodelofev,priuspurchasingisprimarilydonebydriverswithverylowvmtandlesseraversionstomsrp.giventhatthepriusandrav4arebothproducedbythesamemanufacturertheavailabilityofincentives,andthusestimatedpercentagesofvehiclestock,areinterdependent.
http://jasss.soc.surrey.ac.uk/16/2/5.html 9 14/10/2015
EffectsofMPGandRangeAssumptions
5.9 Estimatesofrangehavebeenproducedfrommanufacturers(MFR),thegovernmentthroughtheEnvironmentalProtectionAgency(EPA),andCarandDriver(C&D).Additionally,MPGrateshavebeenmeasuredbymanufacturersaswellastheEPA.Inthissection,thesimulationsareconductedwiththesesixdifferentassumptionsontherangeandMPGofelectricvehicles.Resultsindicatethat,similartothesensitivityanalysisinvolvingthelimitofdailyVMTasrange,theestimatedrangeofBEVshasacriticalimpactonthedeploymentofelectric-onlyvehicles.Theoverallpercentagestockofvehiclesvariesbynearly1.2%fromtheoptimistic(MFR)tothepessimistic(C&D)estimatesofBEVrange.HybridvehiclesarealsoinfluencedbythevariousassumptionsofBEVsandPHEVs;whentherangeestimatescomefromthemostpessimisticsource,hybridvehicles'2030percentageofvehiclestockincreasesbyapproximately0.6%overtheothermostoptimisticrangeassumptionsscenarios.SimilartoBEVs,PHEVsasapercentageofvehiclestockaresignificantlyimpactedbythevariousrangeandMPGassumptionsyettheeffectismuchmoresubstantialbetweentheEPAandC&Dscenarioswhichentailanestimated1%decreaseinpercentageofvehiclestockin2030.
BatteryCharacteristics
5.10 AlthoughseenasapotentialhindrancetoEVdiffusion,thedifferentcharacteristicsofbatterycostshaverelativelysmallimpactsonEVdiffusionincomparisontopolicy,range,MPG,andVMTasrangeassumptions.Incomparisontotheotherprimaryfinancial-centricmodelmechanismofpolicypopulationratios,the2030estimatesforbatterycostsvarybyapproximately$1000-$2000whereasthefinancialincentivesreducetheMSRPby$7500inthebeginningyearswhenEVsarestillrelativelyexpensive,addingtothebaseofinitialadoptersandencouragingWtCdiffusion.
5.11 Eventheoptimisticbatteryscenarioscouldbesignificantlyoverestimatingthefuturecostsofbatteries.ThebatteriesandEVportionoftheAmericanRecoveryAct(USDOE2012)anticipatesthatinvestmentsinbatteriesaloneshouldhelplowerthecostofsomeelectriccarbatteriesbynearly70%bytheendof2015,withthelargestproportionofvehiclecostreductionsoccurringbetween2009and2013.Additionally,theAmericanRecoveryActanticipatesthattheexpectedlifetimeofatypicalEVbatterywillincreaseby350%inthesametimeperiod,allowingthebatterytolastuptoandpotentiallybeyondthevehicles'anticipatedlifetimesanddecreasingthepresentvalueofbatterydepreciationoverownershipperiods.TheUKCCCmodeledchangesinEVbatterycostsaremuchless
optimisticandwiththeAmericanRecoveryAct'sassumptions[11],totalEVdiffusionisincreasedbyapproximately0.85%overtheoptimisticscenariobytheUKCCC.Othervehiclecharacteristicsthemselveswhichwouldbeaffectedinthesescenariosaresimplifiedinthisportionoftheanalysisandwouldlikelyplayamajorroleinconsumerdecision-making.
HighWorldOilPrice(HWOP)Scenario
5.12 Forthisscenario,theratesofchangeofallfuelpriceswereadjustedusingtheHWOPdatafromtheAEO2012(USEIA2012).TheestimatedeffectthattheHWOPconditionshadonEVstockswasminimal,buttheeffectsontheICEportionsofvehiclestockofcarsandlighttrucks(LTs)areworthmentioning.ConsumersubstitutionofLTsforcarsincreasestheestimatedstockofcarsbyapproximately5.1%by2030intheHWOPscenarioovertheBAUscenario.Figure15displaystheCarandLTpercentagesofICEstockundertheBAUandHWOPscenarios.IntheHWOPscenario,thestockofcarsisexpeditedduetotherelativelylargegaspricemarginalratesofchangeestimatedinthefirstfewyearsofsimulations.FollowingtheearlygreaterdivergencefromtheBAUscenarioICEstockpercentages,theHWOPcarandtruckICEstockpercentagesadjustatlessintenseratesbutstillmorerapidlythanintheBAUscenario.
5.13 WithrespecttoFigure5,thereisaninnateinstitutionalandmarketplacelagassociatedwithshiftingstocksandthusgaspriceincreaseswillnotforceimmediateshiftsofvehiclecharacteristics.AlthoughitisintuitivelyexpectedwithrisinggaspricesthatLTpercentagesofvehiclestockwouldimmediatelyfall,agentswithhigheraversiontoDGCareseekinghigherMPGLTsaswellasalternative,non-ICEvehicleswhichmeetthesestandards.Additionally,survivabilityratesfromNHTSAfavorLTsslightlymorethancars.Forsomeagents,inthelateryears,withanincreaseddisutilityofincreasedDGCs,theyarewillingtosacrificethepurchaseofanLTcategorizedvehicleforacar.
http://jasss.soc.surrey.ac.uk/16/2/5.html 10 14/10/2015
Figure5.PercentageofICEStockbyVehicleTypeandFuelPriceScenario[12]
MostOptimisticScenario
5.14 Inthisscenario,themostoptimalconditionsaresettoinvestigatetheupperboundonEVdiffusion;theseconditionsaretohaveoptimisticbatterycosts,eagerWtC,noWtCdecay,0.5PopulationRatio,andmanufacturer-estimatedMPGandrange.TheVMTasRangeLimitisleftat0.75sincethiscomponenthasbeenshowntonothaveaconsiderableimpactonBEVdiffusionoverthefull-rangescenariobutisnecessarytomaintainrealisminagents'BEVconsiderations.Whentheopposite(most-conservative)simulationsareconducted,percentagesofvehiclestockofBEVsandPHEVsapproachzeroandhybridvehicles'percentageofvehiclestockisestimatedatapproximately0.25%bytheyear2030.Inthemostoptimisticscenario,PHEVsandBEVsareestimatedtomaintain12.2%and8.4%ofvehiclestockin2030.
Discussion
6.1 Resultsfromthisanalysisarehighlyvariabledependingonmodelspecificationswiththeunderlyingcoreparametershavingthehighestinfluenceonresultantscenarios.Thisreflectsthewiderangeinpotentialmodelinputsaswellastheagent-basedmodel'sassumptions.RecentsurveyworkhasindicatedthatamajorityofconsumersareeitherlikelyormoderatelylikelytoconsiderthepurchaseofanAFV,withthetwomostdeterringaspectsbeingvehiclepricesandrefuelingoptions(Michniaketal.2012).Thus,itmaybemoreappropriatetoassumeaneagermeanWtC.Neutralscenarioresultsweredisplayedtoprovideconservativeestimatesandthemainintentionofthispaper'sresultssectionistodemonstratemodeldynamicsfromimposedscenarioassumptions.Todevelopthe'true'meanWtCdistribution,furtherpurpose-specificprimaryresearchisrequiredtoquantifyconsumers'willingnesstoexploreAFVoptions;thissourceofuncertaintyisthemodel'slargestshortcomingintermsofexternalvalidity.
6.2 HybridandPlug-inEV(PEV)percentagesofvehiclestockandmarketsharesaredemonstratedtobeinterdependent.WhenthecharacteristicsofPEVsare
improvedtomakesaidvehiclesmoredesirablewithrespecttoconsiderationassumptions[13]andmixedlogitderivedparameters,hybridvehiclesaredisplacedbythemoreattractivePEVs;thisdisplacementisespeciallyevidentinscenariosinvolvingrangeandMPGassumptions.Atthesametime,thismodeldoesnottakeintoaccounthowlearning-by-doinginthesupplychainbyPHEVsandBEVsassistingimprovementsinoneanother.Thatis,evenifPHEVsdisplaceBEVsinearlyyears,theyhelpdevelopandlowerthecostoftechnologythatcanassistBEVs.
6.3 Similartopaststudiesoffinancialpolicyefficacy(GallagherandMuehlegger2011),hybridvehiclepercentagesofvehiclestockareestimatedtoberesponsivetofinancialpolicyavailabilityassumptions.Inthesesimulations,hybridvehiclesareestimatedtobethemostresponsiveintermsofpercentageofvehiclestock.FinancialpolicyefficacyisnotparticularlyevidencedforthePHEVpercentagesofvehiclestockastheirhigherMSRPandlowerconsiderationratesinthebeginningyears,evenwiththeincentive,isamajordeterrenttodiffusion.EstimatesofBEVsasapercentageofvehiclestockrisesubstantiallymorethanPHEVsasapercentageofvehiclestocksincetheirgenerallylowerMSRPrelativetoPHEVsbecomesevenmoreattractivewhentheincentiveisapplied;thisalsohelpstoalleviatethedisadvantageouscategorizationoftheMiEVasahatchbackvehicle.ThisleadstohigheroveralldiffusionofPHEVsandBEVssincetheWtCisabletospreadatanincreasedratefollowingmoreinitialadoptersinthebeginningperiodsofsimulations.
6.4 VehicleselectionofBEVsandPHEVsgenerallyfavorthemodelswiththelowerMSRPandsedancategorization,sincethesearegiventhehigherrelativepart-worthestimatesfromthemixedlogitmodel.GreenautomobilecolumnistsarecurrentlydenotingtheRAV4EVasanichevehicledesignedforwealthyearly-adopters(Voelcker2012;Marchetti2012)butasbatterycharacteristicsimproveconsumerswithloweraversionstoMSRPandhigheraversionstoDGCareestimatedtopurchasethevehicleatsmallyetstillconsiderablerates.Givenitsrelativelylowvalue,sensitivityanalysiswasconductedonthePrius'rangebyadjustingittohypotheticalvalues;thevehiclebecomesmuchmorecompetitivewhenrangesareincreasedtotheaverageofitsPHEVcompetitors.Withinthecontextofthismodel'smixedlogitparametersthereisonlyasmallrangeofagentswillingtospendextramoneyonacarwithlessrange;generallyonlyconsumerswithhigherincomes,muchlessdailyVMT,andhighersafetyparameterpart-worthsareestimatedtopurchasethePriusPHEV.Thesehouseholdsweregenerallythemarried,higher-incomehouseholdswithlowdailyVMT.
6.5 IntheAEO2012,PEVsandAFVsareestimatedtoreachapproximately1.6%and19.81%ofvehiclestock,respectively.HigherestimatesofPEVsasapercentageofvehiclestockareanticipatedinmostscenariossincesubstitutionsofPEVsbyotherAFVsarenotincorporatedintothisanalysis.ThismodelreplicatesamarketplacewheretheonlyalternativeoptiontoanICEisanEVandthusconsumersareassumedtoseekalleviationfromhigherDGCthroughthissinglechannel,creatingthespreadbetweenthismodel'sandtheAEO2012'sestimatesofEVsasapercentageofvehiclestock.Inthemostoptimisticscenario,theEVsasapercentageofvehiclestocknearlymatchbutslightlyexceedtheAEO2012'sestimateofAFVsasvehiclestockin2030.
6.6 SimilartoKarplus(2011)butnotafocalobjectiveofthisanalysis,simulationsdemonstratethatUShouseholdsareestimatedtopurchasemorefuelefficientvehiclesinresponsetogasolinepriceincreases(evidencedbyFigure5).However,dissimilarinthesesimulationstoKarplus'analysisisthenon-distinctincomegroupingsofhouseholdsthatswitchtheirvehicleconsumptionbehavior.Theambiguousbehaviorbyhouseholdcategoricalgroupingslendsitselftotherandomnessinheritinthemixedlogitparameterswhichhaveoverlappingestimateddistributions.
6.7 Overall,ahighamountofvariationisestimatedwithrespecttoallscenarioassumptionsandresults.ThemostoptimisticscenarioestimatesEVsatapproximately22%of2030vehiclestock,muchlessthanmoreoptimisticestimates(Becker,Sidhu,andTenderich2009)butthelandscapeiscontinuallychangingonboththefossilfuelandEVsides.Althoughgasolinepriceshaveincreasedgreatlyoverthepastfewyears,thegrowingextractionratesofunconventionalsourcesofpetroleumintheBakkentightshaleformationandCanadianoilsandshavedramaticallyalteredtheperceptionsofthefutureintheNorthAmericanpetroleummarket.Ingeneral,EVproponentsareoptimisticofthefutureoftheelectricautomobilebutmightfaceanuphillbattleifmajorrisesinpetroleumpricesareaprerequisitetoEVdeployment.
Limitations
7.1 Throughoutthisanalysis,multipleassumptionswereimposedanddemonstratedtoalterresults.Thereisgreatuncertaintyinthefuturecharacteristicsofautomobilesand,althoughthismodelattemptstogrounditselfinstatisticsanddata,aninfinitenumberofdifferentoutcomesarepossible.Inthesesimulationsalone,thepermutationsofpossiblescenariosamountto8,748differentcombinationsofassumptions.Asinanyothermodel,itshouldbethoughtofasasimplifiedversionofrealityandtreatedasawell-informedthoughtexperiment.
7.2 AlargeassumptionisthatconsumersvaluethesevehiclesequallyastheywouldtheirICEcounterparts.Atthetimethisiswritten,EVtechnologieshavenotbeenaroundinsufficientquantitiestofullyassessconsumervaluationfunctionswithrevealedpreferencedata.Anotherapproachtodevelopingthemixedlogit
http://jasss.soc.surrey.ac.uk/16/2/5.html 11 14/10/2015
parametersinthisstudywouldbetoperformprimaryresearchwithstatedpreferencedataacquiredthroughconjointanalysis,orwaitforAFVstobecomemoremainstreamvehiclesandthenre-assessthestatisticalportionofthisanalysis.Performingthistaskwouldprovideaninterestingandvaluableanalysisbutisoutsidecurrenttimeandresourceavailability.
7.3 Asmentioned,EVtechnologieshavenotbeenaroundforasubstantive-enoughamountoftime.Thiscreatesuncertaintyinthefuturecharacteristicsofvehicles.Thisanalysisincorporatedconstantproducerandbatterymanufacturerresponseswhereasamoredynamicdepictionwouldbeaprofit-maximizationresponseofvehicleandcomponentmanufacturers,similartoZhangetal(2011).
7.4 Paststudieshaveindicatedthatseveralothersocialfactorsinfluenceconsumeradoptionandperceptionsofvehicles(ChooandMokhtarain2004;TurrentineandKurani2007).IncorporatingnetworkinteractionsintoaVCMisdonetoincludetheseeffectsbutfurtherresearchisrequiredtoinvestigatehumaninteractionsinamathematicalframework.Themodelpresentedhereisasimplificationoftheseinteractionsandtheassumptionsappliedhaveasubstantialeffectonthemodel'soutcome.
Appendix
AdditionalMixedLogitModelAssumptions
8.1 Discretechoicemodelsrequiredatatobeformattedsothatasinglechoiceismadeamongalternatives.Withrespecttothenewvehiclemarketplace,therearemorepurchaseoptionspresentthancurrentstatisticalsoftwarepackagescanreasonablyhandle.Duetotheserestrictions,thevehicleconsumerwasoriginallydesignatedtohavechosenamong100differentrandomly-determined(withoutreplacement)vehicleoptionsfromthesameyeartheobservedvehiclewaspurchased.Afterseveralstatisticalrunsandbasedonconvergenceandestimationefficiency,thechoicesetwaslimitedto20randomly-includedchosenvehicles.Thefinalamountofpotentialoptionsavailableinthechoicesetalsopresentedassuranceinboththereliabilityandconsistencyofestimationresults.
8.2 Theprobabilitythataspecificvehicleisincorporatedintothechoicesetisbasedontheannualmarketshareofsaidvehicle.Forexample,in2004,theFordF150marketsharewas5.29%(Wards2011),meaningthatineachobservation'schoicesettheprobabilitythattheFordF150isincludedis5.29%.Theaccuracyofthemarketsharesandtheirprobabilitiestobeincludedistothenearest1e-4,anyvehiclebelowthisvaluewasnotincludedinobservations'choicesets.
8.3 ThesoftwarepackagechosenforestimationisStata/SE64-bitandtheprocedureutilizedis'mixlogit'(Hole2007).Weightsfromthe2009NHTSwereincorporatedasfrequencyweightsintheanalysis.Additionally,sincethelargeamountofobservationsandvariabilityinthedatapresentedchallengesforthestandardalgorithmtoreachconvergence,thestandardNewton-RaphsonalgorithmweresupplementedwiththesteepestascentmethoddevelopedbyGouldandScribney(1999)aswellasGould,Pitblado,andScribney(2003).Thisreplacementonlyimpactedthenumberofiterationsrequiredtoreachconvergenceandnottheresultingparameterestimatesormodelqualities.
8.4 Since"[t]heobjectivefunctionoptimizedtoestimatethecoefficientsof[mixedlogit]modelsisgenerallyhighlynonlinear,andthuspronetomultiple,localoptima"(Greene2010,p.12),post-optimizationalgorithmsensitivityanalysiswasperformed.Theproductoftheparametersfollowingconvergenceandrandomly-assignedvalueswereusedasstartingparametersintheoptimizationprocesstoinsurethealgorithmacquiredthesameconvergencevalues.Theparametersobtainedfollowingtheoriginaloptimizationweresubsequentlymultipliedbyrandomly-assignedvaluesfromthefollowingsets:(-1.1,1.1),(-1.5,1.5),and(-2,2).Followingthesemodifications,thealgorithmreachedconvergencewiththesameparameters;theamountofadditionaliterationsrequireddependedontheextentofthedistortionappliedtotheoriginalparameterset.
8.5 Inthismodel,distributionsareimposedusingthebestfitmethoddescribedbyTrainandSonnier(2004).First,themodelswererunwithzeroandthenonlyoneofeachexplanatoryvariablewithanimposedlog-normaldistribution.Ifmultipleexplanatoryvariablessolelyprovidedanincreaseinthelog-likelihoodvalue,theexplanatoryvariablewiththegreatestmarginalcontributiontowardsthelog-likelihoodfunctionwasintroducedasthefirstlog-normallydistributedexplanatoryvariable.Thesecondgreatestcontributingexplanatoryvariabletothelog-likelihoodvaluefromlog-normaldistributionspecificationisthenintroducedandthelog-likelihoodvaluecomparedtothepriormodel'slog-likelihoodestimation,andsoon.Intheanalyses,thehouseholdswithchildrenincurredthebestfitwhenonlythelog-normaldistributionisappliedtothesafetyparameter.Imposingthelog-normaldistributiononthepowerparameteronlyimprovedtheretiredhousehold'smodelfit.Noneofthegroupings'modelfitsimprovedwhenimposingthelog-normaldistributiononeithertheDGCorMSRPparameters.Theresultsforlog-normaldistributedparameterspresentedinTable1haveundergonethenecessaryconversioncalculations.
8.6 Observationsfromthe2009NHTSwithincompleteinformationforvaluesincludedinthisstudywereexcludedresultinginalossofapproximately5%oftheoriginalsample.Also,inordertotakeintoaccountrecentvehiclecharacteristics,onlynewvehiclespurchasedfrom2004tothedateofthesurveytakenwereincludedinthemixedlogitmodel.TheNHTS2009queriedneitherthevehicle'sconditionwhenpurchased(neworused)northepriceofthevehicleandthustwoassumptionsneedtobemade.First,thedateofpurchaseneededtobewithinthenewmodelyear'ssellingperiod,assumedtobeAugusttheprioryearof
releasetoDecemberofthevehicleyear.Second,theconsumerpaidtheinflation-adjustedmanufacturer'ssuggestedretailprice(MSRP)[14].Intheend,thedatasetusedinthemixedlogitmodelcontained33,235observationsofvehiclesmatchedwiththeirprimarydriversandrespectivehouseholds.
8.7 SinceNationalAutomotiveSamplingSystem(NASS)vehiclecodesdonotdifferentiatebetweenvehicletrimlevels,theobservations'vehicletrimlevelwasdeterminedbytheclosestmatchwithinvehiclemodelbasedonthereportedMPGlevelfromthe2009NHTS.
FutureVehicleSpecifications
8.8 Vehicleconsumerswillchooseamongthesubsetofvehiclesavailableforpurchaseinthe2011vehiclefleetbutlimitedtothosewhichmeetthesamemarketsharepercentagesasthoseincludedinthemixedlogitmodeldatasetconstruction.ThesevehiclesarefixedinspecificationsexceptfortheattributesofMPGwhichincreases,foreachvehicle,withrespecttotheannualmarginalratesofchange(ROC)ofeachvehicle'spowertype(ICE,BEV,PHEV,HEV)andclass(carandlighttruck)indicatedinArgonneNationalLaboratory'sVISIONmodel(ANL,2011).
ICE
8.9 AcaveatwiththesimulationsisthatthechoicesetfromwhichthemixedlogitderivedparameterswereestimatedissmallerthanthechoicesetavailabletoconsumerswithintheABMsimulations.
HEV
CarClass LightTruckClassChevroletCruzeEcoFordFusionHybrid
BMWActiveHybridX6CadillacEscaladeHybrid
http://jasss.soc.surrey.ac.uk/16/2/5.html 12 14/10/2015
HondaCivicHybridHondaInsightHondaCR-ZandCR-ZEXHyundaiSonataHybridKiaOptimaHybridLexusCT200hand200hPremiumLexusGS450hLexusHS250hand250hPremiumLincolnMKZHybridMercedesBenzS500HybridNissanAltimaHybridToyotaCamryHybridToyotaPriusI-V
ChevroletSilverado1500SeriesHybridChevroletTahoeHybridFordEscapeHybridGMCSierra1500SeriesHybridGMCYukonHybridMercedesBenzML450HybridPorsche911CayenneSHybridToyotaHighlandHybridVolkswagenTouaregHybrid
PHEV
8.10 ChevroletVolt
ToyotaRAV4Electric[15],[16],[17]
ToyotaPrius[18]
BEV
8.11 NissanLeafFordFocusElectric
Mitsubishii(MiEV)[19],[20],[21]
TeslaModelS[22](with40,60,and85kWhbatterycapacities,atbaseMSRP)
RechargingSystemCosts,PHEVDGC,andBatteryCosts
8.12 FuelingstationavailabilityandrangeanxietyhavebeenasignificantlimitationwithconsiderationtothediffusionofBEVs(Tate,Harpster,andSavagian,2008).
Inthemodel,ifthehouseholdisconsideringanEV,thefixedpriceforalevel2chargingstationis$2000[23]andisincurredasafixcostaddedtotheMSRP.Toaccountforrangeanxiety,aconsumerwillonlyconsideranelectricvehicleiftheiraveragedailyvehiclemilestravelledislessthanorequaltoaspecifiedpercentageoftherangeoftheBEVbeingassessed.ThisspecifiedpercentageisexploredintheResultssection.
8.13 Agenti'sDGCforPHEVmodeljiscomputedas:
(5)
where:
(6)
DPGGEe:Dollarspergallonofgasolineequivalent[24]MPGej:MilespergallonofgasolineequivalentVMTi:AveragedailyvehiclemilestravelledofagentiDPGg:DollarspergallonofgasolineMPGgj:MilespergallonofthegasolineportionofPHEVj
8.14 BEVandPHEVbatteryrangeshavebeenevaluatedbythreedifferentgroups:manufacturers,government(EPA),andtheprivatesector(CarandDriver).Sincetherangeofthevehicleisaseriousconsiderationforbuyers,scenariosareimplementedwhichreducetheagents'rangeofeachBEVandPHEVmodelbydifferentspecifications.Manufacturers,theEPA,aswellasCarandDriverestimatethemaximum,median,andminimumofrangeestimates,respectively.IftheEPAhastestedthevehicle,thatreductiongoesintoplace.Ifnot,theaveragepercentagereductionofthatvehicletype'srangeisapplied.SinceCarandDriverhasonlyreviewedandsuppliedestimatesfortheNissanLeafandChevroletVolt(asofMay2012),thesamepercentagereductionswillbeappliedontherangeofsaidvehicles'class.Additionally,MPGisadjustablebetweenmanufacturerandEPAestimates,andaveragepercentagereductionsifnotavailablefromthepriorsource.
8.15 SincebatterydepreciationandcostsareabarriertoconsumeradoptionofEVs(Morales-Espana2010),thepriceofEVsisadjustedannuallytotherateofchange,byvehicleclass(carandlighttruck)andBEV/PHEVdesignation,accordingtothescheduleestimatedbytheUKCommissiononClimateChange's(CCC)reportregardingfuturebatteryspecifications(UKCCC2012).Thesedataaremanufacturerpricesandthusanassumedmarkupof10%isincluded.Additionally,sincethebatterymaintainsalargereplacementcost,thenetpresentvalueofthetotaldepreciatedvalueofthebatteryattimeofresaleisincludedatthetimeofpurchase.Thisisdonewithanassumedconstantfailurerateat10years.Inequationform:
(7)
Where:
http://jasss.soc.surrey.ac.uk/16/2/5.html 13 14/10/2015
PVi:PresentvalueofbatterydepreciationofvehicleiT:Yearsofownership(pre-determinedattimeorpurchase)ΔCi:Annualchangeinbatteryvalue
r:Discountrate[25]
FFVUtilization
8.16 Althoughotherworkhasresearchedconsumers'choicesinusingE85fuels(GreeneandZhou2008),usageratesforE85arenotsimulatedinthisresearchendeavor.Eventhoughbiofuelsareanimportantcomponentofaless-carbonintensiveandmoreenergysecurefuture(LeibyandRubin2012),severallimitationsonbiofuelutilizationhavebeenontheindustrialstructure,feedstock,andfinancialcomponentsintheindustrialsphere(Rabago2008)whichareoutsideofthismodel'spurpose.Thisisapotentialmodelexpansioncurrentlybeingexplored.
Acknowledgements
Thisworkwouldnotbepossiblewithouttheguidanceandsupportofmyadvisor,Dr.JonathanRubin.ThesiscommitteemembersDr.GaryHuntandDr.TimothyWaringdeserveconsiderableappreciationforhelpingmecompletethistask.Additionally,Iamfortunateandthankfulforthefouranonymousreviewerswhogreatlyhelpedinrefiningthispaper.
Notes
1Rangesforannualincomeareasfollows:Low:≤$45,000Middle:$45,000<x<$125,000high:≥$125,000
2Gas-guzzlertaxisincludedforcarsmeetingthecriteria
3Methodsforestimatingsafetyratingschangedforyear2011vehicles;safetyratingsusedinthisanalysisarethosepresentduringvehiclechoicedecision-making
4Significancelevels:*:0.15**:0.10***:0.05****:0.01
5AlthoughpossibletoreplicatetheentireUSpopulationitisdiscouragedduetoextensivecomputingtimes
6IftheagentownsanEV,theywillstillconsiderthestatusquoofICEvehicles
7Thisvariablefunctionssimilarlytothe'susceptibility'variablefromEppsteinetal(2011)
8ThisparameterbecomesthefocusofincentivepolicystructuresensitivityanalysisintheResultssection.
9After2035,annualratesofchangeforgasolinepricesarefromVISION2011
10Becausethisscenarioonlychangesonce,itisnotnormallypresentedinthesecondarylegend
11Includesextended-lifeonthepresentvaluecalculationsofbatterydepreciationandbatterycostadjustmentsbutdoesnotincludeefficiencygainsfromenergydensityimprovementbenefits.
12Assumedgasolineheatingvalueof125,000BTU
13Forexample,rangeisincreasedandthusVMTasarangelimitisnotasbinding
14Allmonetaryvaluesinthismodelwereadjustedto2011dollarsusingtheConsumerPriceIndex(2011)
15ModelinformationtakenfromToyota(2012a)andEdmunds(2012)
16TargetLaunchDateofSpring2013(Edmunds,2012)
17Volumeassumedtobethesameasthe2011modelICEcounterparts
18ModelinformationtakenfromToyota(2012b)
19ModelinformationtakenfromMitsubishi(2012)andEdmunds(2012)
20TargetLaunchDateofFall2012
21Atthetimethiswaswritten,safetyinformationwasnotavailable;thereforetheratingfromtheEuropeanNewCarAssessmentProgramme(EuroNCAP)offourstarswasassumed(EuroNCAP,2012)
22Sourceforinformation:TeslaMotors(2012);currentlysetonlytobeavailableinCalifornia
23BasedonNissan'sestimatesforchargerandinstallation(Nissan,2012)
24Assumedgasolineheatingvalueof125,000BTU/gallon
25Assumeddiscountrateof7%
References
http://jasss.soc.surrey.ac.uk/16/2/5.html 14 14/10/2015
ALLCOTT,H.andN.Wozney(2010)."Gasolineprices,fueleconomy,andtheenergyparadox."fromhttp://www.nber.org/papers/w18583.
ALTERNATIVEFUELSADVANCEDVEHICLESDATACENTER(AFDC)(2012)."AlternativeFuelStations."RetrievedJanuary,15,2012,fromhttp://www.afdc.energy.gov/afdc/fuels/stations.html.
ARAL,S.andD.Walker(2011a)."Creatingsocialcontagionthroughviralproductdesign:Arandomizedtrialofpeerinfluenceinnetworks."ManagementScience57(9):1623-1639.[doi:10.1287/mnsc.1110.1421]
ARAL,S.andD.Walker(2011b)."IdentifyingSocialInfluenceinNetworksUsingRandomizedExperiments."IEEEIntelligentSystems,September-OctoberInpress.[doi:10.1109/mis.2011.89]
ARGONNENATIONALLABORATORY(2011)."TheVISIONModel—2011AEOBaseCase"fromhttp://www.transportation.anl.gov/modeling_simulation/VISION/.
BEARDEN,W.O.andM.J.Etzel(1982)."ReferenceGroupInfluenceonProductandBrandPurchaseDecisions."JournalofConsumerResearch9:183-481.[doi:10.1086/208911]
BECKER,T.,Sidhu,I.,andTenderich,B.(2009)."ElectricvehiclesintheUnitedStates:Anewmodelwithforecaststo2030."UniversityofCalifornia,BerkeleyCenterforEntrepreneurshipandTechnology.RetrievedMay21,2012from:http://cet.berkeley.edu/dl/CET_Technical%20Brief_EconomicModel2030_f.pdf
BOLDUC,D.,N.Boucher,etal.(2009)."HybridchoicemodelingofnewtechnologiesforcarchoiceinCanada."TransportationResearchRecord,JournaloftheTransportationResearchBoardPlanningandAdministration:63-71.
CHILDERS,T.L.andA.R.Rao(1992)."TheInfluenceofFamilialandPeerBasedReferenceGroupInfluenceonConsumerDecisions."JournalofConsumerResearch19:198-211.
CHOO,S.andP.Mokhtarian(2004)."Whattypeofvehicledopeopledrive?Theroleofattitudeandlifestyleininfluencingvehicletypechoice."TransportationResearchPartA38(2004):201-222.[doi:10.1016/j.tra.2003.10.005]
CUI,X.,H.Kim,etal.(2011).Amultiagent-basedframeworkforsimulatinghouseholdPHEVdistributionandelectricdistributionnetworkimpact.PaperinConferenceProceeding,TransportationResearchPartA:PolicyandPractice,Washington,DC.June2011.
CURTIN,R.,Y.Shrago,etal.(2009)."Plug-inHybridElectricVehicles."fromhttp://www.ns.umich.edu/Releases/2009/Oct09/PHEV_Curtin.pdf.
DASGUPTA,S.,S.Siddarth,etal.(2007)."LeaseorBuy:AStructuralModelofaConsumer'sVehicleandContractChoiceDecisions."JournalofMarketingResearch44(3):490-502.[doi:10.1509/jmkr.44.3.490]
DE-HAAN,P.andM.Mueller(2009)"Howmuchdoincentivesaffectcarpurchase?Agent-basedmicrosimulationofconsumerchoiceofnewcarsPartI:Modelstructure,simulationofboundedrationality,andmodelvalidation."
DEHAAN,P.,M.Mueller,etal.(2009)."Howmuchdoincentivesaffectcarpurchase?Agent-basedmicrosimulationofconsumerchoiceofnewcars-PartII:Forecastingeffectsoffeebatesbasedonenergy-efficiency."EnergyPolicy37(3):1083-1094.[doi:10.1016/j.enpol.2008.11.003]
DUGUNDJI,E.R.,A.PáezandT.A.Arentze(2008).SocialNetworks,Choices,MobilityandTravel.EnvironmentandPlanningB:PlanningandDesign,35(6),2008,956-960.
DUGUNDJI,E.R.,A.Páez,T.A.ArentzeandJ.L.Walker(2011).TransportationandSocialInteractions.TransportationResearchPartA:PolicyandPractice,45(4),2011,239-247.[doi:10.1016/j.tra.2011.01.001]
DUGUNDJI,E.R.andL.Gulyás(2013).Socio-DynamicDiscreteChoiceonNetworksinSpace:ImpactofInitialConditions,NetworkSizeandConnectivityonEmergentOutcomesinaSimpleNestedLogitModel.StudiesinComputationalIntelligenceSeries,Springer,vol.424,2013,pp.149-160.[doi:10.1007/978-3-642-30287-9_16]
EDMUNDS(2012)."CarReviews."fromhttp://www.edmunds.com/car-reviews/.
EPPSTEIN,M.,D.Growever,etal.(2011)."Anagent-basedmodeltostudymarketpenetratioofplug-inhybridelectricvehicles."EnergyPolicy39(2011):3789-3802.[doi:10.1016/j.enpol.2011.04.007]
EUROPEANNEWCARASSESSMENTPROGRAM(EuroNCAP)(2012)."EuroNCAPSafetyRatings."fromhttp://www.euroncap.com.
GALLAGHER,K.S.andE.Muehlegger(2008)."GivingGreentoGetGreen?IncentivesandConsumerAdoptionofHybridVehicleTechnology."JournalofEnvironmentalEconomicsandManagement.61(1):1-15.[doi:10.1016/j.jeem.2010.05.004]
GOULD,W.,J.Pitblado,etal.(2003).MaximumlikelihoodestimationwithSTATA,SecondEdition.CollegeStation,TX,StataPress.
GOULD,W.andW.Scribney(1999).MaximumLikelihoodEstimationwithSTATA.CollegeStation,TX,STATAPress.
GREENE,D.(2010).Howconsumersvaluefueleconomy:aliteraturereview.UAEnvironmentalProtectionAgency(EPA).Washington,DC.
GREENE,D.andJ.Zhou(2008)."ConsumerChoiceofE85:LessonsfromMinnesota'sExperience."TransportationResearchRecord,JournaloftheTransportationResearchBoard,650(TRB09-3018).
HAMLYN,P.(1967).CarsCarsCarsCars.PublishedbyTheHamlynPublishingGroupLTD.Middlesex,England.
HEFFNER,R.,K.Kurani,etal.(2005)."EffectsofVehicleImageinGasoline-HybridElectricVehicles."UCDavisInstituteofTransportationStudies:UCD-ITS-RR-05-08.
HENRYFORDESTATE(2007)."InsideHenryFord'sGarage-1914DetroitElectric."RetrievedJanurary12,2012from:http://www.henryfordestate.org/claracar.htm
HIEBELER,D.(forthcoming)IntroductiontoSimulatingandModelingSpatialPopulationEcologicalandEpidemiologicalDynamics.
HOLE,A.(2007)."Fittingmixedlogitmodelsbyusingmaximumsimulatedlikelihood."TheStataJournal7(3):388-401.
INTERNALREVENUESERVICE(2011)."Plug-inElectricVehicleCredit."RetrievedDecember2,2011from:http://www.irs.gov/businesses/article/0,,id=214841,00.html
STRUBEN,J.andSterman,J(2008)."TransitionChallengesforAlternativeFuelVehicleandTransportationSystems".EnvironmentandPlanningB35(6)1070
http://jasss.soc.surrey.ac.uk/16/2/5.html 15 14/10/2015
–1097.[doi:10.1068/b33022t]
KAHN,M.andR.Vaughn(2009)."GreenMarketGeography:TheSpatialClusteringofHybridVehicleandLEEDRegisteredBuildings."TheBEJournalofEconomicAnalysis&Policy9(2).[doi:10.2202/1935-1682.2030]
KARPLUS,V.(2011)."ClimateandEnergyPolicyforU.S.PassengerVehicles:ATechnology-RichEconomicModelingandPolicyAnalysis."PhDDissertation,MITEngineeringSystemsDivison.
KURANI,K.andT.Turrentine(2004)."AutomobileBuyerDecisionsaboutFuelEconomyandFuelEfficiency,FinalReporttotheDepartmentofEnergy."fromhttp://escholarship.org/uc/item/5hh5k3j3.pdf.
LEIBY,P.andJ.Rubin(2012)."EnergySecurityImplicationsofaNationalLowCarbonFuelStandard."EnergyPolicy,InPress.
MAKGOSA,R.andK.Mohube(2007)."Peerinfluenceonyoungadults'productspurchasedecisions."AfricanJournalofBusinessManagement1(3),64-71.
MARCHETTI,N.(2012)."ToyotaRAV4ElectricSUV:A$50,000toyorstatement?"RetrievedJune19,2012fromhttp://www.tgdaily.com/sustainability-features/63346-toyota-rav4-electric-suv-a-50000-toy-or-statement
MAU,P.,J.Eyzaguirre,etal.(2008)."The'neighboreffect':Simulatingdynamicsinconsumerpreferencesfornewvehicletechnologies."EcologicalEconomics68(2008):504-516.[doi:10.1016/j.ecolecon.2008.05.007]
MCPHERSON,M.,L.Smith-Lovin,etal.(2001)."Birdsofafeather:homophilyinsocialnetworks."AnnualReviewsofSociology27:415-444.
MICHNIAK,B.,R.Morin,etal.(2012)."Consumerattitudesandbehavioursrelatedtoalternativelyfueledvehicles."RetrievedJune19,2012fromhttp://uwaft.com/wp/wp-content/uploads/SS-Consumer-Attitudes-and-Behaviours-Related-to-AFVs.pdf
MITSUBISHIMOTORS(2012)."MiEV."RetrievedMarch4,2012fromhttp://i.mitsubishicars.com/.
MOLOFSKY,J.,R.Durrett,etal.(1999)."LocalFrequencyDependenceandGlobalCoexistence."TheoreticalPopulationBiology55:270-282.[doi:10.1006/tpbi.1998.1404]
MORALES-ESPANA,G.A.(2010)."Electricvehiclediffusion:exploringuncertaintiesusingsystemdynamics."DelftUniversityofTechnologyInstitutionalRepository.RetrievedApril6,2012from:http://repository.tudelft.nl/view/ir/uuid%3A70105630-9dc9-445c-8e0e-71b5987d1384/
NATIONALHIGHWAYTRAFFICSAFETYADMINISTRATION(NHTSA)(2006).VehicleSurvivabilityandTravelMileageSchedules.DepartmentofTransportation.Washington,DC.
NATIONALHIGHWAYTRAFFICSAFETYADMINISTRATION(NHTSA)(2011).NationalAutomotiveSamplingSystem(NASS)GeneralEstimatesSystem(GES)AnalyticalUsersManual1988-2010.U.S.DepartmentofTransportation(DOT).Washington,DC.
NISSAN(2012)."NissanLeafcharging,howtochargeathome."RetrievedFebruary8,2012from:http://www.nissanusa.com/leaf-electric-car/home-charging
POLK(2012)."U.S.consumersholdontonewvehiclesnearlysixyears,anall-timehigh."RetrievedApril14,2012from:https://www.polk.com/company/news/u.s._consumers_hold_on_to_new_vehicles_nearly_six_years_an_all_time_high
RABAGO,K.(2008)."AreviewofbarrierstobiofuelmarketdevelopmentintheUnitedStates."EnvironmentalandEnergyLawandPolicy179.
ROGERS,E.(1963).DiffusionofInnovations.NewYork,NY:FreePress.
ROMERO,F.(2009)"Abriefhistoryoftheelectriccar."RetrievedJanuary12,2012fromhttp://www.time.com/time/business/article/0,8599,1871282,00.html
RUBIN,J.andP.Leiby(2012)."NationalLowCarbonFuelStandard:TradableCreditsSystemDesignandCostSavings."EnergyPolicy,InPress.
SCHELLING,ThomasC.1978.MicromotivesandMacrobehavior.NewYork:W.W.Norton.
SEXTON,S.andA.Sexton(2010)."DRAFT:ConspicuousConservation:ThePriusEffectandWTPforEnvironmentalBonaFides."fromhttp://areweb.berkeley.edu/fields/erep/seminar/Prius_Effect_V1.5.1.pdf.
STRUBEN,J.andJ.Sterman(2008)."TransitionChallengesforAlternativeFuelVehicleandTransportationSystem."EnvironmentandPlanningB35:1070-1097.[doi:10.1068/b33022t]
SULLIVAN,J.Salmeen,I,andSimon,P.(2009)"PHEVMarketplacePenetrationAnAgentBasedSimulation."UniversityofMichiganTransportationResearchInstitute.RetrievedMay19,2012fromhttp://141.213.232.243/bitstream/2027.42/63507/1/102307.pdf
TATE,E.,O.Harpster,etal.(2008)."TheElectrificationoftheAutomobile:FromConventionalHybrid,toPlug-inHybrids,toExtended-RangeElectricVehicles."fromhttp://www.media.gm.com/content/dam/Media/microsites/product/volt/docs/paper.pdf.
TESLAMOTORS(2012)."TeslaMotors,PremiumElectricVehicles."fromhttp://www.teslamotors.com/.
TOYOTA(2012a)."RAV4EV."RetrievedMarch6,2012fromhttp://www.toyota.com/concept-vehicles/rav4ev.html.
TOYOTA(2012b)"2012ToyotaPriusPlug-in."RetrievedMarch7,2012fromhttp://www.toyota.com/prius-plug-in/
TRAIN,K.(2009).DiscreteChoiceMethodswithSimulation,CambridgeUniversityPress.
TRAIN,K.andG.Sonnier(2004)."MixedLogitwithBoundedDistributionsofCorrelatedPartworths."fromhttp://elsa.berkeley.edu/~train/tsaug04.pdf.
TRAIN,K.andC.Winston(2007)."VehicleChoiceBehaviorAndTheDecliningMarketShareOfU.S.Automakers."InternationalEconomicReview48(4):1469-1496.[doi:10.1111/j.1468-2354.2007.00471.x]
TURRENTINE,ThomasS.andKennethS.Kurani(2007)Carbuyersandfueleconomy?.EnergyPolicy35(2007),1213-1223.[doi:10.1016/j.enpol.2006.03.005]
UKCOMMITTEEONCLIMATECHANGE(2012)."CostandperformanceofEVbatteries."RetrievedMay7,2012fromhttp://www.axeon.com/getdoc/77fbadce-7d3e-4f13-b996-e8224330db44/CCC-battery-cost-report_March2012_Public.aspx
UNIONOFCONCERNEDSCIENTISTS(2011).Statesandfederalhybridincentives.RetrievedJanuary9,2012fromhttp://go.ucsusa.org/hybridcenter/incentives.cfm
USCENSUSBUREAU(2012)."The2012StatisticalAbstract."fromhttp://www.census.gov/compendia/statab/.
http://jasss.soc.surrey.ac.uk/16/2/5.html 16 14/10/2015
USDEPARTMENTOFENERGY(2001).TheTransportationSectorModeloftheNationalEnergyModelingSystem,ModelDocumentationReport.DOE/EIA-M070(2001).Retrievedfrom:ftp://ftp.eia.doe.gov/modeldoc/m0702001.pdf
USDEPARTMENTOFENERGY(2012)."TheRecoveryAct:TransformingAmerica'stransportationsector,batteriesandelectricvehicles."RetrievedMay1,2012from:http://www.whitehouse.gov/files/documents/Battery-and-Electric-Vehicle-Report-FINAL.pdf
USDEPARTMENTOFTRANSPORTATION(DOT)(2012)."VehicleSafetyRatings."fromhttp://www.safercar.gov.
USDEPARTMENTOFTRANSPORTATION(DOT)ANDFEDERALHIGHWAYADMINISTRATION(2009)."NationalHouseholdTravelSurvey".fromhttp://nhts.ornl.gov.
USENERGYINFORMATIONADMINISTRATION(USEIA)(2011)."2012AnnualEnergyOutlook."fromhttp://www.eia.gov/forecasts/aeo/er/.
USENERGYINFORMATIONADMINISTRATION(USEIA)(2012)."GasolinePricesbyFormulation,Grade,SalesType."fromhttp://205.254.135.7/dnav/pet/pet_pri_allmg_a_EPM0_PTC_Dpgal_m.htm.
USENVIRONMENTALPROTECTIONAGENCY(2010).EnvironmentalProtectionAgencyFuelEconomyLabelLiteratureReview.U.S.EnvironmentalProtectionAgency(EPA).Washington,DC.
VOELCKER(2012)."2012ToyotaRAV4EVelectricSUV:Fewerthan1,000ayear."RetrievedJune20,2012from:http://www.greencarreports.com/news/1075962_2012-toyota-rav4-ev-electric-suv-fewer-than-1000-a-year
WARD'SAUTOMOTIVEINFORMATIONPRODUCTS(2011).Vehiclespecificationsandcharacteristics.PurchasedJuly,2011.Contact:McLincha,[email protected]
ZHANG,T.,S.Gensler,etal.(2011)."Astudyofdiffusionofalternativefuelvehicle:Anagent-basedmodelingapproach."JournalofProductionInnovationandManagement28:152-168.[doi:10.1111/j.1540-5885.2011.00789.x]
http://jasss.soc.surrey.ac.uk/16/2/5.html 17 14/10/2015