Contents lists available atScienceDirect
Energy Research & Social Science
j o u r n a l h o m e p a g e :w w w . e l s e v i e r . c o m / l o c a t e / e r s s
Original research article
White goods for white people? Drivers of electric appliance growth in emerging economies
Narasimha D. Rao
∗, Kevin Ummel
InternationalInstituteforAppliedSystemsAnalysis(IIASA),Austria
a r t i c l e i n f o
Articlehistory:
Received7November2016
Receivedinrevisedform14February2017 Accepted6March2017
Keywords:
Householdbehavior Residentialenergy Appliancediffusion
a b s t r a c t
Willeverybodywantandhavearefrigerator,televisionandwashingmachineasincomesrise?Consider- ableuncertaintysurroundsthelikelyincreaseinenergyconsumptionandcarbonemissionsfromrising incomesamongtheworld’spoor.Weexaminedriversofandpredictapplianceownershipusingmachine learningandothertechniqueswithhouseholdsurveydatainIndia,SouthAfricaandBrazil.Televisions andrefrigeratorsareconsistentlypreferredoverwashingmachines.Incomeisstillthepredominant driverofaggregatepenetrationlevels,butitsinfluencediffersbyapplianceandbyregion.Theaffordabil- ityofappliances,wealth,raceandreligiontogether,amongotherhouseholdcharacteristics,helpexplain theheterogeneityinapplianceownershipatlowerincomelevels.Understandingnon-incomedrivers canbehelpfultoidentifybarrierstoapplianceuptakeandtobetterforecastneartermresidentialenergy demandgrowthwithincountries.
©2017TheAuthor(s).PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction
Will everybody want and have a refrigerator and washing machineasincomesrise?Considerableuncertaintysurroundsthe energyconsumption and carbonemissionsfromrising incomes amongtheworld’spoor[1].Besidesheatingandcoolingbuildings, householdelectronics,primarilytelevisions,and ‘whitegoods’– largeelectricalhouseholdappliances–increasinglydrivehouse- holdelectricitydemandgrowth[2].Residentialelectricitydemand innon-OECDcountries,whichiscurrentlyslightlylowerthanOECD countries,isexpectedtogrowfasterandexceedOECDcountries’
demandbyupto25percentin2030,reachingover1000Terawatt- hours[3].Globalclimateandenergydemandscenariostypically adoptaveragenationalGDPastheprimarydeterminantofhouse- holdelectricity demandin countries, whilesome alsoconsider relevantsocietaltrends,suchasurbanizationand electrification [3–6].Ineffect,thecurrentthinkingisbasedontheassumption thatallhouseholdsgloballyatacertainincomelevelwouldhavethe sameappliances.Buttheseassumptionshavenotbeenempirically validatedonasystematicbasis.Usingmicro(householdsurvey)
∗Correspondingauthor.
E-mailaddresses:nrao@iiasa.ac.at(N.D.Rao),kevinummel@gmail.com (K.Ummel).
datainthreeemergingeconomies,Brazil,SouthAfricaandIndia,1 weshowthisassumptionoversimplifiesreality.Whileincomeis thedominantdriverinthelongrun,marketaccess,affordability, andwealthtogetherbetterexplainownership,whichdifferconsid- erablyatsimilarincomelevelswithinandacrossthesecountries andfordifferentappliances.Policiestoimproveenergyefficiency and equitable access to decent living conditions can be better designedwithsuchknowledgeofmarketbarriersandhousehold preferences.
The lessons from the study of household cooking choices and electrification suggest that household conditions matter, andexhibitheterogeneityacrossand withincountries[7,8].We attempt to systematically understand the drivers of appliance uptake inmajor emerging economies,taking intoaccount both internalandexternalhouseholdcharacteristics.Westudytelevi- sions,refrigeratorsandwashingmachines.Televisions,whilenot strictlyspeaking‘whitegoods’,aresimilarfromanenergyperspec- tive,beingcapitalintensiveconsumerdurableswithhighelectricity consumption.Otherwhitegoods,suchasovensandtumbledry- ersarenotasprevalentindevelopingcountries.Usingnationally sampledhouseholdsurveydatafromIndia,BrazilandS. Africa, this paperasks:isrising income alonesufficient toexplainthe rateand extent ofelectrical appliance penetration in emerging
1MicrodataforChinawereunavailable.
http://dx.doi.org/10.1016/j.erss.2017.03.005
2214-6296/©2017TheAuthor(s).PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/
4.0/).
economies?Howdoacquisitiontrendsdifferacrossthesecountries andwhy?Whatimplicationsdothesetrendshaveforfutureenergy demandprojections?Thispapercontainsanumberofnovelties.We consideradditionaldriversbesidesincomeanddemographicchar- acteristics,including:marketconditions,suchasapplianceprices andelectricityreliability,whereavailable;socialandculturalfac- torssuchasraceandreligion;andwealth-relatedindicators,such asdwellingqualityandautomobileownership.Wedevelopastan- dardizedhouseholdconsumptionmicro-datasetacrossthethree countriesandtwopointsintime.Weapplymachinelearningalgo- rithmsto identifyand visualize influentialdrivers fromtheset availableinsurveys;andweassessownershippredictionaccuracy withandwithouttheseadditionaldrivers.
Wefindthatbeyondacertainthresholdofincome,itislikely thatmosthouseholdswouldpurchasetelevisionsand refrigera- tors,thoughfewerwouldpurchasewashingmachines.Thereare, however,likelytobemanydifferencesintheinterimtransition pathsincountries,whichwillbeinfluencedbymanynon-income factors,notleastappliancesprices.Ifwithtechnologicalchange appliancepricesshowdramaticchangesovertimeordifferences betweenregions,sowouldtheaffordabilityofappliances,andthe speedandtrajectoriesofappliancepenetration.Otherinfluential non-incomedriversincludewealthandrace–blackhouseholdsare lesslikelythanwhiteandcoloredpeopletoownwashingmachines, ceterisparibus.Withincountries,especiallyatlowerincomelev- els,applianceownershipvariesgreatlywiththesefactors,thoughat anaggregateleveltheimprovementinpredictionoftotalappliance penetrationismodest.
Therestofthepaperisorganizedasfollows:inSection2we discussthestateofknowledge;inSection3wediscussdata;inSec- tion4weillustrateapplianceacquisitiontrends,andthepossible roleofnon-incomefactors;inSection5wepresentaquantitative analysistotesttheinfluenceofhouseholdcharacteristicsonappli- anceownership;andinSection6wediscusstheresultsandpolicy implications.
2. Whatweknow
Earlier work examined drivers of residential energy use, includingspaceconditioningand transport,and highlightedthe importanceofenergyprices,dwellingtypeandtechnologyevo- lution,inadditiontoincomeand climate[9,10].It isclearfrom theliterature,though,thatdifferentapplianceshaveverydiffer- entratesofpenetrationovertime,whosecausesarestillnotwell understood.Thereismuchevidencetosuggestthattelevisionsare thefirstandmostwidelyacquiredappliance[3,10,11].Studiesof Indiaseemtorevealahierarchyintheorderinwhichfurthergoods areacquired[5],but,aswediscusslater,thismaybeparticularto India.
Empiricalresearchonappliancediffusionhaslargelyfocusedon explainingorpredictingappliancestockbasedonbroadsocietal trends,whileveryfewexaminedeterminantsofapplianceown- ershipatahouseholdlevel.Inindustrializedcountries,Howarth etal.[12]examinethedriversofresidentialenergyevolutionin OECDcountriesandshowthatthegrowthofappliancestockslowed fromthesixtiestotheseventies.However,Bayus[13]inanexten- siveexaminationofhomeappliancediffusionratesover several decadesintheUSshowsthatdiffusionratesshownopatternwith time.Incontrast,BowdenandOffer[14]showthatdifferenttypes ofappliancesdoindeedhavedifferentdiffusionrates.
Morerecently,manystudiesdescribethegrowthofappliances inemergingeconomies,particularlyinurbanIndiaandChina,and extrapolatethesetrends[3,15,16],oftenwiththegoalofestimating energygrowthorefficiencypotential.However,thesestudiestypi- callydonotformallyexaminehousehold-leveldriversofappliance
acquisition,otherthanhouseholdsize.Asubsetofthesestudies focusonthe(positive)incomeelasticityofapplianceacquisitionto illustratetherelationshipbetweenpatternsofincomeandenergy growth[17,18].
Macro approaches to estimate future appliance penetration uselogisticcurvesdrivenbyincome,electrificationandurbaniza- tion[4,5,19,20].However,theseestimatesdonotcomprehensively explainhistoricalappliancediffusion,nordotheyattempttoexam- inehousehold-specificfactors,inpartbecauseoftheirfocuson appliancestock,andnottheextentofhouseholdpenetration.Con- spicuouslyabsentisaffordabilityofappliances,whichdependon incomeandapplianceprices,amongotherfactors.2TheUSEnergy InformationAdministration’sNEMSmodel doesconsiderprices, butinappliance-specific paybackperiodsthatdon’tincorporate household-specificpreferencesforwhitegoods[21].
Among the few studies of household-level determinants, O’Dohertyetal.[22]examinethedeterminantsofthetotalstockof appliancesinIrelandtodetermineenergysavingspotential.They findthatmosthouseholdcharacteristicsaresignificant,buthome typeandagearethemostimportantnon-incomedeterminants.
LeahyandLyon[23]inalaterstudyalsofindthathouseholdchar- acteristicsinfluenceappliancesownership,butalsofindthatthe totalstocksignificantlyinfluencestotalenergyuse.Inastudyof rural China,Rong and Yao [24] quantitatively assess drivers of applianceacquisition,andfindsthatbesidesincome,moreedu- cation,femalemembersandpublicservicesincreasethelikelihood ofapplianceownership.Kemmler[8]examinespredictorsofelec- tricityaccessuptake inIndia,and findsa numberof household conditionsinfluenceaccess.Matsumoto[25]examinesandcon- firmstheinfluenceofhouseholdsizeandcompositiononappliance usageinJapanfordifferenttypesofappliances.Acrossallthesearti- cles,wefindnoconsiderationofsocialorculturalfactors.Appliance pricesareaccountedforbyZhaoandYang,butnotincombination withincomeasanexpenditureshare.
In summary, most studies examine macro trends, ignoring within-countryheterogeneity.Consequently,notmuchisunder- stoodabouttherateandextentofdiffusionofdifferentappliances indifferentcountries,particularlyinthedevelopingworld.While the saturation of televisions and mobile phones may seem inevitable,andpossiblypredictablewithrisingincome,thesame maynotbethecaseforotherappliances,suchasrefrigeratorsor washingmachines.
3. Data
Wehaveconstructedadatasetfrompubliclyavailablenation- allyrepresentativehouseholdsurveydataand othersourceson household characteristics, appliance ownership, national aver- ageapplianceprices,andconsumption expenditure.Thedataset includes: India, using the India Human Development Surveys (IHDS) of 2004–05 and 2010–11 (41, 554 and 42, 152 house- holds);Brazil,usingtheConsumerExpenditureSurveys(POF)of 2002–03 and 2008–09 (48,470 and 55,970 households); and S.
Africa,usingtheIncomeandExpenditureSurveysof2005–06and 2010–11(21,144and25,328households).WeselecttheIHDSover theoftenusedIndianNSS(NationalSampleSurvey)becausethe IHDShasmoreappliancesandaquestiononelectricityreliability.
Allthreesurveyscollectinformationonhouseholdconsumption expenditure,andincludedataonappliancesandenergyconsump- tion,generaldemographicsandotherhouseholdcharacteristics.
TheBrazilandS. Africasurveyshavea questiononrace,which includecommoncategoriesofwhiteandblack,anddifferentdefi-
2Electricitypricesalsoinfluenceoperatingcosts,asdotheneedforandcostof credit,thoughwhetherthesefactorsdriveapplianceacquisitionisnotknown.
Fig.1.Televisionpenetrationvsincome,nationalaveragevswithin-country.GraycirclesshowaveragepenetrationbyGDPpercapforallcountries;linesshowwithin-country penetrationbyhouseholdexpenditurepercapforBrazil,IndiaandS.Africa.
nitionsforothercoloredpeople.IntheIHDS,therelatedquestion is onreligion, and includes Hinduism, Christianity, Islam,Bud- dhism,Sikhism, Zoroastrianism,Jainism andothers. Inaddition, weobtainedaveragenational,annualappliancepricesandmar- ket volumesfor different product types for each countryfrom EuromonitorInternational.Euromonitor surveysretailersacross the respective countries to obtain end-use prices for different products.Notably,thepricesarebuiltfromactualpricespaidby households,weightedbytheshareofproductmodelssoldatdif- ferentpricesbydifferentsuppliers. However,theyonlyprovide asinglenationalaveragepriceforeachproduct.Notethatinthe subsequentanalysisweuseonlyappliance pricesand nottotal operatingcostsofappliances.Giventhehighdiscountratestypi- caloflow-incomeconsumers[26],aswouldbetypicalinemerging economies,andofbuyersofwhitegoodsingeneral[27],theupfront costdominatesdecision-making.
Thedatawereinterpretedandprocessedtocreateacommon platformofvariables andunits acrosscountriesand years.This processincluded: convertingall monetaryvaluestopurchasing powerparity(PPP)2010dollars;creatingnewindicatorsforpoten- tialexplanatoryvariables,suchasanaffordabilitymetric(share of appliance price in annual per capital expenditure), head-of- householdyearsofschooling,anddwellingquality.Weconstruct dwellingqualityusingasetoffivehousing-relatedvariablescom- monacrosssurveys:roof material,wallmaterial,floormaterial, toilettype,andwatersource.Responsevaluesaresurvey-specific, butwecategorizeeachaseithermodern(1)ortraditional(0).We didhavetoexercisesomejudgmentincategorizingdwellingqual- ity,becausedifferentmaterialsandhousingtypesexistinthethree countries.However,ourfocus wasondistinguishingsolid from weakconstruction,whichwasstraightforward.Thedwellingqual- ityindexisthemeanvalueacrossthebinaryvariables,multiplied byfive.Avalueoffiveindicatesahouseholdwithmoderndwelling
featuresfor allavailablehousingvariables.Table2shows some ofthedescriptivestatisticsofkeyvariablesexaminedbycountry.
Thelastthreevariables–dwellingquality,numberofroomsand automobileownership–representproxiesforwealth.
4. Applianceacquisitiontrends
AsdiscussedinSection2,thecommonbaseofunderstanding today isthat appliance penetration3 would differ byincome in thesamemanneracrossandwithincountries.Onewouldexpect thatatsimilarincomelevelsindifferentcountries,oneshouldsee similarpenetrationlevels,afteradjustingforelectricityaccessand urban/rurallocation.InFig.1weplotnationaltelevisionpenetra- tionforcountriesversusaverageGDPpercapita(PPP-adjusted) forasetof314country-yearcombinations.Wethenoverlaythe within-country relationshipbetween penetrationand per-capita expenditure asobserved in Brazil, India and S. Africa.4 Several observationsarenoteworthy.First,amongcountriesthereissignif- icantvariationinpenetrationatagivenincomelevel,whichimplies
3Bypenetrationwemeantheshareofhouseholdsthathaveatleastoneappliance, insteadofaveragenumberofunitsperhousehold.Theformerdoesnotprovidean indicationofthetotalstock,whilethelattermasksthepenetration,sincemultiple ownershipamongtherichcanhidenoownershipamongthepoor.Theliterature tendstofocusonstock,dueinparttotheconcernforenergyuseandemissions.
4Householdexpenditureisonlyabout60%ofGDPinourselectedcountries.In ordertopresentcountry-andhousehold-leveldataonasinglex-axis,wedivide eachhousehold’sexpenditurebytheshareofhouseholdfinalconsumptionexpen- ditureinGDP.Eventhoughallthreecountrieshavedifferentlevelsofurbanization andelectrification,wedon’tknowtheaverageincomeofelectrifiedhouseholds, soweareunabletopresentthedataforonlyelectrifiedhouseholds,orcontrol- lingforurban/rural.Nevertheless,forthemicrodata,weseethesamepatternwith electrifiedhouseholdsalone.Similarly,ifweadjustthenationalaveragesforelectri- ficationrates,assumingthatelectrifiedhouseholdshavethesameaverageincome asthenationalaverage,thepatternandvarianceisthesame.
Fig.2. AppliancepenetrationbyhouseholdexpenditurelevelamongelectrifiedhouseholdsinurbanareasofIN,BRAandZAF,2009–2012.Dataaretruncated(left)at PPP$750/cap/yrand(right)at97.5percentile.
country-specificfactorsmatter.Second,weseeclearlythehetero- geneitywithinthethreecountriesintherelationshipoftelevision penetrationtoincome,andthatthethreecurvesdifferinshape fromeachother,andfromtheimpliedshapefornationalaverages.
4.1. Notallhighincomehouseholdownwashingmachines
Animplicitassumptioninliteratureisthatallhouseholdswould eventuallyownwhitegoods,assumingrealincomeskeeprising.
Belowweshowtherateofpenetrationforwhitegoodsandmobile phonesinselectEUcountries,includingtwoofthepoorestcoun- triesinCentralEurope,ArmeniaandAlbania,forwhichdatawere available(Table1).Weshowdataformobilephonesasapointof comparison,sinceithashadthefastestandbroadestproliferation ofanydeviceinhistory.5
The data among developed countries show that saturation levelstypically reachover 90%, butwithexceptions.Almostall households have televisions, even in Albania/Armenia. Among theemergingeconomies, televisionownershipseemstoalready approachsaturationinurbanBrazilandChina,despitehighurban
5InternationalTelecommunicationUnion(ITU).ICTfactsandfigures.Availableat:
https://www.itu.int/en/ITU-D/Statistics/Documents/facts/ICTFactsFigures2015.pdf
poverty.Thesameobservationappliestorefrigerators.Itisnote- worthythatalthoughwashingmachinesarealsoownedbyover 95%ofthenon-poorEUcountriesshownandJapan,theyareowned byonly82%ofUShouseholds.
Lookingatmicrodata(Fig.2)revealsthatinS.AfricaandBrazil differentappliancesreachsaturationatdifferentincomelevels,and atdifferentlevelsofpenetration.Washing machinepenetration appearstoplateauat∼80%,whilerefrigerators,liketelevisions, reachcloseto100%penetrationamonghighincomehouseholds inbothcountries(Fig.3).
4.2. Incomeeffectsdifferbyregionandappliancetype
Aboveweshowthatappliancessaturateatdifferentpenetration levels.Herewedescribetrendswithincomechanges,bothacross householdsandovertime.Atverylowincomes,veryhighsharesof householdshaverefrigeratorsinBrazil,fewerhavetheminS.Africa, andsignificantlyfewerinIndia.Thisisn’texplainedjustbyprice,as pricesinBrazilarenotparticularlylow(Fig.4).WhilepricesinIndia arehighest,thereisnodemandforeventhecheapersmallrefrig- erators(<140l)thatareprevalentisinSouthAfrica,presumably amongthepoorerpopulation.
We alsoseethattheshapesof thepenetration curvesdiffer bycountryfortelevisionsandrefrigerators,butappearsimilarfor
Table1
Householdappliancepenetrationinselectindustrializedandemergingeconomies,variousyears(2009–12).
Country Incomepercap(2010$PPP) Electricityaccess Television Mobilephone Refrigerator Washingmachines
US 48,374 100 98.7 93 99.8 82
UK 35,855 100 100 92 100 97
Germany 39,612 100 100 >90 99 96
France 35,867 100 100 89 100 100
Japan 33,741 100 100 93 100 100
Albania 9298 100 98.9 94.1 94.8 NA
Armenia 6376 99.8 98.7 86.9 78 39–49
UrbanChina NA >95 95 100 83.3 81.8
UrbanIndia 10,713 97 87.9 91.1 46.9 17.3
UrbanBrazil 24,093 99.8 95.9 NA 94.9 49.3
UrbanS.Africa 25,149 91.7 84.0 92.1 78.7 44.1
Sources:Nationalstatistics,Statista2014,Euromonitor2009,DemographicandHealthSurveys.ForIndia,BrazilandS.Africasources,seeDatasectionintext.
Fig.3. Appliancepenetrationlevelsbyaffordability(applianceprice(nationalavg)/percapexpenditure)amongelectrifiedhouseholdsinurbanareasofIN,BRAandZAF, 2009–2012.Dataaretruncatedto2.5and97.5percentile.Valuestotheleftmaybeexaggeratedtotheextentthepoorpaybelow-averageprices.
washingmachines.InIndia,forinstance,televisionsandrefrigera- torsseemtoexhibitatippingpointinpercapitaexpenditure,above whichownershipincreasessteeply.WhereasinS.Africa,penetra- tionismore(andinBrazilalmostcompletely)income-inelastic.
Changesinpenetrationlevelsovertimealsodifferbyappliance andbyregion(Fig.5).Inthe5–10yearperiodbetweensurveys, inIndia,despitedecreasingprices,theuptakeoffridgeshasbeen higherathigherincomelevelsthanatlowerincomelevels,while inBrazilandS.Africa,theuptakehasbeengreateratlowerincome
levelscomparedtomiddleincomelevels(penetrationisalready saturatedatthehighestincomelevels).Absolutechangeshavebeen lowestinIndia,higherinBrazilandhighestinSA,whichcan’tbe explainedbypricechanges(Fig.4).Thatis,priceshavedeclined moreconsiderablyinIndia,whereuptake(inrelativeterms)has beenslowest.
Insum,inthethreecountriestheseappliancesbecomewidely owned at different levelsof income, and exhibit different pat- ternsofownershipatlowerincomelevels.Thereare,therefore,
Fig.4.Applianceprices(2005,2010)andtype(2010).Source:EuromonitorInter- national.
driversotherthanincomethatinfluencewhenandhowhouseholds acquiretheseappliances.
4.3. Affordabilityexplainssomeregionaldifferences
Incomeisintendedtobeaproxyforaffordability,butfailsto accountfordifferentprices.Upfrontpurchasecostsmatterforlow- incomehouseholds,astheymaynothavethecapitaltopurchase largeappliances,orlackthecredittobuyfinancing.InFig.3,we showthepenetrationlevelsbyaffordability,whichwedefineas theappliancepricedividedbypercapitahouseholdexpenditure.
Notethatthesearenationalannualaverageprices,sotheyonlydif- ferentiateaffordabilitybetweencountriesandacrosstime.Prices aregenerallyhighestinIndiaonapurchasingpowerparity(PPP) basis,andlowestinS.Africafortelevisionsandfridges,butlowest inBrazilforwashingmachines(Fig.4).6Comparisonsofhousehold adoptionacrosscountriesonthebasisofexpendituresharesare thereforemoreappropriate,andthoseonthebasisofincomecan bemisleading.Forexample,becausepricesaregenerallyhigherin India,theincomepenetrationcurvesunderestimateIndianhouse- holds’propensitytoowntheseappliances(thecurvesshiftleftin Fig.3relativetoFig.2).
Saturationlevelsofappliancesinthethreecountrieswhenmea- suredagainstaffordabilityratherthanincomearemoresimilar.
Onanincomebasis,televisionsandrefrigeratorssaturateurban householdsinBrazilatunder$5K,butnotuntil$15KinS.Africa.In contrast,asaruleofthumb,itseemsthatinallregionsappliances attainfullsaturationwhenappliancecostsarecloseto1percentof percapitaexpenditure.7
Itisalsorevealingthatmanyhouseholdsarewillingtopayprices for appliancesthat exceedtheirper capitaannualexpenditure.
Morehouseholdsinallregionsarewillingtopayovera100per- centoftheirtotalexpenditureontelevisionsthanonrefrigerators, washingmachinesorevenmobilephones.8Furthermore,penetra- tionlevelsarelowerforwashingmachinesatallaffordabilitylevels inallthreeregions.Notethatthisisdespitethefactthatwashing machinesarerelativelycheaperthantelevisionsinallcases,and byfarinIndia.However,therelativepenetrationoftelevisionsand refrigeratorsdiffersbyregion.OnlyinIndiadoesthereappearsto
6Partofthesedifferencesstemfromdifferencesinthepredominanttechnology soldineachmarket,whichisanotherfactordeservingattentionthatweleavefor futureresearch.
7Wetriedformallytestingforsaturation,butwereunableasdatafortopincomes arepoorlysampled.
8Notethatbecauseweusenationalaverageprices,theextenttowhichthe poorarewillingtopaymaybeoverstated–anecdotally,itisknownthatthepoor buyinexpensiveimportedtelevisionsthatareprobablynotaccountedforinthe Euromonitordata.
beaclearorderinginpenetrationlevelsbetweenrefrigeratorsand televisionsatallaffordabilitylevels(seealsoFig.5).
Given the different saturation levels for washing machines vis-à-visrefrigeratorsandtelevisionsaswell,thesedataprovide furtherevidencethathouseholdsacrossregionsplacelowerprior- ityonwashingmachines.Householdpreferencesbeyondpriceand incomeconsiderationsseemtoexplainthispreferenceordering.
5. Quantitativeanalysis–Methods
In order tounderstandtherelative influence ofincome and otherdriversof applianceuptake byhouseholds,weconducted quantitativeeconometricanalysisonourhouseholdsurveydata usingtwoestimationmethods.Thefirst,whichwouldrepresent thestateoftheart,isatraditionallogistic(orlogit)model,while thesecondusesamachinelearningalgorithm(boostedregression trees(BRT)).Withbothmodels,wepredictapplianceownership foreachofthethreeappliances(television,refrigeratorandwash- ingmachines)foreachcountry,poolingbothsurveyperiods,and onlyincludinghouseholdswithelectricityaccess.Below,wefirst describethemachinelearningalgorithmandtherationaleforits use,thendiscusstheresultsandtheirimplications.
Conventionallogitmodelshavethelimitationofbeingrestricted toaparticularfunctionalform,andrequireapriorispecification ofcovariates.WiththeBRTmodel,bothconstraintsarerelaxed.
Onecanincludea‘kitchensink’ofvariables,whichthealgorithm analyzestodeterminethosethathavethestrongestinfluenceon appliance ownership, includingthrough nonlinear relationships andcomplexinteractioneffects.Theuseofsuchaflexibleapproach is advantageous since there is comparatively little theoretical understandingofpeople’sdecision-makingaroundappliances.
BRTisatree-based,ensemblemachinelearningtechnique,sim- ilartothepopularrandomforests,9 thatusesgradientboosting tobuildanensembleofdecisiontreesthataresequentiallyfitto remainingmodelresiduals.10Theoptimalnumberoftreesistypi- callydeterminedvian-foldcross-validation,soastomaximizethe resultingmodel’sexpectedout-of-sampleperformance.Elithetal.
[28]provideanexcellentreviewoftheBRTtechniqueandapplica- tions.WeemploytheBRTimplementationintheRprogramming languagegbmpackage.11
Weknowofveryfewcasesofmachinelearningbeingemployed in energyresearch. Kuanand White [29] compared theperfor- mance of logit models, neural networks and regression trees in predicting appliance ownership in the US. They found that logit out-performed regression trees for in-sample predictions, but regressiontrees outperformedtheothers for out-of-sample predictions.12Morerecently,wefoundonlyoneresearchgroup usingsimilartechniquestounderstanddriversofurbanenergyuse [30,31].
Inordertoseparatetheeffectofmodelchoicefromtheinflu- enceofarichersetofcovariates,werunbothestimationmethods (logitandBRT)witha‘sparse’and‘rich’setofcovariates(predic- tor/independentvariables).The‘sparse’specificationincludesjust incomeandurbanization–thatis,covariatescommonlyusedin past literatureonappliance penetration.The ‘rich’specification includesabroadsetofpotentialcovariatescommontoallofthe
9http://link.springer.com/article/10.1023%2FA%3A1010933404324.
10https://statweb.stanford.edu/∼jhf/ftp/trebst.pdf.
11Greg Ridgeway with contributions from others (2015). gbm: General- ized Boosted Regression Models. R package version 2.1.1. (https://CRAN.R- project.org/package=gbm)
12Thisisnotsurprisingsincemaximumlikelihood(ML)techniquesregularly employ cross-validation to prevent over-fitting and explicitly maximize out- of-sampleperformance,whileconventionalmodelingtechniquesare proneto overspecification.
Fig.5.Appliancepenetrationovertime,bypercapitaexpenditure.Theperiodsshownincludedifferentyearsforeachcountry.Foractualyearsofeachsurvey,seeData section.
Fig.6. MarginaleffectsofrelevantcovariatesinBoostedRegressionTree(BRT)models,Brazil.
countrysurveys.Thissetincludestheaforementionedaffordability metricinsteadofincome,whichadditionallyaccountsforappliance price,andtherefore captureschanges inaffordabilityover time withineachcountry.Dataonelectricityreliabilitywereavailable, andthereforeincluded,onlyforIndia.Forsocial/culturalfactors,we usedraceforBrazilandS.Africa,andreligionforIndia.Othercovari- atesinclude age(oftheheadofhousehold (HoH)),urban/rural,
dwellingquality(seeDatasection),vehicleownership,household size,education(oftheHoH),numberofrooms,male/femaleHoH, andhomerental/own.Wefitthe‘rich’BRTmodelusingallavail- ablecommoncovariates,asubsetofwhichwasdeemedtohave non-zeroinfluence.Thisinfluentialsubsetwasusedtofitthe‘rich’
logitmodel.
Fig.7.MarginaleffectsofrelevantcovariatesinBoostedRegressionTree(BRT)models,India.
6. Quantitativeresults
Overall,thepredictionaccuracyforaggregateappliancepen- etration is quite strong in all the models (See Supplemental Information).Thedifferenceinpredictionaccuracyfromtherich- nessofcovariatesexceedsthatfrommodelchoice.Thevalueofthe BRTmodelwasmoreinidentifyingthesetofinfluentialvariables toincludeinthelogitmodelinthefirstplace.Acrosssurveysand models,thedifferencebetweenpredictedpenetration ratesand actualsiswithin6percentagepoints.Arichmodelthataccounts forhouseholdcharacteristicsbetterpredictsapplianceownership thanonebasedonjustincomeandlocation(urban/rural)alone.13 However,thebenefitofarichsetofexplanatoryvariablesismodest –themagnitudeoftheimprovementinpredictionis,onaverage, withintwopercentagepointsofthepredictionaccuracywiththe sparsemodels.Atthesametime,themarginalpropensityofown- ershipincreasesbyupto30%foranumberoffactorsrelatedto wealth,cultureandotherhouseholdcharacteristics.Theseresults likelyreflectthefactthattheinfluenceoftheseotherfactorsis higheratlowerincomelevels,wherethecontributiontooverall penetrationisrelativelylow.
6.1. Regionaldifferencesintheincomeeffect
Theanalysisconfirmsthatalthoughincomeisstillthestrongest predictor,itsinfluencediffersbyregionandappliance.Theincome effectisstrongestinIndia,sinceitisthepoorestandhasthelowest
13 Amodelwithonlyincome(andnoturban/rural)hasmarginallydifferentpre- dictions.
penetrationforallappliances.Accordingtothelogitmodel,foran increaseinannualincomeofPPP$1000percapinIndia,theoddsof owningatelevisionincreaseby52%andthatofowningarefrigera- torincrease30%,andthatofowningawashingmachineincreaseby about18%.14Thesameincomechangehasnoeffectontelevision ownershipinS.AfricaandBrazil,andatrivialincreaseinBraziland S.Africaforbothwashingmachinesandrefrigerators.However, notingthatBrazilandS.Africahavefourtimestheaverageincome asIndia,acomparisonbetweentheoddsofownershipbetweenthe 25thand75thpercentileofthepopulationismoreinformative.15 Fortelevision,oddsincreaseby43%inbothcountries;forrefriger- ators,∼60%inBraziland216percentinS.Africa;andforwashing machines∼60%inS.Africabut288%inBrazil.
Insummary,therearecleardifferencesintheelasticityofadop- tiontochangesinincomeacrosscountries.Buttherearefurther differencesintheabsolutelevelsofpenetration.Someofthesedif- ferencesareexplainedbyheterogeneityinnon-incomehousehold characteristics.Thisisdiscussednext.
6.2. Marginaleffectsofnon-incomedrivers
The importance of non-income factors is understated when examiningaggregatepenetration.Thatis,ifonewereinterested in‘who’hasparticularappliances(forthosethatarenotownedby
14Thechangeinprobabilityassociatedwithachangeinoddsiscontingentonthe initialprobability.Probabilitychangesareusuallysmaller.Forachangeinoddsof 5and50percent,themaximumchangeinprobabilityis∼1and∼10percentage pointsrespectively.
15ForIndia,anincomechangeof$1000percapdoes,coincidentally,correspond approximatelytothedifferencebetweenthe25thand75thpercentile.
Fig.8. MarginaleffectsofrelevantcovariatesinBoostedRegressionTree(BRT)models,SouthAfrica.
allhouseholds),ratherthanjust‘howmany’householdshavethese appliances,additionalcovariatesbecomemoreimportant.
Therichmodelrevealssomeofthehouseholdcharacteristics thatmayexplainthisheterogeneousbehavior.Themarginaleffects ofthesevariablesareshownbycountryinFigs.6–8.InBraziland S.Africa,indicatorsofwealthandraceseemtostronglyinfluence ownership.Thepossessionofautomobilesandthequalityandsize ofthedwelling,bothhaveastrongeffect,todifferentdegreesin eachcountry.Thiscouldreflecthouseholds’abilitytogetcredit,or theavailabilityofcommunallaundryfacilities(forexample,sub- urbanhomesvsapartmentbuildings).InIndiaasimilareffectis seenforrefrigeratorownership–asreflectedintheownershipof eithermotorcyclesorcars,anddwellingquality.Theeffectofbetter electricitysupplyisrelativelysmall,andonlyataverageavailabil- ityhigherthan18hday(whichmakessense,giventhebenefitof havingoneisfairlylowiftherefrigeratorisoffformorethanafew hoursinhotweather).Theeffectofdwellingqualityisparticularly strongfortelevisionsinIndia.
Theinfluenceofrace/ethnicity(controllingforallotherfactors) isparticularlyinteresting–inbothS.AfricaandBrazil,beingcol- oredorwhite(overbeingblack)hasastrongmarginaleffecton washingmachineownership(Figs.6–8).InS.Africa,theincreasein marginalprobabilityofownershipforbeingwhiteishigherthan thatfor increased affordability. It is possiblethat theinfluence ofracereflectshouseholdculturalpreferencesorexternal mar- ketconditions,suchasdifferentialaccesstomarketsforcredit,or forappliancesthemselves.InIndia,culturalpreferencesrelatedto religionmayplayaroleinrefrigeratorownership(Fig.7).InIndia, thereisasmaller,albeitnoticeable,effectofreligiononrefriger- atorownership–whereinSikhshaveahigherchanceofowning one.ThismayhavetodowiththefactthatSikhsareknownfora
highconsumptionofmilkproducts.16Therelativeimportanceof thesefindingsdiffersbycountry,sinceSikhscompriselessthan1%
ofthepopulation,blackscomprise9%inBrazil,andblackscomprise 79%inS.Africa.Nevertheless,thesefindingsareillustrativeofthe importanceofnon-economicfactors.
Otherwise,ceterisparibus,urban,morepopulous,larger,more educated,andbetterqualityhomes,arelikelytohavemoreappli- ances.Thesefindingsareconsistentwiththoseofpreviousstudies.
However,ourresultsshowthattheinfluence ofmany ofthese driversisgradualoverbroadsegmentsofthepopulation,rather thanhaving‘tippingpoints’,asinthecaseofaffordability. That is,themarginalprobabilityofownershipincreasessteadilyfrom belowandthroughthemeanlevelsforthepopulation(seeTable2), andflattenoutthereafter.Thismayexplainwhytheaggregatepre- dictionratesdonotshiftsosignificantlyintherichmodel.However, foranygivensetofhouseholdsataparticularincomelevel,the combinationofallthesemarginaleffectswouldmakethepredicted ownershipfarmoreaccuratewiththerichmodelthanthesparse one.
7. Conclusions
Wehaveexaminedpatternsofownershipoftelevisions,refrig- erators andwashing machinesin India,Braziland South Africa usinghousehold-levelsurveydata.Thisstudyforthefirsttimepro- videsquantitativeevidenceonahierarchyofpreferencesamong
16Basedonourcalculationsofmilkproductconsumption intheIndianNSS 2011–12,Sikhsconsumemorethanallotherreligiousgroups,andmorethandouble thatofthepredominantgroupswithhigherpopulations.
Table2
Descriptivestatisticsofkeycovariatesbycountry/survey(forurbanhouseholdswithelectricityaccessonly).
India S.Africa Brazil
04–05 11–12 05–06 10–11 02–03 08–09
Expendpercap 1542 2191 6928 7210 5704 6516
1553 2240 12,106 11,320 9435 12,621
Yearsofeducation 8.1 7.9 9.7 9.8 6.2 10.7
5.1 5.0 4.0 3.8 4.6 2.9
Householdsize 4.9 4.7 3.5 3.6 3.6 3.2
2.1 2.2 2.3 2.2 1.8 1.6
AgeofHoH 46.1 50.0 42.0 45.7 45.9 47.7
12.9 12.8 15.0 14.6 15.3 15.6
Dwellingquality 4.0 4.2 3.9 4.1 4.8 5.0
1.1 1.0 1.4 1.1 0.8 0.3
Numberofrooms 2.6 2.8 4.4 4.5 4.6 4.7
1.5 1.6 2.3 2.5 1.9 1.8
AutomobileOwners(%) 3.7 7.5 26.5 39.8 32.3 35.3
Figuresinitalicsshowstandarddeviations.Expendituresin$PPP2010.HoH:Headofhousehold.
electricappliances.Ifdevelopingeconomiesexhibitthesamepat- ternsasweobserveinoursample,eventuallymosthouseholdswill havetelevisionsandrefrigerators,but,alowersharewouldlikely havewashingmachines.
Aswithpreviousstudies,wefindthathouseholdcharacteristics, includingofthephysicalhouse,andofinhabitants’demographic characteristics,haveaninfluenceonapplianceownership.How- ever,inadditionweidentifynewfactorsrelatedtoaffordability, wealthandidentity,someofwhichmaybeparticulartodeveloping economies.Affordabilityofappliances,definedastheirexpendi- tureshare,providesamorecomparablemetricforcross-country comparisonthandoesjustincome.Indicatorsofwealth,suchas ownershipofvehiclesorhomesizeandquality,alsoinfluencepur- chases.Surprisingly,race(color)seemstoplayadistinctroleinboth BrazilandSouthAfricainexplainingwashingmachineownership, whilereligionwasfoundtoplayaroleinrefrigeratorownership in India.Thesedifferences couldreflect culturalpreferences,or differencesinmarketaccess.Thismeritsfurtherexploration.The technologyandsizeofappliancesthatarepurchasedincountries alsodifferinwaysthatarerelatedto,butnotexclusivelyexplained by,affordability,butwhichhavenotbeenexploredindetailinthis study.
Affordabilitymaybethemostsalientinsightfromthisstudyto incorporateinfuturedemandprojections.Forecastsofappliance uptakebasedonmacroeconomictrendsalonemaybeinaccurateto theextentthatappliancepricetrajectoriesdivergeoverspaceand time.Incorporatingotherfactorsintolong-termforecastsofenergy demandwouldbechallenging,andpossiblylessuseful,giventhe uncertaintyintheirpredictivevalueandintheabilitytoproject trendsataglobalscale.
Theseresultsareonesteptowardbetterunderstandingtheroles ofmarket barriersand householdbehavior inappliance uptake indevelopingcountries.Theseinsightscaninformthedesignof energyefficiencyandequitableaccesspoliciesandforecastnear termresidentialenergydemandgrowth.
Acknowledgements
This research was made possible by funding provided by theEuropean Research CouncilStarting Grant 637462.Support from ERC-STG-2014, European Research Council Starting Grant
#637462,‘Decent LivingEnergy’, and from SCHEMA, Socioeco- nomicHeterogeneityinModelApplications,IIASAcrosscutproject.
AppendixA. Supplementarydata
Supplementarymaterialrelatedtothisarticlecanbefound,in theonlineversion,athttp://dx.doi.org/10.1016/j.erss.2017.03.005.
References
[1]M.Bazilian,R.J.Pielke,Makingenergyaccessmeaningful,IssuesSci.Technol.
29(4)(2013)74–79.
[2]D.Ürge-Vorsatz,etal.,Chapter10-EnergyEnd-Use:Building.InGlobal EnergyAssessment-TowardaSustainableFuture,CambridgeUniversity Press,Cambridge,UKandNewYork,NY,USAandtheInternationalInstitute forAppliedSystemsAnalysis,Laxenburg,Austria,2012,pp.649–760, Availableat:http://www.globalenergyassessment.org.
[3]L.F.Cabeza,etal.,Investigatinggreenhousechallengefromgrowingtrendsof electricityconsumptionthroughhomeappliancesinbuildings,Renew.
Sustain.EnergyRev.36(2014)188–193,Availableat:
http://dx.doi.org/10.1016/j.rser.2014.04.053.
[4]M.A.McNeil,etal.,Bottom-UpEnergyAnalysisSystem(BUENAS)-an internationalapplianceefficiencypolicytool,EnergyEfficiency6(2)(2013) 191–217.
[5]B.J.vanRuijven,etal.,ModelprojectionsforhouseholdenergyuseinIndia, EnergyPolicy39(12)(2011)7747–7761.
[6]M.A.McNeil,V.E.Letschert,Modelingdiffusionofelectricalappliancesinthe residentialsector,Berkeley,CA,2010.
[7]M.Farsi,M.Filippini,S.Pachauri,FuelchoicesinurbanIndianhouseholds, Environ.Develop.Econ.12(06)(2007)757–774.
[8]A.Kemmler,FactorsinfluencinghouseholdaccesstoelectricityinIndia, EnergySustain.Develop.11(4)(2007)13–20.
[9]L.Schipper,A.Ketoff,Explainingresidentialenergyusebyinternational bottom-upcomparisons,Annu.Rev.Energy10(1985)341–405.
[10]H.Nakagami,LifestylechangeandenergyuseinJapan:householdequipment andenergyconsumption,Energy21(12)(1996)1157–1167.
[11]J.A.Rosas-Flores,D.Rosas-Flores,D.M.G??lvez,Saturation,energy consumption,CO2emissionandenergyefficiencyfromurbanandrural householdsappliancesinMexico,EnergyBuild.43(1)(2011)10–18.
[12]R.B.Howarth,etal.,Thestructureandintensityofenergyuse:trendsinfive OECDNations,EnergyJ.14(2)(1993)27–45.
[13]B.L.Bayus,Havediffusionratesbeenacceleratingovertime?Market.Lett.3 (3)(1992)215–226.
[14]S.Bowden,A.Offer,Householdappliancesandtheuseoftime:TheUnited StatesandBritainsincethe1920s,Econ.Hist.Rev.47(4)(1994)725–748.
[15]S.C.Bhattacharyya,InfluenceofIndia’stransformationonresidentialenergy demand,Appl.Energy143(2015)228–237.
[16]N.Zhou,etal.,AnalysisofpotentialenergysavingandCO2emissionreduction ofhomeappliancesandcommercialequipmentsinChina,EnergyPolicy39 (8)(2011)4541–4550.
[17]C.D.Wolfram,O.Shelef,P.J.Gertler,HowWillEnergyDemandDevelopinthe DevelopingWorld?2012.
[18]M.Auffhammer,C.D.Wolfram,PoweringupChina:incomedistributionsand residentialelectricityconsumptionpoweringupchina:incomedistributions andresidentialelectricityconsumptionAm.Econ.Rev.104(May)(2014) 575–580.
[19]M.A.McNeil,V.E.Letschert,Futureairconditioningenergyconsumptionin developingcountriesandwhatcanbedoneaboutit:thepotentialof
efficiencyintheresidentialsector,2008,Availableat:
http://escholarship.org/uc/item/64f9r6wr.pdf.
[20]H.Batih,C.Sorapipatana,Characteristicsofurbanhouseholds’electrical energyconsumptioninIndonesiaanditssavingpotentials,Renew.Sustain.
EnergyRev.57(2016)1160–1173.
[21]USEIA,ResidentialdemandmoduleoftheNationalEnergyModelingSystem (NEMS),ModelDocumentation,WashingtonD.C,2014.
[22]J.O’Doherty,S.Lyons,R.S.J.Tol,Energy-usingappliancesandenergy-saving features:determinantsofownershipinIreland,Appl.Energy85(7)(2008) 650–662.
[23]E.Leahy,S.Lyons,EnergyuseandapplianceownershipinIreland,Energy Policy38(8)(2010)4265–4279.
[24]Z.Rong,Y.Yao,Publicserviceprovisionandthedemandforelectric appliancesinruralChina,ChinaEcon.Rev.14(2)(2003)131–141.
[25]S.Matsumoto,Howdohouseholdcharacteristicsaffectapplianceusage?
ApplicationofconditionaldemandanalysistoJapanesehouseholddata, EnergyPolicy94(2016)214–223.
[26]EmilyC.Lawrance,Povertyandtherateoftimepreference:evidencefrom paneldata,J.PoliticalEcon.99(1)(1991)54–77.
[27]KennethTrain,Discountratesinconsumers’energy-relateddecisions:a reviewoftheliterature,Energy10(12)(1985)1243–1253.
[28]J.Elith,J.R.Leathwick,T.Hastie,Aworkingguidetoboostedregressiontrees, J.Anim.Ecol.77(4)(2008)802–813.
[29]C.-M.Kuan,H.White,Predictingapplianceownershipusinglogit,neural network,andregressiontreemodels,1990.
[30]F.Creutzig,etal.,Globaltypologyofurbanenergyuseandpotentialsforan urbanizationmitigationwedge,Proc.Natl.Acad.Sci.112(20)(2015) 6283–6288.
[31]G.Baiocchi,etal.,AspatialtypologyofhumansettlementsandtheirCO2
emissionsinEngland,GlobalEnviron.Change34(2015)13–21.