Transcription of Any Way You Slice It: Issues of Behavior and Influence in ...
1 AnyWayYouSliceIt:IssuesofBehaviorandInfl uenceinNetImpactAnalysisChristieToro~Qua ntumConsultingInc.,Berkeley,CAJohnCavall i,QuantumConsultingInc.,Berkeley,CAMaryO net-to-gross isdefinedastheprocessofadjustingestimate sofgrossprogramimpactfortheeffectsoffree ridershipandspillover,resultinginanestim ateof netimpact. Threeapproacheswereusedtoestimatenetprog ramimpacts:billingdataanalysis,self-repo rteddataanalysis, , , (CEEI)Program,runbyPacificGas&Electric, ,andcustomizedfinancialincentivestocusto merswhoundertakelarge, ,QuantumConsultinghasconductedanevaluati ontodeterminethefirst-yeargrossandnetene rgy, ,inboththeoryandpractice, net-to-gross isdefinedastheprocessofadjustingestimate sofgrossprogramimpactfortheeffectsoffree ridershipandspillover,resultinginanestim ateof netimpact.
2 Acts(thathavebeeninteractedwiththeMillsr atios)arethenusedtoadjusttheengineeringe stimatesofexpectedannualenergyimpacts(th eoriginalSAEcoefficients) ,J. TheCommonStructureofStatisticalModelsofT runcation,SampleSelectionandLimitedDepen dentVariablesandaSimpleEstimatorforSuchM odels. ,AnnalsofEconomicandSocialMeasurement, , , ,J. SampleSelectionBiasasaSpecificationError . Econometrics, , , ,MiriamandKennethTrain. NetSavingsEstimation:AnanalysisofRegress ionandDiscreteChoiceApproaches ,preparedfortheCADMACS ubcommitteeonBaseEfficiencybyXenergy, ,WI, , , ,itwasfoundthatthenetbillingmodelresults weresignificantatthe95percentlevelinallc asesbutone, , ,wecannow backout anestimateoffreeridership, :(1-FR)~= iii;*8 mWhere,MillsmisthemeanMillscoefficientfo rallcustomerswithtechnologym;PMistheSAEc oefficientfromtheGrossBillingmodelfortec hnologym;and,i?
3 , (1-FR)Resulting(1-FR) , , , , ,largercustomersaregenerallymorelikelyto befreeriders, , ,theInverseMillsRatiomethoddoesnotproduc eanestimateofspillover, ,theresultsofthenetbillinganalysisinclud eestimatesofthemagnitudeofnetimpacts, ,behaviorshouldnotbeassumedtoberelatedon lytoconcentrationofenergyuse, selfreport methodthatkbasedoncustomers , ,surveyquestionsregardingprogramawarenes s,programinfluence,andwhattherespondent sbehaviorwouldhavebeenintheabsenceofthep rogram, , ,808participantsand4, ,atotalof860participantsand4, , , ,theparticipantwouldhaveinstalledhigheff iciencyequipment,butnotassoon(morethanon eyearlater) ,theparticipantwouldhaveinstalledhigheff iciencyequipmentatthesametime(withinthey ear) ,thetechnologygroupwiththelowestrateoffr eeridershipwastheDelampFluorescentFixtur escategory, , , ~ ,andoccurredasadirectresultoftheprogram ,thefollowingfourconditions,whichreflect thisdefinitionofspillover, , , :ParticipantSpillover:NTGpart_spill=SP_R ATE part*POPpart*IMPACTpti_spill/IMPACTpopWh ere,NTGpti_spill=theparticipantcontribut ionofspillovertothenet-to-grossratioSP_R ATE part=theparticipantspilloverratePOPpart= theparticipantpopulation,innumberofsites IMPACTpti_spill=theperparticipantsiteimp actassociatedwithspilloverIMPACTpop=thet otalCEEIP rogramimpactNonparticipantSpillover.
4 NTGnp_spill=SP_RATEnp*POPnp*IMPACTnp_spi ll/IMPACTpopWhere,NTGnp_spill=thenonpart icipantcontributionofspillovertothenet-t o-grossratioSP_RATEnp=thenonparticipants pilloverratePOPnp=thenonparticipantpopul ation,innumberofsitesIMPACTnp_spill=thep ernonparticipantsiteimpactassociatedwith spilloverIMPACTpop= , ,thenumberofinstancesofspilloveroccurrin gwithinthesampledpopulationislikelytober elativelysmall, ,200completedtelephonesurveysin1996and19 97, ,thesamplemustbeverylargetoensurerobuste stimates, ,fornonparticipants,asampleof4,200issuff icient, ,themoreresourcesdevotedtothedatacollect ioneffort, $15, ,230 NumberContributingtoSpillover168 SpilloverAvoidedCost$2,622,950Li~htin~Av oidedCost$51,077,333.., $18, ,308 NumberContributingtoSpilloverSpilloverAv oidedCost$1,730,77;LightingAvoidedCost$5 9,140,572 NTGC ontributlonfromParticipantSpilloverZ.
5 !)S /o1996 NonparticipantSpilloverEstimateAvoidedCo stPerNonparticipant$8,473 Spi]] ,668 NumberContributingtoSpillover320 SpilloverAvoidedCost$2,710,747 LightingAvoidedCost$51,077, $10, ,188 NumberContributingtoSpillover378 SpilloverAvoidedCost$4,137,013 LiRhtinRAvoidedCost$59,140, , ,incommercialsettings, ,nomatterhowsmallorlargeasinglecustomeri s, ,thestraightforwardmethodologyallowsitto beusedinawidevarietyofsituations, , , ,thosewhothesupplysideaffectareexcludedf romtheanalysis,Forexample,arespondentmay havebeeninfluencedbyvendorstockingpracti ces,orarecommendationfromacontractorwhow asinfluencedbytheprogram, , , :Prob(Purchase&EquipmentA)=Prob(Purchase )*Prob(EquipmentA\Purchase) , , , :PURCHASE=u+~ X+~ Y+11 Z+ , resultsareconsistentwitheachother, , , ,thosethatdidnotmakeanypurchaseshavealow estimatedpurchaseprobability, , , ~ AWARE+~ PREDISP+~ SQFEET+fl CINDEX+~ SAVINGS+Z~ BLDTYPE-f-eWhere,AWARE=Awarenessvariable sPREDISP= (cost rebate)/costSAVINGS=Annualdollaramountof electricitysavingsexpectedfromequipmentB LDTYPE=BuildingsizeandbusinesstypeE= , , , , :EXPECTEDIMPACTW=ZPwj*IMPACTjWherePwj=Jo intprobabilityofchoosingequipmentoptionj withtheprogramIMPACTj= :EXPECTEDIMPACTwO=ZPwOj*IMPACTjWherePwOj = :NETIMPACT=EXPECTEDIMPACTW-EXPECTEDIMPAC TwOjThenet-to-grossratioisthenthenetimpa ctdividedbytheexpectedimpactwiththeprogr am.
6 NTG= ,theoverallratesestimatedfor1996and1997w ereidenticaltotwodecimalplaces, sanswerstoquestionsregardingtheirattitud estowardshighefficiency, ,oronrespondents correctlyreportingthedegreetowhichthepro graminfluencedtheirbehavior, ,themodelcontrolsfortheeffectsofbusiness type,buildingcharacteristics,size, , , , ,threeseparatemodelswereimplementedtoest imatethecomponentsofthenet-to-grossratio (freeridershipandspillover).Table6belowp resentthe1996and1997resultsofeachmethod, ,bothwithinbusinesstypeandoverall, ,aswellasthetwoprimarycomponents, ,onlyfreeridershipisestimated, , , , , , , , , < , , < ; , , ,theself-reportresultswerewithin20%ofthe discretechoicemodelresultsforallbuttwobu sinesstypes, , ,The1996evaluationproducedself-reportedr esultsthatwerewithin10% ,theself-reportedresultswerewithin20% , , , ,andpreferred,ofthethreeapproachesisthet wo-stagediscretechoicemodel, , , , ,tocapturethemarkettransformationeffects oncustomerpurchases,an outsideserviceterritory controlgroupwouldbeusedinadditiontoan in-serviceterritory nonparticipantsample.
