Citation
Residential Energy Consumers Response to Energy Efficiency Rebates, Incentives, and Prices

Material Information

Title:
Residential Energy Consumers Response to Energy Efficiency Rebates, Incentives, and Prices
Creator:
Boampong, Richard
Place of Publication:
[Gainesville, Fla.]
Florida
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (108 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Economics
Committee Chair:
HAMILTON,JONATHAN H
Committee Co-Chair:
SAPPINGTON,DAVID
Committee Members:
AI,CHUNRONG
MOSS,CHARLES BRITT
Graduation Date:
8/6/2016

Subjects

Subjects / Keywords:
Average prices ( jstor )
Billing ( jstor )
Consumer prices ( jstor )
Datasets ( jstor )
Electricity ( jstor )
Energy ( jstor )
Energy consumption ( jstor )
Marginal pricing ( jstor )
Price elasticity ( jstor )
Sales rebates ( jstor )
Economics -- Dissertations, Academic -- UF
airconditioner -- autopay -- difference-in-difference -- efficiency -- gainesville -- matching -- rebate -- rebound -- utilities
Genre:
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Economics thesis, Ph.D.

Notes

Abstract:
This dissertation is comprised of three chapters. In the first chapter, I estimate the energy savings effects of a Demand-Side Management (DSM) program, specifically Gainesville Regional Utility's (GRU) high-efficiency central Air Conditioner (AC) rebate program, in which GRU offers incentives to its customers to replace their old low-efficiency AC unit with a high-efficiency model. This research combines a Coarsened Exact Matching (CEM) methodology with a Difference-in-Difference (DD) approach. I estimate the impact of GRU's 2009 high-efficiency AC program on annual energy consumption. Also, because the primary reason for a DSM program is to reduce peak period consumption, I disaggregated the energy savings effects of the program into summer peak effects, winter peak effects, and non-peak effects The results show substantial annual energy savings of the high-efficiency AC program. While the program has substantial effects on summer peak and non-peak consumption, it has no significant effects on winter peak usage. The second chapter investigates if there is a "rebound effect" (increased energy consumption) once consumers learn of their energy cost savings after participating in a DSM program. I find no statistically significant rebound effects of the AC rebate program. The third project analyzes whether automatic bill payment (or autopay) renders consumers electricity demand less price sensitive. I propose a conceptual model in which consumers make behavioral rules to their electricity consumption only after observing their bill from the previous month. An implication of this model is that a consumer's current demand for electricity responds to the average price he paid for electricity in the preceding month. Another implication of the model is that automatic bill payment which reduces the likelihood that a consumer will examine the charges on their bill makes him less price salient. We demonstrate, empirically, that consumers respond to their one-month lagged average price, consistent with the conceptual model. I also show that automatic bill payment users are 10\% less price elastic than non-autopay users. I show further that enrolling in automatic bill payments makes a consumer's electricity demand elasticity 5\% less (in absolute terms) than it was before he enrolled in autopay. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (Ph.D.)--University of Florida, 2016.
Local:
Adviser: HAMILTON,JONATHAN H.
Local:
Co-adviser: SAPPINGTON,DAVID.
Statement of Responsibility:
by Richard Boampong.

Record Information

Source Institution:
UFRGP
Rights Management:
Copyright Boampong, Richard. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Classification:
LD1780 2016 ( lcc )

Downloads

This item has the following downloads:


Full Text

PAGE 1

RESIDENTIALENERGYCONSUMERSRESPONSETOENERGYEFFICIENCYREBATES,INCENTIVES,ANDPRICESByRICHARDBOAMPONGADISSERTATIONPRESENTEDTOTHEGRADUATESCHOOLOFTHEUNIVERSITYOFFLORIDAINPARTIALFULFILLMENTOFTHEREQUIREMENTSFORTHEDEGREEOFDOCTOROFPHILOSOPHYUNIVERSITYOFFLORIDA2016

PAGE 2

c2016RichardBoampong

PAGE 3

Tomyparents

PAGE 4

ACKNOWLEDGMENTSIamindebtedtomydissertationchairandco-chair,ProfessorJonathanHamiltonandProfessorDavidSappingtonfortheirusefulguidance,comments,andconstructivecriticism.Theirusefulsuggestions,notonlyonmyanalysis,butalsoonmywritinghasusheredmeontoapathofsuccessfulandeectiveacademicwriting.I'malsogratefultoothermembersofmydissertationcommittee,ProfessorChunrongAiandProfessorCharlesMoss,aswellasDr.TedKuryandotherprofessorsoftheEconomicsDepartmentfortheirinvaluablesupportandmotivation.Mysinceregratitudealsogoestomyparents,Mr.andMrs.Boachiefortheirsupportandencouragementthroughoutmyacademiccareer. 4

PAGE 5

TABLEOFCONTENTS page ACKNOWLEDGMENTS ................................... 4 LISTOFTABLES ...................................... 7 LISTOFFIGURES ..................................... 9 ABSTRACT ......................................... 10 CHAPTER 1EVALUATINGTHEENERGYSAVINGSEFFECTSOFAUTILITYDEMAND-SIDEMANAGEMENTPROGRAMUSINGADIFFERENCE-IN-DIFFERENCECOARSENEDEXACTMATCHINGAPPROACH 12 Introduction ....................................12 Background:GainesvilleRegionalUtilitiesRebatePrograms ...........15 EmpiricalStrategyandMethod ..........................16 Data ........................................22 Self-SelectionBasedonIndividualPre-TreatmentCharacteristicsforTheTreatmentandControlGroups ................................26 Results .......................................30 Conclusion .....................................48 2THE\REBOUND"EFFECT 49 Introduction ....................................49 ConceptualFramework ...............................52 GraphicalAnalysisofTheReboundEect .....................53 EstimationandResults ...............................54 Conclusion .....................................59 3EFFECTSOFAUTOPAYPROGRAMONPRICESENSITIVITY 60 Introduction ....................................60 ABriefonAutomaticBillPaymentandABackgroundtoGRU'sAutopayProgram 64 ElectricityDemandandConceptualFramework ..................66 Data ........................................70 EmpiricalStrategyandResults ..........................72 Conclusion .....................................88APPENDIX AAPPENDIXTOACCOMPANYEVALUATINGTHEENERGYSAVINGSEFFECTSOFAUTILITYDEMAND-SIDEMANAGEMENTPROGRAMUSINGADIFFERENCE-IN-DIFFERENCECOARSENEDEXACTMATCHINGAPPROACH 91 BAPPENDIXTOACCOMPANYTHE\REBOUND"EFFECT 99 5

PAGE 6

REFERENCES ........................................ 104 BIOGRAPHICALSKETCH ................................. 108 6

PAGE 7

LISTOFTABLES Table page 1-1GRU'sRebateProgramsandIncentives ......................16 1-2VariablesfromEachDataset ............................23 1-3SummaryStatistics ................................25 1-4MatchingSummary{High-EciencyACRebateProgram .............32 1-5AnnualEnergySavingsEectofThe2009High-EciencyACRebateProgram 35 1-6Non-PeakMonthsEectsOfThe2009ACRebateProgram ...........38 1-7SummerPeakEectsofThe2009ACrebateprogram ..............40 1-8WinterPeakEectsofThe2009ACRebateProgram ..............43 1-9Summer,Winter,andNon-PeakMonthsEectsofTheHighEciencyACRebateProgramforTheElectric-OnlyHouseholds ....................45 1-10Summer,WinterandNon-PeakMonthsEectsofThe2009ACRebateProgramonElectricityConsumptionforHouseholdswithBothElectricityandNaturalGasinTheirEnergyMix ..............................47 2-1EectofThe2009High-EcientRebateProgramon2011EnergySavings ...57 2-2SummerPeak,WinterPeak,andNon-peakReboundEectsofTheACRebateProgram ......................................58 3-1VariablesfromEachDataset ............................70 3-2SummaryStatistics ................................73 3-3EncompassingTest:CurrentVs.One-MonthLaggedAveragePrice .......77 3-4WhatAectsParticipationinAutopay ......................78 3-5AutomaticBillPaymentEectsonPriceElasticity ................86 3-6SummaryofDierence-in-DierenceEstimates ..................87 3-7EectsofEnrollinginAutomaticBillPaymentonPriceElasticity ........88 A-1DIDCEMEstimateofTheEectsofThe2009HighEciencyACProgram ..91 A-2AnnualEnergySavingsEectOfThe2010HighEciencyACRebateProgram 92 A-3CEMDIDEstimateofTheEectsofThe2010HighEciencyACProgram ..93 7

PAGE 8

A-4Non-PeakMonthsEectsOfThe2009ACRebateProgramusingCEMDIDwithZipCodesasNeighborhoods .........................94 A-5DIDCEMEstimateofTheEectofTheHighEciencyACRebateonSummerPeakEnergyConsumption .............................95 A-6DIDCEMEstimateofTheEectofTheHighEciencyACRebateonWinterPeakEnergyConsumption .............................96 A-7Summer,Winter,andNon-PeakMonthsEectsoftheHighEciencyACprogramforHouseholdswithElectricitybutNoNaturalGasUsingTheDIDCEMwithZipCodesasNeighborhoods ............................97 A-8Summer,Winter,andNon-PeakMonthsEectsoftheHighEciencyACprogramforHouseholdswithElectricityandNaturalGasUsingTheDIDCEMwithZipCodesasNeighborhoods ..............................98 B-1SummerPeakReboundEectsbyFuelMixUsingCEMDDwithZipCodesasNeighborhoods ...................................100 B-2WinterPeakReboundEectsbyFuelMix ....................101 B-3Non-PeakReboundEectsbyFuelMix ......................102 B-4SummerPeak,WinterPeak,andNon-peakReboundEectsofTheACRebateProgramUsingDDCEMwithZipCodesasNeighborhoods ...........103 8

PAGE 9

LISTOFFIGURES Figure page 1-1PercentageofProgramParticipantsandNon-ParticipantswithinEachEnergyConsumptionQuartile ...............................27 1-2DistributionofAgeofBuildingforParticipantsandNon-Participants ......29 1-3DistributionofHeatedAreaSquareFootageofBuildingforParticipantsandNon-Participants ..................................30 1-4AverageElectricityConsumptioninTheSummer,Winter,andNon-peakMonthsbyHouseholdEnergyComposition ........................42 1-5AverageElectricityConsumptionintheSummer,Winter,andNon-peakmonthsbyHouseholdEnergyComposition ........................44 2-1AverageEnergyConsumptionbyParticipantsandNon-Participants .......55 3-1Increasing-BlockResidentialElectricityPricingScheduleofGainesvilleRegionalUtilities(GRU),2009 ...............................67 3-2BillingPeriodsandBillArrivalTimes:AnIllustrationofWhatLaggedAveragePriceConsumersRespondDuringTheBillPeriod ................69 B-1AverageEnergyConsumptionbyParticipantsandNon-Participants .......99 B-2AverageEnergyConsumptionbyParticipantsandNon-Participants .......99 9

PAGE 10

AbstractofDissertationPresentedtotheGraduateSchooloftheUniversityofFloridainPartialFulllmentoftheRequirementsfortheDegreeofDoctorofPhilosophyRESIDENTIALENERGYCONSUMERSRESPONSETOENERGYEFFICIENCYREBATES,INCENTIVES,ANDPRICESByRichardBoampongAugust2016Chair:JonathanHamiltonCochair:DavidSappingtonMajor:EconomicsThisdissertationiscomprisedofthreechapters.IntheChapter 1 ,IestimatetheenergysavingseectsofaDemand-SideManagement(DSM)program,specicallyGainesvilleRegionalUtility's(GRU)high-eciencycentralAirConditioner(AC)rebateprogram,inwhichGRUoersincentivestoitscustomerstoreplacetheiroldlow-eciencyACunitwithahigh-eciencymodel.ThisresearchcombinesaCoarsenedExactMatching(CEM)methodologywithaDierence-in-Dierence(DD)approach.IestimatetheimpactofGRU's2009high-eciencyACprogramonannualenergyconsumption.Also,becausetheprimaryreasonforaDSMprogramistoreducepeakperiodconsumption,Idisaggregatedtheenergysavingseectsoftheprogramintosummerpeakeects,winterpeakeects,andnon-peakeects.Theresultsshowsubstantialannualenergysavingsofthehigh-eciencyACprogram.Whiletheprogramhassubstantialeectsonsummerpeakandnon-peakconsumption,ithaslittleornosignicanteectsonwinterpeakusage.Chapter 2 investigatesifthereisa\reboundeect"(increasedenergyconsumption)onceconsumerslearnoftheirenergycostsavingsafterparticipatinginaDSMprogram.IndnostatisticallysignicantreboundeectsoftheACrebateprogram.Chapter 3 analyzeswhetherautomaticbillpayment(orautopay)rendersconsumerselectricitydemandlesspricesensitive.Iproposeaconceptualmodelinwhichconsumersmakebehavioralrulestotheirelectricityconsumptiononlyafterobservingtheirbillfromtheprevious 10

PAGE 11

month.Themodelsuggeststhataconsumer'scurrentdemandforelectricityrespondstotheaveragepricehepaidforelectricityintheprecedingmonth.Animplicationofthemodelisthatautomaticbillpaymentwhichreducesthelikelihoodthataconsumerwillexaminethechargesontheirbillmakeshimlesspricesalient.Wedemonstrate,empirically,thatconsumersrespondtotheirone-monthlaggedaverageprice,consistentwiththeconceptualmodel.Ialsoshowthatautomaticbillpaymentusersare10%lesspriceelasticthannon-autopayusers.Ishowfurtherthatenrollinginautomaticbillpaymentmakesaconsumer'selectricitydemandelasticity5%less(inabsoluteterms)thanitwasbeforeheenrolledinautopay. 11

PAGE 12

CHAPTER1EVALUATINGTHEENERGYSAVINGSEFFECTSOFAUTILITYDEMAND-SIDEMANAGEMENTPROGRAMUSINGADIFFERENCE-IN-DIFFERENCECOARSENEDEXACTMATCHINGAPPROACH 1.1IntroductionSincethelate1970's,therehasbeenawidevarietyofUtilityDemand-SideManagement(DSM)programstoreduceenergyconsumption.Price-basedprogramssuchaspeak-loadpricingandincentive-basedDemandResponse(DR)programssuchasdirectloadcontrol,demandbidding,andinterruptibleprogramsareconsideredmosteectiveinreducingpeakperiodenergydemand.However,mostutilitiesnditdiculttoimplementthesemeasuresduetoprogramcostandproblemswithoverpaymentorunderpaymentofincentivesduetounveriablebaselinemechanismsforobtainingconsumptionreductions( Bushnelletal. , 2009 ).Residentialhomeretrottingprogramsthusappearasanalternativeforenergysavingsthatcanavoidtheproblemsofprice-basedorincentives-baseddemandresponseprograms.Also,thesetraditionalenergyeciencyretrotprogramscanhelpinstalltheautomationsystemsneededtoallowcustomerstoparticipateinanautomateddemand-responseprograms( Violette , 2008 ).Anothersignicantadvantageofresidentialretrottingprogramsisthatunlikepriceorincentive-baseddemandresponseprograms,they\donotinvolvemajoradjustmenttoconsumers'lifestylesandoerpotentialeconomicreturnstoconsumers"( GamtessaandRyan , 2007 ).Currently,almostallelectricutilitiesintheUnitedStatesoerrebateprogramstoencouragecustomerstoparticipateinretrotprograms.Astheseenergyeciencyretrotprogramsgrowinsizeandcost,thereistheneedtounderstandbettertheireectsandcost-eectiveness.Sincethe1990's,therehasbeenamultitudeofevaluationmethodologiesrangingfromthecrystalballmeasuresofsavings(e.g.monthlyenergysavingsinCalifornia's20-20programinthesummerof2005wascalculatedasthedierenceinenergyconsumptionrelativetothesamemonthinthepreviousyear.1)to 1See Koichiro ( 2012 )forabackgroundoftheprogram. 12

PAGE 13

engineeringsimulationmodels2,andtovariouseconometricmodelscombiningmonthlymeterreadingsandavailabledataoncustomercharacteristicstodetermineenergysavings(e.g. Jonesetal. ( 2010 ); Cohenetal. ( 1991 )).Engineeringmethodsusesimulationstopredictenergysavingsfromspecicmeasuresattheindividualbuildinglevelorattheend-useequipmentlevel.Sincetheseengineeringmethodsdonotrequirecustomers'consumptiondata,theyaretheoreticallyappealingwhencustomerinformationisnotavailable.However,predictionsfromengineeringmodelsarenormallyawedandmisrepresentstheactualenergysavingssincetheydonotaccountfortheinuencesofconfoundingfactorssuchasbehavioranddemographicsofahousehold( FelsandKeating , 1993 ).Econometricmethods,ontheotherhand,usecustomersbillinginformationwhilecontrollingforweatherandhousehold-levelandbuildinglevelfactorsthatmightaectconsumers'energyconsumption.MosteconometricevaluationsoftheeectsofaDSMprogramusetheclassicdierence-in-dierence(DD)methodologyoravariantofitwheretheimpactoftheDSMprogramisestimatedasthedierenceinmeanoutcomesbetweenallhouseholdsparticipatingintheprogramandthosenotparticipating(e.g. Godberg ( 1986 )).Thisapproachleadstobiasifthereareunobservedcharacteristicsthataecttheprobabilityofparticipatingintheprogramthatarealsocorrelatedwiththeoutcomeofinterest.Further,theresultmightalsobebiasedifprogramparticipantsareverydierentfromnon-participantsintermsofpre-treatmentcharacteristics.Evencontrollingforpre-treatmentcharacteristicsintheDDregressiondoesnotnecessarilyreducethisbiassincetheestimatedeectdependsontheexactfunctionalformused.Inthischapter,Ievaluatetheenergysavingseectofaresidentialretrottingprogram,GRU'shigh-eciencyACrebateprogram.Icombineadierence-in-dierence(DD) 2See ArchitecturalEnergyCorporation ( 1992 )forareviewofvariousengineeringsimulationprogramsforestimatingDSMprogramenergysavings. 13

PAGE 14

methodologywithaCoarsenedExactMatching(CEM)approach3describedin Iacusetal. ( 2008 )toovercomethebiasfromconfoundingpre-treatmentcharacteristics.Suchanestimationapproachisnoveltotheevaluationofenergysavingsfromdemand-sidemanagementprogramsand\theyarearguablymoreappropriatecomparedtoasimpleinstrumentalvariableapproach(fordealingwiththeselectionbias4)asnostrongexclusionrestrictionsareneeded"( GirmaandGorg , 2007 ).ThismethodisparticularlyimportanttoevaluatingDSMprogramsforotherreasons;matchingonneighborhoodsallowsustocompareparticipantsandnon-participantsinthesameneighborhood.Hence,weareabletodisentangletheeectsofweatherfromprogrameectssincehousesinthesameneighborhoodaremorelikelytoexperiencethesameweather.Thismethodisparticularlyusefuliftheareaunderstudyhasoneorjustafewweatherstationswhichmakeitimpossibletocontrolfortheeectsofweatheronenergyconsumption.Also,sincehousesbuiltinthesameyearorafewyearsapartandinthesameneighborhoodarelikelytobebuiltwiththesamebuildingmaterialsandhavesimilarcharacteristics,usingneighborhoodsandageofbuildinginthematchingmethodologycontrolsfortheeectsofbuildingcharacteristicsandmaterialsonenergyconsumption.AnaddedimportanceoftheCEMmethodisthatsincetherebateprogramhadaverylowparticipationrate,itprovidesawaytoselectareasonablecontrolgroupfromthehighpercentageofnon-participatinghouseholds.Forexample,onlyabout6%ofhouseholdsparticipatedinatleastoneofGRU'srebateprogramsintheyear2009.Thispercentageismuchlower(about2%)whenweconsideronlythehigh-eciencyACprogram.Usingallthe98%ofhouseholdsthatdidnotparticipateasacontrolgroupmaybiastheenergysavingsestimateasthetreatmentgroupdoesnotincludeallsectionsofthepopulation. 3Theideaofcoarsenedexactmatchingisdescribedundertheempiricalstrategyandmethodologysection(Section 1.3 ).4Selectionbiasoccurswhenparticipationinaprogramisnotrandomanddependsonsomeobservableorunobservablecharacteristicsthatarecorrelatedwiththeoutcomeofinterest. 14

PAGE 15

IusedataonhouseholdelectricityandnaturalgasconsumptionandretrotprogramparticipationfromGainesvilleRegionalUtilitiesfrom2008to2012.Specically,Ievaluatethesavingseectofthe2009high-eciencyACrebateprogram.First,Iestimatetheenergysavingeectsonannualenergyconsumption.Next,sincethemainaimofDSMorenergyeciencyprogramsistoreducepeakperiodconsumption,Idisaggregatetheannualeectintosummerpeakeect,winterpeakeect,andnon-peakmonthseecttostudythesavingsimpactoftheprogramonpeakperiodenergyconsumption.Theresultsindicatethatwhiletheprogramledtosubstantialenergyconsumptionreductionsinthesummerpeakandnonpeakmonths,winterpeakreductionsarestatisticallyandeconomicallyinsignicant.Theremainderofthischapterisasfollows:Section 1.2 givesabriefbackgroundofGRU'senergyrebateprograms,Section 1.3 describestheempiricalstrategy,Section 1.4 givesabriefdescriptionofthedata,andSection 1.5 investigatesselectionintotreatmentbasedonpre-treatmentcharacteristics.Section 1.6 presentstheresultsoftheprogramonannualenergyconsumptionandpeakperiodconsumptionwhileSection 1.7 concludes. 1.2Background:GainesvilleRegionalUtilitiesRebateProgramsGainesvilleRegionalUtilities(GRU)oersitsconsumersamixofrebatesandincentivestopromoteenergyeciency.GRUoersrebatesforhigh-eciencycentralairconditioners,roomairconditioningunits,heatpumps,waterheaters,insulation,ductsealing,refrigeratorrecycling,poolpumps,installationofsolarwaterheaters,andatticmeasures.GRUalsooersincentivesforacomprehensivewholesystemmeasurethroughitsEnergyStarHomePerformanceProgramandLow-IncomeEnergyEciencyProgram.Inthischapter,Ievaluatetheenergysavingseectofthehigh-eciencycentralairconditionerrebateprogram.Thehigh-eciencycentralairconditionerprogramencourageshomeownerstoreplaceold,low-eciencyHeatingVentilationandAir-Conditioning(HVAC)systemwithanewhigh-eciencyunit.Toqualifyfortherebate,householdsmustuseapartneringFloridastatelicensedHVACmechanicalcontractorinallretrottingwork.In2009about3,226singlefamilyhouseholds(representingabout6%ofallsinglefamilyhomesinGainesville)voluntarily 15

PAGE 16

participatedinatleastoneoftherebateprogramsoeredbyGRU.Participantswereallowedandevenencouragedtoparticipateinmultiplerebateprogramstomaximizetheenergysavings.Table 1-1 liststherelevantnancialincentivesinGRU's2009rebateprograms.5 Table1-1. GRU'sRebateProgramsandIncentives RebateProgramAmountMaximumIncentive HeatPumpWaterHeater$200CentralAC$550HomePerformancewithEnergyStar$775-1400LowIncomeEnergyEciencyProgram$3200Insulation$0.125persquarefoot$375DuctleakRepair50%ofcost$375Poolpumps$250RefrigeratorBuybackandRecycling$50WindowReplacement$1.125persquarefoot$300WindowFilm/SolarScreen$1persquarefoot$100 Source:DatabaseofStateIncentivesforRenewablesandEciency,http://www.dsireusa.org Note:1.OneductleakrepairperHVACsystem,3perlocation 1.3EmpiricalStrategyandMethodThissectionmotivatesandsummarizesourmethod.TheaimistoovercomeproblemsintheestimationofenergysavingsinthepreviousliteratureandalsotoprovideasimplemethodofcontrollingfortheeectsofweatheronEnergyconsumptionwhenthereisnoproxyforhousehold-specicweather.Iuseadierence-in-dierence(DD)strategyincombinationwiththeCoarsenedExactMatching(CEM)methodologydescribedin Iacusetal. ( 2008 ).Lettreatit2f0,1gbeanindicatorofwhetherhouseholdiparticipatedintherebateprogramunderconsiderationinperiodtandletyitbetheenergyconsumptionofhouseholdiinperiodt.Lety1it+sbetheenergyconsumptionofhouseholdi,speriodsafterparticipatingintherebateprogram.Also,lety0it+sbethecounterfactualenergyconsumptionofhouseholdiinperiodt+shaditnotparticipatedintherebateprogram.Thusthegainorenergysavings 5Iprovideinformationforthe2009rebateprogramsinceIspecicallyevaluatethe2009program. 16

PAGE 17

fromparticipatingintherebateprogramforhouseholdiis: i=y1it+s)]TJ /F9 11.955 Tf 11.95 0 Td[(y0it+s.(1{1)Ifwecouldsimultaneouslyobservey1it+sandy0it+sforthesamehousehold,thenprogramevaluationwouldbestraightforward.WecouldestimateiforeveryhouseholdthatparticipatedintherebateprogramandaverageouttondtheAverageTreatmentEectontheTreated(ATT).TheAverageTreatmentEectontheTreatedisdenedintheevaluationliteratureas: E(y1it+s)]TJ /F9 11.955 Tf 11.96 0 Td[(y0it+sjtreatit=1,X)=E(y1it+sjtreatit=1,X))]TJ /F8 11.955 Tf 11.95 0 Td[(E(y0it+sjtreatit=1,X).(1{2)Xisavectorofcontrolvariables.SinceE(y0it+sjtreatit=1,X)isunobserved,weneedtoconstructanapproximationforthisvalue.Thedierence-in-dierenceliteratureusestheoutcomeofacontrolgroupofhouseholdsthatdidnotparticipateintherebateprogram,E(y0it+sjtreatit=0,X),asanapproximationtotheaverageoutcomeofthosewhoparticipatedintherebateprogram.Onefundamentalproblemwiththedierence-in-dierenceapproachisthecreationofacomparisongroupofhouseholdswhointheabsenceoftheprogramwouldhavesimilaroutcomestothosewhoparticipated.Normallyinexperimentalprograms,participationintheprogramisrandomized,andacrediblecomparisongroupisselectedbeforehand.Whentheprogramisvoluntary,thenthosewhoparticipatedintheprogrammaydierfromthosewhodidnotparticipatebasedonthepre-treatmenthouseholdcharacteristics.Thisimbalancebetweenparticipantsandnon-participantscanleadtoselectionbias.Inaddition,iftreatitiscorrelatedwithsomeunobservablecharacteristicsthataecttheprobabilityofparticipationintheprogram,thentheanalysisisplaguedwithendogeneityandsimultaneitybias.Controllingforpre-treatmentvariablesinthedierence-in-dierencestrategydoesnotcompletelyovercometheselectionbiasnortheendogeneitybias.Thereisalsotheproblemofcommonsupport(e.g.,programparticipantsmaybelongtoaparticularsetofneighborhoods. 17

PAGE 18

Includingnon-participantsinotherneighborhoodsoutsidethissetintheestimationleadstoacommonsupportproblemthatmightbiastheresults).Thecommonsupportproblem,particularlywithrespecttoneighborhoods,cangreatlybiastheestimatedeectsofDSMprogramonenergyconsumption.Thisisbecausewecannotaccuratelydisentangletheeectsofweatherfromprogrameectswhentherearejustoneorjustafewweatherstationsintheareaunderstudydespitethefactthatmuchofthevariationinhouseholdenergyconsumptioncanbeexplainedbychangesintheweather( Actonetal. ( 1976 ); PartiandParti ( 1980 ); ReissandWhite ( 2005 , 2003 )).Byincludingparticipantsandnon-participantsfromcompletelydierentneighborhoodsandwithnoproxyforhouse-specicweatherinformation,theestimatedtreatedeectislikelytobebiased.Weathercandierfromonelocationtoanothereveninthesamecity.Theidealwayofcontrollingfortheeectsofweatheronenergyconsumptionistocontrolforhouseholdspecicweather.However,suchinformationisnotavailable.In ReissandWhite ( 2003 )'sstudyoftheSanDiegoservicearea,theauthorsmappedeachhouseholdtooneof21weatherstationsinSanDiegoconsideringbothproximityandelevationandusedtheweatherinformationofthenearestweatherstationasaproxyforhouseholdspecicweather.Whenthereareonlyafewweatherstations,thismethoddoesnotallowforenoughvariationintheweathervariabletoobtainanaccurateestimateoftheeectsofweatheronenergyconsumption.Thealternativetocontrollingfortheeectsofweather,inthiscase,istocompareonlyhouseholdsinthesameneighborhood.Further,bycomparinghouseholdsbasedonneighborhoodsandageofbuilding,weareabletocontrolforbuildingcharacteristicsandhousebuildingmaterialssincehousesbuiltinthesameyearorafewyearsapartandinthesameneighborhoodareusuallybuiltwiththesameconstructionmaterialsandhavesimilarcharacteristics.Inthisstudy,IemploytheCoarsenedExactMatching(CEM)methodologydescribedin Iacusetal. ( 2008 )withadierence-in-dierence(DD)methodologyinordertosolvethecommonsupportproblem,theselectionbiasproblem,andalsocontrolfortheeectsofweather.Thepurposeofmatchingistoconstructanaccuratecontrolgroupwhoseoutcomes 18

PAGE 19

willbeusedasthecounterfactualconsumptionofparticipantsinthetreatmentgroup.Thematchingmethodologypairseachtreatedhouseholdwithagroupofhouseholdsinthecomparisongroupbasedonpre-treatmentcharacteristicssothatthecomparisongroupofhouseholdshavesimilarpre-treatmentcharacteristicsasthetreatedhouseholdswithwhomtheyarepaired.Ispecicallyemploythecoarsenedexactmatchingmethodologyinordertocircumventthecurse-of-dimensionalityprobleminherentinexactmatching(addingonecontinuousvariabletoanexactmatchingmethodologyeectivelykillsthematching,sinceweareunlikelytondtwoobservationswiththesamevalueonacontinuousscale).TheideaofCoarsenedExactMatchingistotemporarilygroupeachvariableintomeaningfulstrataandpairprogramparticipantstonon-participantswhobelongtothesamestrataoneachcoarsenedvariable.6Theoriginal(uncoarsened)variablesare,however,retainedforanalysis.TheCoarseningExactMatchingalgorithmasdescribedin Blackwelletal. ( 2009 )isasfollow: 1. BeginwiththecovariatesXandmakeacopy,whichwedenoteasX. 2. CoarsenXaccordingtouserdenedcutpointsorCEM'sautomaticbinningalgorithm. 3. CreateonestratumperuniqueobservationofX,andplaceeachobservationinastratum. 4. Assignthesestratatotheoriginaldata,X,anddropanyobservationwhosestratumdoesnotcontainatleastonetreatedandonecontrolunit.Ithenperformexactmatchingonthematchedstrata.LetA=fA1,A2,,Akgbeasetofmatchedstratawithcoarsenedexactmatchingmethodology.LetNTandNCbethetotalnumberoftreatedobservationsandcontrolobservationsrespectively.Also,letNTAjandNCAjbethenumberoftreatedandcontrolobservationsinstratumAj.Letyijbethepost-treatment 6Notallvariablesneedtobecoarsened,somevariablescanberestrictedfromcoarsening. 19

PAGE 20

energyconsumptionofhouseholdiinstratumj.AstandardmatchingestimatorfortheAverageTreatmenteectontheTreated(ATT)ofaDSMprogramis: ATT=Xi2NT8>><>>:yij)]TJ /F14 11.955 Tf 17.75 32.63 Td[(Pi2NCAjyij NCAj9>>=>>;.(1{3)ThevalueinparenthesisinEquation 1{3 istheindividualtreatmenteectofaprogramparticipantinstratumAj.SummingandaveragingoveralltreatedparticipantsgivestheaveragetreatmenteectonthetreatedoftheDSMprogram.Equation 1{3 usesonlythepost-treatmentenergyconsumptiontoestimatedtheprogrameects.However,sincewehavepaneldata,wedonotemploytheCoarsenedExactmatchingestimatorinlevels.Weuseadierence-in-dierencecoarsenedexactmatchingestimatoronthematchedobservationineachstratum.Thedierence-in-dierencecoarsenedexactmatchingrelaxesthestrongselection-on-observablesassumptioninherentinmatchingestimators.Combiningadierence-in-dierencemethodologywithamatchingmethodologyhastheadditionaladvantageofeliminatingunobservabletime-invariantdierencesinenergyconsumptionbetweentreatedcontrolhouseholdsthatstandardmatchingestimatorsfailtoeliminate( GirmaandGorg , 2007 ; SmithandTodd , 2005 ).Letyijbethedierenceinenergyconsumptionbetweenthepost-andpre-treatmentperiodsofhouseholdiinstratumAj.Thenthedierence-in-dierencecoarsenedexactestimatorisdenedas: =Xi2NT8>><>>:yij)]TJ /F14 11.955 Tf 17.75 32.63 Td[(Pi2NCAjyij NCAj9>>=>>;.(1{4)IfIhadperformedexactmatching,thentherewouldbenoimbalanceleftandEquation 1{4 perfectlyestimatestheenergysavingseectsofthedemand-sidemanagementprogram.However,sinceIusedcoarsenedvariables,IuseavariantofEquation 1{4 byalsocontrollingfortheactual(uncoarsened)valuesofthevariablesinalinearregression.Thisistheestimateweemployintheanalysisbelow.Istartwithaninitialequationoftheform: 20

PAGE 21

logyit=0+0.d2+1treatit+2treatitXi+1Xi+2d2Xi+i+uit,t=1,2(1{5)whereyitistotalenergyconsumptionforhouseholdiinperiodt.Totalenergyconsumptionforhouseholdiisdenedasthesumofelectricityconsumptionandnaturalgasconsumption.7d2isadummyvariableforthesecondperiod,iistheindividualheterogeneitythatisconstantacrosstime,anduitistheidiosyncraticerrorthatvarieswithtime.Xiisavectorofhouseholdcharacteristics.Allthehouseholdcharacteristicsinoursampledonotvaryacrosstime.However,Iincludeaninteractiontermbetweenthesecharacteristicsandthesecond-perioddummyvariable,d2,sothatthehouseholdcharacteristicswouldhavedierenteectsonenergyconsumptionindierentperiods.8Ialsoallowforthetreatmentvariabletohavevariedeectswithrespecttothehouseholdcharacteristicsbyincludinganinteractiontermbetweenthehouseholdcharacteristicsandthetreatmentdummy.Firstdierencingthetwoequationsacrossthetwotimeperiodsremovestheindividualheterogeneityaswellasallthetimeconstantexplanatoryvariablessothenalequationonwhichweappliedthecoarsenedexactmatchingmethodologyisoftheform: yi=0+1.treati+2treatiXi+2Xi+uit(1{6) 7Naturalgasconsumptionisoriginallymeasuredinthermswhileelectricityismeasuredinkilowatthours.Inordertocombineelectricityconsumptiontonaturalgasconsumption,bothwereconvertedtoequivalentkWh(ekWh)usingtheconversionrate1therm=29.300111ekWhand1kWh=1ekWh.8Anidealwaywouldbetoincludeaninteractionbetweenthehouseholdcharacteristicsandthetotalheatingandcoolingdegreedayssothatthehouseholdcharacteristicshavedierenteectsbasedontheseverityoftheweatherindierentperiods.Whilethereareabout26weatherstationsinandaroundGainesville,theNationalOceanicandAtmosphericAdministrationwebsitehasdailymaximumandminimumtemperatureonlyfortheweatherstationattheGainesvilleAirport.Onlyprecipitation,windand/orwaterinformationisavailableattheotherweatherstations.Hencethetotaldegreedays,usinginformationfromonlyoneweatherstationwouldbeconstantforallhouseholdsinthedataset. 21

PAGE 22

Itshould,however,benotedthatEquation 1{6 isonlyanestimatingequationtogetridoftheindividualheterogeneityandanyotherconstant(time-invariant)unobservablefactorsthataectenergyconsumption.Theestimatesfromtheseequationsshould,therefore,beinterpretedinthecontextoftheoriginallevelequation(Equation 1{5 ).9Theinterceptinthisequationmeasurestheaveragedierenceinconsumptionbetweenthetwotimeperiodsthatcanbeattributedtothedierencesintheseverityoftheweatheroranyotherunobservedtime-variantfactorthatleadstochangesinenergyconsumptionacrossperiods. 1.4DataIusedatafromthreedierentsources:GainesvilleRegionalUtilities(GRU),theAlachuaCountyPropertyAppraiser(ACPA)database,andtheCensusBureau.TheGRUdatasetswereobtainedfromtheProgramforResourceEcientCommunities,andtheycontaintwodistinctdatasets.TherstGRUdatasetgivesthemonthlyelectricityandnaturalgasconsumptionforeachresidentialhouseholdfrom2008to2012.ThesecondGRUdatasetincludesinformationaboutrebateprogramparticipantsthrough2011.IextractedtheACPAdatafromtheACPAwebsite,anditcontainsinformationonthephysicalcharacteristics,location,andsalesdatesofallpropertiesinAlachuaCounty.Ialsogeocodeeachpropertylocationaddressinoursampletolinkeachpropertylocationaddresstoacensustractortoazipcodetowhichtheybelong.Thecensustractsandzipcodesserveasneighborhoodsforeachpropertysothatbymatchingonthecensustractorzipcodeswecancontrolforeectsofweatheronenergyconsumptionwithoutactuallyhavingweatherdata.10Since GamtessaandRyan ( 2007 )founddemographic 9Thus,thecoecientsonthehouseholdcharacteristicsinEquation 1{6 donotmeasuretheeectsofthecharacteristicsonenergyconsumptionortheeectsofthecharacteristicsonthedierencedenergyconsumption,butratherthedierencesintheeectsofthecharacteristicsonenergyconsumptionacrossthetwoperiodsasinterpretedinEquation 1{5 .ForexamplethecoecientonthenumberofbedroomsinEquation 1{6 measuresdierenceintheeectsofbedroomsonenergyconsumptionacrossthetwoyears.10Thebestwayofcontrollingforweatheristoincludehousehold-specicweatherinformation.However,sincehousehold-specicweatherinformationisnotavailable,an 22

PAGE 23

informationsuchasincomeandhouseholdcharacteristicstoplayaroleinthedecisiontoundertakerebateprograms,Iextractedthemeanincomeandmeanhouseholdsizeforeachcensustractandimputedthosevaluestoallhouseholdsinthecensustractorzipcode.Table 1-2 givesthevariablescontainedineachdataset. Table1-2. VariablesfromEachDataset GRUConsumptionGRURebateACPADatasetCensusDatasetDatasetDatabase ParcelNumberParcelNumberParcelNumberCensustractcodesMonthofconsumptionRebatetypePhysicalAddressMeanincomeYearofconsumptionYearinstalledYearbuiltMedianincomeMonthlyconsumptionMonthinstalledNumberofbedroomsAveragehouseholdsizeDaysofconsumptionNumberofbathroomsNumberofStoriesBaseAreasquarefeetTotalAreasquarefeetHeatedAreasquarefeetPrevioussalesdatePoolownership TheGRUconsumptiondataset,theGRUrebatedataset,andtheACPAdatasetwerelinkedtogetherbytheparcelnumberidentierwhichispresentinthethreedatasets.IgeocodethelocationaddressofeachhouseusingArcGISandmapthegeocodedaddressesintooneofthe47censustractsinGainesville.11Thebasedatasetwiththegeocodedaddressescontainsapproximately28000singlefamilyhouseholds.Becausethepurposeofthisresearchisto approximationtocontrollingforweatherinformationistomapeachhouselocationtoathenearestweatherstationandusetheweatherinformationforthatweatherstationasanimputedvalueforthehouse-specicweather( ReissandWhite , 2003 ).Suchapproachisonlypossibleifthereareenoughweatherstationsintheareaunderstudytoallowforvariationintheimputedweatherinformation.Asstatedearlier,theNationalOceanicandAtmosphericAdministrationwebsitewhereIcollectedthetemperatureinformationhastemperatureinformationforonlyoneweatherstationinGainesville.Includingthetemperatureinformationfromonlyoneweatherstationintheanalysiswillberedundant.11ArcGISsuccessfullymapped98%oftheaddressesintotheirrespectivecensustracts.Fortheremaining2%thatwereunsuccessfulorwherethelocationaddressismissing,IsearchedfortheparcelnumberinGoogleEarthtondtheaddressandcensustract. 23

PAGE 24

evaluatetheenergysavingseectsofthehigh-eciencycentralairconditionerrebateprogram,allhouseholdsthatparticipatedinotherrebateprogramsweredroppedfromthedataset.Allhouseholdsthatparticipatedinmultipleprogramswerealsodropped.Theremainingdatasetcontains24794households.Further,householdswhomadehomeimprovementsovertheperiodthatarelikelytoaectsignicantlytheirenergyconsumptionwerealsodropped.Forexample,householdsthataddedapoolorsolarheaterduringtheperiodweredroppedfromthenaldataset.Droppingtheseobservationsmayleadtoanunderestimationoftheenergysavings;forexample,householdsthatparticipatedinmultipleprogramsaremorelikelytobetheoneseagertosaveenergy.Italsomakesourestimatedsavingseectalocaltreatmenteectonthosewhoparticipatedonlyinthehigh-eciencyACprogram.Nonetheless,sincethosewhoparticipatedonlyinoneprogramaremorelikelytohavethesamepre-treatmentcharacteristicsasnon-participants,weareabletoreducethebiasfromconfounding,unobservablecharacteristics.Thenaldatasetfortheevaluationofthe2009high-eciencyACrebateprogramcontained24010households(about40%ofallsinglefamilyresidentialhouseholdsinGRU'sserviceareainGainesville).Energyconsumptionwascalculatedasthesumofelectricityconsumptionandnaturalgasconsumption(convertedtoekWh).12Table 1-3 givesasummarystatisticsofthedata. BillingTimingandStandardizationofMonthlyEnergyConsumptionDierenthouseholdsusuallyhavedierentbillingperiodsbasedonwhentheirutilitymeterisread.Whenahousehold'sbillingmeterisread,theirbillingperiodclosesandanewbillingperiodstarts.Sincethebillingmeterisreadondierentdaysfordierenthouseholds,the\monthly"electricityandnaturalgasconsumptionfordierenthouseholdsinthesame\month"havedierentdatesofconsumption.Onewaytoexploitthebillingmethodinouranalysissothatnobiasesareresultingfromthevariousbillingperiodsistofollow Reissand 12Weconvertednaturalgasconsumption,originallymeasuredintherms,toequivalentkilowatt-hour(ekWh)usingtheconversionrate1therm=29.300111ekwh. 24

PAGE 25

Table1-3. SummaryStatistics VariableNMeanStd.Dev.MinMax Electricity2008(kwh)2401012485.236662.762271.59137633.7NaturalGas2008(ekwh)160358707.534887.22114.65006110128.3TotalEnergy2008(ekwh)2401018300.538976.362507.57164589.2Electricity2009(kwh)2401012583.856712.384120.75184895.7NaturalGas2009(ekwh)160349327.9354946.78929.3001192101.09TotalEnergy2009(ekwh)2401018813.19147.951324221120.9Electricity2010(kwh)2401013256.477686.592197.88191181.8NaturalGas2010(ekwh)1603511963.6911221.91142.1055327575.8TotalEnergy2010(ekwh)2401021246.3713136.05502.44352763.3Electricity2011(kwh)2401012317.827128.0344.81173417NaturalGas2011(ekwh)160788490.7378889.2893.809014323516.3TotalEnergy2011(ekwh)2401018003.5411379.96137.86431242Electricity2012(kwh)2401011609.137115.8471157079.6NaturalGas2012(ekwh)160707176.27410121.627.032026316227.9TotalEnergy2012(ekwh)2401016412.2511622.78102.22336224.8AgeofBuilding2401024.9293610.89751289Bedrooms240103.1591420.643531615Bathroomss240102.0277180.6515361110TotalArea(squarefeet)240102346.1081009.16439920639HeatedArea(squarefeet)240101784.903732.867139910855MeanIncome(dollars)2401073702.1933029.6817087160823MeanHouseholdSize240102.3809010.24176561.343Pool240100.14360680.350697901 White ( 2005 )andgrouphouseholdsintobillingcohorts(agroupofhouseholdswiththesamebillingdatesforallmonthsinayear).Thebillingcohorts(restrictedfromfurthercoarsening)couldbeaddedtothevariablesonwhichthematchingisperformedsothatwecomparetheenergyconsumptionofhouseholdsacrosscohorts.Inthedata,thebillingcohorttowhichahouseholdbelongssometimeschangesacrossyearssothatIamnotabletofollowthesamehouseholdinaparticularbillingcohortfortwoormoreyears.Also,sinceIhaveonlyafewprogramparticipants,comparingacrossbillingcohortswillleadtomorestratawithonlytreatedornon-treatedobservations.Imightloseasignicantpercentageofthealreadylimitedtreatedgroup.TheapproachItookinthispaperistostandardizeenergyconsumptionbycalculatingtheaverageconsumptionforeachcalendarmonth.Weachievethisbydividingeachhousehold'sconsumptioninabillingperiodbythetotalnumberofdaysofconsumption 25

PAGE 26

tondadailyaverageenergyconsumption.Ifconsumptioninacalendarmonthspanstwobillingperiods,thenumberofdaysinthemonththatareineachbillingperiodaremultipliedbytheaveragedailyconsumptionineachperiodandsummedtogethertocalculatetheenergyconsumptionforthemonth.13Forexample,supposeahousehold'sbillingperiodstartsontheseventhofeachmonthsothatthehousehold'selectricitybillfortwoconsecutivebillingperiods7May2010{6June2010and7June2010{6July2010are868kWhand780kWhrespectively.Thereare31daysintherstbillingperiodand30daysinthesecondbillingperiod.Hence,thedailyaverageelectricityconsumptionforthetwobillingperiodsare28kWhand26kWhrespectively.Therstbillingperiodcontains6daysinJunewhilethesecondbillingperiodcontains24daysinJune.Hence,theaveragemonthlyusageforthemonthofJuneis286+2624=792kWh.Asimilarcalculationismadefornaturalgasconsumption.The\monthly"electricityandnaturalgasconsumptionforeachhouseholdarethenaddeduptoobtainthe\monthly"energyconsumptionforthehousehold. 1.5Self-SelectionBasedonIndividualPre-TreatmentCharacteristicsforTheTreatmentandControlGroupsAsstatedabove,oneconcernwithusingonlyadierence-in-dierencemethodologyintheestimationofthetreatmenteectofavoluntaryprogramisthebiasfrompre-treatmentcharacteristics.Participantsoftheprogrammaybethosemorelikelytosaveenergyfromparticipating.Forexample,sincenewerhomesarealreadymoreecientandhavemorestringentbuildingcodesthanolderhomes14,weexpectolderhomestosavemoreenergyfromretrottingprogramsthannewerhomes.Therearealso\halo"eectswherepeopleparticipate 13Thismethodwasparticularlyusefulwhenestimatingthesummerpeakandwinterpeakeects.IthankNickTayloroftheProgramforResourceEcientCommunitiesattheUniversityofFloridaforprovidinguswiththealreadystandardizeddata.14Floridaincreasedthestringencyofitsenergycodein2002whichisexpectedtomakehousesmoreenergyecient. 26

PAGE 27

inaprogramiftheirneighborsarealreadyinvolvedintheprogram.Sucheectsconcentrateprogramparticipantsinafewneighborhoodssothatincludingcontrolobservationsfromareaswithnoorfewerprogramparticipantsmaybiastheresultsoftheestimation.Inthissection,weexaminetheextenttowhichprogramparticipantsandnon-participantsdierregardingtheirpre-treatmentcharacteristicsandpre-treatmentenergyusage. Pre-TreatmentUsagePattern Figure1-1. PercentageofProgramParticipantsandNon-ParticipantswithinEachEnergyConsumptionQuartile Weexpectthathighenergyconsumerswillbetheonesmorewillingtosaveenergybyswitchingtomoreenergyecientappliances.ThisisthecaseinFigure 1-1 .Thegureshowsthepercentageofprogramparticipantsandnon-programparticipantsineachenergyconsumptionquartile.Itshowsthatmajorityofthetreatmentgroup(about40%)areinthefourthquartile.Thefourthquartilecontainsabouttwicethenumberoftreatedobservationastheotherquartiles.Thus,consumerswithhighenergyusagearemorelikelytoparticipateintherebateprogramthanconsumerswithlowusage.Thegurealsoshowsthatthenumberoftreatedobservationsisnondecreasing(orincreasing)aswemovefromthelowerquartiletothehigherquartile. 27

PAGE 28

Self-SelectionBasedonAgeofTheBuildingRecenthousesaremoreenergyecientandnormallycontainsmoreenergyecientappliancesthanolderhomes.Achangeinanapplianceinanolderhomeisthusexpectedtohavehighersavingseectsontotalenergysavings(sinceolderACunitsconsumemoreenergythanmodernACunits).Weexpectthathouseholdsinolderbuildingaremorelikelytoparticipateinenergyeciencyprogramstotakeadvantageofthehighenergysavingsthanhouseholdsinnewerhomes.In2002,Floridaincreasedthestringencyofitsenergycodestomakebuildingsmoreenergyecient.Thisincreaseinstringencyisassociatedwithadecreaseinelectricityconsumptionby4%andnaturalgasby6%( JacobsenandKotchen , 2013 ).Thus,sincenewerbuildingsarealreadyenergyecient,thereislessroomforimprovementinenergysavingsfromrebateprograms.Furthernewerbuildingsarelikelytobeinstalledwithahigh-eciencyACunit,soreplacingthealreadyinstalledACwithaslightlymoreecientACwillsaveonlyafractionoftheenergysavingsexpectedfromreplacingaveryoldACinanoldbuilding.InFigure 1-2 ,wedividedtheageofthebuildingintoageofbuildingquartiles.Eachquartilecontainsabout6000houses.Therstquartilecontainsbuildingagedlessthan16years,thesecondcontainsbuildingsbetweentheagesof16and28years,thethirdcontainsbuildingsagedbetween28and33yearswhilethefourthquartileismadeupofbuildingsagedmorethan33years.Thegureshowshouseholdsinnewerhouses(lessthan16years)arelesslikelytoparticipateintherebaterebateasexpected.Householdsinoldhousesover33yearsarealsolesslikelytoparticipateintheprogram.Perhapsthesehouseholdshadparticipatedinasimilarrebateprograminthepastsothatthereislowerexpectedenergysavingsfromparticipatinginthesameprogramagain.Programparticipationisratherhighamonghouseholdsinhouseswithanageofbuildingunder28years(rstandsecond).Thesehousesmakeupabout63%ofalltreatedhouseholdsinthedata.Only13.7%ofthetreatedhouseholdsareinthefourthquartile. 28

PAGE 29

Figure1-2. DistributionofAgeofBuildingforParticipantsandNon-Participants Self-SelectionBasedonSizeofTheBuildingBiggerhomesusemoreenergythansmallerhomes.Ahome'sheatingorcoolingareasquarefootagedeterminestheamountofenergythehouseholdwilluseonairconditioningorheating.Thus,weexpecthouseholdsinbiggerhousestoparticipatemoreintheACrebateprogramtoreducetheirenergyusagethanthoseinsmallerhouses.Figure 1-3 showsthedistributionofthecontrolandtreatmentgroupsamongquartilesoftheheatedareasquarefootageofthehouse.Inthegure,biggerhousesasdeterminedbythesizeoftheheated/coolingareaparticipatedmoreintheACrebateprogramthansmallerhouses.Housesweregroupedintoquartilesbasedonthesizeoftheheatedareasquarefootage.Fromthegure,alargepercentageofthetreatedhouseholds(42%)liesinthefourthquartilewhilethesecondandthirdquartileeachhasabout22%.Therstquartilecontainstheleastnumberoftreatedobservations(13%).Thegure,therefore,supportstheargumentthathouseholdsinbiggerhousesaretheonesmoreeagertoreduceenergysincetheyconsumemore. 29

PAGE 30

Figure1-3. DistributionofHeatedAreaSquareFootageofBuildingforParticipantsandNon-Participants 1.6ResultsInthissection,wepresenttheresultsofthedierence-in-dierencecoarsenedexactmatchingmethodology.Wematchedonneighborhoods,theageofthebuilding,pre-treatmentenergyconsumption,numberofbathrooms,numberofbathrooms,thenumberofstories,heatedareasquarefootage,andtypeofheatingfuel.Weassumedthatpre-treatmentconsumptioncanbeexplainedbybuildingcharacteristics.Thus,bymatchingonpre-treatmentconsumptionweareineectmatchingonasinglevariablethataggregatestheeectsofallthebuildingcharacteristicsonenergyconsumption.15.Wearemainlyinterestedinthematchingonneighborhoodstocontrolfortheeectofweatherandtheageofbuildingto 15Wedidn'tincludeallotherbuildingcharacteristicsinthematchingsoastoreducethenumberofstrataandincreasethenumberoftreatedobservationsineachstratum. 30

PAGE 31

controlfortheeectsofbuildingcharacteristics.Weusetwoneighborhoodvariablesseparatelyinthematchingmethodology:zipcodesandcensustracts.Zipcodesandcensustractswererestrictedfromfurthercoarseningsothatnotwoneighborhoodscanbeinthesamestratum.Thatis,wecompareonlyhouseholdsinthesamecensustractorzipcodes.Thealgorithmautomaticallyimposesthecommonsupportcondition,soallobservationswithinanystratumthatdoesnothaveatleastoneobservationforeachuniquevalueofthetreatmentvariablearediscarded.Thenumberofbathroomswasrecodedwithone-halfbathroomscountingasfullbathroomsinthematchingmethodology.Heatedareasquarefootagewascoarsenedintotenequalgroupswhereaspre-treatmentenergyconsumptionwasdividedbythe1st,5th,10th,25th,50th,75th,90th,95th,and99thpercentilesofthedistributionofpretreatmentconsumption. AnnualEectsofTheHigh-EciencyACRebateProgramTable 1-4 showsthetotalnumberofobservations,andthenumberoftreatedanduntreatedobservationintheoriginaldatasetbeforeandafterthecoarsenedexactmatching.Thetabledisplaysasummaryoftheresultofthematchingmethodologywithzipcodesandcensustractsasneighborhoods.Matchingwithcensustractsasneighborhoodsledto135matchedstratawith1413controlobservationand139treatedobservation.Thematchingwithzipcodesproduced159matchedstratawith7650controlobservationsand193treatedobservations.Whilematchingonathecensustractsisexpectedtoprovideanaccuratematch,sincecensustracksaresmaller,itproducesfewermatchesthanmatchingonzipcodesandwelosealargernumberofthealreadyfewtreatedobservations.Thematchingonthezipcodesasneighborhoods,ontheotherhand,producedahighmatchsincezipcodesarebiggerthancensustracts,buthouseswithinmatchedcellorthedemographicsofthehouseholdswithinthematchedcellsaremorelikelytobeverydierence.Theymayalsoexperiencedierentweatherconditionswhichmightbiastheresults.Thus,whilesmallerneighborhoodcategoriesproducefewermatchesthanbiggerneighborhoodcategories,resultsusingsmallerneighborhoodcategoriesarelikelytomemoreaccuratethanmatchingonbiggerneighborhoods. 31

PAGE 32

Table1-4. MatchingSummary{High-EciencyACRebateProgram CensusTractsZipCodesNumberofstrata:9771Numberofstrata:3653Numberofmatchedstrata:135Numberofmatchedstrata:159 ControlTreated controlTreated All23785225 All23785225Matched1413139 Matched7650193Unmatched2237286 Unmatched1613532 TheresultsincolumnIandIVofTable 1-5 (andColumnIofTable A-1 intheappendix)showthatthehigh-eciencyACrebateprogramledtostatisticallysignicantenergysavingsunderallthreeregressions.Thedierence-in-dierencewithoutmatchingshowsaverageenergysavingsofabout8.5%pertreatedhousehold.TheDDCEMwithcensustractsasneighborhoodshowsarelativelyhighersavingsof9.5%pertreatedhousehold.16Table A-1 intheappendix,whichuseszipcodesasneighborhoods,alsoshowsthattheACrebateprogramledtosavingsof8.6%peryear.Thusevenwhenmatchingonafewvariablesbecauseofdataunavailability,itcanbeseenthattheregularDDmarginallyunderstates(oroverstates)theeectofenergysavingsmainlybecauseitfailstoaccountforthedierencesinweather.Theresultsalsosuggestthatusingsmallerneighborhoods(inthiscaseusingcensustracts)reducesbiasandimprovesthetreatmenteect.IncolumnsII,III,V,andVI,weallowedtheeectsoftreatmenttovarybythesizeoftheheated/coolingareasquarefootageofthehouseand/ortheageofthebuilding.Theageofahousedoesnotsignicantlyaectthesavingseect(usingbothDDCEMandtheregularDD).Thesizeoftheheatedareasquarefootage,ontheotherhand,reduces 16MatchingonarenedneighborhoodvariablesuchasactualneighborhoodsorsubdivisionsusedbytheAlachuaCountypropertyappraisermightfurtherreducebiasandleadtoamoreaccurateestimateofthetreatmenteect.Thisisbecausehousesinthesamesubdivisionsaretypicallybuiltinthesameyear,areusuallyconstructedwiththesameconstructionmaterial,andhavesimilarcharacteristics.Housesinasmallerneighborhoodcategoryarealsolikelytohavesimilarweatherconditionsthanhousesfarapart,hence,wecanaccuratelydisentangleweathereectsfromprogrameects. 32

PAGE 33

theenergysavingsfromtheprogram(undertheDDCEMmethodology).Itis,however,notstatisticallysignicantusingtheregularDD.FromColumnII,thecoecientonTreatsigniesthatatreatedhouseholdwithzeroheatedareasquarefootagereducesenergyusageby23.1%onaverage.17Moreimportantly,aone-standarddeviationincreaseintheheatedareasquarefootage(about733squarefeet)reducestheenergysavingsfromtheprogramby5.4percentagepoints.Theenergysavingseectsoftheprogram,therefore,becomesnon-existentforahousewithaheatedareasquarefootageofabout3080squarefeet(aboutthe92ndpercentileofthedistributionofheatedareainthesample).Thisnegativeeectoftheheatedareaontheenergysavingsslightlyincreaseswhentreatmentisfurtherallowedtovarywiththeageofthebuilding(thecoecientonTreat*HeatedAreaincreasedfrom0.075inColumnIIto0.077inColumnIII),buttheeectsofageontheenergysavingsisstatisticallyinsignicant.Infact,thecoecientonTreat*Ageiszerotothreedecimalplaces.TheresultswereunexpectedsincehouseholdsinbiggerhousesaremorelikelytoparticipateintheprogramthanthoseinsmallerhousesasseeninFigure 1-2 .IncolumnsVandVI,bothheatedareaandageofbuildinghasnostatisticallysignicanteectonthesizeofthetreatmenteectusingtheregularDD.Thus,theregularDDwithoutmatchingagainfailstohighlightthedierenttreatmenteectsbasedonthesizeoftheheatedarea.Sincewematchedononlyafewvariableswithoutanyhouseholddemographics,muchofthedierencebetweentheDDestimateandtheDDCEMestimatescanbeattributedtothedierencesinhowwellthetwomethodologiesaccuratelydisentangleweathereectsfromprogrameects.Asstatedearlier,thecoecientsonthehouseholdandhousingcharacteristicsintheregressionmeasurethedierencesintheimpactofthesecharacteristicsonenergyconsumptionbetweenthetwoyears.Thedierencesintheeectsofthesescharacteristicsonenergyconsumptionaremainlyattributedtothedierencesinweatherbetweentheyears.For 17Thisgureisnotinterestingbyitselfsincethereisnohousewithazeroheatedareasquarefootageinthesample. 33

PAGE 34

example,householdsinbiggerhousesareexpectedtoincreasetheirenergyconsumptionfarmorethanthoseinsmallerhousesduringayearofsevereweatherconditions.Apartfromtheelectricandgasdummythatissignicantinallthreeregressions,thecoecientsontheothervariablesarenotsignicantintheDDCEMregression.ThisismainlybecausetherewasonlyaslightimbalanceinamatchedstratausingtheDDCEM.TheregularDD,however,hasalotofvariableswithsignicanteects.Theageofthebuilding,heatedarea,poolownership,electricandgasdummy,andmeanincomeareallstatisticallysignicantintheregularDD.Coecientsonthenumberofbedrooms,numberofbathrooms,numberofstories,meanhouseholdsizeandhottubownershipare,however,notstatisticallysignicant.ThelackofstatisticalsignicanceofthesehousecharacteristicsintheregularDDregressionandtheDDCEMsuggeststhattheweatherinthetwotimeperiodswasnotstatisticallysignicantlydierentfromeachothertoallowthesecharacteristictohavedierenteectsonenergyconsumption.Ialsoevaluatethe2010ACrebateprogramtochecktherobustnessof2009programresults.The2009ACrebateparticipantsweredeletedfromthedatasetsoasnottobiastheresultsoftheestimation.The2011participantswerealsodeleted.Ourcontrolgroupthusconsistofhouseholdswhofrom2008to2011neverparticipatedinanyrebateprogram.TheresultsoftheestimationarepresentedinTables A-2 and A-3 intheappendix.Theresultsofthe2010programaresimilartothatofthe2009program.Theprogramledtoastatisticallysignicantenergysavingsof8.1%usingtheDDCEMwithcensustractasneighborhoods,9.8%usingtheregularDDapproach,and8.9%usingtheDDCEMwithzipcodesasneighborhoods.Themaindierenceinthetreatmenteectofthe2010programwiththe2009programisthat,whileheatedareasquarefootagehadanadverseeectonthesizeoftheenergysavingsinthe2009program,theeectsoftheheatedareaontheenergysavingsinthe2010programisstatisticallyinsignicant.Rather,theageofbuildingthathadastatisticallyinsignicanteectontheenergysavingsofthe2009programhasastatisticallysignicanteectonthe2010programintheregularDDmethodology(thecoecienton 34

PAGE 35

Table1-5. AnnualEnergySavingsEectofThe2009High-EciencyACRebateProgram log(EnergyUsage) CEMDD(CensusTract)RegularDD (I)(II)(III) (IV)(V)(VI) Treat -0.0946***-0.2314***-0.2424** -0.0852***-0.0983**-0.0494 (-4.11)(-3.40)(-2.64) (-5.77)(-2.77)(-0.97)Treat*HeatedArea 0.0748*0.0766** 0.00600.0036(1000squarefeet) (2.52)(2.68) (0.50)(0.30)Treat*Age 0.0003 -0.0019 (0.13) (-1.14)Bedrooms 0.04770.04750.0475 -0.0066-0.0066-0.0066 (1.76)(1.75)(1.75) (-1.55)(-1.53)(-1.53)Stories 0.01750.01800.0179 -0.0073-0.0072-0.0072 (0.76)(0.78)(0.78) (-1.42)(-1.42)(-1.42)HeatedArea -0.0475-0.0546-0.0547 0.0143***0.0141***0.0142***(1000squarefeet) (-1.63)(-1.84)(-1.85) (3.43)(3.34)(3.35)Age 0.00250.00250.0025 0.0005*0.0005*0.0005* (1.84)(1.83)(1.71) (2.23)(2.22)(2.27)Pool -0.0300-0.0296-0.0297 -0.0437***-0.0437***-0.0436*** (-0.82)(-0.81)(-0.81) (-7.51)(-7.50)(-7.49)ElectricandGas 0.1014**0.1013**0.1013** 0.0865***0.0866***0.0865*** (3.25)(3.25)(3.25) (18.49)(18.49)(18.47)MeanIncome($1000) 0.00030.00030.0003 -0.0002**-0.0002**-0.0002** (0.54)(0.54)(0.54) (-3.10)(-3.09)(-3.09)MeanHouseholdSize 0.05840.05920.0592 0.00300.00300.0029 (1.00)(1.01)(1.01) (0.33)(0.32)(0.32)HotTub -0.0375-0.0347-0.0347 -0.0110-0.0110-0.0111 (-0.88)(-0.82)(-0.82) (-1.19)(-1.20)(-1.21)cons -0.2174-0.2063-0.2055 0.0892***0.0895***0.0892*** (-1.44)(-1.36)(-1.35) (3.36)(3.36)(3.36) N 155215521552 240102401024010 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 35

PAGE 36

Treat*AgeinColumnVIofTable A-2 ).Thiseectishowevernoteconomicallysignicant.A100-yearchangeintheageofthebuildingwillreducethetreatmenteectby0.46percentagepoints. PeakandNon-peakMonthEectThemainreasonfordemandmanagementprogramsisto\allowautilitytocontrolthebalanceofitsresourcesanddemandsforenergybymanagingtheconsumers'needsforenergyratherthanbysimplyaddingmoresupply"( FelsandKeating , 1993 ).Sinceutilitiesoperateundercapacityduringnon-peakperiods,andthereisnoneedtoworryaboutaddingmoresupplyorbuyingpoweratthemarketrate,Utilitiesareparticularlyinterestedinhowdemand-sidemanagementprogramsaectpeak-perioddemands.Floridahastwopeakperiods:thesummerpeakwhichstartsinmid-mayandendsinSeptemberandthewinterpeakwhichbeginsinDecemberandendsinFebruary.AhighpercentageofFlorida'senergyisconsumedinthosetwopeakperiods.Inthispartofouranalysis,weallowtheeectsofthedemand-sidemanagementprogramtodierbyseason:summerpeak,winterpeakandnon-peaktoevaluatetheimpactoftheprogramonpeakperiodconsumption.December,January,andFebruarywereconsideredasthewintermonths.June,July,August,andSeptemberwereconsideredassummermonths.Theremainingmonthswereconsideredasnon-peak.Thisclassicationisbasedonthehistoricaldistributionofcoolingandheatingdegreedaysinnorth-centralFloridaintheliterature. Non-peakmonthsThenon-peakmonthsinFloridacomprisetheSpringmonthsofMarch,April,andMayandtheFallmonthsofOctoberandNovember.Thesemonthsarenormallythemostpleasantmonthsinthestateintermsofthecomfortindex,thesumofheatingandcoolingdegreedays.Householdsrequirelessenergyforheatingorcoolingthanintheothermonths.Airconditionersareusuallyusedonlyminimallyduringthisperiod.We,therefore,expectlowerenergysavinginthenon-peakmonthsfortheACrebateprogram.Theresultsoftheenergysavingseectoftheprograminthenon-peakmonthsarepresentedinTable 1-6 .Thetable 36

PAGE 37

showsthattheACrebateprogramledtoastatisticallysignicantenergyreductionofabout4.8%inthenon-peakmonthsusingboththeDDCEMwithcensustractsasneighborhoodsandtheregularDD(ColumnsIandIII).The4.8%energysavingsinthisperiodishowevernotsurprisingsincethereisabout58%percentofdaysinthisperiodinGainesvillewherethedailymaximumtemperatureisabove800F,andthushouseholdsrequiretheACtocooltheirhomes.18InregularDDregression,ageandheatedareasquarefootagedonotaectthesizeoftheenergysavings.However,intheDDCEMregressionwithcensustracksasneighborhoods,heatedareahasastatisticallysignicantlynegativeeectonthemagnitudeoftheenergysavingsoncetheageofthebuildingisalsoallowedtovarywiththetreatmenteect(coecientofTreat*HeatedAreainColumnIII)ofTable 1-6 ).Theeectisnoteconomicallysignicantasaonestandarddeviationincreaseintheheatedareasquarefootageofahouse(anincreaseofabout733squarefeet)reducestheenergysavingsbyonly0.047percentagepoints.TheresultsusingDDCEMwithzipcodesastheneighborhoodsinshowninTable A-4 intheappendix.TheresultspresentedinTable A-4 aresimilartotheresultsusingtheregularDD. SummerpeakFlorida'shot,humidsummerbeginsinmid-Mayorlaterwithaveragemaximumtemperaturesreachingabout900Fduringthedaybutwithhighhumidity,therealfeelofthetemperatureisabout1080F.Nighttimeprovideslittlerelieffromtheheatasthetemperaturereducesbyjustalittleduringthenight,withnighttimetemperaturesstillabout760Fandwitharealfeelofabout870F.Airconditioningthusbecomesthemaindriverofenergyusageandcostduringthesummer.\WhencombinedpeakmonthlydemandforthemonthsofJune,July,andAugust(thehottestmonths)iscomparedtothatofthecombinedmonthsofDecember,JanuaryandFebruary(thecoolestmonths),exceptforahandfulofpowercompanies,the 18Thereisalsoabout83%ofdaysinwhichthedailyminimumtemperatureisbelow650FinGainesvillesothathouseholdswouldrequiretheirACforheating. 37

PAGE 38

Table1-6. Non-PeakMonthsEectsOfThe2009ACRebateProgram log(EnergyUsage) DDCEM(CensusTract)RegularDD (I)(II)(III) (IV)(V)(VI) Treat -0.0488*-0.1439*-0.2196* -0.0487***-0.0524-0.0456 (-2.15)(-2.15)(-2.16) (-3.36)(-1.53)(-0.82)Treat*HeatedArea 0.05200.0646* 0.00170.0013 (1.72)(2.11) (0.14)(0.12)Treat*Age 0.0023 -0.0003 (0.83) (-0.15)Bedrooms 0.04330.04320.0432 -0.0050-0.0050-0.0050 (1.47)(1.47)(1.47) (-1.06)(-1.06)(-1.06)Stories 0.02370.02400.0238 -0.0056-0.0056-0.0056 (0.96)(0.98)(0.97) (-1.04)(-1.04)(-1.04)HeatedArea -0.0401-0.0451-0.0459 0.0109*0.0109*0.0109*(1000squarefeet) (-1.42)(-1.55)(-1.58) (2.40)(2.35)(2.35)Age 0.00160.00160.0014 -0.0004-0.0004-0.0004 (1.21)(1.20)(1.01) (-1.83)(-1.83)(-1.82)Pool -0.0639*-0.0636*-0.0644* -0.0340***-0.0340***-0.0339*** (-2.04)(-2.03)(-2.05) (-5.36)(-5.35)(-5.35)ElectricandGas 0.0729*0.0729*0.0727* 0.0528***0.0528***0.0528*** (2.43)(2.42)(2.42) (10.33)(10.33)(10.33)MeanIncome($1000) 0.0015**0.0015**0.0015** 0.0005***0.0005***0.0005*** (2.88)(2.88)(2.88) (6.40)(6.40)(6.40)MeanHouseholdSize 0.05740.05800.0583 0.0255*0.0255*0.0255* (0.94)(0.94)(0.95) (2.50)(2.49)(2.49)HotTub -0.0182-0.0163-0.0163 -0.0164-0.0165-0.0165 (-0.43)(-0.38)(-0.38) (-1.63)(-1.63)(-1.63)cons -0.4047*-0.3969*-0.3914* -0.1076***-0.1075***-0.1075*** (-2.47)(-2.41)(-2.38) (-3.68)(-3.67)(-3.67) N 155215521552 240102401024010 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 38

PAGE 39

demandforelectricityduringthethreehottestmonthsisabout20%higherthanforthethreecoolestones.Airconditioningaccountsformostofthatdierence"( WinsbergandSimmons , 2009 ).GRUgeneratesabout30%moreelectricitypermonthinthesummermonthsthaninthewintermonthsand40%moreelectricitypermonththaninthenon-peakmonths.TheresultsofthesummerpeakeectsoftheprogramareshowninTable 1-7 .AsexpectedtheACrebateprogramledtosubstantialenergysavingsabout20%inthesummermonthsusingtheDDCEMwithcensustractsasneighborhoods(Column(I)ofTable 1-7 ).TheregularDDwithoutmatchingalsoproducedastatisticallysignicantbutrelativelylowerenergysavingsestimateof16.77%(ColumnIVofTable 1-7 ).Thesehigherestimatesinthesummermonthsareexpectedsinceairconditioningaccountsforagreaterpercentageoftheenergyusageduringthesummermonths.InColumnsII,III,V,andVI,weallowedthesummerpeakeectstovarybythesizeofthehouseasmeasuredbytheheatedareasquarefootage(ColumnsIIandVI)andbyheatedareaandageofthebuilding(ColumnsIIIandVI).Aswasthecaseinthenon-peakmonths,theageofthebuildinghasnosignicanteectsontheenergysavings.ThecoecientofTreat*AgeinboththeregularDDandtheDDCEMregressions(ColumnsIIIandVI)isstatisticallyandeconomicallyinsignicant.Heatedareasquarefootage,however,hasastatisticallysignicantinverseeectontheenergysavingsinthesummer.IncolumnII,anincreaseintheheatedareabyonestandarddeviation(732.866squarefeet)reducesthetreatmenteectby7percentagepoints.IntheregularDDregression(ColumnV),aonestandarddeviationincreaseintheheatedareasquarefootagereducesthetreatmenteectby2.2percentagepoints.Thus,bothregressionsshowthathouseholdsinsmallerhouseshadahigherenergysavingsratethanthoseinbiggerhouses.Nonetheless,householdsinlargerhousesaremorelikelytohaveagreaterimpactontheirbillthanhouseholdsinsmallerhouses. WinterpeakFlorida'srelativelylowerenergypeakperiodisinthewintermonthswhenthemaximumtemperaturesaveragesabout700F.ThewinterinFloridaorGainesvilletobespecicisnot 39

PAGE 40

Table1-7. SummerPeakEectsofThe2009ACrebateprogram log(EnergyUsage) DDCEM(CensusTract)RegularDD (I)(II)(III) (IV)(V)(VI) Treat -0.2063***-0.3829***-0.4227*** -0.1677***-0.2362***-0.2447*** (-6.77)(-4.18)(-3.36) (-8.68)(-5.38)(-3.83)Treat*HeatedArea 0.0966*0.1032* 0.0310*0.0315* (2.17)(2.25) (2.01)(2.05)Treat*Age 0.0012 0.0003 (0.36) (0.16)Bedrooms 0.04900.04870.0487 -0.0020-0.0017-0.0017 (1.45)(1.44)(1.44) (-0.35)(-0.31)(-0.31)Stories 0.02030.02090.0208 0.00020.00040.0004 (0.69)(0.71)(0.71) (0.04)(0.06)(0.06)HeatedArea -0.0565-0.0656-0.0660 0.00210.00120.0012(1000squarefeet) (-1.36)(-1.57)(-1.57) (0.38)(0.21)(0.21)Age 0.00400.00400.0039 0.0000-0.0000-0.0000 (1.95)(1.95)(1.77) (0.02)(-0.01)(-0.01)Pool 0.02000.02050.0201 -0.0271***-0.0269***-0.0269*** (0.42)(0.44)(0.43) (-3.79)(-3.76)(-3.76)ElectricandGas 0.01850.01840.0183 0.0197**0.0197**0.0197** (0.43)(0.43)(0.43) (3.16)(3.16)(3.16)MeanIncome -0.0000-0.0000-0.0000 -0.0003**-0.0003**-0.0003**($1000) (-0.04)(-0.04)(-0.04) (-2.66)(-2.62)(-2.62)MeanHouseholdSize 0.04520.04620.0464 0.00070.00040.0004 (0.75)(0.76)(0.77) (0.06)(0.03)(0.03)HotTub -0.1084*-0.1049*-0.1049* -0.0140-0.0142-0.0142 (-2.25)(-2.21)(-2.21) (-1.30)(-1.32)(-1.32)cons -0.2178-0.2034-0.2005 0.06550.06680.0669 (-1.38)(-1.28)(-1.26) (1.87)(1.91)(1.91) N 155215521552 240082400824008 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 40

PAGE 41

assevereandrequireslessenergyforheatingthaninthesummerforcooling.ThepercentageofdaysfromDecemberthroughFebruarythatthedailymaximumtemperatureexceeded800Fisabout70%inGainesvilleand50%inotherpartsofcentralFlorida.MostFloridiansusestheirACsystemtoheattheirhomesinthewinter.Therefore,theACrebateprogramisexpectedtohaveasignicantlyhighereectsinthewintermonthsthaninthenon-peakmonths.TheresultsoftheimpactoftherebateprogramonenergyconsumptioninthewinterpeakarepresentedinTable 1-8 .Whiletheimpactonwinter-peakconsumptionisstillnegativeasexpected,themagnitudeissmallcomparedeventotheeectsinnon-peakmonths.ItisagainstatisticallyinsignicantusingtheDDCEMwithcensustractasneighborhoodsandsignicantatthe5%levelusingtheregularDD.19Therelativelysmallmagnitudesandthestatisticalinsignicanceoftheestimatedeectsaresurprisingasweexpectthehigh-eciencyACtohavealargeimpactinthewintermonthsparticularlybecausemostFloridiansdependontheirairconditionertoheattheirhomesduringwinter.About53%ofhouseholdsinoursampleuseselectricityastheirprimaryheatingfuel.Nevertheless,theresultlendssupporttotheobservationthatthetemperaturesinFlorida'swinteraresucientlymildtorequiremuchlessenergyforheatingthaninthesummerforcooling( WinsbergandSimmons , 2009 ).OnereasonforthenoorlittlewinterpeakeectmaybethatthereisfuelswitchinginthewintermonthssothathouseholdsinGainesvilleusegasforheatinginthewinterandelectricityforcoolinginthesummer.BecausetheACrebateprogramisexpectedtohaveahigheectonelectricityconsumptionthanongasconsumption,thissubstitutionofgasforelectricityinthewintermonthsreduceselectricityconsumptionandthusthereislittleroomforenergysavingscomparedwithifhouseholdscontinuedtousetheirelectricityforheating.Figure 1-4 comparesthewinter,summerandnon-peakmonthselectricityconsumptionofhouseholdswhouseelectricityonlytohouseholdswhousebothelectricityandnaturalgas.The 19AsimilarresultsinobtainedwiththeDDCEMusingzipcodesasneighborhoodsinTable A-6 intheappendix. 41

PAGE 42

barchartsshowthatwhileelectricityconsumptioninthesummerbetweenthetwohouseholdsgroupsissimilar,thereisalargedierenceinelectricityconsumptionbetweenthetwogroupsinthewintermonths.Thereisalsoasignicantdierenceintheelectricityconsumptionbetweenthehouseholdsinthenon-peakmonthsbutasexpected,thisdierenceislowerthanthedierenceintheelectricityconsumptioninthewintermonths.Tofurthertestthefuel Figure1-4. AverageElectricityConsumptioninTheSummer,Winter,andNon-peakMonthsbyHouseholdEnergyComposition switchingproposition,wecomparedtheelectricityandnaturalgasconsumptioninthesummer,winter,andnon-peakmonthsforhouseholdsthatusebothelectricityandnaturalgas.ThisisshowninFigure 1-5 .Thegureshowsthatwhileagreaterproportionofthehouseholdsenergyusagecomesfromelectricityinthesummer,naturalgasbecomesmainfuelintheenergymixinthewinter.Thereisstillrelativelyhighelectricityusageinthenon-peakmonths.Thisfurtherlendsmoresupporttotheresultsthatthehigh-eciencyACreducedenergyconsumptioninthesummerandnon-peakmonthsbuthasnostatisticallysignicanteectonwinterpeakconsumption.Whilethehigh-eciencyACprogramisexpectedtohaveahigheectonelectricityusagethanonnaturalgasusage,households,especiallythosewithbothelectricityandnaturalgasreducestheirconsumptionofelectricityandincreasestheirnaturalgasusagesothattheenergysavingsonthe"little"energyconsumptionisalmostnegligible. 42

PAGE 43

Table1-8. WinterPeakEectsofThe2009ACRebateProgram log(EnergyUsage) DDCEM(CensusTract)RegularDD (I)(II)(III) (IV)(V)(VI) Treat -0.0404-0.1599-0.1118 -0.0403*-0.02830.0605 (-1.44)(-1.92)(-0.93) (-2.26)(-0.71)(0.99)Bedrooms 0.05560.05540.0554 -0.0043-0.0043-0.0043 (1.78)(1.77)(1.77) (-0.86)(-0.87)(-0.87)Stories 0.00810.00850.0086 -0.0159**-0.0159**-0.0159** (0.24)(0.25)(0.25) (-2.70)(-2.71)(-2.70)HeatedArea -0.0453-0.0515-0.0509 0.0171***0.0173***0.0174***(1000squarefeet) (-1.19)(-1.31)(-1.30) (3.56)(3.54)(3.56)Age 0.00150.00150.0016 0.00040.00040.0004 (0.90)(0.90)(0.92) (1.62)(1.63)(1.70)Pool -0.0530-0.0526-0.0521 -0.0532***-0.0532***-0.0532*** (-1.15)(-1.15)(-1.13) (-8.15)(-8.14)(-8.13)ElectricandGas 0.1567***0.1567***0.1568*** 0.1375***0.1375***0.1374*** (4.07)(4.08)(4.08) (25.24)(25.24)(25.21)MeanIncome -0.0010-0.0010-0.0010 -0.0009***-0.0009***-0.0009***($1000) (-1.63)(-1.63)(-1.63) (-10.08)(-10.08)(-10.09)MeanHouseholdSize 0.09790.09860.0984 -0.0235*-0.0235*-0.0236* (1.15)(1.16)(1.15) (-2.16)(-2.16)(-2.16)HotTub -0.0048-0.0024-0.0024 -0.0017-0.0016-0.0018 (-0.07)(-0.03)(-0.04) (-0.16)(-0.15)(-0.17)Treat*HeatedArea 0.06530.0573 -0.0055-0.0098 (1.72)(1.47) (-0.39)(-0.72)treatage -0.0015 -0.0034 (-0.48) (-1.74)cons -0.0195-0.0097-0.0132 0.3742***0.3740***0.3735*** (-0.09)(-0.04)(-0.06) (12.15)(12.12)(12.11) N 155215521552 240102401024010 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 43

PAGE 44

Thetwogures(Figure 1-4 andFigure 1-5 )suggestthatthereisfuelsubstitutionamong Figure1-5. AverageElectricityConsumptionintheSummer,Winter,andNon-peakmonthsbyHouseholdEnergyComposition householdswithbothnaturalgasandelectricityconsumptionwhichaccountedfortheloworstatisticallyinsignicantenergysavingseectsinthewintermonths.Totestwhetherenergysavingswouldhavebeenhigherwithoutthesubstitutionofgasforelectricity,weconsideronlythesubsetofhouseholdswhouseonlyelectricityandestimatethetreatmenteectinthesummerpeak,winterpeaks,andnon-peakmonths.TheestimatedeectsareshowninTable 1-9 .TheresultsinTable 1-9 showstatisticallysignicant14%energysavingsinthesummerusingtheCEMDDregression(ColumnI).TheregularDDgivesa11%energysavingsinthesummer.Theseestimatesarequitelesswhencomparedtotheestimatesusingallhouseholds.Thisimpliesthatthehigh-eciencyACrebateprogramhasgreaterenergysavingseectsonhouseholdswithbothelectricityandgasthanonhouseholdswithelectricityastheironlyenergysourceduringthesummermonths.I.e.,thehigh-eciencyACsignicantlyreducesthealreadylownaturalgasconsumptionduringthesummermonths.TheelectricitysavingsestimatesinthewinterpeakaredoubledandstatisticallysignicantintheDDCEMregressionwithanestimated8%reductioninwinterpeakconsumption(ColumnII).Theestimated 44

PAGE 45

Table1-9. Summer,Winter,andNon-PeakMonthsEectsofTheHighEciencyACRebateProgramforTheElectric-OnlyHouseholds log(ElectricityUsage) DDCEM(CensusTract)RegularDD SummerWinterNon-PeakMonths SummerWinterNon-PeakMonths Treat -0.1391***-0.0804*-0.0376 -0.1110***-0.0659**-0.0348 (-3.73)(-2.16)(-1.20) (-3.51)(-2.64)(-1.60)Bedrooms 0.04310.03950.0617 -0.00120.00140.0017 (1.14)(0.82)(1.65) (-0.12)(0.14)(0.20)Stories -0.0088-0.0032-0.0199 0.0063-0.0227*-0.0050 (-0.39)(-0.11)(-0.75) (0.65)(-2.55)(-0.63)HeatedArea -0.0090-0.0523-0.0272 -0.00330.01310.0096(1000squarefeet) (-0.24)(-1.88)(-1.14) (-0.33)(1.48)(1.17)Age -0.00220.0016-0.0015 0.00020.0003-0.0002 (-1.06)(0.74)(-0.78) (0.34)(0.70)(-0.50)Pool 0.04800.04230.0064 -0.0051-0.0454***-0.0234* (0.88)(0.94)(0.16) (-0.41)(-3.84)(-2.09)MeanIncome($1000) -0.0013-0.0011-0.0001 -0.0003*-0.0011***0.0000 (-1.65)(-1.23)(-0.24) (-1.96)(-7.28)(0.08)MeanHouseholdSize 0.07480.17770.1121 0.01540.01080.0310 (0.76)(1.16)(1.27) (0.70)(0.61)(1.87)HotTub -0.0937-0.00030.0229 -0.01560.0100-0.0068 (-1.39)(-0.00)(0.50) (-0.77)(0.42)(-0.32)cons -0.0936-0.1560-0.3755 0.02720.3067***-0.1127* (-0.41)(-0.38)(-1.74) (0.45)(6.10)(-2.39) N 502502502 797579757975 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 45

PAGE 46

energysavingsinthewintermonthsusingtheregularDDisalsohigher(6.6%)thanpreviouslyestimatedusingthecombinedgroupofallhouseholds.TheseindicatethattheACrebateprogramhadalargeeectonelectricityconsumptioninthewintermonths.However,thesubstitutionofelectricityforgasinthewintermonthsbythosewithgasintheirenergymixovershadowsthiseectsothattheaggregateeectisstatisticallyinsignicant.Theestimatedeectsinthenon-peakmonthsarestatisticallyinsignicantintheDDCEMwithcensustracksaswellasintheregularDD.Itis,however,signicantatthe5%levelintheDDCEMregressionwithzipcodesasneighborhoods.20WealsoestimatetheeectoftheACrebateprogramonelectricityconsumptionforthosehouseholdswithbothgasandelectricityintheirenergymixtogivemoresupporttoourfuelswitchinghypothesis.Wepositthatsincethosewithnaturalgasswitchtonaturalgastoheattheirhomesinthewinter,thereislittleACorelectricityusageinthewinter,sothereshouldbelittleornoenergysavingsforthisgroupofhouseholdsinthewinter.WealsoexpectarelativelyhighersummerpeakeectsincetheACrebateprogramisexpectedtohavebiggereectsonbothsummerelectricityandnaturalgasconsumption.Table 1-10 summarizesourresults.Fromthetable,thesummerpeakeectoftheACrebateprogramisabout24%intheCEMDDwithcensustracksasneighborhoods(Column1)andaabout21%intheregularDID(ColumnIV).Therearealsoenergysavingsinthenon-peakmonths(7.8%reductionusingtheDDCEMwithcensustractsasneighborhoods(ColumnIII)and6.5%usingtheregularDID(ColumnVI).However,asexpected,therearenostatisticallysignicantenergysavingsinthewinterpeakusingtheDDCEMregression.TheregularDDshowsthatelectricityusageincreasedinthewinterfortheprogramparticipants,buttheresultisonlymarginallysignicantatthe5%level. 20SeeTable A-7 intheappendix. 46

PAGE 47

Table1-10. Summer,WinterandNon-PeakMonthsEectsofThe2009ACRebateProgramonElectricityConsumptionforHouseholdswithBothElectricityandNaturalGasinTheirEnergyMix log(EnergyUsage) DDCEM(CensusTract)RegularDD SummerWinterNon-PeakMonths SummerWinterNon-PeakMonths Treat -0.2408***0.0262-0.0780*** -0.2168***0.0552*-0.0658*** (-7.62)(0.91)(-3.58) (-7.52)(1.96)(-3.50)Bedrooms 0.0514-0.0324-0.0212 0.0005-0.0107-0.0063 (1.32)(-0.88)(-0.79) (0.07)(-1.65)(-1.01)Stories 0.08140.04160.0145 -0.00790.0237*0.0043 (1.63)(1.06)(0.46) (-0.92)(2.58)(0.53)HeatedArea -0.04510.05920.0372 0.00180.0109-0.0028(1000squarefeet) (-1.71)(1.75)(1.22) (0.27)(1.77)(-0.48)Age -0.00130.0008-0.0019 -0.00000.0016***-0.0006 (-0.89)(0.64)(-1.73) (-0.13)(4.79)(-1.70)Pool -0.0245-0.1126**-0.0810 -0.0479***-0.0388***-0.0304*** (-0.54)(-2.65)(-1.92) (-5.53)(-4.44)(-3.70)MeanIncome($1000) 0.0006-0.0011*-0.0006 -0.0002-0.0009***0.0000 (1.20)(-1.97)(-1.29) (-1.90)(-7.65)(0.07)MeanHouseholdSize -0.02410.13580.1407* -0.00860.02590.0297* (-0.38)(1.69)(2.53) (-0.55)(1.61)(2.10)HotTub -0.0849-0.0075-0.0157 -0.00220.0061-0.0085 (-1.61)(-0.22)(-0.46) (-0.19)(0.47)(-0.69)cons -0.0404-0.2477-0.2671* 0.1347**0.0046-0.0514 (-0.20)(-1.33)(-2.08) (3.15)(0.11)(-1.31) N 410744374437 160301603516035 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 47

PAGE 48

1.7ConclusionUnderstandingtheactualenergysavedatthewholebuildinglevel,andnotjustattheappliancelevelisimportanttoconsumers,theutilitycompanyandregulators.Allstakeholderswanttoknowtheexactenergysavingstodeterminethecost-eectivenessoftheprogramfromtheirperspective.ConsumersareinterestedinwhetherthediscountedmonthlysavingswouldoutweightheinitialcostofparticipatingintheDSMprogram.Regulatorsandutilitycompaniesareinterestedintheoverallcost-eectivenessoftherebateprogramandwhethertheprogramshouldbecontinuedinfutureyears.Thisstudyprovidesananalysisoftheenergysavingseectsofademand-sidemanagementprogramparticularlyGRU'shigheciencyACprogramwhereGRUoersincentivestoitscustomerstoreplacetheiroldlow-eciencyACunitwithahigh-eciencymodel.Theresultsshowthatthehigh-eciencyACprogramhassignicanteectsonannualenergysavingsinbothourproposeddierence-in-dierencecoarsenedexactmatchingmethodologyortheregulardierence-in-dierencemethodologywithoutmatching.Theresultsalsoshowthatwhilethehigh-eciencyACprogramhadsignicanteectsonsummerpeakenergyconsumptionandnon-peakmonthsconsumption,ithadlittleornostatisticallysignicanteectonwinterpeakconsumptionWhiletheempiricalanalysispresentedhereisspecictoGainesvilleandtothehigheciencyACrebateprogram,andtheanalysisislimitedbyproblemsofdataavailabilityandreliability,usingthedierence-in-dierencecoarsenedexactmatchingapproachtoreducetheimbalancebetweenthetreatedanduntreatedobservationaswellasmatchingonneighborhoods(withoutfurthercoarsening)tocontrolfortheeectsofweatheronenergyconsumptionisoneofthecontributionsofthispaper.Theresultsalsoindicatethatwhentheareaunderstudyhasonlyoneweatherstationsothatthereisnoproxyforhouseholdspecicweather,thedierence-in-dierencemethodology(withoutmarching)overstates(orunderstates)theenergysavingseectsofDSMprograms. 48

PAGE 49

CHAPTER2THE\REBOUND"EFFECT 2.1IntroductionInthischapter,Ipresentanestimateofthereboundeect.ReboundoccurswhenDSMprogramparticipationresultsinadeclineinparticipants'energycostsothatparticipantsadjusttheirthermostatsettingorotherenergyuselevels,therebydecreasingenergysavings.Reboundeects,therefore,implythatDSMinvestmentswouldnotleadtoproportionatereductionsinenergyconsumption.ThereasonforthisisthatDSMmeasuresreducetheeectivepriceofoperationofenergy-consumingequipment.Hence,consumersusesomeofthemoneysavedtopurchaseincreasedcomfort,increasingtheuseofenergy-consumingequipment(e.g.adjustingthermostatsettingorincreasedhoursofoperation).Theterm\reboundeect"rstappearedinaseminalpaperbyDanielKhazzoominwhichtheauthorarguedthatmandatedenergyeciencystandardsforhouseholdapplianceswouldnotleadtoaproportionatedecreaseinenergyconsumption( Khazzoom , 1980 ).Sincetheterm'sappearanceintheliterature,therehasbeenextensiveresearchonthesizeofthereboundeect.However,thereisawiderangeandvariationofestimatesofthereboundeectintheliterature.Dependingontheenergyeciencymeasure,thetheoreticalliteraturepositsreboundeectsofbetweenzero(noreboundeect)and100%rebound(backre),whileestimatedreboundeectsintheempiricalliteratureliebetween0%andabout75%.Thestarkvariationintheestimatesofthereboundeectstemsfromthedenitionandtheempiricalmethodologyused.Someempiricalstudiesusesurveydatawhereconsumers'responsestoquestionnairesareusedtoinvestigatethereboundeect(e.g. Fowleretal. ( 2015 )).Otherstudiesuseobservedthermostatsettingsandhotwatertemperaturestoestimateadirectreboundeect(e.g. Dubinetal. ( 1986 )).Adirectreboundeectmeasuresincreasesintheconsumptionoftheappliancethathasreceivedtheenergyeciency 49

PAGE 50

improvement.1Amajorityoftheempiricalstudies,especiallyliteratureinthetransportationsector,however,relyonobservationaldataonenergyuseandenergyprices.Inthesecases,thereboundeectisinvestigatedusingvariationinenergypricesratherthanvariationinenergyeciency.Theintuitionforusingpricevariationisthatenergyeciencyimprovementsreducethecostofusingtheenergy-consumingappliance,inthesameway,anenergypricereductionwould.Therefore,wecanexpectconsumerstoresponsetoadecreaseinenergycostasaresultofenergyeciency,inthesameway,theywouldrespondtoadeclineinenergyprices.Further,pricesareexogenouslyxed,andconsumershavenocontroloverthemcomparedtoenergyeciencyimprovementsthatareendogenouslychosenbyhouseholds.Althoughusingpricevariationhelpstocircumventtheproblemofendogeneitywithenergyeciencyinvestments,theelasticitiesforpricesandeciencycanbestatisticallydierentfromeachother( Greene , 2012 ).Anotherrecentempiricalstudyfurthersuggeststhatconsumersrespondcomparativelylesstochangesinthefueleconomyofvehiclesthantofuelprices( Gillingham , 2011 ).Againintheelectricitysector,sinceachangeinpriceaectstheconsumptionofallotherenergy-consumingappliances,usingthepriceelasticitytoestimatetheresponsetoenergyeciencyforjustoneappliancemayoverstatetheresponsetoenergyeciency.Whilethetheoreticalliteraturemodelshouseholddemandasasumofelectricitydemandforelectricityconsumingappliances(e.g.,( ReissandWhite , 2005 )),oneproblemwithempiricalestimationishowtoseparatethetotalhouseholddemandintoitspartswithoutsmartmetersthatcanmeasureenergyconsumptionattheappliancelevel.Someempiricalstudiesmeasurechangesinthermostatsettingandusethischangetoestimatetheenergyrequiredtomaintainthenewsetting.Forexample, Dubinetal. ( 1986 )approximateddailyheatingloadasaquadraticfunctionofthedierencebetweenoutdoorandindoortemperatureandestimatedthepriceof 1Thereisalsoanindirectreboundeectwhichmeasurestheimpactoftheenergyeciencyimprovementontheconsumptionofotherenergy-consumingappliancesorallotherproducts. 50

PAGE 51

comfortasthechangeinbillingperiodutilizationassociatedwithadegreechangeinhouseholdthermostatsetting.Thereisalsotheproblemofwhichpriceconsumersrespondtowhenutilitycompaniesusethecomplexincreasing-blockpricingschedulethathasbecomeverycommonamongutilities.Priceelasticitiescalculatedusingmarginalpricesimplicitlymaketheassumptionthatconsumers,atanypointintime,knowtheirlevelofconsumption,andtherefore,theirmarginalprice.Thisreasoningseemsunrealisticasthiswouldmeanconsumersvisittheirelectricitymeteronadailyorhourlybasis.Priceelasticitiesestimatedwiththeseassumptionseitheroverstatesorunderstatesthetruepriceelasticityandthustheresponsivenesstoenergyeciency. Dubinetal. ( 1986 )estimatedthepriceelasticitiesofspaceheatinginJanuary,February,andMarchtobe-0.52,-0.81,and-0.73respectivelyforFloridaPowerandLightCustomers.Usingtheseelasticities,theauthorsestimatedtheresponsivenessofspaceheatingandcoolingtodecliningunitenergyconsumptionofappliancesandconcludedthatthereboundeectisabout8-12%belowengineeringestimatesforthosemonths.AnimportantelementoftheFloridaPowerandLightstudyisthatspaceheatingwasmeteredseparatelywhichallowedforanaccurateestimationofthedirectreboundeect.However,itignorestheimpactofthelowenergyconsumptionofoneapplianceontheenergyconsumptionofotherenergy-consumingappliances.Someofthereboundmaybedesirableforoverallenergyusage.Forexample,theanalysisinsection6ofthispapersuggeststhatthereisfuelswitchinginthewintermonthssothatelectricityconsumptioninthewintermonthsislowforhouseholdswithbothelectricityandnaturalgas.AnincreaseinACeciencymeanselectricitybecomescompetitiveorevenbetterthannaturalgasastheprimaryheatingfuel.Thus,whilewemightobserveanincreasedACusageinthewinter,naturalgasusageandtheuseofotherstand-aloneroomheatersmayhavereducedsothattheoverallenergyconsumptionforthehouseholdislower.Insuchacase,wemighthaveasignicantdirectreboundeectoftheACprogrambutthetotalreboundeect(sumofdirectandindirectrebound)maybesmallornonexistent. 51

PAGE 52

2.2ConceptualFrameworkAsstatedearlier,themainreasonforusingpricevariationtoestimatereboundofenergyeciencyimprovementsistheendogeneityofeciencyimprovements( Westetal. , 2015 ).Thatis,whilepricesaretakenexogenouslybyconsumers,householdsthatparticipateinretrottingprogramsarenotselectedrandomlyfromthesetofallhouseholds.Inthispartoftheanalysis,wemakeaselection-on-observablesassumption.Weassumethatparticipationintheenergyeciencyimprovementdependsonhouseholdspre-treatmentcharacteristics.Hence,byselectingacontrolgroupthathassimilarcharacteristicsasprogramparticipants,weareabletominimizethebiasfromtheendogenousselectionintotreatment.Weusedthecoarsenedexactmatchingmethodologytoselectareasonablecontrolgroupthathassimilarpre-treatmentcharacteristicsasthetreatedhouseholdssothatintheabsenceoftheprogram,thetrajectoryofaverageelectricityconsumptionoftheprogramparticipantswouldbesimilartothatofthecontrolhouseholds.Iftheselection-on-observablesassumptionholds,thenthecontrolgroupwouldhavethesimilarlikelihoodofparticipatingintheprogramasthetreatedhouseholdsbutratherchosenottoparticipate.Again,incontrasttoestimatingreboundinthefueleconomyofnewcarsinwhichconsumerscaneasilycomparethefuelcostpermiletraveledgiventhefueleconomyofacar,theexactcostsavingsofecientelectricappliancesarenotreadilyknowntotheconsumer.Thecomplexpriceschedulesusedbyutilitycompaniesfurthermuddlethecalculationofthecostsavings.We,therefore,positthateventhoughconsumersexpectareductionintheirbillafterinstallinganewecientAC,theexacteectontheirbillisnotknown.Itisonlyafterobservingtheimpactofparticipatingintheprogramontheirtotalenergybillthatconsumersinfertheirenergycostsavings.Thus,weassumethatintheperiodthatconsumersundertaketheenergyeciencyimprovements,thereisnobehavioralchangeinanticipationofthemoneysavingsfromparticipatingintheretrottingprogram.Changesinenergyconsumptionabovethecounterfactualconsumptionintherstperiodoftheprogramcan,therefore,beconsideredas\pure"programeects.However,afterconsumerslearnoftheirenergycostsavingsthrough 52

PAGE 53

theirenergyusagebill,theymakechangestotheirbehaviororlifestylewhichmightreducetheirenergysavingsorincreaseenergyconsumption.Inthiscase,thereboundmayoccuronamonth-on-monthbasisasconsumerslearntheirelectricitycostsavingsfromthepreviousmonth'sbillorcanoccurinthenextwinterorsummerafterobservingtheircostsavingsinthepreviouswinterorsummer.Forexample,householdsthatparticipateinaweatherizationassistanceprogrammightchangetheirthermostatsettinginthesecondwinterafterobservingtheirenergycostreductionintherstwinteraftertheprogram.Inthispaper,weconsiderthewholeyear,orthewinterpeak,summerpeak,orthenon-peakmonthsasourperiod.Thus,participants,afterobservingtheirenergycostdecreaseintherstperiodaremorelikelytoengageinactivitiesthatleadtoenergysavingsreboundinthesecondyear.Thisframeworkissimilartotheresearchdesignof Fowleretal. ( 2015 )inestimatingtemperaturetakebackeects.Theauthorssurveyedhouseholds'indoorairtemperatureatleastoneyearafterthehouseholdshavereceivedeciencyimprovementsto\allowplentyoftimeforresidentstoobservehowtheretrotaectedwinterheatingcost"( Fowleretal. , 2015 ).Wefollowtheparticipantsofthe2009programforanotheryeartoobservethechangesintheirenergysavingsinthesecondyearaftertheprogram.Thisallowedparticipantstoseetheirsummerandwinterenergycostforatleastoneyear. 2.3GraphicalAnalysisofTheReboundEectUsingthecontrolgroupandthetreatmentgroupobtainedusingtheCEMmethodology(withcensustractsasneighborhoods),wegraphtheaveragemonthlyconsumptionofbothgroupsbeforetreatment(2008),inthetreatmentyear(2009),andthreeyearsaftertreatment(2010,2011,2012).Thisgivesaroughestimateofthereboundeect.Figure 2-1 showsthegraphofaveragemonthlyenergyconsumptionofparticipantsandnon-participants.Thechartshowsthatbeforetreatment,i.e.in2008,theparticipantshadahigherenergyconsumptionthanthenon-participantswithasignicant(t-statistic=9.86)dierenceofabout168kWhandwiththeprogramparticipantsbeingthehighenergyusers.Howeverintheyearofthetreatment,thisdierencediminishesto96.2kWh.Thus,therewasimmediateeectofthe 53

PAGE 54

programevenintheyearoftreatment.Theyear2010istherstfullyearofparticipatingintheprogram,andwhiletheenergyconsumptionofbothgroupsincreased,theenergyconsumptionofthecontrolgroupincreasedsharplycomparedtothatofthetreatedhouseholdssothataveragemonthlyenergyconsumptionisalmostthesameforbothgroupswithnostatisticallysignicantdierence.Againwhilethemonthlyaverageenergyconsumptionofbothgroupsdecreasedin2011and2012,theprogramparticipantshadasharpdeclineofabout17.3%comparedtothenon-participantswhohadareductionof13.8%.Theaverageenergyconsumptionofthetreatmentgroupisbelowthatofthecontrolgroupin2011withastatisticallysignicantdierenceof60.3kWh.Theaverageconsumptionofbothgroupsdecreasedbythesamepercentage(11%)intheyear2012.Thegraph,therefore,suggestthatevenafterconsumerslearnoftheirenergycostsavingsintheyearofprogramparticipationandaftertherstfullyearofprogramparticipation,theirenergysavingsinthesubsequentyearsisevenmuchhigher.ThegraphthussuggestsnoreboundeectoftheACrebateprogrambutratheracontinuedincreaseinenergysavingsinthesubsequentyears.Figures B-1 and B-2 intheappendixshowsthegraphofaverageenergyconsumptionofthetreatedandcontrolconsumerswhowerematchedintheCEMmethodologywhenzipcodeswereusedasneighborhoodsandwhennomatchingmethodologywereused.WhilethetwoguresdierslightlyfromFigure 2-1 andfromeachother,themainobservationfromFigure 2-1 thattheprogramparticipantsevenincreasedtheirenergysavingsmuchhigherinthesubsequentyearsaftertheprogramispresentinallthreegures. 2.4EstimationandResultsOurestimationofthereboundeectsfollowsthesameprocedureastheevaluationoftheenergysavingseects.Weestimatewhethercomparedtothecontrolgroup,thetreatmentgroupincreasedconsumptionintheyear2011aftertheirrstfullyear(2010)ofprogramparticipationandobservingtheirenergycostsavings.Anyreboundafterobservingtheirusagecostwillleadtoareductionofenergysavingsoranincreaseinenergyconsumptionbythetreatmentgroupcomparedtothecontrolgroup.We,therefore,comparedtheenergy 54

PAGE 55

Figure2-1. AverageEnergyConsumptionbyParticipantsandNon-Participants consumptionofthetreatmentgrouptothecontrolgroupacrosstheyears2010and2011.WeusedEquations 1{5 and 1{6 withandwithouttheCEMmethodology.Irepeattheseequationsbelowforeasyreference.Equation 1{5 modelsthelogoftotalenergyusageoftheform: yit=0+0.d2+1treatit+2treatitXi+1Xi+2d2Xi+i+uit,t=1,2(2{1)whereyitisthelogoftotalenergyconsumptionforhouseholdiinperiodt.Totalenergyconsumptionforhouseholdiisdenedasthesumofelectricityconsumptionmeasuredinkilowatthours(kWh)andnaturalgasconsumptionconvertedtoequivalentkilowatthours(ekWh).2d2isadummyvariableforthesecondtimeperiod,iistheindividualheterogeneitythatisconstantacrosstime,anduitistheidiosyncraticerrorthatvarieswithtime.Xiisavectorofhouseholdcharacteristics.Firstdierencingthetwoequationsacrossthetwotime 2Naturalgasconsumptionisoriginallymeasuredintherms.Inordertocombineelectricityconsumptiontonaturalgasconsumption,naturalgasconsumptionwereconvertedtoequivalentkWhusingtheconversionrate1therm=29.300111ekWh. 55

PAGE 56

periods(2010and2011)removestheindividualheterogeneityaswellasallthetimeconstantexplanatoryvariablessoweobtaintheestimationequation: yi=0+1.treati+2treatiXi+2Xi+uit(2{2)Table 2-1 showstheresultoftheestimationusingEquation 1{6 .ColumnIofthetablegivestheresultsoftheDDCEMwithcensustractsasneighborhoods,ColumnIIgivestheresultsoftheDDCEMwithzipcodesasneighborhoods,andColumnIIIgivestheresultsoftheregularDDmethodologywithonlythematchedsamplefromtheCEM(withcensustractsasneighborhoods).Allthreeregressionsgiveanegativesignonthetreatmentgroupwhichimpliesthatthetreatmentgroupfurtherdecreasedenergyconsumptionin2011abovethe2010consumption(asobservedinFigure 2-1 ).However,theeectisnotstatisticallysignicantintheDDCEMwithcensustractsasneighborhoodsandintheregularDDbutsignicantatthe5%levelintheDIDCEMwithzipcodesasneighborhoods.Allthreeresults,therefore,showthereisnoreboundeectofthehigh-eciencyACrebateprogram,ratherprogramparticipantsfurtherincreasedtheirenergysavingsinthesubsequentyearaftertheprogram. ReboundinPeakandNon-peakPeriodsTheresultsaboveconcludesthattherearenosubstantialannualreboundeects.Weinvestigateiftherearereboundeectsinsomeperiodsthaninsomeotherperiods.TheresultsinTable 2-2 showthatthereisnostatisticallysignicantreboundeectinallperiods.3Thecoecientsinbothregressionsforallperiodshavenegativesignsimplyingthatenergysavingswereevenmorepronouncedin2011,buttheestimatesarenotstatisticallysignicant.Thisresultishoweverincontrasttotheresultby Dubinetal. ( 1986 ). Dubinetal. ( 1986 )conductedananalysisoftheeectsofhigh-eciencyACorahigh-eciencyheatpump 3ColumnsI,III,VshowstheresultsusingtheDDCEMwithcensustractsasneighborhoods,whileColumnsII,IV,andVIshowstheresultsusingtheregularDDonthematchedCEMsample. 56

PAGE 57

Table2-1. EectofThe2009High-EcientRebateProgramon2011EnergySavings log(EnergyUsage) CEMDDCEMDDRegularDD (CensusTracts)(ZipCodes)(withCEM) Treatmentgroup -0.0261-0.0300*-0.0166 (-1.74)(-2.30)(-0.64)Bedrooms -0.00440.0005-0.0034 (-0.17)(0.03)(-0.14)Stories -0.0217-0.01430.0060 (-0.96)(-0.57)(0.15)HeatedArea(1000squarefeet) -0.0158-0.0037-0.0121 (-0.60)(-0.21)(-0.52)Age -0.0032**-0.0014*-0.0023* (-3.01)(-2.25)(-2.34)Pool 0.0359-0.0036-0.0046 (1.20)(-0.20)(-0.17)ElectricandGas -0.0759**-0.0674***-0.0612** (-3.26)(-6.15)(-3.06)MeanIncome($1000) 0.00030.0004-0.0002 (0.59)(1.58)(-0.71)MeanHouseholdSize -0.0327-0.05990.0169 (-0.68)(-1.76)(0.38)HotTub -0.0785*-0.0666-0.0478 (-2.17)(-1.88)(-0.98)cons 0.08820.0543-0.0618 (0.74)(0.53)(-0.49) N 155278431552 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. programoperatedbyFloridaPowerandLightin1981.Theauthorscombinedstatisticalanalysistoengineeringstudyintheirmodelandconcludedthatlittlereboundtakesplaceinthesummer(about1-2%belowengineeringestimates)butasignicantreboundinthespringandfall(about13%belowengineeringestimates).IntheDDCEMregressionwithzipcodesasneighborhoods,theresultsshowthattherewereratherincreasedenergysavingsofabout3%inthesummerandnon-peakperiods.4 4SeeTable B-4 intheappendix. 57

PAGE 58

Table2-2. SummerPeak,WinterPeak,andNon-peakReboundEectsofTheACRebateProgram log(EnergyUsage) SUMMERWINTERNON-PEAK DDCEMregularDD DDCEMregularDD DDCEMregularDD Treat -0.0154-0.0089 -0.0059-0.0001 -0.0328-0.0235 (-0.69)(-0.50) (-0.30)(-0.00) (-1.87)(-1.53)Bedrooms -0.0219-0.0162 0.0123-0.0076 0.01330.0359 (-0.60)(-0.51) (0.32)(-0.41) (0.44)(1.56)Stories -0.03730.0031 -0.0231-0.0152 -0.01330.0166 (-1.05)(0.14) (-0.63)(-0.75) (-0.49)(0.86)HeatedArea(1000squarefeet) 0.0039-0.0101 -0.00290.0148 -0.0461-0.0470* (0.07)(-0.37) (-0.11)(0.82) (-1.62)(-2.16)Age -0.0045**-0.0012 -0.0033*-0.0043*** -0.0019-0.0012 (-2.72)(-0.86) (-2.49)(-4.96) (-1.34)(-0.98)Pool 0.0445-0.0035 0.02830.0077 0.04100.0119 (1.01)(-0.11) (1.07)(0.41) (1.21)(0.57)ElectricandGas -0.00010.0201 -0.1084***-0.0877*** -0.1005***-0.0821*** (-0.00)(0.74) (-3.46)(-5.28) (-3.92)(-4.03)MeanIncome($1000) 0.0006-0.0002 0.00140.0011*** -0.0008-0.0014** (0.92)(-0.36) (1.93)(3.41) (-1.29)(-3.13)MeanHouseholdSize 0.03110.1149 -0.1819-0.1392** 0.05230.0617 (0.40)(1.80) (-1.36)(-3.20) (0.95)(1.37)HotTub -0.1100-0.0075 -0.0150-0.0654 -0.0990-0.0799* (-1.55)(-0.17) (-0.26)(-1.84) (-1.94)(-2.23)cons 0.0393-0.2303 0.18680.1408 -0.0484-0.1620 (0.20)(-1.54) (0.70)(1.14) (-0.35)(-1.28) N 15521552 15521552 15521552 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 58

PAGE 59

2.5ConclusionThesectioncontributessignicantlytothegrowingliteratureonreboundeectsofenergyeciencypolicies.Itshowsthatinthecaseofthehigh-ecientACrebateprogram,thereisnoreboundeect.Infacttheprogramhadincrementalenergysavingstwoyearsaftertheprogramparticipation.Theadditionalenergysavingsishowevernotstatisticallysignicant.ReboundeectsareimportanttotheutilityandregulatorstodetermineiftherstyearenergysavingswouldpersistorwhetherthesupplyresourcesthattheDSMprogramwasdesignedtodisplacewillindeedbeavoidedoverthelongrun.AnaccuratemeasurementofthereboundeectthereforehelpsinestimatingtheavoidedcostofDSMprogramsinordertogarnermorestakeholdersupportfortheseprograms. 59

PAGE 60

CHAPTER3EFFECTSOFAUTOPAYPROGRAMONPRICESENSITIVITY 3.1IntroductionEconomictheoryisbuiltontheassumptionthatrationalagentsmakeconsumptiondecisionsbychoosingabundleofgoodstomaximizeutilitygiventhepricesofeachgood.Itfurtherassumesthatthepriceofthegood,aswellasitsobservableattributes,isperfectlyknowntoconsumerswhenmakingconsumptiondecisions.Also,themethodofpaymentdoesnotaectaconsumer'sevaluationofproductcharacteristics.Theoptimizationdecisionofaconsumerwhofacesaconstantmarginalpriceforagoodisstraightforward.Nonlinearpricing,however,complicatestheoptimizationdecisionbycreatingmultiplemarginalpricesforthesamegoodbasedonthelevelofconsumption.Nonetheless,aperfectly-optimizing,perfectlyinformedconsumerwouldstillmaketheoptimizationdecisionbyconsuminguptothepointwherethemarginalpriceofthegoodequalstheconsumer'smarginalvalueforthelastunitofthegood.However,empiricalevidenceshowsthatinanonlinearpricingschedule,consumersmaketheoptimizationdecisionwithlimitedinformation,attention,andcognitiveability.Forexample,subjectsinanexperimentrespondtoaveragetaxratesbutrespondtothetruemarginaltaxrateifthetaxtableisredesignedtostressthemarginalrates( DeBartolome , 1995 ).Taxpayershavelowawarenessoftheirmarginaltaxrates( Gensemeretal. , 1965 ). LiebmanandZeckhauser ( 2014 )suggestthatwhenpeoplearefacedwiththecomplexpriceschedule,insteadofrespondingtothediscretejumpsinthepricingschedulewithdiscretejumpsinconsumption,theysmooththeirconsumptionovertheentirerangeoftheschedule.Anothersetofempiricalevidencefurthersuggeststhattheuseofcreditcardsasapaymentmechanism,forexample,increasesthepropensitytospendmoreascomparedtocashinotherwiseidenticalpurchasesituations( PrelecandSimester , 2001 ; Soman , 2001 ).Also,consumers'perceptionandevaluationoftheproductsdieracrosspaymentmechanisms( ChatterjeeandRose , 2012 ). 60

PAGE 61

Thecontributionofthispaperistwo-fold.First,wediscusswhatpriceconsumersrespondtoinanonlinearpricingofelectricity.Second,weanalyzewhetherautomaticbillpaymentsinwhichaconsumer'selectricitybillistakendirectlyfromhisbankaccountsincreaseshisinattentionandmakehimlesspricesensitive.Manyutilitiesusetheincreasing-blockpricing,anexampleofnonlinearpricinginwhichthepriceoftheutilityservicefollowsastep-functionwiththemarginalpriceincreasingatspeciedlevelsofconsumption.1SeeFigure 3-1 forGainesvilleRegionalUtilities(GRU)'sincreasingblockpricingschedule.Normally,thereisalsoaconstantupfrontfeeorcustomerchargethatdoesnotdependonthelevelofconsumptionandaconstantper-unitfuelcharge.Thecomplexityofthispricingschedulemakesthepriceslesssalienttoconsumers.Theconsumptiondecisionofmostelectricityconsumers,thus,tendstodeviatesystematicallyfromthatofaperfectlyoptimizingconsumer.Anelectricconsumer'selectricityusagedoesnotexhibitdiscretejumpsinconsumptioninresponsetothediscretejumpsinthemarginalprices( Borenstein , 2009 ).Furthermore,householdelectricityconsumptiondoesnotbuncharoundthecut-opointsoftheincreasingblockpricingscheduleaswouldbeexpectedifconsumersrespondtothetruemarginalprice( Ito , 2014 ).Thecomplexpriceschedule,aswellasthefactthatconsumersdonotknowtheirelectricityconsumptionatanypointintheirbillingcycle,combinetomaketheprevailingmarginalpriceofelectricitylesstransparenttotheconsumer.Empiricalevidenceshowsthatconsumersrespondtotheaveragepriceoftheincreasing-blockpricingschedule( Ito , 2014 ),oranexpectedmarginalprice( Borenstein , 2009 )insteadofthemarginalprice.Onereasonoftencitedforwhyconsumersrespondtothe(current)averagepriceoftheincreasing-blockpricinginsteadofthemarginalprices,astheorypredicts,isthattheexpectedcostoflearning 1Declining-blockpricingwherebythemarginalpriceofelectricityislowerforhigherlevelsofconsumptionhadbeencommoninthe1970s.However,sincethe1980s,utilitieshavebeenswitchingtoincreasing-blockpricingbecausethedecliningblockpricingwasseenaspromotingwastefulness. 61

PAGE 62

theprevailingmarginalpriceexceedsthecorrespondingexpectedbenet( FosterandBeattie , 1979 ).However,consumersrespondingtocurrentaverageprices2stillrequireslotsofeortonthepartofconsumers.Itrequiresthemtoknowthepricescheduleorthestartandenddatesoftheirbillingperiod.Itfurtherrequiresconsumerstobeabletoanticipatealldemandshockswithinthebillingperiodorhaveknowledgeoftheirtotalconsumptionattheendoftheperiod.Theseassumptionsseematoddswithreality.\Itseemssafetosaythatnotonlydomostconsumersnotknowhowmuchpowerorwatertheyhaveusedsincetheirbillingperiodbegan,butmostconsumersalsodon'tknowwhentheircurrentbillingperiodbegan"( Borenstein , 2009 ).I,therefore,proposeaconceptualmodelinwhichconsumersmakebehavioralrulesaboutconsumptiononlyafterobservingtheirpreviousbill.Iftheirelectricitybillinthepastmonthishigh,consumersmakerulesaimedatreducingelectricityconsumptioninthecurrentmonth.Ontheotherhand,iftheirbillislow,consumersmakesrulestoeaseontheirconservationpractices.Thismodelsuggeststhataconsumer'scurrentdemandforelectricityrespondstotheaveragepricehepaidforelectricityintheprecedingmonth.Iempiricallytestthismodeltodeterminewhethertheproposedconceptualmodelisanadequatedescriptionofreality.Iusetheencompassingtestofnon-nestedmodelstoshowthatconsumersrespondtotheirone-monthlaggedaveragepriceinsteadofthecurrentaverageprice,consistentwiththeconceptualmodel.Theresulthasimplicationsforutilityratedesign,astheempiricalpriceelasticityplaysaroleinutility'sratedesignandthechoiceofpriceandnon-priceconservationstrategies.Italsohasimplicationsforenergyeciencypoliciesbecauseinafullinformationworldinwhichconsumersrespondtomarginalprices,increasingthemarginalpricesespeciallythetoptiermarginalpriceswouldencourageconsumptionreduction.However,ifconsumersinsteadrespondtotheiraverageorlaggedaverageprices,thenmean-preservingchangestothe 2currentaveragepriceincalculatedasthetotalbillinabillingperioddividedbythetotalusageinthatbillingperiod 62

PAGE 63

priceschedulewouldnotaectelectricityconsumption.Itisimportanttoknowtheactualpriceconsumersrespondtoifchangestothepricescheduleareexpectedtoinuenceconsumersbehaviorinapredictableandconsistentway.Animplicationoftheconceptualmodelisthatconsumerswhoinspecttheirbillcarefullywillrespondmoretopricesthanconsumerswhoonlytakeacursorylookattheirbillorfailtoexaminetheirbill.Theelectricitybillofautopayusersisdeductedfromtheirbankaccountaftertheirbillissenttothem,somostautopayusersdonotbothertoreviewthechargessincetheydonothavetowriteacheck.Hence,automaticbillpaymentsincreasetheprobabilitythatuserswillforgobillinspections,whichincreasesinattentiontothelaggedaverageprice,andmakeconsumersrespondlesstoprices.Furthermore,utilitycompaniesencourageautopayuserstoparticipateinonlinebilling3tocomplementtheirenrollmentinautopay.Participatinginbothprogramsincreasesfurthertheprobabilitythatconsumerswillignoretheirbillandthusrespondlesstopricesthannon-autopayusers.Mysecondempiricalmethodologytestthisconceptbyallowingthepriceelasticityofelectricityconsumptiontodierbetweenautopayusersandnon-autopayusersinanInstrumentalVariable(IV)estimationofhouseholdelectricitydemand.Theresultsshowthatautomaticbillpaymentusershaveanelasticityofelectricitydemandthatis10%lower(inabsolutevalue)thannon-autopayusers.Also,usingadierence-in-dierenceframework,Ishowthatenrollinginautomaticbillpaymentsreducesconsumers'elasticityofelectricitydemandby5%.While Sexton ( 2015 )suggeststhatanautopayprogrammayreducepricesensitivity,tothebestofmyknowledgethispaperisthersttoestimatetheimpactofenrollinginautomaticbillpaymentsonpriceelasticity.Wealsond(contraryto Sexton 's( 2015 )ndings)thatthelowerpriceelasticityforautopayusersdoesnotnecessaryleadto 3Onlinebillingimpliesthatconsumersdonotreceivetheirbillinthemail.Theyreceiveanoticationintheire-mailthattheirbillisready,andtheycansecurelyviewtheircompletebilldetailsonline. 63

PAGE 64

increasedelectricityconsumption.Weshowthatnon-autopayusersreducetheirconsumptioninmonthsfollowingthereceiptofahighelectricitybill.However,autopayusershadlowerconsumptioninmonthsfollowingthereceiptofalowerelectricitybill.Theneteectcanbeambiguous.Thesendingsalsohaveimplicationsforenergyconsumptionpolicies.Thepricesalienceeectsofautopayprogramsinterferewiththeelectricityutility'seortofusingpricesignalstosteerenergyconsumption( Sexton , 2015 ).Itfurthermakesastrongpointfortheneedtoincreaseinformationprovisionoruseadvancedtechnologytohelpconsumersperceivetheiractualmarginalpriceofelectricity.IusealargepanelofmonthlyelectricitybillingdatafromGainesvilleRegionalUtilities(GRU)'ssinglefamilyresidentialhouseholdsfrom2009to2014.IalsouseGRU'sautomaticbillpaymentenrollmentinformationfrom2007through2014.Imatchedeachhouse'sparcelnumbertohousinginformationdatafromtheAlachuaCountyPropertyAppraiser(ACPA)website.IalsouseweatherinformationfromtheNationalOceanicandAtmosphericAdministrationwebsitewhichweusedtocalculatetheHeatingDegreeDays(HDD)andCoolingDegreeDays(CDD)foreachbillingperiodforeachhousehold.Theremainderofthepaperisasfollows:Section 3.2 givesabriefhistoryofautomaticbillpaymentprogramsandabackgroundofGRU'sautopayprogram.Section 3.3 providesatheoryofhouseholdelectricitydemandanddescribestheconceptualframework.Section 3.4 describesthedatasets.TheempiricalanalysisandresultsarepresentedinSection 3.5 .Section 3.6 concludes. 3.2ABriefonAutomaticBillPaymentandABackgroundtoGRU'sAutopayProgramAutomaticBillPayment,aconvenientmethodofpaymentinwhichacustomer'sbillistakendirectlyfromtheirbankaccounthasbecomeapopularpaymentsystemforthepaymentofalmostallmonthlyrecurringbillssincethelastdecade.Accordingtothe2010FiservConsumerBillingandPaymentstrendssurvey,thenumberofhouseholdsthatusesonline 64

PAGE 65

billpaymentintheUSincreasedabouteight-foldbetween2000and2010( Fiserv , 2010 ).Accordingtothesurvey,asof2010,80percentofallhouseholdswithInternetaccessuseonlinebankingwhile40percentofallhouseholdswithInternetaccessuseonlinebillpayment( Fiserv , 2010 ).Automaticbillpaymentshavebecomeparticularlypopularinthepaymentofcellphonebills,cableorsatellitebills,majorcreditcards,insurance,electricity,naturalgas,water,andalmosteveryrecurringmonthlybillortransaction.Oneprimaryimportanceofautomaticbillpaymentisthatitsavesconsumersthehassleofhavingtoschedulepaymenteverymonthforsuchrecurringpaymentssuchascarpaymentsorutilitybills.Theycanalsohelpcustomersavoidlatefeesduetoforgetfulness.Further,itgivesconsumersachoiceofavarietyofpaymentchannels.Italsosavestheconsumertimeofgoingthroughaclusterofpaperbillsandwritingcheckseverymonth.Vendorsorretailersandserviceprovidersalsoenjoybenetsofautopaysuchasreducedbillingtransactioncostandgreatercertaintyoftimelypayment( Sexton , 2015 ).IntheUS,almostallelectricutilitycompaniesnowallowandencouragetheircustomerstopaytheirbillthroughautomaticbillpaymentmethods.Somefurtherincentivizetheircustomerstouseautomaticbillpayments.Forexample,GainesvilleRegionalUtilitieswaivestheinitialdepositfornewresidentialcustomerswhosignupforautopay.Often,themajordownsidesforautomaticbillpaymentsarethatthecustomermightincurareturnedfeepaymentiftherewereinsucientfundsintheaccountoranoverdraftfeefromthebank.Thevendors,ontheotherhand,donotfaceanydownside.However,intheutilitysector,automaticbillpaymentmakesconsumerslesspricesensitive,whichactsagainsttheutility'sabilitytousepricesignalstosteerenergyconservation.InGainesville,enrollmentintheautopayprogramisopentoallGRUcustomerswithsatisfactorypaymenthistoryandallnewGRUcustomers.GRUstillsendscustomersbilldirectlytothemthroughthetraditionalmailore-billing.Toincreaseparticipation,GRUoers 65

PAGE 66

towaivetheinitialdepositfeeofabout$2704forallnewcustomerswhoenrollinautomaticbillpayments.Enrollmentwasinitiallyslowwithonly10%ofcustomersparticipatingintheprogram.However,becauseofthenancialincentive,about80%ofallnewcustomersenrollinautomaticbillpayments.Currently,morethanhalfofGRU'scustomersuseautomaticbillpayments 3.3ElectricityDemandandConceptualFrameworkManyutilitycompaniesuseanincreasing-blockpricingscheduleforelectricitydemand.Undertheincreasing-blockpricingschedule,electricitypricefollowsastepfunctionwiththemarginalpriceincreasingatspeciedlevelsofconsumption.Normally,thereisalsoaconstantupfrontfeeorcustomerchargethatdoesnotdependonthelevelofconsumptionandaconstantper-unitfuelcharge.Forexample,considerathree-tieredincreasingblockpricingschedulewhereconsumerspaymarginalperunitchargesofp1,p2,p3perkilowatt-hourforconsumptionintheintervals[0,x1],(x1,x2],(x2,1)respectively.Assumeaper-unitfuelchargeoffperkWhandaconstantcustomerchargeCperbill.Assumefurtherthataconsumerusedxunitsofelectricityduringtheperiod.Thentheconsumer'stotalelectricitybillisgivenby:B=8>>>><>>>>:C+p1x+fx,if0xx1C+p1x1+p2(x)]TJ /F9 11.955 Tf 11.96 0 Td[(x1)+fx,ifx1x2Figure1showsthegraphaGainesvilleRegionalUtilities'increasingblockpricingschedule.Thenonlinearpricingimpliesthatconsumersfaceanonlinearbudgetconstraint( Gabor , 1955 ; ReissandWhite , 2005 ; Actonetal. , 1980 ; HerrigesandKing , 1994 ).Whileeconomictheoryassumesthathouseholdsoptimizebasedonmarginalprices5, ReissandWhite ( 2005 ) 4ThisamountisthetotalfeeforallGRUservices.ThebreakdownofthefeeamongGRU'sservicesareasfollows:electricity-$145,gas-$50,water-$35,andwastewater-$40.5 Diamond ( 1998 )and Mirrlees ( 1971 )bothassumethatpeoplerespondthemarginaltaxrateofanon-lineartaxschedule. 66

PAGE 67

Figure3-1. Increasing-BlockResidentialElectricityPricingScheduleofGainesvilleRegionalUtilities(GRU),2009 250 500 750 0.102 0.067 0.028 PriceperKilowatt-hour($) Quantity(kWh) notesthat,\Thedemandbehaviorofautility-maximizingconsumerthusdependsnotontheaverageprice,noranysinglemarginalprice,butontheentirepriceschedule"( ReissandWhite , 2005 ).Nonetheless, ReissandWhite ( 2005 )stillusedmarginalpricesinestimatingpriceelasticitiesofelectricitydemandundernonlinearprices.Otherempiricalstudiessuchas Huang ( 2008 )inestimatingdemandforcellularphoneserviceundernonlinearpricingand Hausman ( 1985 )inapplyingchoiceunderuncertaintywithanonlinearbudgetsettoamodelofdisabilityinsurancestillusemarginalpricesorassumedconsumersrespondtomarginalprices.Thisoversimplicationontheuseofthemarginalpriceasthepricetowhichconsumersrespondinestimatingdemandwithnonlinearpricesisbecausetherehasnotbeenanyeconometrictechniquetoincorporatethewholenonlinearpriceschedulethatconsumersrespondtointoademandspecication( ReissandWhite , 2005 ).Incontrasttotheassumptionthatconsumersrespondtomarginalpricesusedinsomeofthestudies,recentempiricalevidence,however,showsthatconsumersrespondtotheaveragepriceofanonlinearpricingschedule.Forexamplesubjectsinanexperimentrespondtoaveragetaxrateinsteadofthemarginalrate( DeBartolome , 1995 ). Ito ( 2014 )showsthatconsumersrespondtotheaverage 67

PAGE 68

priceofthenonlinearpricingofelectricitywhile Borenstein ( 2009 )suggestthatelectricityconsumersareprobablyrespondingtotheexpectedmarginalpriceunderincreasing-blockpricing.Becausethedataconsistofelectricityconsumptionandpricinginformationfromonlyoneutilitycompanyandonlyoneutilityservicearea,thereisnotenoughvariationinthemarginalpricestoallowforaconsistentestimationofdemandelasticities.I,therefore,followthepredictionsoftherecentempiricalstudiesthatconsumersmayberespondingtoaverageprices.However,consumersrespondingtocurrentaveragepricingstillseemsatoddswithhowconsumersbehaveinreality.Theaveragepriceofanincreasing-blockpricingchangeswithrespecttothelevelofconsumption.Consumersrespondingtotheircurrentaveragepricesimplythatconsumersknowtheirexactlevelofconsumption,arefullyawareofthemarginalprices,andknowthecutopointsinthepricingschedule.\Itseemssafetosaythatnotonlydomostconsumersnotknowhowmuchpowerorwatertheyhaveusedsincetheirbillingperiodbegan,butmostconsumersalsodon'tknowwhentheircurrentbillingperiodbegan"( Borenstein , 2009 ).Consumersrespondingtothecurrentaveragepricefurtherassumesthatconsumershaveaperfectornearperfectideaabouttheirtotalconsumptionattheendofthebillingperiodorcananticipateanydemandshocktheymightfaceinthebillingperiod.Suchanassumptionagainseemstobeatoddswithreality.Weproposeabehavioralorconstrainedoptimizationmodelofhowelectricityconsumersrespondtononlinearpricinginwhichhouseholdsdonotknowthestartorendoftheirbillingperiodortheamountofusageattheendoftheperiodoratanypointintimeduringthebillingperiod.Weproposethatafterconsumersreceivetheirbill,dependingonthetotalamountoftheirbill,theymakebehavioraladjustmentstotheirlifestylethataecttheirelectricityconsumptioninthedaysfollowingtheirbillarrival.Thesebehavioraladjustmentsorrulesremaininplaceuntilconsumersreceivetheirnextbill,afterwhichtheyupdatetheirbehavioralrules.Ifconsumersreceivetheirbillattheendofeachbillingperiod,thenthe 68

PAGE 69

behavioralrulesbasedonthepreviousmonth'sbillwillaecttheelectricityconsumptionofonlythecurrentmonthsothatconsumersrespondtotheirpreviousmonth'sbillortheaveragepriceoftheirpreviousmonthbill.However,consumersnormallyreceivetheirpreviousmonth'sbillmid-wayintheircurrentbillingperiod.Thus,thebehavioralrulesthatconsumersmakebasedonthepreviousbill,onlyaectaportionoftheircurrentbillingperiodandaportionofthenextbillingperiod.Thisimpliesconsumersrespondtotheirtwo-monthlaggedaveragepriceuntiltheyreceivetheirpreviousmonth'sbill.Afterthereceiptoftheirpreviousmonth'sbill,theyswitchtorespondingtotheaveragepriceofthatbill.Figure 3-2 givesanillustrationofwhatlaggedaveragepriceconsumersmightrespondinthecurrentbillingperiod. Figure3-2. BillingPeriodsandBillArrivalTimes:AnIllustrationofWhatLaggedAveragePriceConsumersRespondDuringTheBillPeriod billingperiod billarrivaldate b0 b1 b2 b3 d0 d1 d2 d3 d0,d1,d2,d3representsthestartorendofbillingperiods.Billingperiodiextendsfromdateditodatedi+1.Billforbillingperiodiarrivesondatebiinbillingperiodi+1. Theconceptualmodelsuggeststhatconsumersmightrespondtoanexpectedvalueoftheirprevioustwobillingperiods'averagepriceinsteadofthecurrentperiod'saverageprice.Nevertheless,sincetheelectricitybillsarrivequicklyenough,consumersrespondtotheirone-monthlaggedaveragepriceformostdaysinthebillingperiod.6We,therefore,usethepreviousmonth'saveragebillasthepricetowhichconsumersrespond.UnderSection 3.5 ,wetestthismodelbyusinganencompassingtestofnon-nestedmodelstoinvestigatewhetherconsumersrespondtothecurrentaveragebillortheone-monthlaggedaveragebill. 6ForexampleinGRU'sservicearea,customersreceivetheirbillusuallybetween3to5daysaftertheirbillingperiodends. 69

PAGE 70

3.4DataWeusethefollowingvedatasetsforouranalysis.Table 3-1 givesalistofsomeofthevariablescontainedineachdataset. 1. GRU'smonthlyelectricityconsumptiondataforsinglefamilyhouseholdsfrom2008to2014. 2. GRU'shistoricratesandfueladjustmentdata. 3. GRU'sautomaticpaymentenrollmentdatafrom2007to2014. 4. HouseCharacteristicsandinformationdatafromtheAlachuaCountyPropertyAppraiser(ACPA). 5. WeatherinformationfromtheNationalOceanicandAtmosphericAdministrationwebsite. Table3-1. VariablesfromEachDataset GRUConsumptionACPADatasetGRU'sAutopayWeatherandRatesDatasetDatasetDataset ParcelNumberParcelNumberParcelNumberMin.Temp.MonthofconsumptionPhysicalAddressDateenrolledMax.Temp.YearofconsumptionYearbuiltDateexitedMonthlyconsumptionNumberofbedroomsServiceAddressNumberofbathroomsStartdateofbillingperNumberofStoriesEnddateofbillingperiodBaseAreasquarefootageMarginalpricesTotalAreasquarefootagecustomerchargeHeatedAreasquarefootagefueladjustmentrateSubdivisionGainesvilletaxratePoolownershipCountytaxrate 1.Min.Temp.istheminimumaveragedailytemperature2.max.Temp.isthemaximumaveragedailytemperature GRUprovidedthemonthlybillingdataforhouseholdsinGainesvillefrom2008to2014,whichIcomplementwithdataalreadyprovidedbytheProgramforEnergyEcientCommunitiesattheUniversityofFlorida.Imergedthetwodatasetstoformacomprehensivedatasetthatcontainsmoreinformationthanthetwoindividualdatasets.Forexample,monthlyconsumptiondatasetfromGRUdidn'thavethebillingperiodswhilethedatafromtheProgramforEnergyEcientCommunitiesdidn'thavethebillingaddresses.First,Imergedthe 70

PAGE 71

monthlyconsumptiondatasetwiththehousecharacteristicsdataobtainedfromtheAlachuaCountyPropertyAppraiserwebsite.Iselectedhousesthatweredesignatedassinglefamilyresidential.Theinitialelectricityconsumptiondatasetisapanelofelectricityconsumptiondatafor35,248singlefamilyresidentialhouseholdsoveraperiodof84monthsfrom2008to2014.Thedatasetcontainsinformationonhousehold'smonthlyelectricityconsumption,billingdates,andserviceaddress.Theautomaticbillpaymentenrollmentdata,alsofromGainesvilleRegionalUtility,containstheenrollmentandexitdatesforGRU'sautopayprogramfrom2007to2014.Thebasedatasetcontained21,174dierentparcels.However,thisdatasetalsocontainsenrollmentandexitdatesformultifamilyresidentialhouseholdsandapartments.Werestrictedtheautopaydatatoonlyhouseholdsintheconsumptiondataset.Therestrictedautopaydatasetcontains13138singlefamilyresidentialhouseholdsthatparticipatedinGRU'sautopayprogramfrom2007to2014.Somehouseholdsenrolledmultipletimesduringtheperiod.Forexample,twohouseholdsenrolled6timesduringtheperiod2007to2014.Ideletedhouseholdsthatenrolledmultipletimesfromthedatasetandalsofromtheelectricityconsumptiondataset.7Thenalautopayenrollmentdatasetcontained10,335singlefamilyhouseholdsthatparticipatedintheautopayprogramonlyonceduringtheperiod2008through2013.Alsoatotalof32,080households,with72monthsofbillingdataforeachhousehold,remainedinthemonthlyconsumptiondataset.Therewere,however,3,103missingmonthlybillinginformation,sothebasedatasetusedintheanalysiscontains2,926,565observations.TheGRU'shistoricratedata,extractedfromtheGRU'swebsite,containsinformationaboutGRU'spricingschedule,marginalpricesforthedierentconsumptionlevels,fueladjustmentrateforeachmonth,theconstantcustomercharge,Gainesvilletaxrate,andtheAlachuaCountytaxrate.Theinformationalsoincludedasummaryofhowtocalculatethe 7Multipleenrollmentdatessuggeststhatthebuildingchangedhandsseveraltimesduringtheperiod. 71

PAGE 72

electricitybillforeachhouse.Usingtheinformation,Icalculatedeachconsumer'sbillforeachbillingperiodandcomparedittothebillingdataalreadyobtainedfromGRU.ThedailyminimumandmaximumtemperatureforeachdayfromJanuary1,2008,toDecember31,2014wereobtainedfromtheNationalOceanicandAtmosphericAdministration(NOAA).Foreachday,thedailyminimumandmaximumtemperatureswereaddedtogetheranddividedbytwotondthedailyaveragetemperature.IthencalculatedtheHeatingDegreeDays(HDD)andtheCoolingDegreedays(CDD)foreachbillingperiodforeachhousehold.8 3.5EmpiricalStrategyandResultsLetqit9bedailyaverageelectricityconsumptionforhouseholdiduringbillingperiodt.Letpitbethepricethathouseholdirespondtoduringbillingperiodt.Thisprice,pit,canbethecurrentperiod'smarginalprice,thecurrentperiod'saverageprice,thelaggedaverageprice,orthelaggedmarginalprice.Weassumethatconsumersrespondtopricepitwithconstantelasticity,1.Wefurtherassumethattheconstantpriceelasticityisthesameacrossallhouseholdsandallbillingperiods,sothatthedemandforhouseholdiinperiodtcanbedescribedas: logqit=+1logpit+2wit+3Xi+4Xiwit+t+i+uit(3{1) 8Heatingdegree-daysforahouseholdinabillingperiodisthenumberofdegreesthatthemeantemperatureforadayfallsbelow650Fsummedoveralldaysinthebillingperiod.Coolingdegreedays,ontheotherhand,istheisthesumoveralldaysinabillingperiodofthenumberofdegreesthatthemeantemperatureisabove650F.Thus,ahousehold'scoolingdegree-daysforthebillingperiodbeginningattimet1andendingattimet2iscalculatedasCDD=Pt2t=t1maxfA(t))]TJ /F8 11.955 Tf 10.29 0 Td[(65,0gwhereA(t)istheaveragetemperatureofthedailymaximumandminimumtemperaturesmeasured0F.Heatingdegree-daysarecalculatedinasimilarwayasHDD=Pt2t=t1maxf65)]TJ /F9 11.955 Tf 11.95 0 Td[(A(t),0g.9Thedatacontainsthetotalhouseholdelectricityconsumptionforeachbillingperiod.Howeversincethedurationofthebillingperiodvariesacrossmonthsandhouseholds.Westandardizethedatabydividingthetotalconsumptionoverthebillingperiodbythedurationofthebillingperiodtondthedailyaverageelectricityconsumptionforeachbillingperiod. 72

PAGE 73

Table3-2. SummaryStatistics Variable ObsMeanStd.Dev.MinMax Monthlyusage2008(Kwh) 3740821069.45689.82115120Monthlyusage2009(Kwh) 3731421061.07690.93119720Monthlyusage2010(Kwh) 3743961107.13744.57120640Monthlyusage2011(Kwh) 3767061036.44698.73121680Monthlyusage2012(Kwh) 379301964.95621.71116280Monthlyusage2013(Kwh) 379343942.88609.82115360MarginalPrice($) 22591980.080.020.030.102Averageprice($) 22569470.150.210.1014.92435HDD(1000) 23097600.000.0000.0040CDD(1000) 23097600.000.0000.0064Autopay 23097600.180.3801Autopayever 320800.280.4501Bedrooms 320753.170.6615Bathrooms 320752.050.67110Stories 320801.120.4103TotalArea(1000squarefeet) 320802.391.060.4022.685HeatedArea(1000squarefeet) 320801.810.770.3714.844PoolOwnership 320800.100.3001AgeofBuilding(asof2013) 3208028.6911.420113 Autopayeverisadummyvariableequalto1ifahouseholdeverparticipatedintheAutomaticbillpaymentprogrambetween2008and2013. wherewitisavectorofweatherinformationforhouseholdiinperiodperiodt.witincludesthecoolingdegreedays(CDD)andheatingdegreedays(HDD)forhouseholdiinbillingperiodt.10iisthesubdivisionofhouseholdi'sbuilding.11trepresentsyearxed-eectsandcontrolsforunobservabledierencesinelectricityconsumptionacrossyears.ThedemandfunctioninEquation( 3{1 )canbeobtainedfromaquasilinearutilityfunctionsotherearenoincomeeectsofpricechanges. 10Sincehouseholdshavedierentbillingperiods,theCDDandHDDvariesforeachhouseholdforeachmonth.11TheAlachuaCountyPropertyAppraiserdenessubdivisionasthelegallyrecordednameofadevelopedarea.Itincludesparcelsorbuildingthatareclosetoeachotherandhavesimilarcharacteristics.Thereareovera1,500subdivisionsinGainesville.Addingsubdivisionasacontrolintheregressionfurthercontrolsfortheeectsofweatheronelectricityconsumptionsinceweathercandiersubstantiallyfromoneneighborhoodtoanothereveninthesamecity. 73

PAGE 74

Fromastandarddemandequation,pitisendogenousbecauseofthesimultaneitybetweenpricepitandquantityqit.Anothersourceofendogeneityisfromthefactthatthemarginalpricedependsonthelevelofconsumption.Theaveragepriceisalsoobtainedbydividingthetotalbillamountbythetotalquantityontheleft-hand-sideofEquation 3{1 andthatpresentsanothersourceofendogeneity.OrdinaryLeastSquare(OLS)estimation,therefore,leadstoinconsistentestimatesofthepriceelasticityofdemand.Hence,weestimateEquation( 3{1 )usinginstrumentalvariableestimation.Weusethetwo-stageleastsquareestimationmethod.ValidinstrumentsshouldbecorrelatedwiththepricepitbutuncorrelatedwiththeerrortermofEquation( 3{1 ).Wefollow Ito ( 2014 )inourchoiceofinstruments.Weuseapolicy-inducedpricepit(~qit),alsocalledsimulatedinstrumentasourinstrumentalvariableforpit.Thisinstrument,pit(~qit),computesthepredictedpriceinperiodtforaconsumptionlevel~qit.Tobeavalidinstrument,~qitshouldnotbecorrelatedwithuit.Wecouldusetheconsumptionofhouseholdiinthesameperiodinthepreviousyearqit)]TJ /F10 7.97 Tf 6.59 0 Td[(12for~qit.Hence,oursimulatedinstrumentforpitisthepredictedprice,usingthecurrentyear'spricescheduleontheconsumptionofthesamebillingperiodinthepreviousyear.Thatis,oursimulatedinstrumentforpitiswhatthepricewouldhavebeenhadconsumersusedthesameamountofelectricityastheydidinthesamebillingperiodofthepreviousyear.SinceGRUchangesitspricescheduleinOctoberofeveryyear,suchaninstrumentwouldensurethattheinstrumentandthecurrentperiod'saveragepricearedierentevenwhenconsumptioninthetwoperiodsisthesame.Inthecasewhenconsumptioninthesamemonthofthepreviousyearisthesameastheconsumptioninthecurrentbillingperiod,thedierencebetweenthecurrentperiod'saveragepriceandtheinstrumentisinducedbythepolicyofpriceschedulechange.But,asnotedby( Ito , 2014 ),qit)]TJ /F10 7.97 Tf 6.59 0 Td[(12islikelytobecorrelatedwithuitbecauseofmeanreversionofconsumption.Wethereforeusedaquantitylevelqit)]TJ /F10 7.97 Tf 6.59 0 Td[(13for~qitsothatthepolicy-inducedpriceforperiodt,pit(qit)]TJ /F10 7.97 Tf 6.59 0 Td[(13)isoneinstrumentforpit.Sincethetwoperiodst,andt)]TJ /F8 11.955 Tf 12.78 0 Td[(13areonlyonemonthapart,theyarelikelytobelongtothesameseasonand 74

PAGE 75

havesimilarelectricityusageforeachhousehold.Understableweatherconditionsthedierencesinconsumptionbetweenthetwoperiodsisinducedbythepolicychange.Weaddanotherinstrumentsbychoosingqit)]TJ /F10 7.97 Tf 6.59 0 Td[(6for~qit.tandt)]TJ /F8 11.955 Tf 11.97 0 Td[(6arelikelytohavethesamepriceschedulebutarefromdierentseasons.Hence,pit(qit)]TJ /F10 7.97 Tf 6.59 0 Td[(6)istheprice(current)fortheweatherinducedconsumptioninperiodt)]TJ /F8 11.955 Tf 12.58 0 Td[(6.Usingthesetwoinstrumentsimplythatourmodelisover-identied. EncompassingTestofCurrentAveragePriceversusLaggedAveragePriceWhileourconceptualframeworkinSection 3.3 predictsthatconsumersrespondtolaggedaveragepriceinsteadofthecurrentaverageprice,undertheempiricalsectionweallowedforthepossibilitythatconsumersmightrespondtothecurrentaveragepriceorboththecurrentandtheone-monthlaggedaverageprice.WeestimateEquation( 3{1 )usingthecurrentperiod'saveragepriceasthepriceconsumersrespond.Next,weestimatethesameequationusingthepreviousmonth'saveragepriceasthepriceconsumersrespond.Thusweestimatethefollowingtwomodels: logqit=0+logAPit+2wit+3Xi+4Xiwit+t+i+uit(3{2)and logqit=0+logAPit)]TJ /F10 7.97 Tf 6.58 0 Td[(1+2wit+3Xi+4Xiwit+t+i+uit(3{3)whereandaretheelasticitieswithrespecttothecurrentaveragepriceandtheone-monthlaggedaveragepricerespectively.Next,usingtheencompassingtestfornonnestedmodelssuggestedby MizonandRichard ( 1986 ),wetestthetwononnestedmodelsagainsteachotherbyconstructingacomprehensivemodelthatcontainseachmodelasaspecialcaseandtestingtherestrictionsthatwouldleadustoeachmodel( Wooldridge , 2012 )and( Ito , 2014 )).Ourcomprehensivemodelisgivenby: 75

PAGE 76

logqit=+logAPit+logAPit)]TJ /F10 7.97 Tf 6.59 0 Td[(1+2wt+3Xi+4Xiwit+t+i+uit(3{4)InestimatingthecomprehensiveEquation 3{4 ,weaddanotherpolicy-inducedinstrumentbychoosinganotherconsumptionlevel,qit)]TJ /F10 7.97 Tf 6.59 0 Td[(11for~qit.TheresultsoftheestimationusingEquations 3{2 , 3{3 ,andEquation 3{4 arerepresentedinTable 3-3 .ColumnIofthetableshowstheresultsoftheestimationfromEquation 3{2 whenthecurrentaveragepriceisusedasthepriceconsumersrespondwhileColumnIIshowstheresultswhenlaggedaveragepriceisused.ColumnIIIshowstheencompassingtestresultsusingthecomprehensiveEquation 3{4 .Inallthreeregressions,standarderrorswereclusteredatthehouseholdleveltocorrectforserialcorrelation.Thersttwocolumnsshowthatconsumersrespondtoboththecurrentandtheone-monthlaggedaveragepricewithpriceelasticitiesof-0.65and-0.56respectively.Theseelasticitiesarestatisticallysignicantevenatthe1%level.TheencompassingtestresultinColumnIIIshowsthatthecoecientonlaggedaveragepriceisstillsignicantatthe1%level,butthecoecientoncurrentaveragepricebecomespositiveandstatisticallyinsignicant.Thatis,thecoecientofcurrentaveragepricebecomesstatisticallyinsignicantoncethelaggedaveragepriceiscontrolled.Henceconditionalonthelaggedaverageprice,thecurrentaveragepricedonotaectcurrentelectricityconsumption. Self-SelectionIntoAutopay?Theautopayprogramisvoluntaryhenceparticipationintheprogrammaybebasedonthephysicalcharacteristicsofthehouseholdandthehouse.Householdswithhigherincomes,forexample,aremorelikelytoparticipateintheautopayprogramthanlowerincomehouseholds.Thisisbecauselowerincomehouseholdsaremorelikelytohaveinsucientbalanceintheirbankaccount,soaremoresusceptibletofeesonreturnedchecks.Whilehouseholdincomeisnotpresentinthedata,householdincomeisknowntobecorrelatedwiththesizeandsubdivisionofthebuilding.HenceIexpectnumberofbedroomsandtotalareasquarefootagetohaveapositiveeectsofautopayparticipation.AlsosinceGRUoerstowaivetheinitial 76

PAGE 77

Table3-3. EncompassingTest:CurrentVs.One-MonthLaggedAveragePrice log(dailyusage) IIIIII log(AP) -0.6443***0.1506 (-7.64)(1.39)log(AP)]TJ /F10 7.97 Tf 6.59 0 Td[(1) -0.5526***-0.5975*** (-6.29)(-7.66)CDD 1359.0720***1369.4750***1373.7198*** (44.74)(47.97)(47.87)HDD 84.8559111.5930*120.6977** (1.71)(2.41)(2.59)HeatedArea 0.2369***0.2407***0.2398*** (24.74)(25.22)(25.30)Stories 0.01300.01550.0158* (1.61)(1.94)(1.98)Bedrooms 0.0292***0.0312***0.0322*** (4.35)(4.65)(4.75)Bathrooms 0.0453***0.0408***0.0402*** (5.37)(4.90)(4.81)age -0.0008-0.0009-0.0010 (-1.35)(-1.60)(-1.81)pool 0.2917***0.2931***0.2919*** (27.77)(28.23)(28.30)cons 1.0380***1.2049***1.4100*** N 181736018129221805575R2 0.33990.31260.3029 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. Thecoecientsonthesubdivisiondummies,theyeareects,andtheinteractiontermbetweenCDD,HDD,andthehouseholdcharacteristicsareleftoutfromthetable. depositfeeforallnewservices,householdsinnewerbuildingsareexpectedtoparticipateintheautopayprogrammorethanhouseholdsinolderbuilding.Dierenthouseholdsparticipatedintheautopayprogramondierentdates.Thus,thereisnoparticulardateofautopayprogramparticipation.Ithereforecreatedanewdummyvariable,autopayever,equalto1,ifahouseholdparticipatedintheautopayprogramirrespectiveofdateofparticipation.AutoypayeverequalszeroforhouseholdsthatneverparticipatedintheautopayprogrambetweentheJanuary2007throughDecember2008.Thisvariableinusedasthedependentvariableinestimatingtheprobabilityofautopayprogramparticipation.Theprobabilityofparticipatingthetheautopayprogramisassumedtodependonageof 77

PAGE 78

thebuilding,totalareasquarefootage,numberofstories,numberofBedroomsandpoolownership.Sincetotalannualelectricityusageiscorrelatedwithincome,Iincludedtotalannualusagein2010.Theprobitregressionequationfortheestimationisgivenby: Autopayeveri=0+1X+j+ui(3{5)whereXiisavectorofhousecharacteristicswhilejisasubdivisiondummytocontrolforthesubdivisionofthehouse.uiistheidiosyncraticerrortermthatisassumedtohaveanormaldistributionwithmeanzero.TheresultsoftheprobitregressioningiveninTable 3-4 . Table3-4. WhatAectsParticipationinAutopay autopayever III AnnualElectricityUsagein2010(Kwh) -0.0211***-0.0241*** (-14.73)(-18.71)TotalAreaSquarefeet(1000sq.feet) 0.1382***0.2648*** (8.98)(24.42)Stories -0.0328-0.0128 (-1.59)(-0.67)Bedrooms -0.0230-0.0896*** (-1.37)(-6.12)AgeofBuilding -0.0075***-0.0041*** (-5.13)(-5.68)Pool 0.01670.0498 (0.59)(1.85)Cons -0.3281-0.5102*** (-1.41)(-10.77) N 3143232075PseudoR2 0.07750.0271 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesisThecoecientsonsubdivisionsinColumnIhavebeenleftoutfromthetable. ColumnIofTable 3-4 showstheresultsfromEquation 3{5 whileColumnIIshowstheresultsfromthesameequationbutwithoutthesubdivisionxedeects.WhileIexpectsubdivisionstoaectautopayparticipationbecausethesubdivisionsofone'shouseisrelatedtohisincome,ColumnIIwasaddedtoshowthattheresultsoftheestimationisnotprimarilydrivenbytheinclusionbythesubdivisionxedeects.ThemaindierentbetweenthetwocolumnsisthehigherpseudoR2inColumnI.Thecoecientsandtheirsignsare 78

PAGE 79

similarinthetwocolumns.TheonlyexceptionisthecoecientonBedroomswhichisnotstatisticallysignicantinColumnIbutstatisticallysignicantinColumnII.TheresultsfromColumnIshowthata1000squarefeetincreaseinthetotalareaincreasestheprobabilityofprogramparticipationby0.1points.Thecoecientonbedroomsis,however,notstatisticallysignicant.Asexpected,theageofthehousereducestheprobabilityofautopayparticipation.A10yearincreaseintheageofthehouseincreasestheprobabilityofprogramparticipationby0.08points.GRUwaivestheinitialdepositfeefornewcustomerswhoopttoparticipateinautopay.Sincenewhomesrequiresnewservice,householdsinnewhomesaremorelikelytoparticipateintheautopayprogramthanhouseholdsinolderhouses.Surprisingly,annualelectricityusageisnegativelyrelatedtotheprobabilityofautopayprogramparticipation.A1000kwhincreaseinannualelectricityusagedecreasesprobabilityofparticipatingby0.1points.Thus,largeresidentialenergyconsumersarelesslikelytoparticipateintheautopayprogram.Thisresultinrobustirrespectiveoftheyearusedforcalculatingtheannualelectricityusage. AutomaticBillPaymentEectsonPriceSalience-TheoreticalMotivationLetlogq=+logp+ybethedemandfunctionelectricityoranygoodwhenconsumersperfectlyperceivetheprice.qisthequantityofthegoodconsumed,pisthepriceandyinincome.Weassumeaquasilineardemandfunctionofthegoodonlyforsimplicity.Letbeaninattentionparameter,sothatinsteadofconsumersobservingprice,p,theyperceiveprice(1+)p,where2[)]TJ /F8 11.955 Tf 9.3 0 Td[(1,1].Thisimplythatconsumersperceiveahigherthanactualpricefor2(0,1],andapricelowerthanactualpricefor2[)]TJ /F8 11.955 Tf 9.3 0 Td[(1,0).Thedemandequationwithinattentionisthereforegivenby: logq=+log(1+)p+y(3{6)Thepriceelasticityofdemandunderfullattention(=0)is.whilethepriceelasticityofdemandunderinattentionisgivenby: 79

PAGE 80

=@logq @log(1+)p@(1+)p @p=(1+)(3{7)Thus,thepriceelasticityunderinattentioncouldbegreaterorlessthanthepriceelasticityunderfullattentiondependingontheinattentionparameter.Howeversincepriceshaveatendencytobeadjustedupward,theinattentionnormallyimplythatconsumersperceivealowerpricethanitactuallyissothat<0.Insuchacase,thepriceelasticityoftheinattentiveconsumerwouldbelower(inabsoluteterms)thanthatoftheattentiveconsumer.Ourempiricalanalysisconsidersautomaticbillpaymentasaproxyofinattention.Itthereforeassumesthatnon-autopayusershavefullattentionwhileautopayusersmaketheirconsumptiondecisionwithsomedegreeofinattention.Thusweassume,withrespecttoEquation 3{7 ,thatnon-autopayusershavepriceelasticitywhilenon-autopayusershavepriceelasticity(1+).Inthecasethatautomaticbillpaymentdoesnotcauseinattention,thetwoelasticitieswouldbethesame,thatis=0.Thus,ournullhypothesisofnopricesalienceeectsofautomaticbillpaymentis=0. AutomaticBillPaymentEectsonPriceSalienceTheprevioussubsectionsuggestthatconsumersrespondtotheirlaggedaveragepriceinsteadoftheircurrentaverageprice.Thereasoningbehindthisisthatconsumershavelessinformationorareinattentivetotheirlevelofconsumptionineachperiod.Moreover,mostconsumersdonotknowthecut-opointsofthepricescheduleorthestartorendoftheirbillingperiod.Thus,consumersdonotrespondtotheirmarginalpriceortheircurrentaveragepricebutmakebehavioralrulesaboutconsumptiononlyafterreceivingandobservingtheirbillsothattheyrespondtotheirone-monthlaggedaverageprice.Animplicationofthisresultisthatconsumerswhoinspecttheirbillcarefullywillrespondmoretopricesthanconsumerswhoonlytakeacursorylookattheirbillorfailtoexaminetheirbill.Theelectricitybillofautopayusersisdeductedfromtheirbankaccountevenbeforetheyreceivetheirbill,somostautopayusersdonotbothertoreviewthechargessincetheydonothavetowriteacheck.Hence,automaticbillpaymentsincreasetheprobabilitythatuserswillforgobillinspections, 80

PAGE 81

whichincreasesinattentiontothelaggedaverageprice,andmakeconsumersrespondlesstoprices.Utilitycompaniesfurtherencouragesautopayuserstoparticipateinonlinebillingtocomplementtheirenrollmentinautopay.Participatinginbothprogramsincreasesfurthertheprobabilitythatconsumerswillignoretheirbillandthusrespondlesstopricesthannon-autopayusers.Wenowinvestigatewhetherautopayusersarelesspriceelasticthannon-autopayusers.While( Sexton , 2015 )suggeststhatanautopayprogrammayreducepricesensitivity,tothebestofourknowledgeourpaperisthersttoestimatetheimpactofenrollinginautomaticbillpaymentsonpriceelasticity.UsingthedemandEquation( 3{3 )weestimatetheelasticityofdemandforelectricityconsumers.Weallowedthepriceelasticityofdemandtobedierentforthegroupwithautopayandthegroupwithoutautopay.Weaddedadummyvariableautopaydenedas: autopayit=8><>:=1,ifhouseholdiisenrolledinautopayintimeperiodt=0,otherwise(3{8)WemodifyEquation( 3{3 )toobtaintheequation, logqit=0+logAPit)]TJ /F10 7.97 Tf 6.59 0 Td[(1+autopayitlogAPit)]TJ /F10 7.97 Tf 6.58 0 Td[(1+2wt+3Xi+uit(3{9)TheresultsoftheestimationusingEquation(3{9)ispresentedinTable 3-5 .Inallcolumnsofthetable,Iusedthelaggedaveragepriceasthepriceconsumersuse.ThecoecientofinterestisthecoecientonlogAP)]TJ /F10 7.97 Tf 6.58 0 Td[(1autopay.Thecoecientispositiveandstatisticallysignicantacrossallthecolumns.InColumnsIandII,onlythepriceinassumedtobeendogenous.Autopayassignmentisassumedtoberandom.ColumnIfurtherassumesthataftercontrollingfortheeectsofautopayonprice,theautopayassignmentdoesnotaectelectricityconsumption.Thatis,autopayonlyaectselectricityconsumptionthroughit'seectsonthepriceelasticitysothataftercontrollingfortheeectsofautopayonprice 81

PAGE 82

elasticity,theautopayassignmentitselfdoesnotaectelectricityconsumption.Equation 3{9 ,therefore,doesnotexplicitlyincludeautopayasanindependentvariableexcepttheinteractiontermbetweenautopayandtheprice.ThisassumptionishoweverrelaxedinColumnII.Weallowedautopaytohaveadirecteectonelectricityconsumptionevenaftercontrollingforitseectsonthepriceelasticity.WeaddedautopayasanindependentvariableinEquation 3{9 toobtainEquation 3{10 : logqit=0+logAPit)]TJ /F10 7.97 Tf 6.59 0 Td[(1+autopayit+autopayitlogAPit)]TJ /F10 7.97 Tf 6.58 0 Td[(1+2wt+3Xi+uit(3{10)ColumnI,thus,showstheresultsoftheregressionformEquation 3{9 whileColumnIIshowstheregressionfromEquation 3{10 .TheresultsfromColumnIshowsthatwhilethepriceelasticityonnon-autopayusersisabout-0.58,thepriceelasticityofdemandfortheautopayusersis-0.53.Thatis,thepriceelasticityofdemandfortheautopayusersisabout10%lowerthanthatofthenon-autopayusers.ColumnIIshowsamorepronounceddierenceinpriceelasticitiesbetweenautopayandnon-autopayusers.Whilethepriceelasticityofnon-autopayusersis-0.65,autopayusershaveapriceelasticityof-0.1312.Thatis,thepriceelasticityoftheautopayusersisabout78%lower(inabsoluteterms)thanthatofthenon-autopayusers.ThecoecientonautopayinColumnII,however,showsthatanautopayuserconsumesabout88%moreelectricitythanacomparablenon-autopayuser.Thisresultislikelytobeoverestimatedsincetheautopayassignmentisnotrandom.Highincomeearners,forexample,whohavehigherelectricityconsumptionthanlowincomeearners,aremorelikelytoparticipateinanautopayprogramthanlowincomeearners.Onedrawbacktoautomaticbillpaymentenrollmentisthatitrequiresconsumerstohavesomeminimumamountofmoneyintheirbankaccount 12ThepriceelasticityofautopayusersisthesumofthecoecientsonlogAP)]TJ /F10 7.97 Tf 6.59 0 Td[(1andlogAP)]TJ /F10 7.97 Tf 6.59 0 Td[(1autopay 82

PAGE 83

everymonth.Whileautomaticbillpaymentsavescustomersfromlatepaymentfees,overdraftfeesfrombanks,whencustomershaveinsucientfundscanbemorecostlythanthesavingsonlatefees.Thisisparticularlytrueforpoorcustomerswhoaremorelikelytooccasionallyhaveinsucientfundstocovertheirbills.Thefearofoverdraftfeesmeanslowincomepeople(whoareusuallylowelectricityusage),arelesslikelytouseautopaythanhighincomeearners.Autopayis,therefore,likelytobeendogenoussincethemissingvariableintheregression,income,iscorrelatedwithautopay.In Boampong ( 2014 ),Imappedeachhouseholdtothecensustracttowhichtheybelongedandusedtheaverageincomeofthecensustractasanimputedincomeforeachhousehold.Thisapproach,however,didnotgiveaconsistentestimateofthecoecientoftheincomevariablebecausethecensustractswerelargeanddidnotallowenoughvariationintheincomevariable.Thesubdivision/neighborhoodvariableinEquations 3{9 and 3{10 whilecontrollingforthedierencesinenergyconsumptionasaresultoftheslightdierencesinweatheracrossneighborhoods,alsoserveasaproxyvariableforincome.TheneighborhoodvariableusedinEquations 3{9 and 3{10 isthereneddenitionofneighborhoods/subdivisionsusedbytheAlachuaCountypropertyAppraiser.AccordingtotheAlachuacountypropertyAppraiserwebsite,subdivisionsarethemostcommonwayofdescribingapropertylocationandcontainshouseholdbasedonproximity.ThesubdivisionsareusuallynotdierentfromtheareadesignationalsousedbytheAlachuaCountyPropertyAppraiser(ACPA).TheACPAdenesareaasagroupofparcelshavingsimilarlycharacteristicsconcerningassessment,oneofwhichmaynotbethephysicallocationorsubdivision.However,inmostcasestheareaandthesubdivisionarealmostthesameorthathousesinthesamesubdivisionsarenormallyinthesamearea.Thesehousesnotonlyhavesimilarcharacteristicsconcerningassessment,butalsoareinhabitedbypeopleofsimilarincomes.Thuscontrollingforsubdivisionseectivelycontrolsforincomelevels.Also,dierenthouseholdsusuallyhavedierentbillingperiodsbasedonwhentheirelectricitymeterisread.Whenahousehold'sbillingmeterisread,theirbillingperiodcloses 83

PAGE 84

andanewbillingperiodstarts.Sincethebillingmeterisreadondierentdaysfordierenthouseholds,the\monthly"electricityconsumptionofhouseholdshavedierentdatesofconsumption.Inthedataset,householdsinthesamesubdivisionhadthesamebillingperiod.Addingasubdivisionsdummyfurthercontrolsforcohortseectsonelectricityconsumption.InColumn(III)oftheTable 3-5 ,IcombinedtheanalysisinColumnsIwithaCoarsenedExactMatchingApproachinordertosolveanypossibleendogeneityresultingfromautopayparticipationevenaftercontrollingforsubdivisions.13.Thepurposeofthematchingistopairautopayhouseholdswithnon-autopayhouseholdsbasedoncharacteristicsthataectautopayadoption.First,Igroupedhouseholdsintoparticipantsandnon-participants.Allhouseholdsthatparticipatedintheautopayprogramatleastonceintheperiodwereconsideredasautopayparticipantsandthosethatneverparticipatedintheautopayprogramwereconsideredasnonparticipants.Ithenperformacoarsenedexactmatchingbasedontheparticipants/non-participantsassignment.TheideaoftheCoarsenedExactMatchingistotemporarilygroupeachvariableintomeaningfulstrataandpairprogramparticipantstonon-participantswhobelongtothesamestrataonallcoarsenedvariables.14Theoriginal(uncoarsened)variablesarehoweverretainedforanalysis.Ichoosecharacteristicsthataremorelikelytoaectahousehold'sparticipationinautomaticbillpayment.Thusweperformedtheexactmatchingonsubdivisions(controlsforincomeandneighborhoodeects),numberofbedrooms(controlsforthesizeofthesize),andtheagegroupofthehouse.15 13See Boampong ( 2014 )or Iacusetal. ( 2008 )foradescriptionofCoarsenedExactMatching14Notallvariablesneedtobecoarsened,somevariablescanberestrictedfromcoarsening.15Wedividedallhousesinthedatabasedonwhethertheywerebuildbefore2002orafter2002.ThisisbecauseFloridaincreasedthestringencyofitsbuildingcodesin2002whichhavebeenshowntoreduceenergyconsumptionby4%( JacobsenandKotchen , 2013 ). 84

PAGE 85

Householdsinthesamegroup(bothparticipantsandnon-participants)havesimilarcharacteristicsonthefactorsthataectautopayparticipation.TheeectsoftheautopayprogramisthenestimatedusingonlythematchedsamplefromtheCEMmethodology.Suchanestimationapproachis\arguablymoreappropriatecomparedtoasimpleinstrumentalvariableapproach(fordealingwiththeselectionbias16)asnostrongexclusionrestrictionsareneeded"( GirmaandGorg , 2007 ).TheresultsinColumn(III)includesautopayasanindependentvariable,butitisstatisticallyinsignicant.Thisisexpectedasthematchingpairsparticipantsandnon-participantswithsimilarcharacteristicssothatanyimbalancebetweenthetwogroupsinthesamematchedcellissmallornegligible.TheresultsfromColumn(III)showsthatwhilethepriceelasticityofdemandis-0.5fornon-autopayhouseholds,thepriceelasticityofdemandfortheautopayhouseholdis-0.07.Thatis,thepriceelasticityoftheautopayhouseholdsisabout86%lower(inabsoluteterms)thanthatofthenon-autopayhouseholds. DoesEnrollinginAutomaticBillPaymentMakeConsumersLessPriceSensitive?Theprevioussubsectionrevealsthatautopayusersarelesspriceelasticthannon-autopayusers.Inthissubsection,wefurtherinvestigatewhetherenrollinginautomaticbillpaymentsmakesconsumerslesspricesensitivebycomparingthepriceelasticityofautopayusersbeforeandafterenrollinginautomaticbillpayments.Thedierenceinthepre-enrollmentelasticityofdemandandthepost-enrollmentelasticityoftheautopayusersiscomparedtothatofacontrolgroupinadierence-in-dierenceestimation.Treatmentinthiscaseisenrollinginanautomaticbillpayment.Householdsinthedataenrolledintheautopayprogramatdierenttimesfrom2007to2014.Tousethedierence-in-dierencemethodologywedenednewtimevariables~tiand~i.Lettibethetimethathouseholdienrolledintheautopayprogramthen,~tiisthenumberofbillingperiodsormonthselapsedsincehouseholdienrolledintheautopayprogram,denedas~ti=t)]TJ /F8 11.955 Tf 11.57 1.62 Td[(ti.~iisadummyvariableequaltooneforthersttwelvemonths 16Selectionbiasoccurswhenparticipationisaprogramisnotrandomanddependsonsomeobservableorunobservablecharacteristicsthatarecorrelatedwiththeoutcomeofinterest. 85

PAGE 86

Table3-5. AutomaticBillPaymentEectsonPriceElasticity log(dailyusage) IIIIII logAP)]TJ /F10 7.97 Tf 6.58 0 Td[(1 -0.5787***-0.6441***-0.5047*** (-6.66)(-7.28)(-5.06)logAP)]TJ /F10 7.97 Tf 6.58 0 Td[(1 0.0563***0.5073*0.4297* (17.97)(2.45)(1.96)Autopay 0.8760*0.7296 (2.17)(1.71)CDD 1370.6753***1375.9023***1398.8295*** (48.02)(48.00)(46.54)HDD 108.8295*111.8534*40.3222 (2.35)(2.41)(0.81)HeatedArea 0.2444***0.2447***0.2426*** (25.58)(25.68)(27.17)Stories 0.01440.01400.0202** (1.81)(1.77)(2.62)Bedrooms 0.0299***0.0300***0.0254*** (4.47)(4.48)(3.38)Bathrooms 0.0423***0.0424***0.0426*** (5.11)(5.11)(5.01)Age -0.0010-0.0010-0.0000 (-1.83)(-1.83)(-0.03)Pool 0.2906***0.2899***0.3020*** (28.03)(28.03)(31.42)cons 1.1774***1.0491***1.2949*** (6.28)(5.52)(6.11)N 181292218129221653691R2 0.31600.31610.3153 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. Thecoecientsonthesubdivisiondummies,theyeareects,andtheinteractiontermbetweenCDD,HDD,andthehouseholdcharacteristicsareleftoutfromthetable. afterhouseholdienrolledinautopayandequaltozeroforthe12monthsbeforehouseholdienrolledinautopay.Thatis ~i=8><>:=1,if0~ti<12=0,if)]TJ /F8 11.955 Tf 11.96 0 Td[(12~ti<0(3{11)Thusforeachautopayuser,weusedthe12monthsbeforetheyenrolledinautopayasthepre-treatmentyearsandthe12monthsafterenrollmentintheprogramasthepost-treatmentyears.Forautopayuserswhowereintheprogramforlessthan12months,wechoosethe 86

PAGE 87

pre-treatmentperiodstocorrespondtothesamemonthsasthetreatmentperiodsbutinathepreviousyear.Nextweassignedthebillingperiodsofthecontrolgroup(householdswhoneverenrolledinautopayprogramsduringtheperiod2008-2013)topre-treatmentandpost-treatmenttimes.Weusedthemediandateofenrollmentintheprogramforthetreatedgrouptodividethebillingperiodsofthecontrolgroup.Thusforeachnon-treatedhousehold,allbillingperiodsbeforethemedianenrollmentdatewasassignedtopre-treatmentwhileallbillingperiodsafterthemedianenrollmentdatewasassignedtopost-treatment.WemodifyEquation( 3{9 )toobtainthedierence-in-dierenceestimationEquation( 3{12 ). logqit=0+0logpit+1autopayitlog(pit)+2~ilog(pit)+3~iautopayitlog(pit)+2wit+3Xi+4Xiwit+t+i+uit(3{12)Thecoecientonlogpit,0measuresthepre-treatmentpriceelasticityofdemandforthenon-autopaywhilethecoecientonautopayitlog(pit),1,measuresthedierenceinpriceelasticitybetweennon-autopayusersandtheautopayusersbeforetreatment.Thus,thepriceelasticityoftheautopayusersbeforetheyenrolledintheautopayprogramis0+1.Thecoecienton~ilog(pit),2,measuresthechangeinpriceelasticitybetweenthepre-andpost-treatmentyearsthatiscommontoboththetreatedgroupandthecontrolgroup.Ourcoecientofinterestis3,thecoecienton~iautopayitlog(pit),whichisthedierence-in-dierenceestimate.Itmeasurestheadditionalchangeinelasticityfortheautopayusersafterenrollingintheautopayprogram. Table3-6. SummaryofDierence-in-DierenceEstimates PriceElasticities BeforeAfter After-Before Control(Non-autopayusers) 00+2 2Treated(Autopayusers) 0+10+1+2+3 2+3 Treated-Control 11+3 3 87

PAGE 88

Table 3-7 givestheresultsofthedierence-in-dierenceestimation.Thetableshowsonlythedierence-in-dierenceestimatespresentedinTable 3-6 .Theresultsusingtheone-monthlaggedaverageprice(ColumnI)andthecurrentaverageprice(ColumnII)areverysimilar.ColumnIofthetableshowsthatbeforetreatment,thepriceelasticityofdemandforthenon-autopayuserswasabout-0.96whilethepriceelasticityofdemandforthetreatmentgroupwas-0.92.Aftertreatment,thepriceelasticityofthecontrolgroupincreasedby0.084pointsto-1.04,whilethatofthetreatedgroupincreasedby0.034pointsto-0.96.Thecounterfactualpriceelasticityforthetreatmentgroupisintheabsenceoftreatmentis-0.96-0.0834+0.049=-1.01.Enrollinginautomaticpayment,therefore,reducedpriceelasticitybyapproximately5%. Table3-7. EectsofEnrollinginAutomaticBillPaymentonPriceElasticity log(dailyusage) III logAP)]TJ /F10 7.97 Tf 6.58 0 Td[(1 -0.9636*** (-13.31)logAP)]TJ /F10 7.97 Tf 6.58 0 Td[(1~ -0.0837*** (-7.49)logAP)]TJ /F10 7.97 Tf 6.58 0 Td[(1autopaygroup 0.0350*** (4.55)logAP)]TJ /F10 7.97 Tf 6.58 0 Td[(1~autopaygroup 0.0491*** (4.85)logAP -0.9439*** (-14.11)logAP~ -0.0815*** (-7.21)logAPautopaygroup 0.0366*** (4.73)logAP~autopaygroup 0.0502*** (4.90) N 311463312576R2 0.31100.3665 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 3.6ConclusionInthispaper,weanalyzedwhetherconsumerrespondtothecurrentpriceortheone-monthlaggedaveragepricewhentheyfaceanincreasing-blockpricescheduleusedby 88

PAGE 89

manyutilitycompanies.Usingtheencompassingtest,wefoundsupportfortheideathatconsumersrespondtotheone-monthlaggedaveragepriceinsteadofthecurrentaverageprice.Thishaspolicyimplicationsforutilityratedesignastheempiricalpriceelasticityplaysaroleinutility'sratedesignandthechoiceofpriceandnon-pricedemandstrategies.Italsohelpstoinformpricespecicationdebateastowhichpriceofthenonlinearpricingscheduleconsumersrespond.Currentempiricalresearchshowsthatconsumersrespondtotheaveragepricesoftheincreasingblockpricingschedule,ratherthantothemarginalpriceastheorypredicts.Themainreasonsusuallycitedforwhyconsumersrespondtothemarginalpricesarethatmarginalpricesarenotworththetroubletolearntheratescheduleandtrackconsumptionlevels.However,respondingtothecurrentaveragepricesalsorequiregreatdealofeortfromconsumers:itstillrequiresconsumerstoknowthepricescheduleorattheleastknowexactlywhentheirbillperiodstartsandends.Againsinceaveragepricesiscalculatedfromthetotalbill,itrequiresconsumerstoknowtheirtotalconsumptionattheendofthebillingperiod,orthatconsumersbeabletoanticipatealldemandshockswithinthebillingperiod.Weproposedaconceptualmodelinwhichconsumersmakesbehavioralrulesaboutconsumptiononlyafterobservingtheirpreviousmonth'sbill.Therstempiricalanalysisshowsthatconsumersrespondtotheirone-monthlaggedpricesinsteadofthecurrentaverageprice.Thisisbecausethelaggedaveragepricerequirestheleastcalculationandeortfromconsumers.Thusintheabsenceoftechnologicaldevelopmentstoprovideinformationtoconsumersabouttheirtruemarginalprice,energypolicyonpriceelasticityshouldbebasedontheassumptionofconsumersrespondingtothelaggedaverageprice.Consumersrespondingtolaggedaveragepricesalsosuggestthatautomaticbillpaymentswhichincreasethelikelihoodthataconsumerwillforgoinspectiondecreasespricesalienceandhencepriceresponsiveness.Theanalysisshowsthatautomaticpaymentusersare10%lesselasticthannon-automaticpaymentusersandthatenrollinginautopaymakesconsumers5%lesspricesensitive.Theseresultsimplythatautomaticbillpaymentsmayactagainst 89

PAGE 90

theutility'sconservationeortofusingpricesignalstosteerenergyconservation.Thus,thepapersuggeststhatwhileautomaticbillpaymentisaconvenientmethodofpaymentforbothconsumersandtheutility,thereistheneedformoreinformationprovisiontohelpconsumersrespondtoprices.Advancesintechnologythathelpsconsumersperceivetheiractualmarginalcostshouldbeacomplementtoautomaticbillpaymentmethods.Forexample, Wolak ( 2011 )showsthatsmartmeteringwithdynamicpricingprovidesastableandsizabledemandreductionsinresponsetocriticalpeakpricing(CPP)andhourlypricing(HP)warnings. 90

PAGE 91

APPENDIXAAPPENDIXTOACCOMPANYEVALUATINGTHEENERGYSAVINGSEFFECTSOFAUTILITYDEMAND-SIDEMANAGEMENTPROGRAMUSINGADIFFERENCE-IN-DIFFERENCECOARSENEDEXACTMATCHINGAPPROACH TableA-1. DIDCEMEstimateofTheEectsofThe2009HighEciencyACProgram log(EnergyUsage) IIIIII Treat -0.0858***-0.1382**-0.1048 (-5.15)(-2.83)(-1.59)Treat*HeatedArea(1000squarefeet) 0.02620.0240 (1.32)(1.22)Treat*Age -0.0012 (-0.63)Bedrooms 0.01000.01000.0100 (0.69)(0.69)(0.69)Stories -0.0007-0.0007-0.0007 (-0.04)(-0.04)(-0.04)HeatedArea(1000squarefeet) -0.0049-0.0056-0.0055 (-0.32)(-0.37)(-0.36)Age 0.00030.00030.0003 (0.35)(0.35)(0.38)Pool -0.0279-0.0278-0.0278 (-1.83)(-1.83)(-1.83)ElectricandGas 0.0991***0.0991***0.0990*** (7.80)(7.81)(7.80)MeanIncome($1000) -0.0006*-0.0006*-0.0006* (-2.36)(-2.35)(-2.36)MeanHouseholdSize 0.02760.02750.0275 (0.92)(0.92)(0.92)HotTub 0.03270.03310.0331 (0.83)(0.84)(0.84)cons 0.03110.03250.0318 (0.38)(0.39)(0.38) N 784378437843 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 91

PAGE 92

TableA-2. AnnualEnergySavingsEectOfThe2010HighEciencyACRebateProgram log(EnergyUsage) CEMDID(CensusTract)RegularDID (I)(II)(III) (IV)(V)(VI) Treat -0.0817***-0.1463*-0.2742** -0.0981***-0.1463**-0.2860*** (-4.00)(-2.12)(-3.19) (-6.91)(-3.19)(-4.75)Treat*HeatedArea 0.03500.0578 0.02310.0366 (1.01)(1.74) (1.20)(1.87)Treat*Age 0.0036 0.0046** (1.67) (3.04)Bedrooms -0.0341-0.0340-0.0341 -0.0054-0.0054-0.0053 (-1.13)(-1.13)(-1.13) (-1.16)(-1.15)(-1.14)Stories 0.03820.03830.0386 0.00380.00370.0036 (0.64)(0.64)(0.65) (0.73)(0.73)(0.71)HeatedArea 0.01570.01290.0117 -0.0061-0.0065-0.0067(1000squarefeet) (0.55)(0.44)(0.39) (-1.27)(-1.34)(-1.38)Age -0.0002-0.0002-0.0005 -0.0015***-0.0015***-0.0015*** (-0.18)(-0.17)(-0.36) (-6.50)(-6.51)(-6.63)Pool -0.0218-0.0217-0.0220 -0.0139*-0.0139*-0.0139* (-0.67)(-0.67)(-0.68) (-2.34)(-2.34)(-2.34)ElectricandGas 0.02090.02120.0211 -0.0084-0.0084-0.0084 (0.67)(0.67)(0.67) (-1.68)(-1.68)(-1.66)MeanIncome($1000) 0.00060.00060.0006 0.00010.00010.0001 (1.42)(1.43)(1.43) (1.72)(1.74)(1.76)MeanHouseholdSize -0.1557-0.1558-0.1557 -0.0210*-0.0211*-0.0209* (-1.57)(-1.57)(-1.57) (-2.01)(-2.02)(-2.00)HotTub 0.01680.01770.0173 -0.0086-0.0088-0.0086 (0.37)(0.39)(0.38) (-0.96)(-1.00)(-0.97)cons 0.27840.28260.2911 0.05380.05440.0550 (1.00)(1.01)(1.04) (1.81)(1.83)(1.85) N 250425042504 240572405724057 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 92

PAGE 93

TableA-3. CEMDIDEstimateofTheEectsofThe2010HighEciencyACProgram log(EnergyUsage) IIIIII Treat -0.0890***-0.1681**-0.2846*** (-5.68)(-2.92)(-4.20)Treat*HeatedArea(1000squarefeet) 0.04010.0553* (1.54)(2.23)Treat*Age 0.0036* (2.11)cons 0.11030.11150.1136Bedrooms -0.0236-0.0237-0.0237 (-1.91)(-1.92)(-1.92)Stories -0.0075-0.0074-0.0074 (-0.33)(-0.33)(-0.33)HeatedArea(1000squarefeet) 0.00130.00050.0002 (0.09)(0.03)(0.01)Age -0.0003-0.0003-0.0004 (-0.58)(-0.57)(-0.71)Pool 0.00360.00360.0034 (0.22)(0.22)(0.21)ElectricandGas -0.0210-0.0209-0.0210 (-1.66)(-1.65)(-1.65)MeanIncome($1000) 0.00000.00000.0000 (0.04)(0.05)(0.05)MeanHouseholdSize -0.0305-0.0303-0.0300 (-1.13)(-1.12)(-1.11)HotTub 0.03620.03630.0362 (1.43)(1.43)(1.43) (1.26)(1.27)(1.30) N 961696169616 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 93

PAGE 94

TableA-4. Non-PeakMonthsEectsOfThe2009ACRebateProgramusingCEMDIDwithZipCodesasNeighborhoods log(EnergyUsage) IIIIII Treat -0.0461**-0.0964*-0.0948 (-2.69)(-2.03)(-1.31)Treat*HeatedArea 0.02520.0251 (1.31)(1.31)Treat*Age -0.0001 (-0.03)Bedrooms 0.02110.02110.0211 (1.41)(1.41)(1.41)Stories 0.01340.01350.0135 (0.86)(0.87)(0.87)HeatedArea -0.0251-0.0258-0.0258(1000squarefeet) (-1.71)(-1.73)(-1.73)Age -0.0006-0.0006-0.0006 (-0.73)(-0.73)(-0.71)Pool -0.0153-0.0152-0.0152 (-0.91)(-0.90)(-0.90)ElectricandGas 0.0572***0.0572***0.0572*** (4.11)(4.11)(4.11)MeanIncome($1000) 0.00030.00030.0003 (1.21)(1.21)(1.21)MeanHouseholdSize 0.05290.05280.0528 (1.73)(1.72)(1.72)HotTub 0.01310.01350.0135 (0.43)(0.44)(0.44)cons -0.1896*-0.1882*-0.1882* (-2.19)(-2.18)(-2.17) N 784378437843 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 94

PAGE 95

TableA-5. DIDCEMEstimateofTheEectofTheHighEciencyACRebateonSummerPeakEnergyConsumption log(EnergyUsage) IIIIII Treat -0.1710***-0.2754***-0.2447*** (-8.09)(-4.38)(-3.83)Bedrooms 0.02450.0245-0.0017 (1.25)(1.25)(-0.31)Stories 0.02520.02530.0004 (1.22)(1.22)(0.06)HeatedArea -0.0380-0.03940.0012(1000squarefeet) (-1.87)(-1.92)(0.21)Age -0.0002-0.0002-0.0000 (-0.20)(-0.20)(-0.01)Pool 0.01680.0169-0.0269*** (0.93)(0.94)(-3.76)ElectricandGas 0.01990.01990.0197** (1.26)(1.26)(3.16)MeanIncome -0.0009**-0.0009**-0.0003**($1000) (-2.68)(-2.67)(-2.62)MeanHouseholdSize 0.04950.04920.0004 (1.18)(1.17)(0.03)HotTub 0.03200.0327-0.0142 (0.58)(0.60)(-1.32)Treat*HeatedArea 0.05240.0315* (1.85)(2.05)treatage 0.0003 (0.16)cons -0.0435-0.04060.0669 (-0.41)(-0.38)(1.91) N 7843784324008 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 95

PAGE 96

TableA-6. DIDCEMEstimateofTheEectofTheHighEciencyACRebateonWinterPeakEnergyConsumption log(EnergyUsage) IIIIII Treat -0.0432*-0.0756-0.0096 (-2.14)(-1.32)(-0.13)Bedrooms 0.00540.00540.0053 (0.37)(0.37)(0.37)Stories -0.0198-0.0198-0.0198 (-0.87)(-0.87)(-0.87)HeatedArea 0.00980.00940.0095(1000squarefeet) (0.61)(0.58)(0.59)Age 0.00070.00070.0008 (0.88)(0.88)(0.93)Pool -0.0589**-0.0588**-0.0587** (-2.97)(-2.97)(-2.97)ElectricandGas 0.1590***0.1590***0.1589*** (10.72)(10.72)(10.71)MeanIncome -0.0009***-0.0009***-0.0009***($1000) (-3.81)(-3.80)(-3.81)MeanHouseholdSize -0.0273-0.0274-0.0273 (-0.83)(-0.83)(-0.83)HotTub -0.0085-0.0083-0.0082 (-0.23)(-0.23)(-0.22)Treat*HeatedArea 0.01620.0119(1000squarefeet) (0.66)(0.49)treatage -0.0024 (-1.16)cons 0.3675***0.3684***0.3671*** (3.64)(3.64)(3.62) N 784378437843 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 96

PAGE 97

TableA-7. Summer,Winter,andNon-PeakMonthsEectsoftheHighEciencyACprogramforHouseholdswithElectricitybutNoNaturalGasUsingTheDIDCEMwithZipCodesasNeighborhoods log(EnergyUsage) SummerWinterNon-PeakMonths Treat -0.1307***-0.0753**-0.0524* (-3.72)(-2.85)(-2.23)Bedrooms -0.00030.01880.0154 (-0.02)(1.05)(1.03)Stories 0.0032-0.0220-0.0057 (0.19)(-1.06)(-0.36)HeatedArea(1000squarefeet) 0.02830.00050.0106 (0.91)(0.03)(0.78)Age 0.00110.0018*0.0003 (0.90)(1.96)(0.39)Pool -0.0466-0.0613**-0.0493* (-1.45)(-2.68)(-2.33)MeanIncome$1000) -0.0002-0.0009***0.0002 (-0.63)(-3.43)(0.81)MeanHouseholdSize 0.01500.03660.0127 (0.38)(1.16)(0.45)HotTub -0.02820.0201-0.0127 (-0.68)(0.46)(-0.30)cons -0.04540.1639-0.1280 (-0.37)(1.69)(-1.48) N 340734073407 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 97

PAGE 98

TableA-8. Summer,Winter,andNon-PeakMonthsEectsoftheHighEciencyACprogramforHouseholdswithElectricityandNaturalGasUsingTheDIDCEMwithZipCodesasNeighborhoods log(EnergyUsage) SummerWinterNon-PeakMonths Treat -0.2015***0.0596-0.0689** (-6.11)(1.81)(-3.03)Bedrooms 0.0537-0.0492-0.0127 (1.45)(-1.47)(-0.50)Stories -0.04770.10070.0214 (-0.64)(1.91)(0.66)HeatedArea -0.0630*0.03750.0203(1000squarefeet) (-2.35)(1.55)(0.86)Age 0.00010.0030*-0.0004 (0.04)(2.16)(-0.37)Pool -0.0032-0.0960**-0.0687** (-0.10)(-2.98)(-2.65)MeanIncome($1000) -0.0000-0.0007-0.0003 (-0.05)(-1.28)(-0.84)MeanHouseholdSize 0.09810.0028-0.0346 (1.08)(0.04)(-0.65)HotTub -0.00860.0182-0.0026 (-0.24)(0.42)(-0.09)cons -0.17080.00880.0913 (-0.76)(0.05)(0.69) N 347734793479 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 98

PAGE 99

APPENDIXBAPPENDIXTOACCOMPANYTHE\REBOUND"EFFECT FigureB-1. AverageEnergyConsumptionbyParticipantsandNon-Participants FigureB-2. AverageEnergyConsumptionbyParticipantsandNon-Participants 99

PAGE 100

TableB-1. SummerPeakReboundEectsbyFuelMixUsingCEMDDwithZipCodesasNeighborhoods log(EnergyUsage) Electricity-onlyHouseholdsElectricandGasHouseholdsTreat 0.0251 0.0735* (0.57) (2.38)Bedrooms -0.0168 -0.0412 (-0.47) (-1.23)Stories -0.0351 -0.0961 (-1.08) (-1.66)HeatedArea(1000squarefeet) 0.0158 0.0193 (0.57) (0.53)Age 0.0004 0.0001 (0.14) (0.05)Pool -0.0833* -0.0204 (-1.96) (-0.66)MeanIncome($1000) 0.0009 0.0014* (1.67) (2.37)MeanHouseholdSize -0.0706 -0.1328 (-0.88) (-1.72)HotTub -0.0554 -0.1002 (-1.25) (-1.33)cons 0.1157 0.3650 (0.55) (1.75)N 2231.0000 5611.0000 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 100

PAGE 101

TableB-2. WinterPeakReboundEectsbyFuelMix log(EnergyUsage) Electricity-onlyHouseholdsElectricandGasHouseholds CEMDDDCEMDDDregularDDD CEMDDDCEMDDDregularDDD (censustracks)zipcodes(withCEM) (censustracks)zipcodes(withCEM) Treat -0.0000.0200.028 0.0260.0110.016 (-0.00)(0.56)(0.71) (0.74)(0.39)(0.50)Bedrooms -0.171-0.020-0.065 -0.021-0.006-0.058* (-1.45)(-0.50)(-0.96) (-0.56)(-0.26)(-2.19)Stories -0.0630.046-0.013 0.032-0.0160.036 (-0.85)(0.87)(-0.29) (0.60)(-0.34)(0.81)HeatedArea(1000squarefeet) 0.1380.073-0.035 -0.003-0.0260.003 (1.18)(1.30)(-0.52) (-0.08)(-1.40)(0.09)Age -0.009**-0.000-0.003 -0.003-0.004**-0.004* (-2.81)(-0.24)(-1.41) (-1.44)(-2.81)(-2.44)Pool -0.0240.005-0.019 0.081*0.064**0.057 (-0.36)(0.09)(-0.43) (2.36)(2.66)(1.88)MeanIncome($1000) -0.0000.0010.001 0.003***0.002***0.002*** (-0.26)(1.01)(1.30) (3.58)(4.55)(4.09)MeanHouseholdSize -0.1240.125-0.025 -0.381*-0.055-0.123 (-0.69)(1.62)(-0.27) (-2.37)(-0.99)(-1.50)HotTub 0.579***0.0080.268 -0.0500.031-0.039 (3.98)(0.11)(1.53) (-1.13)(0.76)(-0.94)cons 0.563-0.820**-0.067 0.235-0.382*-0.175 (1.11)(-3.21)(-0.23) (0.68)(-2.53)(-0.79)N 296.0002232.000296.000 1256.0005611.0001256.000 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 101

PAGE 102

TableB-3. Non-PeakReboundEectsbyFuelMix log(EnergyUsage) Electricity-onlyHouseholdsElectricandGasHouseholds CEMDDDCEMDDDregularDDD CEMDDDCEMDDDregularDDD (censustracks)zipcodes(withCEM) (censustracks)zipcodes(withCEM) Treat -0.004-0.008-0.008 -0.029-0.012-0.035 (-0.11)(-0.21)(-0.22) (-0.91)(-0.45)(-1.24)Bedrooms -0.0130.0200.029 -0.028-0.0500.012 (-0.18)(0.39)(0.43) (-0.63)(-1.87)(0.37)Stories -0.071-0.062-0.032 -0.033-0.083*0.014 (-1.85)(-1.35)(-0.92) (-0.57)(-1.99)(0.32)HeatedArea(1000squarefeet) -0.0800.002-0.091 0.0070.003-0.006 (-1.14)(0.05)(-1.37) (0.16)(0.12)(-0.18)Age -0.0040.001-0.002 -0.003-0.000-0.001 (-1.17)(0.40)(-0.59) (-1.56)(-0.18)(-0.31)Pool 0.017-0.0220.039 0.0640.0280.014 (0.31)(-0.46)(0.66) (1.45)(1.16)(0.48)MeanIncome($1000) -0.0000.0000.000 -0.003**-0.000-0.003*** (-0.15)(0.18)(0.26) (-2.78)(-0.99)(-4.82)MeanHouseholdSize -0.255*-0.085-0.127 0.005-0.173*0.044 (-2.14)(-1.25)(-1.22) (0.06)(-2.40)(0.57)HotTub 0.0890.0020.016 -0.096-0.046-0.066 (1.36)(0.05)(0.21) (-1.48)(-1.05)(-1.32)cons 0.915*0.1230.397 0.2990.604**0.034 (2.55)(0.51)(1.26) (1.24)(3.09)(0.17)N 296.0002232.000296.000 1256.0005611.0001256.000 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 102

PAGE 103

TableB-4. SummerPeak,WinterPeak,andNon-peakReboundEectsofTheACRebateProgramUsingDDCEMwithZipCodesasNeighborhoods log(EnergyUsage) SUMMERWINTERNON-PEAK Treat -0.0390*-0.0026-0.0319* (-2.41)(-0.17)(-2.12)Bedrooms -0.00790.0080-0.0074 (-0.38)(0.61)(-0.39)Stories -0.04500.0081-0.0300 (-1.85)(0.25)(-1.13)HeatedArea(1000squarefeet) 0.00060.0161-0.0071 (0.03)(0.74)(-0.42)Age -0.0000-0.0020**-0.0013 (-0.04)(-2.68)(-1.84)Pool -0.02950.00160.0108 (-1.41)(0.07)(0.56)ElectricandGas 0.0164-0.0844***-0.0904*** (1.14)(-6.63)(-7.35)MeanIncome($1000) 0.00060.0013***-0.0007** (1.76)(5.49)(-2.59)MeanHouseholdSize -0.0218-0.0862*-0.0511 (-0.68)(-2.24)(-1.25)HotTub -0.0865-0.0074-0.0421 (-1.62)(-0.31)(-1.84)cons 0.0453-0.13220.1782 (0.54)(-0.96)(1.53) N 784378437843 *p<0.05,**p<0.01,***p<0.001.t-statisticsareinparenthesis. 103

PAGE 104

REFERENCES Acton,Jan,Mitchell,Bridger,andMowill,Ragnhild.\ResidentialDemandForElectricityinLosAngeles:AnEconometricStudyofDisaggregatedData."1976. Acton,JanPaul,Mitchell,BridgerM,andSohlberg,Ragnhild.\Estimatingresidentialelectricitydemandunderdeclining-blocktaris:aneconometricstudyusingmicro-data."AppliedEconomics12(1980).2:145{161. ArchitecturalEnergyCorporation.\EngineeringMethodsforEstimatingtheImpactsofDemand-SideManagementPrograms.Volume1:FundamentalsofEngineeringSimulationsforResidentialandCommercialEndUses."1992. Blackwell,Matthew,Iacus,Stefano,andKing,Gary.\cem:CoarsenedExactMatchinginStata."TheStataJournal9(2009).4:524{546. Boampong,Richard.\EvaluatingtheEnergySavingsEectofaUtilityDemand-SideManagementProgramusingaDierence-in-DierenceCoarsenedExactMatchingApproach.PURCWorkingPaper,2014."Tech.rep.,2014. Borenstein,Severin.\Towhatelectricitypricedoconsumersrespond?Residentialdemandelasticityunderincreasing-blockpricing.",2009. Bushnell,James,Hobbs,BenjaminF.,andWolak,FrankA.\WhenItComestoDemandResponse,IsFERCItsOwnWorstEnemy?"TheElectricityJournal22(2009):9{18. Chatterjee,PromotheshandRose,RandallL.\Dopaymentmechanismschangethewayconsumersperceiveproducts?"JournalofConsumerResearch38(2012).6:1129{1139. Cohen,Sam,Goldman,Charles,andHarris,Je.\EnergySavingsandEconomicsofRetrottingSingle-FamilyBuildings."EnergyandBuildings17(1991):297{311. DeBartolome,CharlesAM.\Whichtaxratedopeopleuse:Averageormarginal?"JournalofpublicEconomics56(1995).1:79{96. Diamond,PeterA.\Optimalincometaxation:anexamplewithaU-shapedpatternofoptimalmarginaltaxrates."AmericanEconomicReview(1998):83{95. Dubin,JereyA.,Miedema,AllenK.,andChandran,RamV.\PriceEectsofEnergy-EcientTechnologies:AStudyofResidentialDemandforHeatingandCooling."TheRandJournalofEconomics17(1986).3:310{325. Fels,MargaretF.andKeating,KennethM.\MeasurementofEnergySavingsfromDemandSideManagementProgramsinU.S.ElectricUtilities."AnnualReviewofEnergyandtheEnvironment18(1993):57{88. Fiserv.\2010AnnualBillingHouseholdSurvey."FiservResearchPaper(2010). Foster,HenrySandBeattie,BruceR.\UrbanresidentialdemandforwaterintheUnitedStates."LandEconomics(1979):43{58. 104

PAGE 105

Fowler,Meredith,Greenstone,Michael,andWolfram,Catherine.\DoEnergyEciencyInvestmentsDeliver?EvidencefromtheWeatherizationAssistanceProgram."NBERWorkingPaperSeries(2015). Gabor,Andre.\Anoteonblocktaris."TheReviewofEconomicStudies23(1955).1:32{41. Gamtessa,SamuelandRyan,DavidL.\UtilizationofResidentialEnergy-SavingRetrotProgramsinCanada:Who,What,andWhy?"2007. Gensemer,BruceL,Lean,JaneA,andNeenan,WilliamB.\Awarenessofmarginalincometaxratesamonghigh-incometaxpayers."NationalTaxJournal(1965):258{267. Gillingham,Kenneth.\TheConsumerResponsetoGasolinePriceChanges:EmpiricalEvidenceandPolicyImplications."StanfordUniversityPh.DDisertation(2011). Girma,SourafelandGorg,Holger.\EvaluatingTheForeignownershipWagePremiumUsingaDierence-in-DierenceMatchingApproach."JournalofInternationalEconomics72(2007):97{112. Godberg,MiriamL.\AMidwestLow-IncomeWeatherizationProgramSeenThroughPRISM."EnergyandBuildings9(1986).1-2:37{44. Greene,DavidL.\Rebound2007:AnalysisofUSlight-dutyvehicletravelstatistics."EnergyPolicy41(2012):14{28. Hausman,JerryA.\Theeconometricsofnonlinearbudgetsets."Econometrica(1985):1255{1282. Herriges,JosephAandKing,KathleenKuester.\Residentialdemandforelectricityunderinvertedblockrates:Evidencefromacontrolledexperiment."JournalofBusiness&EconomicStatistics12(1994).4:419{430. Huang,Ching-I.\Estimatingdemandforcellularphoneserviceundernonlinearpricing."QuantitativeMarketingandEconomics6(2008).4:371{413. Iacus,StefanoM.,king,Gary,andPorro,Giuseppe.\CausalInferencewithoutBalanceChecking:CoarsenedExactMatching.",2008.URL http://gking.harvard.edu/les/gking/les/political analysis-2011-iacus-pan mpr013.pdf Ito,Koichiro.\Doconsumersrespondtomarginaloraverageprice?EvidencefromNonlinearElectricityPricing."AmericanEconomicReview(2014):537{563. Jacobsen,GrantD.andKotchen,MatthewJ.\AreBuildingCodesEectiveatSavingEnergy?EvidencefromResidentialBillingDatainFlorida."TheReviewofEconomicsandStatistics95(2013):34{49. Jones,PierceH.,Taylor,NicholasW.,Kipp,M.Jennison,andKnowles,HaroldS.\QuantifyingHouseholdEnergyPerformanceUsingAnnualCommunityBaselines."InternationalJournalofEnergySystemsManagement4(2010).4:593{613. 105

PAGE 106

Khazzoom,Daniel.\EconomicImplicationsofMandatedEciencyinStandardsforHouseholdAppliances."TheEnergyJournalVolume1(1980):21{40. Koichiro,Ito.\DoesConservationTargetingWork?EvidencefromaStatewideElectricityRebatePrograminCalifornia.WorkingPaper.",2012. Liebman,JereyB.andZeckhauser,RichardJ.\Schmeduling."(2014).URL http://www.hks.harvard.edu/fs/rzeckhau/Schmeduling Oct172004.pdf Mirrlees,JamesA.\Anexplorationinthetheoryofoptimumincometaxation."Thereviewofeconomicstudies(1971):175{208. Mizon,GrayhamEandRichard,Jean-Francois.\Theencompassingprincipleanditsapplicationtotestingnon-nestedhypotheses."Econometrica(1986):657{678. Parti,MichaelandParti,Cynthia.\TheTotalandAppliance-SpecicConditionalDemandforElectricityintheHouseholdSector."TheBellJournalofEconomics11(1980).1:309{321. Prelec,DrazenandSimester,Duncan.\Alwaysleavehomewithoutit:Afurtherinvestigationofthecredit-cardeectonwillingnesstopay."Marketingletters12(2001).1:5{12. Reiss,PeterC.andWhite,MatthewW.\DemandandPricinginElectricitymarkets:EvidencefromSanDiegoDuringCalifornia'sEnergyCrisis.NBERworkingpaper9986.",2003. Reiss,PeterCandWhite,MatthewW.\Householdelectricitydemand,revisited."TheReviewofEconomicStudies72(2005).3:853{883. Sexton,Steven.\Automaticbillpaymentandsalienceeects:Evidencefromelectricityconsumption."ReviewofEconomicsandStatistics97(2015).2:229{241. Smith,JereyA.andTodd,PetraE.\DoesMatchingOvercomeLaLonde'sCritiqueofNonexperimentalEstimators?"JournalofEconometrics125(2005):305{352. Soman,Dilip.\Eectsofpaymentmechanismonspendingbehavior:Theroleofrehearsalandimmediacyofpayments."JournalofConsumerResearch27(2001).4:460{474. Violette,Daniel.\DemandResponseinOrganizedElectricMarkets."FederalEnergyRegulatoryCommission(FERC)TechnicalConference(2008). West,Jeremy,Hoeskstra,Mark,Meer,Jonathan,andPuller,StevenL.\VehicleMiles(Not)Traveled:WhyFuelEconomyRequirementDon'tIncreaseHouseholdDriving."NBERWorkingPaperSeries(2015). Winsberg,MortonD.andSimmons,Melani.\AnAnalysisoftheBegining,End,Length,andStrengthofFlorida'sHotSeason."(2009).URL http://coaps.fsu.edu/climate center/specials/hotseason.shtml 106

PAGE 107

Wolak,FrankA.\Doresidentialcustomersrespondtohourlyprices?Evidencefromadynamicpricingexperiment."TheAmericanEconomicReview(2011):83{87. Wooldridge,Jerey.Introductoryeconometrics:Amodernapproach.CengageLearning,2012. 107

PAGE 108

BIOGRAPHICALSKETCHRichardBoampongearnedhisBachelorofArtsdegreeinmathematicsandeconomicsfromtheUniversityofGhanain2006.HereceivedaMasterofArtsdegreeinbusinesseconomicsfromtheUniversityofSouthFloridain2010.RichardstartedtheEconomicsPh.D.degreeattheUniversityofFloridain2011andspecializedinIndustrialOrganization,Econometrics,andEconomicTheory.Whilepursuinghisdegree,heservedasaninstructorofEconometricsandEnvironmentalEconomics.HewasalsoaresearchassistantatthePublicUtilityResearchCenter.HisresearchinterestincludesIndustrialOrganization,AppliedEconometrics,EnergyEconomics,andEnvironmentalEconomics.Richardhasacceptedapositionofpost-doctoralscholarforthenextacademicyearatFloridaStateUniversity. 108