Contents lists available atScienceDirect
Resource
and
Energy
Economics
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 / r e e
Exploring
the
spatial
and
temporal
determinants
of
gas
central
heating
adoption
Daire
McCoy
a,b,∗,
John
Curtis
b,caTheGranthamResearchInstitute,LondonSchoolofEconomics,UK
bTheEconomicandSocialResearchInstitute,Ireland
cTrinityCollegeDublin,Ireland
a
r
t
i
c
l
e
i
n
f
o
Articlehistory:
Received5December2016 Receivedinrevisedform 20December2017 Accepted26December2017 Availableonline6January2018
JELclassification:
Q40 C31 C36
Keywords:
Residentialfuelchoice Spatialeconomics
Instrumentalvariablesestimation
a
b
s
t
r
a
c
t
Inordertobetterunderstandthepotentialforbothpolicyandtechnologicalimprovements toaidcarbonabatement,long-termhistoricalinformationonthetime-pathoftransition frommoretraditionaltocleanerfuelsisuseful.Thisisarelativelyunderstudiedelementof thefuelswitchingliteratureinbothdevelopedandemergingeconomies.Thisresearchadds tothisliteraturebyexaminingtheadoptiontime-pathofnetworkgasasaheatingfuel.We mergeauniquedatasetongasnetworkroll-outovertime,withothergeo-codeddataand employaninstrumentalvariablestechniqueinordertosimultaneouslymodelsupplyand demand.Resultsindicateanon-linearrelationshipbetweentheproportionofhouseholds usinggasastheirprimarymeansofcentralheatingandthelengthoftimethenetwork hasbeeninplaceineacharea.Proximitytothegasnetwork,peatbogs,andareaswhich havebannedtheconsumptionofbituminouscoalalsoaffectgasconnections.Variationsin socioeconomicanddwellingcharacteristicsatarealevelcanalsohelpexplainconnections tothegasnetwork.Abetterunderstandingofthisvariationiscrucialindesigningtargeted policiesandcanaidnetworkexpansiondecisions.
©2018TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCC BY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction
Policyinterestinresidentialfuelchoiceandconsumptionhasalonghistory(e.g.Halvorsen,1975;Houthakker,1951).
Inrecentyearspolicyfocushascentredonassociatedhealthoutcomesandeconomicgrowthindevelopingcountriesand
moregenerallyonthecontributiontogreenhousegasemissions.WithapproximatelyonequarteroftheEU’stotalprimary
energyrequirementintheresidentialbuildings,1thesectorisafocalpointgiventheEU’sambitiontoreducegreenhouse
gasemissions(EuropeanCommission,2014).Fuelswitchingawayfromcarbonintensivefuels,suchaspeatandcoaltoless
carbonintensivefuels,suchasgasorrenewablesisonewaytheresidentialsectorcanreduceemissionsyetsatisfyenergy
servicedemands.
Abodyofresearchwithindevelopmenteconomicsfocusesontheso-called‘energyladder’,inwhichhouseholdstransition
fromtraditionalheatingandcookingfuels,suchasbiomassorwood,tofuelssuchasgasorelectricityastheirincomelevels
increase(HosierandDowd,1987).Ashouseholdswillcontinuetousetraditionalfuelssuchasfirewoodalongwithmodern
fuels,switchingbackinresponsetorelativepricesandotherfactors(Wickramasinghe,2011;VanderKroonetal.,2013)
∗ Correspondingauthor.Presentaddress:GranthamResearchInstitute,LondonSchoolofEconomicsandPoliticalScience,LondonWC2A2AE,UK.
E-mailaddress:[email protected](D.McCoy).
1 Seehttp://ec.europa.eu/eurostat/web/energy/data/energy-balancesfordetails. https://doi.org/10.1016/j.reseneeco.2017.12.004
somehavearguedthatamultiplefuelmodelismoreappropriate(Maseraetal.,2000).Amongthekeydeterminantsof
fuelchoiceamonghouseholdsindevelopingcountriesarefuelprices,income,andeducation,aswellassecurityofsupply
considerationsforfuelssuchasgasandelectricity(Alemetal.,2016;Beheraetal.,2016;MensahandAdu,2015;Zhangand
Hassen,2017).
Residentialfuelchoice2 andfuelswitching,arealsoaresearchfocusindevelopedeconomies.Forexample,therehas
beenaparticularinterestinrecentyearsintothedecisiontoadoptrenewableormoreefficientresidentialheatingsystems
(MahapatraandGustavsson,2008;SophaandKlockner,2011;MichelsenandMadlener,2012).Acrossnumerousstudiesand
countriesthereisageneralconsensusontherangeoffactorswhichdetermineresidentialfuelchoice.Thesearedescribed
indetailinSection2.Ourresearchaddstoboththisliteratureandtheliteratureexaminingtheacquisitionofenergyusing
assets.3
OurfocusinthispaperisontheadoptionofgascentralheatinginIreland.Irelandprovidesaveryinterestinglensthrough
whichtoexaminethediffusionofanenergy-usingassetovertime.Aculturallegacyofsolidfuelusage,drivenbyplentiful
localendowmentsofpeat,createdareluctancetoswitchtomoremodernheatingsystems.Thiscontrastswithastrong
policypushinrecentyearstoencouragegreaterusageofrenewableenergy,andrecentlegislationprohibitingthesaleand
useofbituminouscoalfordomesticheatinginurbanareas.Accesstonetworkgashasbeenavailableinsomelocationsin
Irelandformorethancentury,however,networkconnectionscanstillberelativelylowinsomelocationsadjacenttothe
gasnetwork.
Weareparticularlyconcernedwithunderstandingmoreabouttheadoptiontime-pathofnetworkgasasadomestic
heatingfuel.Theremaybeseveralreasonswhyatime-lagintheadoptionofmoreefficientheatingmethods,once
avail-able,exists.Therangeoffactorsincludefinancialbarriers,spatialproximitytoalternativeenergysources,culturallegacies
resultinginpreferencesforcertainfuels,misinformationoralackofinformationonalternatives,oruncertaintyaboutfuture
energyprices.Heterogeneityofpreferencesinthepopulationcanalsoexplainvariationsinthetimingofadoption,evenin
caseswherethenewtechnologyisqualitativelybetterthantheexistingone.
Thetime-laginadoptionisarelativelyunderresearchedaspectoffuelswitching,whichgenerallyconsidernetwork
accessasabinaryvariableatapointintime.Ourfocusisenabledbyaccesstoararedatasetcomprisingdetailedinformation
onthelocationandtimingoftheexpansionofeachindividualsegmentoftheoftheIrishgasnetworkover100years.Thisis
linkedtoinformationonthelocationofeveryresidentialdwellinginthecountryandcombinedwithspatialcross-sectional
dataonarea-levelfuelchoicealongwithinformationondwellingattributesandthesocio-demographiccharacteristicsof
households.Suppliersarelikelytoextendthegasnetworktoareasofhighdensity,orthosewithahigherprobabilityof
adoption,andonlythosehouseholdsincloseproximitytothegasinfrastructurecanadopt.Nottakingaccountofthiscould
potentiallybiasourestimates.Toaccountforthissimultaneity,weestimateatwo-stageleastsquaresspecification,allowing
ustoidentifythetime-pathofnetworkroll-outinthesecond-stagegasadoptionequation.
Resultsindicateanon-linearrelationshipbetweenthelengthoftimethenetworkhasbeeninplaceandtheproportion
ofgasusersineacharea.Eachyearthenetworkhasbeeninplaceisassociatedwitha3percentagepointincreaseingas
connectionsonaverage,andthiseffectdecreasesovertime.Variationindistancetothenetworkisasignificantdeterminant
ofconnections,evenforareasincloseproximitytothenetwork.Proximitytopeatsources,suchasbogsisnegatively
associatedwithgasconnections,whileabanonthesaleandburningofbituminouscoalwhichwasinplaceinvariousurban
locationsinIrelandin2011,ispositivelyassociatedwithgasadoption.Oureconometricapproachallowsustoalsoprovide
somescenarioanalysiswhichsimulatesgasnetworkexpansionsyettobeundertakenandthepotentialimpactoftheseon
uptakeandtheassociatedchangesinCO2emissions.
Thelayoutoftheremainderofthepaperisasfollows.Thenextsectionplacesourresearchinaninternationalcontext.
FollowingthisinSection3weprovidesomebackgroundonthehistoricaldevelopmentofresidentialfuelusageinIreland,
includingthegrowthinnetworkgasusage.Section4outlinesthemodelandestimationstrategywepropose,whichis
followedbyanoverviewofthedataused.Estimationresultsandascenarioanalysissimulatinggasnetworkexpansionare
presentedinSections6and7.Section8outlinesarangeofrobustnesschecksundertaken.Section9concludesandprovides
someinsightsforpolicy.
2. Relatedliterature
Asmentioned above,ofprimaryrelevancetothisresearchistheliteratureconcerningfuelchoiceandswitching.In
additiontothis,wealsodrawonotherresearchexaminingtheacquisitionofenergyusingassets.Withregardtotheformer,
thekeydeterminantsareconsideredinturninthefollowingparagraphs.
Buildingattributes,particularlypropertyageandtypeaffectfuelchoice.MichelsenandMadlener(2016)findthatolder
homesarelesslikelytoswitchtorenewableheatingsystems,possiblyreflectingunsuitableexistingheatinginfrastructure.
Theinhabitantsofolderpropertiesaremorelikelytouseoil,firewoodandcoal,whereasthoseinmorerecentlybuilt
propertiesaremorelikelytousegasorheatpumps(LauretiandSecondi,2012;MichelsenandMadlener,2012),though
2Theterm‘residentialfuelchoice’isusedinterchangeablywith‘residentialheatingsystem’,assomeoftheliteraturefocusesonparticularheating technologies,e.g.heatpumps,ratherthanthefueltypes.
Lillemoetal.(2013)findthatdwellingsizeandtypeimpactonheatingsystemchoicebutnottheproperty’sage.Larger
sizedpropertiesaremorelikelytousegasforheatinginsteadofsolidfuels(Lillemoetal.,2013;MichelsenandMadlener,
2012).Determinantsoffuelorheatingsystemsinnewlybuiltpropertiesgenerallydifferstothatfortheexistinghousing
stock.MichelsenandMadlener(2012)concludethatchoiceofaheatingsysteminnewlybuilthomesishighlyinfluenced
bytheoccupants’environmentalpreferences.
Occupants’socio-economiccharacteristicsalsoimpactonfuelchoice,withincome,age,educationandeconomicstatus
beingparticularlyrelevant.Anumberofstudiesfindthatlowerincomesareassociatedwithoilandsolidfuels,whichare
moreemissionsintensive(Fuetal.,2014;LauretiandSecondi,2012;Özcanetal.,2013)thoughtherearemanyotherstudies
thatfindonlyaminorincomeeffectornone(e.g.Braun,2010;Lillemoetal.,2013;Coutureetal.,2012).Theeffectsofhigher
educationandeconomicstatusonfuelchoicearegenerallysimilartothoseassociatedwithincome.InthecaseofageÖzcan
etal.(2013)findthathouseholdheadsaged50andabovearemorelikelytochoosegas,oilandelectricitycomparedtocoal
andothersolidfuelsforreasonsofeaseofuseandforhealthconcerns.Onthecontrary,DeckerandMenrad(2015)findthat
neitherage,educationnorincomeareimportantvariablesinexplainingchoiceofresidentialheatingsystemsinGermany.
Inertia,peereffectsandmotivational impactshavealsobeenfoundtoimpactonfuelchoice.Householdsareoften
reluctanttoadoptmoreenergyefficientoptions,evenifitisfinanciallyadvantageousfor themtodo so.This
energy-efficiencygapalsocharacterisesthereluctancetoadoptothertypesofenergyefficientappliancesthatofferseemingly
positivebenefit(AllcottandGreenstone,2012;Blumsteinetal.,1980;JaffeandStavins,1994).Theinfluenceofpeersisan
importantdeterminantofdecisionsrelatingtoheatingsystemchoice(DeckerandMenrad,2015;MichelsenandMadlener,
2013).Otherimportantmotivationalfactorsincludeattitudetoparticularheatingsystemsorfuels,personalcomfortand
externalthreats,thelaterofwhichreferstoeitheranapprehensionrelatingtodependencyonfossilfuelsorclimaterelated
environmentalconcerns.
Regionalorculturaldifferences,includingthelocalavailabilityofparticularfuelssuchasfirewood,canimpactonfuel
choice(Braun,2010;Fuetal.,2014;LauretiandSecondi,2012).Whileweatherisfrequentlyincludedasacovariatein
modellingenergyconsumptionitalsohasanimpactonfuelchoice,similartoaregionaleffect.Inanumberofcases30-year
meanweatherdataisfoundtohaveastronginfluenceonfuelchoice,withhighertemperaturelocationslesslikelytouse
oilorsolidfuels(Fuetal.,2014;Mansuretal.,2008).
Fuelpricesandheatingsystemcapitalcostshavesubstantialimpactsonhomeheatingdecisions.Thecapitalcostof
heatingsystemequipmentcanactasabarrierinfuelchoicedecisionsduetobudgetconstraints,however,itisdifficultto
captureempiricalevidenceinrevealedbehaviourdata.MichelsenandMadlener(2016)findthatcapitalcostsratherthan
fuelpricesareanimportantmotivationalfactorinsuchdecisions.Inanumberofstatedpreferencestudiescapitalcosts
areanimportantattributeorpotentialbarrierassociatedwithresidentialheatingsystemchoicedecisions(Rouvinenand
Matero,2013;ScarpaandWillis,2010).Therearerelativelyhighimplicitdiscountratesassociatedwithelectricityandoil
basedheatingsystemscomparedtodistrictheating,geothermalorwood-basedsystems.Fuelpricesarecertainlyimportant
considerationsinhouseholdfuelchoicedecisionsindevelopingcountries(Alemetal.,2016;MensahandAdu,2015;Zhang
andHassen,2017)butthereismixedevidenceindevelopedcountries.Inthestated-preferencestudiesfuelpriceshavea
significantimpact(RouvinenandMatero,2013;ScarpaandWillis,2010)butonlyasmallnumberofotherempiricalstudies
includefuelpricesasapotentialdeterminantoffuelchoice.Mansuretal.(2008)findclearown-priceandcross-priceeffects
onfuelchoicedecisions,whileCoutureetal.(2012)findapriceeffectassociatedwithfirewood,theonlyfuelpricethey
consider.Numerouspapers examiningdeterminantsofhouseholdfuelchoicedonotincludefuelpricesasexplanatory
variables,thoughthatmayreflectdifficultyofacquiringsuchinformationforcross-sectionaldatasets(e.g.Fuetal.,2014;
MichelsenandMadlener,2012;Özcanetal.,2013;LauretiandSecondi,2012).
Accesstothenaturalgasnetworkisalsoanimportantfactorthataffectsresidentialfuelchoice,thoughtheissuehas
receivedrelativelylittleattentionintheliterature.Mansuretal.(2008)findthatUShouseholdswithnetworkaccessmake
differentconsumptionchoices comparedtothose withoutaccess.Theyareunabletodetermineifthosedifferencesin
consumptionchoicesaresolelyduetonetworkaccessandconsequentlyanalysefuelchoice(andconditionaldemand)
separatelyforhouseholdswithandwithoutnetworkaccess.Coutureetal.(2012)takeadifferentapproachandinclude
networkaccessasacovariatewithinamultinomiallogitmodeloffuelchoiceintheMidi-PyrénéesregionofFrance.Grid
accessincreasesthelikelihoodthatapropertyusesgasastheprimarysourceofenergyby8percentagepoints,withoil
beingthefuelthatisdisplacedtothegreatestextent.InIrelandFuetal.(2014)findthatthelikelihoodofsolidfuelsbeing
theprimaryresidentialheatingsourcedeclinesby4percentagepointsinareaswithinathresholddistanceofthenatural
gasnetwork.
Inadditiontotheliteratureonfuelchoicedecisionsthereisaparallelliteratureontheacquisitionofenergy-usingassets,
e.g.aresidentialheatingsystem,thatisalsorelevant.OnesideofthatliteraturehasitsorigininBassdiffusionmodels
(Bass,1969)whereadoptionismodelledasasigmoidalfunctionovertime,withadoptionslowatfirst,thenaccelerating
beforereachingaplateau.Applicationsincludemodellinghouseholds’adoptionofheatpumpsandphoto-voltaicpanelsasa
functionofage,education,information,andfinancialincentives(Hlavinkaetal.,2016;Islam,2014).Energyassetacquisition
isalsostudiedinthecontextofenergyconsumptionwithDubinandMcFadden(1984)amongthefirsttohighlightthat
assetownershipisendogenousinanenergydemandmodel.RecentapplicationsincludeDavisandKilian(2011)whomodel
naturalgasdemandintheUSandalsoMansuretal.(2008),whichmodelsfuelchoiceratherthanheatingsystemchoicein
thecontextofmodellinghouseholdfuelconsumption.Gertleretal.(2016)havemodelledtheeffectofhouseholds’income
Fig.1. ResidentialfuelsharesinMtoe1990–2014.
Source:Datafromhttp://statistics.seai.ie/
theyfindthatcreditconstrainedhouseholdsaremorelikelytopurchaseenergyassetsoncetheirincomepassesathreshold
levelandfurthermorethatthethresholdlevelvariesdependingonthetimingofacquisition.Thissuggeststhattheimpact
ofnetworkgasavailabilityonheatingsystemorfuelchoiceisnon-linearandcannotbeadequatelycapturedwithadummy
variableindicatingavailabilityofanetworkconnection.
3. Background:residentialfuelusageinIreland
Irelandhasalonghistoryofsolidfuelusage,andinparticularpeatusageintheresidentialsector.Mokyr(2013)cites
reportsfromthe1830’sdescribingthegeographicalubiquityofpeatandtheintensityofitsusage.Whilecertainplaces,
suchasSouthAntrimandLimerickhaddepletedtheirreservesbythispoint,itwassoplentifulthroughouttherestofthe
countrythatitwastakenforgranted,and“peoplelivingaslittleas4milesawayfromasourceofturfalreadyconsidered
themselvesinconvenienced”.Peatcontinuedtobetheprimarysourceoffuelforhomeheatinguntilrelativelyrecentlyand
thegeographicalrelationshipbetweenthelocationofsolidfuelresourcesanditsusagepersists(Fuetal.,2014).Peatisstill
commonlyharvestedfrompeatbogsbythepublicandalsosoldaspeatbriquettes.
Asrecentlyas1990,theproportionofhouseholdsusingsolidfuelastheirprimarymeansofspaceheatingwasashighas
60%.Thishadfallento16%by2014.4However,62%ofhouseholdscontinuetouseastove,rangeoropenfireasasecondary
heatingsource,andthemajorityoftheseusesolidfuel.5
Comparableresidentialfuelusagetrendsinkilotonnesofoilequivalentfrom1990–2014aredemonstratedinFig.1.The
fallingshareofsolidfuelisevident,whichhasbeenreplacedbyariseingas,oilandelectricityusageprimarily.Renewable
energyhasnotyetestablisheditselfdirectlyindomesticheating,howeverrenewablesourcesaccountedfor14.5%ofenergy
inputstoelectricitygenerationby2014(SEAI,2015).
IntermsofCO2emissions,eventhoughfinalenergyuseinthedomesticsectorincreasedby26%between1990and2011,
energy-relatedCO2emissionsfellby2.7%,reflectingthedecreasingshareofsolidfuelusageandtheimprovedefficiencyof
oilandgascentralheatingboilers(SEAI,2013).
3.1. Gasusage
Roganetal.(2012)provideacomprehensivesummaryofgasnetworkexpansionandusagetrendsinIrelandbetween
1990and2008.Thegastransmissioninfrastructurehadextendedtoanumberoflargetownsandcitiesby1990,however
90%ofgascustomerswerestillresidentinthetwolargestcitiesofDublinandCork.Thatdecadesawanexpansionofthe
transmissioninfrastructureoutwardfromDublin,alongbothnortheastandsoutheastcoastsandwesttothefast-growing
commutertownsinthegreaterDublinarea.Themid-2000ssawanextensionwestwardslinkingDublinwithGalway,from
hereitwasfurtherextendedtothenorthwestbythelate2000s.Thisextensionresultedinaconstantannualcustomergrowth
4http://statistics.seai.ie/
rateof9%overtheperiod1990–2008.Thereissignificantspatialvariationhowever,andby2014,naturalgascustomers
werestillaslowas5%insomewesternareas(CSO,2016).
Overthisperiodconsumptionincreasedby470%(Roganetal.,2012).Thiswasmainlythroughagrowingcustomerbase,
changesinthedwellingstock,andchangingintensityofusage.Weathereffectsarealsoimportant.Fromamicroeconometric
pointofview,Conniffe(1996)andHaroldetal.(2015)alsofindweatherastrongpredictorofseasonaldemand.Thisresearch
alsotiesinwithinternationalresearchofgasconsumptionandmoregeneralspaceheating,whichfindthatdwelling
char-acteristicsandthesocioeconomiccharacteristicsofinhabitantshaveasignificantimpactondemand(Rehdanz,2007;Meier
andRehdanz,2010;Wyatt,2013).
Thefollowingsectionoutlinesourmethodologyandsomeempiricalconsiderationsonemustconsiderwhenmodelling
adoptionatarealevel.
4. Methods
Theutilityconsumersreceivefromadoptinggascentralheatingislikelytobeafunctionofarangeoffactorssuchas
therelativepriceofgascomparedwithalternatives,alongwiththeirsocioeconomicanddwellingcharacteristics.Physical
constraintsonadoptionexistandwillrelatetoeachhousehold’sproximitytothegasinfrastructure.Thekeypricevariable
ataspatiallevelistheconnectioncost.Thisisafunctionofdistancetothenetworkandiscapturedbyavariablewhich
measurestheaveragedistanceofalldwellingsineachareatothenearestpointonthenetwork.Unfortunatelyrelativefuel
pricedatadoesnotexistatacross-sectionallevel.However,providedthisdoesnotvaryacrossareasforagivenperiodit
willbeincludedinourintercept,andasdiscussedaboverelativefuelpricesmayplaylessofarolethanotherfactorsin
developedeconomics.
Economictheorysuggeststhathouseholdswilladoptmainsgascentralheatingifthebenefitsderivedfromadoption
exceedthecostsandthereisanexpectedutilityincreasefromdoingso.However,innovationstaketimetodiffuse,and
householdsregularlymakesuboptimalchoices.Thiscanberelatedarangeoffactors,suchasuncertaintyabouttherelative
costsorbenefitsofadoption,indifference,heterogeneityinconsumerpreferencesorlackofaccesstofinancing.
Inordertoestimatethedeterminantsofgasconnectionsatalocalarealevel,itisnecessarytoconsiderdemandand
supplysimultaneously.Suppliersarelikelytoextendthegasnetworktoareaswithahigherprobabilityofadoption,and
onlythosehouseholdsincloseproximitytothegasinfrastructurecanadopt.Previousresearchhasindicatedthatdwellings
withpipedgasinIrelandhavehigherincomes,partlyduetotheirurbanlocation(Watsonetal.,2003).Thisendogeneity
couldpotentiallyleadtoourcoefficientsbeingbiasedifwesimplyestimateademandequation.Therefore,wefirstestimate
asupplyequationinatwo-stageleastsquareregression.Thechoiceofinstrumentandidentificationaredescribedindetail
inSection4.3andinstrumentvalidityinSection4.4.
Weassumethattheproportionofgasusersinanyareajwillbeafunctionoftheaggregatesocioeconomiccharacteristics
ofthatareaXj,aggregatedwellingcharacteristicsDj,spatialfactorswhichwillvarybylocationSjandthelengthoftimethe
gasnetworkhasbeenlocatedinanareat−t0j.Thiscanbesummarisedasfollows:
Njj=1Gijt
Nj =
f(Xj;Dj;Sj;t−tj0) (1)
whereGijtisabinaryvariableequaltooneifhouseholdiinareajusesgasattimetandequaltozerootherwise.Njisthe
numberofhouseholdsineacharea.
4.1. Supplyequation
Asadoptionmighthaveanon-linearrelationshipwiththelengthoftimethenetworkhasbeeninplace,weestimatetwo
supplyequations.Inthefirstequation,thedependentvariableisthelengthoftimethenetworkhasbeeninplaceineach
area,theseconddependentvariableisthesquaredlengthoftimethenetworkhasbeeninplaceineacharea.
Ourfirst-stagesupplyequationsaresummarisedbelow:
Tj=˛+ˇZZj+ˇXXj+ˇDDj+ˇSSj+
(2)T2
j =+ZZj+XXj+DDj+SSj+ı (3)
WeregresstimeandtimesquaredonourinstrumentsetZjconsistingofhouseholdcount,householdcountsquared,area
andareasquared.
Thisgeneratespredictedvaluesfortimeandtimesquaredwhichwecanusetoidentifytheeffectofthesefactorsin
oursecond-stagedemandequation.Allothervariablesfromthesecondstagearealsoincludedinthefirststageregression.
Weimplementatwo-stage, generalisedmethodofmomentsspecification(GMM),withcommonintercepts(˛,)and
4.2. Demandequation
Thedependentvariableinthisregressionistheproportionofhouseholdsineachareathatusegasastheirprimarysource
ofcentralheating.Whencompletingthe2011Census,householdswereaskedtoselectfromarangeofoptionstheonethat
bestdescribestheirprimarymeansofcentralheating.ThisissummarisedinTable1inSection5.1.
Thedemandequationtakestheestimatedtimeandtimesquaredfromthesupplyequations,alongwitharangeof
socioeconomicanddwellingcharacteristics,somespatialvariablesrepresentingtheproximitytothegasnetwork,proximity
toalternatefuelsourcesandpolicyvariablesprohibitingthesaleandburningofbituminouscoal.
Njj=1Gijt
Nj =
+ıTˆTˆj+ıTˆ2Tˆ
2
j +ıXXj++ıDDj+ıSSj+ (4)
Weincludearangeofsocioeconomicfactorsatarealevel,whichmightinfluencethedecisiontoadoptgascentralheating.
Thesearerelatedtoeconomicstatus,age,educationlevelsandtenuretype.Dwellingcharacteristicsincludehousetype,
ameasureofenergyefficiency(BuildingEnergyRating–BER),anddwellingage.Allofthesevariablesareexpressedas
proportionsforeachSmall-Area.
4.3. Identification
Asdescribedabove,weusehouseholdcount,areasizeandtheirsquaredtermsinourfirst-stagesupplyregressionto
generatepredictedvaluesfortimeandtimesquaredinthesecondstagedemandregression.Thisisbecausethenetwork
operatorislikelytoexpandthenetworkfirsttothoseareaswithahigherprobabilityofadoption.Thismightbiasourresults
unlessaccountedfor.
Therationalebehindthisinstrumentisthatthetotalcostsofextendingthenetworktoanareashouldbeinverselyrelated
tothenumberofcustomersinanarea.Ifdiminishingeconomiesofscaleexist,anegativerelationshipwillalsoexistwith
thesquareofthenumberofcustomers.Inadditiontothis,thedensityofhouseholdswillalsobeanimportantfactorin
drivingnetworkextensions.6Thisinstrumentcapturesthekeyelementbeingtheutilities’decisiontoextendthenetwork
tocertainareasbasedonlocaleconomiesofscale.ThisempiricalstrategydrawsfromLyons(2014)inhisestimationofthe
timinganddeterminantsoflocalbroadbandadoptioninIreland.
Asthenetworkhasbeendevelopedoveralongperiodoftime(approx.100years)usingpopulationdatafrom2011isnot
aperfectmeasure.HoweverwedonothavehistoricalseriesforpopulationatSmall-Arealevel,andthegeographicspread
ofpopulationinthecurrentperiodislikelytobehighlycorrelatedwithpastperiods.
Onthedemandside,thekeyfactordrivingadoptionwillbethecostandavailabilityofthenetworkconnection,thisis
capturedbyoursupply-sideinstrumentsandthevariablemeasuringdistancetothenetwork.Onecouldarguethatuptake
mayalsobeaffectedbyneighbourhoodspillovers,forexampleifahouseholdobservesanumberofneighboursconnecting
tothenetworkandthendecidestoconnect.Further,imperfectinformation,neighbourhoodeffectsorotherfactorsmay
affectthetimingofadoption.Thisunderlinestheimportanceofexaminingthetimelaginadoption.Whilewecanmeasure
themagnitudeofthetime-lagandhowisvariesbyarea,ourdatadonotallowustounpicktheunderlyingreasonsbehindit.
4.4. Instrumentvalidity
Regardinginstrumentrelevance,Baumetal.(2007)suggestusingKleibergen–Paaprkstatistictotestfor
underidentifica-tionwhenusingarobustcovarianceestimator,andthecorrespondingWaldFstatisticwhentestingforweakidentification.
Inbothcasestheresultsofthesetestsfailtorejectthenullhypothesisthatourinstrumentsareunderidentifiedandweakly
identified,asperTableB1.Thisislikelytobethecasebecauseweareincludinginteractionsofendogenousvariables(linear
andquadraticterms)inourestimationsandthesearehighlycorrelated.Wooldridge(2010)suggeststhatwhenthisisthe
case,oneshouldcheckwhetherthemostgenerallinearversionofthemodelisidentifiedandifthisisnotthecase,
pro-ceedwithcaution.Inourcaseboththelinearandquadraticendogenousvariablesarestronglyidentifiedwhenestimated
separately,TableB1,andweproceedonthatbasis.
TheresultfromtheHansenJtestofoveridentifyingrestrictionssuggeststhatwedonotrejectthenullhypothesisthat
theoveridentifyingrestrictionsarevalidforthe100yearsample.Ata5%levelwewouldrejectthenullforthe20year
sample.However,somedoubthasbeencastontheabilityofthistesttoprovideinformationonthevalidityofthemoment
conditionsimpliedbytheunderlyingeconomicmodel(Deaton,2010;ParenteandSilva,2012).ParenteandSilva(2012)in
particularsuggestthatthisshouldmoreaccuratelybeconsideredatestofinstrumentcoherence,asopposedtovalidity.
5. Data
Thedatainthispapercomefromarangeofsources.Theproportionofnaturalgasuserswithineacharea,alongwitharea
proportionsofsocio-demographicanddwellingcharacteristicswereobtainedfromtheCentralStatisticsOffice(CSO)Census
ofPopulation,Small-AreaPopulationStatistics2011.GasNetworksIreland(GNI)provideddetailedGISmaps,includingthe
timingandgeographiclocationofthehigh-pressure(HP),mediumpressure(MP)andlowpressure(LP)gasnetwork.The
EnvironmentalProtectionAgency’s(EPA)websiteprovideGISmapsofsoiltypesinIreland,fromthiswecalculatedthe
averagedistancetobogsforalldwellingsineachlocation.TheEPAalsoprovideinformationonthetimingandlocationof
smoky-coalbansinIrishurbanareas7.Fordescriptivestatisticsofallvariablesusedinestimations,pleaseseeAppendixA.1.
TheanalysisisconductedatSmall-Arealevel.Thisisthemostdisaggregatedunitforwhichonecanobtainpublicly
availableCensusdatainIreland.Theserangeinpopulationfrom8to549dwellings.Thereareover18,000Small-Areasin
Ireland.Oursampleconsistsof9638Small-Areaswhichareallincloseproximitytothegasnetwork.
5.1. Dependentvariable
ThedependentvariableistheproportionofgasuserswithineachSmall-Area.Thiswasself-reportedbyhouseholdsas
perTable1.Naturalgasusageaccountedforalmostathirdofallprimarycentralheatinginthenationalpopulationin2011.
Weexplorehowthisvariesbyrecalculatingtheproportionsofeachfuelusedastheaveragedistanceofalldwellingsin
aSmall-Areagetclosertothegasnetwork.Theaverageproportionofgasusersjumpsto57.5%inareaswithin1000mof
thenetwork(oursample),andincreasesastheaveragedistancetothenetworkfalls.Themainfueldisplacedisoil,and
electricityisincreasingusedasanalternative.Thisreflectsthegreaterproportionofelectricheatinginurbanapartment
buildingsclosetothegasnetwork.
However,evenwithintheseareas,considerablevariationexistsintheproportionofusers.Fig.2illustratesthatevenfor
areasinwhichtheaveragedistanceofallhouseholdstothenearestpointonthelowormediumpressurenetworkisless
than100m,asignificantproportiondonotusegasastheirprimarymeansofheating.
ThisisillustratedgeographicallyinFig.3.AsexampleswechoosefourmetropolitanareasinIreland,allofwhichhave
accesstothegasnetwork.OutsideofDublin,Corkhasboththegreatestnumberandhighestproportionofhouseholdsusing
naturalgasastheirprimarymeansofcentralheating,howeverthereisstillsignificantlocalvariation.Galwayhasarelatively
lowproportionofgasusersinmostareas,reflectingtherecentextensionofthenetworktothiscity.
5.2. Gasnetwork
ThelocationofthegasinfrastructureinIrelandisdisplayedinFig.4.AsdescribedinSection3.1,thenetworklocation
wasconcentratedmainlyinlargercitiessuchasDublinandCorkuntilrelativelyrecently.Thehigh-pressurenetworkwas
expandedtolinkLimerickandGalwayintheearly2000s.
Detailednetworkmaps,whichalsocontainthedateeachindividualsegmentwaslaid,wereobtainedforeachsegment
ofthegasnetwork.FromthiswecalculatewhenthegasnetworkwasputinplaceforeachSmall-Area.
5.2.1. MeandistancetoLPorMPnetwork
ThisdistancevariablewasgeneratedbycalculatingthedistanceofeverydomesticresidenceintheCSO2011Censusto
thenearestpointontheLPorMPgasnetwork(Krahetal.,2016).WethenaggregatedbySmall-Area,tocalculatetheaverage
Table1
Census2011primarycentralheatingproportions.
Whatisthemaintypeoffuelusedbythe
centralheatinginyouraccommodation?
Nationalpopulation Within1000m Within500m Within100m Within50m Within10m
Nocentralheating 1.6% 1.3% 1.3% 1.3% 1.3% 2.8%
Oil 43.1% 23.9% 22.3% 18.2% 16.2% 3.2%
Naturalgas 33.4% 57.5% 59.2% 64.0% 67.1% 69.7%
Electricity 8.5% 11.5% 11.7% 11.8% 11.1% 19.5%
Coal(includinganthracite) 4.8% 2.5% 2.4% 2.0% 1.8% 1.5%
Peat(includingturf) 4.8% 0.8% 0.7% 0.4% 0.2% 0.1%
LiquidPetroleumGas(LPG) 0.6% 0.3% 0.3% 0.2% 0.2% 0.2%
Wood(includingwoodpellets) 1.3% 0.3% 0.3% 0.2% 0.2% 0.1%
Other 0.5% 0.3% 0.3% 0.2% 0.2% 0.3%
Notstated 1.4% 1.6% 1.6% 1.7% 1.7% 2.7%
Notes:Author’scalculationsbasedonCSOCensus2011data.
Datapresentedfornationalpopulationandforvaryingdistancesfromgasnetwork.
Fig.2. Proportionofhouseholdsusinggasastheirprimaryfuelincloseproximitytothelowpressuregasnetwork.
Source:Author’scalculationusingCensus2011data.
Fig.3.SpatialvariationingasconnectionsatSmall-ArealevelinfourIrishmetropolitanareas.
Fig.4. LocationofIrishgasnetworkinfrastructure2011.
Source:DataprovidedbyGasNetworksIreland–pleaseseethedisclaimerattheendofthisdocument.
distanceforeacharea.Thisvariablewillreflecttherelativeeaseofconnectionforvariousareas.Thiscanvaryevenwithin
closeproximitytothenetwork–ascanbeseenfromFig.5.
5.2.2. Datenetworkwaslaid
Eachsegmentofthegasnetwork8hasadateidentifiermarkingthedaythatportionofthenetworkwaslaid.Using
GISsoftware,wemapeachnetworksegmenttoanySmall-Areaitisfullywithinorintersectsatanypoint,illustratedin
Fig.6.ThisgeneratesadistributionofdatevariablesforeachSmall-Area.Asthe2011Census(fromwhichwetakeourgas
proportionsdata)tookplaceonApril10th2011,weconsiderthisastimet.Fromthiswecalculatethelengthoftimein
yearssinceeachsegmentwaslaidast−t0
j,wheret0j isthedateeachsegmentwaslaid.Thisgeneratesadistributionof
year-lengthvariablesforeachSmall-Area.Asaproxyforthelengthoftimegaswasavailabletohouseholdsineachareawe
choosemaximumtimelength,i.e.thedatethefirstsegmentwaslaidineacharea.Howeverwealsorunestimationswith
variousothertimevariables,suchastheaveragetimeandlatesttimegasbecameavailableineacharea.9
8 Thelowpressurenetworkcontains135,195separatesegments,themediumpressurenetworkcontains123,048segments.
Fig.5.Variationinhouseholddensityandlocationincloseproximitytothelowpressuregasnetwork.
Source:DataprovidedbyCSOPopulationCensus;GasNetworksIreland–pleaseseethedisclaimerattheendofthisdocument.
Fig.6.ExampleofSmall-Areaboundariesandgasnetwork.
Fig.7. Locationofpeatbogsandareaswherepeatburningistheprimarymeansofcentralheating.
Source:DataprovidedbyCSOPopulationCensus;EPAGISportal.
5.3. Spatialfuelsourceandpolicyvariables
FromCensus20114.8%ofhouseholdsinIrelandusepeatastheirprimaryheatingsource,howeverasizeableproportion
alsohaveanopenfireorpeatburningstoveasasecondarysource.AscanbeseenfromFig.7,thispatternishighlycorrelated
withthelocationofpeatbogs.UsingGISsoftwarewecalculatethedistanceofeveryhouseholdtothenearestraisedand
blanketbog.Again,weaggregatethesevariablestoSmall-Arealevel,allowingustodeterminetherelativeproximityof
dwellingsineachareatodifferentbogtypes.
Abanonthemarketing,saleanddistributionofbituminousfuel(or“smokycoalban”)wasintroducedinDublinin1990.10
Thiswasinresponsetosevereinstancesofwintersmog.Thisbanwasextendedtoanincreasingnumberoftownswitha
populationinexcessof15,000peoplebetween1990and2013,andaprohibitiononburningwasintroducedinadditionto
thebanonmarketing,saleanddistribution.By2011thiswasinplacein19townsinIreland.Informationonthelocation
ofthesebansallowustooverlaythisontoourSmall-Areas.Dummyvariablesarethencreatedfortheseareas.Whilewe
cannotinferacausalrelationshipbetweenthispolicyandgasusage,wecanexaminethecorrelation,holdingotherfactors
constant.
5.4. Censusandotherdata
Supplementingthespatialandtemporaldataonfuelsourcesandpolicyvariables,weincludearangeofsocioeconomic,
demographicanddwellingvariablesfromtheCensusinourestimations.ThesevariablesareallatSmall-Arealevelandthus
willreflecttheaggregatecharacteristicsofeacharea.Wealsoincludeinformationontheenergyefficiencyofdwellings.
ThisdatawasestimatedusingtheSEAIBERdatabaseandtheCensusofpopulation2011.FormoreinformationseeCurtis
etal.(2015).Weusetheproportionoflow-rated(E,F,G)dwellingsineacharea.
6. Results
Weestimateageneralisedmethodofmoments(GMM)instrumentalvariablesspecification,withhouseholdcount,area
andtheirsquaredtermsasinstruments.Thelengthoftimethenetworkhasbeeninplacemightaffecttheproportionof
usersinanon-linearmanner.Forexample,torunmainsgastocertainhousingestatesadjacenttotheexistingnetwork
GNIrequireaminimumproportionofhouseholdswithinthatareatoadoptimmediately.11Thiswouldresultinalarge
initialuptakewhichmitigatesovertime.Alternativelyforone-offconnections,certainhouseholdsmightbeslowtoswitch
tomainsgaswhenitfirstbecomesavailable,duetosunkcostsrelatedtotheircurrentheatingsystem.Thismightresultin
aslowinitialuptake,followedbymorerapidswitching.Toaccommodatethis,wespecifytwofirststageregressions,with
timeandtimesquaredasthedependentvariables.Standarderrorsarerobusttoheteroskedasticity.Areasareweightedby
populationinallspecifications.Werestrictouranalysistoareasinwhichtheaveragedistanceofalldwellingsislessthan
1kmfromthenearestpointonthelowormediumpressuregasnetwork.Otherareasarenotrelevantforouranalysis,as
itwouldnotbefeasibleforhouseholdswithinthemtoconnecttomainsgas.12Wereporttheresultsfromourfirststage
supplyequationsfirst,followedbythesecondstagedemandequation.
6.1. Supplyequations
Theseequationsareprimarilyusedtoidentifythelengthoftimethenetworkhasbeeninplaceinourdemandequation.
Theinstrumentsareallsignificantandhavetheexpectedsigns.Thegasnetworkwaslocatedfirstinareasofhighdensity.We
includeallothercovariatesfromoursecondstageinthefirststageregressions,asthereisnoefficiencylossfromdoingthis.
However,asmanyofthem,particularlythoserelatedtosocioeconomiccharacteristics,reflectcurrentfactorsandthegas
networkwasconstructedovermanyyears,theirinterpretationissubjecttocaution.TheresultsarereportedinSectionA.1
oftheappendix.
Onevariableofinterestthoughistheproportionofhousesbuiltinvarioustimeperiodsineacharea.Thiswillreflect
changesinthehousingstockovertime.Asonemightexpectthecoefficientsonthesetermsarehighestforthoseareas
withhighproportionsofpre1945dwellings,decreasesforareaswithhigherproportionsofbuildingconstructedbetween
1945–1980,andrisesagainforbuildingsconstructedbetween1980–2000.Thiseffectisindicativeoftheoutwardsprawlof
networkinfrastructurefromareasofhistoricallyhighdensityovertime.
6.2. Demandequation
Predictedvaluesforlengthoftimeandlengthoftimesquaredaregeneratedfromthefirststageestimation.Theproportion
ofgasconnectionsineachareaisthenregressedontheseandothervariables.Wereportresultsforboththewhole100
yearsampleandthemorerecent20yearsampleinTable2.Theresultsindicatethateachadditionalyearthenetworkhas
beeninplaceresultsina3.2percentagepointincreaseintheproportionofhouseholdswithinthatareawhousegasastheir
primaryfuel.Thiseffectmitigatesovertime,asindicatedbythenegativeeffectonthesquaredterm.Wecangraphically
illustratethetime-pathtoadoptionincludingbothlinearandsquaredterms,aspertheleft-handpanelofFig.8.Bothof
theseeffectsarehighlystatisticallysignificant.Onaverage,forallareasinourestimationthereisanincreasingadoption
uptoabout25yearsinthefullsample.Thelimitingfactorisduetocertainareashavinghadaccesstothegasnetworkfor
uptoacentury,butwhichstilldonothaveaveryhighproportionofconnections.Whentheanalysisisrestrictedtomore
recentperiodstherateofadoptionappearstobemuchfaster.Thisisgraphedintheright-handpanelofFig.8.Whenthe
sampleisrestrictedtotheprevious20years,eachadditionalyearisassociatedwithapproximatelya12percentagepoint
increaseingascustomers,againthiseffectappearstoreduceovertime.ThisisbroadlyinlinewithRoganetal.(2012),
whoreportedanannualincreaseof9%between1980and2010.Onaverage,penetrationratesarereachingabout50%after
8–10years.
Theresultsforboth100yearand20yearsamplearebroadlysimilarformostoftheremainingvariables.Wewilldiscuss
bothtogether,unlessotherwisestated,withthe20yearresultsinparenthesis.Thecoefficientonthevariablerepresenting
distancetothegasnetworkissignificantatthe1percentlevel.Eveninareasthatarerelativelyclosetothenetwork,distance
stillmatters.Interpretingthisresultimpliesthata1percentincreaseinaveragedistancetothenetworkisassociatedwith
a12(13)percentagepointreductionintheproportionofusersinanarea.Thisreflectsthecostofdomesticconnections.
Forhouseswithin15mofthenetworkconnectioncostsareD 220,withachargeofD 45foreachadditionalmetrebeyond
this.13
Thedistancetoacutbogs(thisincludesbothraisedandblanketbogs)hasapositivecoefficient,indicatingthatthefurther
awayanareaisfromacutbog,allelsebeingequal,thehighertheproportionofgasusersinthatarea.Thecoefficienton
11Seefordetailshttp://www.cer.ie/document-detail/Gas-Networks-Ireland-Connections-Policy-Review/1007 12Wetestthesensitivityofthisparametertovariousdistancesfrom100mupwards.
Table2
Secondstagedemandequation.
DepVar:proportionofgasusersbySAin2011
Variablecategory Variable 100yearsample 20yearsample
Coefficient RobustSE Coefficient RobustSE
Spatialfuelandpolicy
variables
Maxlengthyearshat 0.032*** (0.004) 0.119*** (0.022)
Maxlengthyearssquaredhat −0.001*** (0.000) −0.006*** (0.002)
log(distancetocutbog) 0.019*** (0.004) 0.038*** (0.006)
log(distancetobktbog) −0.006 (0.005) −0.020*** (0.005)
log(meandistancetogasnetwork) −0.120*** (0.006) −0.129*** (0.007)
Coalbandummy 0.063*** (0.011) 0.106*** (0.016)
Socioeconomic EconWorking [REF] [REF]
EconLookingforfirstjob −0.289 (0.313) 0.076 (0.243) EconUnemployed −0.427*** (0.123) −0.659*** (0.128)
EconStudent −0.398*** (0.130) −0.240** (0.103)
EconHome −0.303* (0.155) −0.438*** (0.127)
EconRetired −0.774*** (0.198) −0.564*** (0.144)
EconDisabled −0.710*** (0.175) −0.438*** (0.121)
EconOther −0.150 (0.198) −0.415 (0.275)
Age25–44 [REF] [REF]
Age0–14 0.560*** (0.117) 0.405*** (0.084)
Age15–24 0.188 (0.156) 0.106 (0.118)
Age45–64 −0.293*** (0.099) −0.039 (0.109)
Age65plus 1.046*** (0.239) 0.458*** (0.165)
Socioeconomic
EduSecondary [REF] [REF]
EduPrimary 0.187** (0.082) 0.400*** (0.076)
EduTechnical −0.124 (0.100) −0.225*** (0.086)
EduDegreeplus 0.253*** (0.059) 0.198*** (0.042)
EduRefused −0.029 (0.128) −0.041 (0.086)
TenOwnmortgage [REF] [REF]
TenOwnNomortgage −0.351*** (0.064) −0.536*** (0.069)
TenRentland −0.153*** (0.044) 0.001 (0.046)
TenRentlocal 0.119** (0.049) 0.073** (0.031)
TenRenvol −0.009 (0.081) 0.078 (0.078)
TenRentfree −0.252 (0.203) 0.177 (0.271)
Dwelling DwellBungalow [REF] [REF]
DwellFlat −0.219*** (0.026) −0.166*** (0.020)
DwellBedsit −0.192 (0.358) −0.847*** (0.120)
DwellOther −0.317** (0.133) 0.285 (0.181)
ProportionEFG −0.499*** (0.036) −0.602*** (0.034)
AgePost2006 [REF] [REF]
AgePre1945 0.446*** (0.075) 0.713*** (0.108)
Age1945–60 0.363*** (0.056) 0.577*** (0.072)
Age1960–80 −0.126*** (0.033) 0.235** (0.101)
Age1980–2000 −0.126*** (0.025) 0.156* (0.082)
Constant 0.840*** (0.097) 0.592*** (0.108)
Diagnostics N 9638 7965
F(34,9603),(34,7930) 461.38 (0.00) 416.62 (0.00)
Overid–HansenJ 1.341 (0.512) 6.836 0.0328
Notes:ResultsfromIV-GMMspecification.Cluster-robuststandarderrorsinparenthesis.
* p<0.1. ** p<0.05. ***p<0.01.
uncutblanketbogsisnegative,butnotsignificant.14Thisislikelytobethecasebecausethecurrentproportionofhouseholds
usingsolidfuelinanareawillreflectpastincentivesinthatarea.Thereforeproximitytocutbogsmightbeabetterindicator
offuelusageasthiswillreflectareaswherepeathasbeenharvestedovermanyyears.Thebanonthesaleandburningof
bituminousfuelappearstoalsohavehadaneffect.Allelsebeingequal,theseareashavea6(11)percentagepointhigher
proportionofgasusers.Wecannotinfercausalityhowever.
Consideringthesocioeconomicanddwellingvariablesnext,ourreadingofthecoefficientschanges.Foreachsetof
vari-ables,weinterprettheeffectrelativetothereferencecategory.Allofthesevariablesareareaproportions.Theemployment
statusvariableindicatesthat,allelseequal,comparedtoareaswithhigherproportionsofpeopleinemployment,allother
Fig.8.ProportionofgasadoptersatSmall-Arealevelovertime.
categorieshavereducedgasconnections,althoughnotallcoefficientsarestatisticallysignificant.Takingthe
“EconUnem-ployed”variableasanexample,ourinterpretationisthatallelseequal,a10percentincreaseintheproportionunemployed,
relativetothereferencecategory(thoseinemployment),isassociatedwitha4.27(6.59)percentagepointdecreaseinthe
proportionusingnaturalgas.
Areaswithhighproportionsofyoungfamiliesandelderlypeoplearealsoassociatedwithgreatergasconnections,
com-paredtothosewithhighproportionsof25–44yearolds.Consideringtenuretypenext,thoseareaswithhigherproportions
ofoutrighthomeownersandprivaterentersarelesslikelytohavegasconnectionsthanthosewithhighproportionsof
mortgageholders.However,localauthorityareashavehigherproportionsofgasconnections.
Areaswithhighproportionsofhouses,asopposedtoflatsorbedsits(studioapartments)aremorelikelytousegas.
ThisreflectsthelargeproportionofelectricalheatinginapartmentcomplexesinIreland.Theproportionoflow-ratedBER
dwellingsinanareaisstronglynegativelyassociatedwithgasconnections.15Finally,whenlookingatthe100yearsamplewe
canseethatbothverynew(post2000)andveryold(pre1960)constructedhousesaremorelikelytohavehighproportions
ofgasconnections.Thislikelyreflectstheurbanlocationofahighproportionoftheolderbuildingstock.Thecoefficients
differslightlyforthe20yearsample,withmorerecentnetworkexpansionsextendingtoahigherproportionofdwellings
builtfrom1960–1980.
Thereisahighdegreeofcollinearitybetweensomeofthesocioeconomicanddemographicvariables.Forexample,areas
withhighproportionsofretiredpeoplealsohavehighproportionsofpeopleagedover65,andhaveahighproportionof
owneroccupierswithoutanyremainingmortgageobligations.Whileeachofthesevariablesiscomparedwiththereference
categoryineachclass,cautionisadvisableininterpretingsomeofthesecoefficients.Forexample,theresultsindicatethat
areaswithgreaterproportionsofretiredhouseholdsarelesslikelytohavehighconnectionstothegasnetworkthanareas
withgreaterproportionsemployed.However,areaswithgreaterproportionsagedover65aremorelikelytohavehighgas
connectionsthanareaswithgreaterproportionsof25–44yearolds.Thisresultseemscontradictory,butisdrivenbythe
referencecategorychangingineachcase,andasmallnumberofareas,withveryhighgasconnections,whichalsohave
householdsagedover65onaverage,thatarenotinretirement.
7. Scenarioanalysis–gasnetworkexpansion
Themodelmaybeusedasatooltopredictresidentialuptakeoffuturegasnetworkexpansionaswellasassessthe
associatedimpactongreenhousegasemissions.Networkexpansionisstillongoingwithanumberofprovincialtowns
earmarkedforconnection.Wexfordtown,whichislocatedinthesouth-eastofthecountry,isonetownwherethenetwork
Table3
Projectedproportionsofgasnetworkconnections.
After: SmallAreas %Householdsconnected Minimum% Maximum% changeemissionstCO2/year
2years 74 0.19 0.01 0.21 −23,065
4years 74 0.36 0.17 0.38 −44,617
6years 74 0.48 0.29 0.50 −61,651
8years 74 0.55 0.36 0.57 −67,614
10years 74 0.57 0.38 0.59 −70,815
hasrecentlyexpandedandisausefulcasestudyformodelsimulations.Thetowncomprises74SmallAreas,whichare theobservationunitintheestimatedmodel.Basedonthe2011populationcensusthereare17,684peoplelivinginthese SmallAreaswithinahousingstockof8437residentialunits.Theseareasincludeaspectrumofbuildingtypes,aswellas socio-demographiccharacteristicsoftheoccupants.Housesarethemostfrequentresidentialunit,withameanacrossthe 74smallareasof87%,thoughthisvariesfromaminimumof19%toamaximumof100%acrossSmallAreas.Themean shareofolderpre-1945propertiesis17%rangingfromaminimumof0%toamaximumof79%,whichreflectsboththeolder centralpartsoftownandmorerecentlybuildareasontheperiphery.Whilethepropertytype,andagemaypresentdifferent engineeringchallengesconnectingtothegasnetworkanimportantadditionalconsiderationistheproperty’soccupants. Themodelincorporatedfoursocio-demographicvariablescoveringtheheadofhousehold’ssocio-economicstatus,ageand education,aswellasdetailsonpropertytenure.Theproportionofhouseholdswithaworkingheadofhouseholdvaries between22%and70%acrossSmallAreas;thosewithauniversitydegreerangefrom6–42%withsimilarlybroadvariations inage.Approximately25–32%ofpropertiesareeitherowner-occupied(with/withoutamortgage)orrentedfromaprivate landlordwhile10%arerentedfromapubliclandlord.Themaximumproportionofeachofthosetenurecategoriesisbetween 63–73%,whiletheminimumvariesbetween0–6%.Theadoptionofgasasafuelislikelytodiffersubstantiallygiventhewide variationbothinbuildingcharacteristicsandtheiroccupants.Theestimatedmodelisanidealtooltopredictgasconnections bySmallAreawiththepassageoftime,whichshouldaidinplanningnetworkexpansion.
Tocompletethesimulationwemakeanumberofassumptions.First,weusethemodelestimatesbasedongasnetwork connectionsinthepast20years,asthisislikelytohavemorerelevanceforpredictingnetworkconnectionsoverashort-term horizon.Tocalculatetheimpactonemissions,dataonthefuelusedpriortogasconnection(i.e.coal,oil,etc.)aswellasthe quantityconsumedisrequired.Meanhouseholdfuelconsumptionbyheatingsourcetypearebasedonfiguresreportedin
LeahyandLyons(2010).Thepre-switchingfueltypeassumptionisbasedonananalysisofthecompositionoffuelsconsumed
inSmallAreaswithnetworkgasconnectionsat5%incrementsinshareofgasnetworkconnections.
TheprojectednetworkconnectionsarereportedinTable3.Withthemodel’sestimatedinflectionpointoccurringat
approximately10years,implyingthatthelevelofconnectionsreachesaplateauafterthattime,wedonotreportpredictions
beyond10years.Within2yearsthemeanshareofconnectionsis19%,whichisrelativelyhighbutastheSmallAreasdifferin
sizethemeanshareisnotequaltotheproportionofallhouseholdsconnected.Thelevelofconnectionsdiffersconsiderably
withtheconnectionsharebeingaslowas1%orashighas21%insomeSmallAreas.Thelevelofconnectionsincreasesquite
rapidlyoverthefirst8yearsreachingaplateaujustbelow60%meanshareofgasconnectionsbySmallArea,thoughthe
shareofconnectionsissubstantiallylowerinsomeplaceswith12%ofSmallAreasnotexceedinga50%connectionrate.The
switchtogasisprimarilyfromoil(mostlykerosene)andalsosolidfuelssuchascoal.Theprojectedreductioninemissions
associatedwithfuelswitchingisalsoreportedinTable3.Emissionsreductionsreaches70ktCO2perannumaftertenyears,
whichisapproximately1%ofemissionsfromtheentireIrishresidentialsectorin2015.16Basedontheseprojectionsthe
furthergasificationofresidentialheatingrepresentsamajoropportunitytosubstantiallyreducegreenhousegasemissions
inthecomingyears.
8. Robustnessandsensitivityanalysis
Thissectionexploresarangeofalternativemodelspecifications.Forreasonsofbrevitythefocusisonourbaseline100
yearsample,butresultsholdforbothunlessotherwisestated.Reportedsecondstagedemandequationresultsarereported
intheAppendix.
8.1. Alternativemodelspecifications
Thefirstsetofadditionalestimationsexaminealternativemodelspecifications.Column1inTableC1presentsresults
wherehouseholddensityandhouseholddensitysquaredareusedinsteadofhouseholdcountandhouseholdcountsquared
fortheinstruments.Theresultsremainquitestablecomparedtoourmainestimation.Thisislikelybecausebyinstrumenting
withhouseholdcount,areaandtheirsquaretermsinthemainestimationweimplicityaccountfordensity.
Anothersourceofconcernwithourmainestimationsisthatwedonotexplicitlyaccountforhouseholdincome.
Informa-tiononincomesisnotavailableatSmall-Arealevelandalthoughwecaptureawiderangeofsocioeconomicfactorscorrelated
withincomesomebiasmayexistduetoitsomission.Toaccountforthisweestimatetwoadditionalmodelswhichinclude
proxiesforincome.Column2presentsresultsusingtheTrutzHaaseHPDeprivationIndex(HaaseandPratschke,2012)
foreachSmall-Area.ThisisacompositemeasurecreatedbycombiningarangeofCensusvariables,someofwhichwe
hadincludedinourmainspecification.Toavoidpotentialmulticollinearityweomitthesevariablesfromthissetofresults
(employmentstatus,age,educationandtenuretype).Column3includesavariablewhichcapturesaveragerelative
employ-mentcompensationforeachcountyfrom1995–2011.Standarderrorsareclusteredatcountylevelfortheseestimations.
Inbothcasesthesevariablesaresignificantandhavetheexpectedsign.Coefficientsonthepredictedlengthoftime
vari-ablesremainstableinbothestimations.ThestatisticalsignificancereducesforsomeofthedistancevariablesinColumn
3–otherwiseresultsremainquitestable.WhiletheTrutzHaaseeisausefulmeasureinitsownright,itisessentiallyan
aggregationofvariablesalreadyincludedinourmodelanddoesnotprovidemuchadditionalinformation.Includingcounty
levelrelativeincomeisausefulmeasure,andgiventhatcountiesareadministrativeboundaries(asopposedtoSmall-Areas)
clusteringthestandarderrorsatthislevelwouldmakesensetocontrolforanyfactorsthataffectgroupsofobservations
uniformlywithineachcounty.However,wearelessconfidentabouttheaccuracyofthisspecificationaswhenclusteringat
countylevelthenumberofclustersareinsufficienttocalculatearobustcovariancematrix.
ToaccountforthisinColumn4weprovideafurtherrobustnesscheckinwhichweclusterstandarderrorsatElectoral
Division(ED)level.17Thisallowsforacalculationofarobustcovariancematrix.HoweverEDsareanarbitraryaggregationof
Small-AreasusedforCensuspurposesanditisnotclearwhyintra-groupcorrelationwouldexistatthislevel.Giventhemain
resultsarequitestableacrossalladditionalmodelsestimatedourmainreportedmodelsremainourpreferredspecification.
8.2. Sensitivityanalysisonmissinggassegmentdateidentifiers
Aspreviouslydescribedadateidentifierismissingfor20%oftheLPandMPnetworksegments.Whilewemitigate
thisproblemthroughaggregationattheSmall-Arealevel,measurementerrormaystillbiasourresults.Toaccountforthis
weconductsensitivityanalysisonvarioussub-setsofthedata.Columns1–4ofTableC2presentresultswherewesetan
acceptablethresholdofmissingdateidentifiersforeachSmall-Areaat0%,5%,10%and20%.TakingColumn1forexample
weomitanySmallAreawithamissingdateidentifier.Thisisquiterestrictiveandreducesoursampleto5072Small-Areas.
Aswemoveacrossthecolumnsthesamplesizeincreasesandresultsconvergetowardsourmainestimates.However,in
allcasestheyarequitestable.Ourmainmodelremainsthepreferredspecificationasitprovidesaconservativeestimateof
theeffectoftimeongasheatingadoption.
8.3. Sensitivityanalysisonaveragedistancefromgasnetwork
ForthemainestimationsweexamineuptakeofgascentralheatinginSmall-Areasinwhichtheaveragedistanceofall
dwellingsiswithin1000mofthegasnetworkinfrastructure.Thisthresholdischosenasitwouldbeprohibitivelyexpensive
toconnectoverdistancesmuchlongerthanthis.AnotherissueisthatasmallproportionofdwellingsuseLPGandwhile
thisisconsideredseparatelybytheCSOandshouldnotbeincludedinourdependentvariable,somehouseholdsmayhave
answeredthisquestionincorrectly–particularlyiftheystatetheyareusingnaturalgasbutarefarfromthenetwork.Results
ofarangeofsensitivitychecksarereportedinTableC3.Asonemightexpect,resultsarequiteunstableatdistancesfarfrom
thenetwork.Householdsintheselocationswouldhavenorealisticchanceofconnecting.Asthethresholdmovescloserto
thenetworkresultsconvergetowardsthemainestimations.Again,ourmainmodelremainsthepreferredspecificationand
itprovidesaconservativeestimateoftheeffectoftimeongasheatingadoption.
9. Conclusion
We haveexaminedthedeterminantsofgascentralheatingadoptionatSmall-ArealevelinIreland, simultaneously
modellingsupplyanddemandinordertoaccountforpotentialendogeneityinnetworkinfrastructureroll-outandadoption.
Weexplicitlymodelthetime-pathindiffusion,whichisimportantinordertobetterunderstandthepotentialforbothpolicy
andtechnologicalimprovementstoaidcarbonabatement.Irelandisinterestingfromaninternationalperspectiveasithas
alegacyandcultureofpeatusageforhomeheating.Thegasnetworkhasbeeninplaceinthetwolargestcitiesforacentury,
butonlyrecentlyextendedtootherpartsofthecountry.Ouruniquetimeandlocationcodeddataallowustoexamine
adoptionoveranextendedperiod.
Onaveragetheresultsshowthatoverthepastcentury,eachyearthenetworkhasbeeninplaceisassociatedwitha3%rise
inconnections.Whenmorerecentperiodsareexamined,theconnectionrateismuchhigher,about12%riseperyearover
thepasttwentyyears.Thereappearstobeanon-linearityintheseestimatesandthiseffectdiminishesovertime.Proximity
tothenetworkisalsoanimportantdeterminantofconnections,andreflectsthecostofconnectionforalldwellingsinthat
area.
Thewidespreadavailabilityofpeatasasourceoffuelhasclearlyinhibitedthetransitiontocleanerfuels.Aspeatusage
ishighlycorrelatedgeographicallywiththelocationofpeatbogs,itisusefultoseehowgasnetworkroll-outinteractswith
theproximityofotherfuelsourcesindetermininggascentralheatingadoption.Proximitytopreviouslycutpeatbogsis
negativelyassociatedwithgasconnections.Recentpolicydevelopmentssuchasthebanonthesaleandconsumptionof
bituminouscoalisassociatedwitha6percentagepointhigherproportionofgasconnectionsintheseareas,allelsebeing
equal.Wecan’tattributecausalityherehowever,asthisbanwasfirstintroducedinurbanareas,whichwouldalreadyhave
hadhigherproportionsofgasconnectionsbeforethebanswereintroduced.
Inthecontextoffuturenetworkgasexpansiontheanalysisprovidesanumberofusefullessons.Asnotedabove,domestic
gasnetworkconnectionsareneitheruniformnorinstantaneousfollowingnetworkexpansion.However,connectionsdo
occurrelativelyrapidlyreachingaplateauwithin10years.Thereisalsoconsiderableheterogeneitybysocio-demographic
characteristicsandbuildingattributesacrossSmallAreasintermsofnetworkconnections.Thisinformationisusefulfor
networkplannersindecidingwheretonextextendthenetwork,andalsoforcommercialsuppliersofgasindeterminingwhy
certainareasincloseproximitytothenetworkhavelowlevelsofconnections.Areasthataremoresociallydeprived,with
fewer‘working’householdsorlowerlevelsofeducation,havinglowerratesofnetworkgasconnection.Networkexpansion
insuchareasmaybeunprofitableorhavelongerpay-backperiods.Ifgasnetworkexpansionisconsideredsociallydesirable
insuchareaspublicsubventionmaybenecessary.
Onereasonwhygasnetworkexpansioncouldbeconsideredapublicpolicyobjectiveisbecauseitcancontributeto
thede-carbonisationoftheresidentialsector.Thecasestudysimulationdemonstratestheshorttermbenefitsofnetwork
expansionforgreenhousegasemissionreductionsassociatedwithfuelswitching.Expansionofthenaturalgasnetworkis
alsoconsistentwiththelongertermambitionsofreducingEUgreenhousegasemissionsby2050byover80%(European
Commission,2011),aslongertermambitionstoinjectbiomethaneintothenaturalgasnetworkhasthepotentialtoreduce
emissionsby74%comparedtonaturalgas(O’Sheaetal.,2017).
Tofullyexaminethefactorsinfluencingthechoiceofhome-heatingsystem,wewouldideallyhavehadaccessto
indi-vidualhouseholdleveldata,asevenaggregatingtoSmall-Arealevelcanmaskimportantheterogeneity.Also,asidefromthe
networkroll-outdata,weonlyhavedataforonepointintime.Apaneldatasetonhowgasproportionsandvarious
charac-teristicschangeovertime,wouldhavegivenusgreaterabilitytoidentifyeffects.Similarly,theinclusionofotherspatially
codedinformation,suchasrelativepricesofalternatefuels,orthelocationofkerosenesuppliers,forexample,wouldhave
significantlybenefitedthispaper.Thesearealllimitationsoftheresearch.However,ourabilitytoexaminetime-trendsin
adoptionisquitenovelandmakesauniquecontributiontothewiderfuelswitchingliterature.
Disclaimer
“GasNetworksIreland(GNI),itsaffiliatesandassigns,acceptnoresponsibilityforanyinformationcontainedinthis
document(“theInformation”).GNImakesnorepresentationsorwarrantiesofanykind,expressorimplied,inrelationto
theInformationandherebyexcludesallsuchrepresentationsorwarrantiesinrelationtotheInformationtothefullest
extentpermittedbylaw.Noliabilityshallbeacceptedforanylossordamageincluding,withoutlimitation,direct,indirect,
special,incidental,punitiveorconsequentiallossincludinglossofprofits,arisingoutoforinconnectionwiththeuseofthe
Information.”
Acknowledgements
ThismaterialisbaseduponworkssupportedbytheScienceFoundationIrelandunderGrantNo.12/RC/2302.Theresearch
waspartfundedbyGasNetworksIrelandthroughtheGasInnovationGroupandbyScienceFoundationIreland(SFI)through
MaREI–MarineRenewableEnergyIrelandresearchcluster.FundingfromtheESRI’sEnergyPolicyResearchCentreisalso
gratefullyacknowledged.ThisresearchhasalsobeensupportedbytheGranthamInstituteforClimateChangeandthe
EnvironmentandtheESRCCentreforClimateChangeEconomicsandPolicyundergrantnumberES/K006576/1.Wearealso
gratefultotheCentralStatisticsOffice,GasNetworksIrelandandTheEnvironmentalProtectionAgency(EPA)forproviding
data.Seminarparticipantsatthe5thAnnualESRI-UCCSeminaronEnergyPolicyResearchandGlenDimplexprovidedhelpful
comments.Thispaperalsobenefitedgreatlyfromcommentsbytheeditorofthisjournalandtwoanonymousreviewers.
AppendixA.
A.1. Descriptivestatisticsforallvariablesincludedinestimations
TableA1
Descriptivestatistics.
Variablecategory Variable Obs Mean Std.Dev. Min Max
Gasvariables Gas 9638 0.565 0.328 0 1
Maxlengthyears 9638 14.523 13.134 0 111
log(meandistancetogasnetwork) 9638 3.252 1.074 0.732 6.906
Peatproximity log(distancetobktbog) 9638 8.772 1.167 −3.037 10.638
log(distancetocutbog) 9,638 9.315 0.639 6.147 10.786
Coalbanareas Coalbandummy 9638 0.829 0.377 0 1
Density Householdcount 9638 94.853 22.256 21 252
Areakm 9638 3.691 13.026 0.0163 417.358
Socioeconomic EconWorking 9638 0.516 0.141 0 0.942
EconLookingforfirstjob 9638 0.010 0.012 0 0.489
EconUnemployed 9638 0.110 0.064 0 0.440
EconStudent 9638 0.117 0.083 0 0.980
EconHome 9638 0.084 0.036 0 0.297
EconRetired 9638 0.116 0.090 0 0.727
EconDisabled 9638 0.042 0.037 0 0.494
EconOther 9638 0.003 0.016 0 0.595
Age0–14 9638 0.195 0.090 0 0.594
Age15–24 9638 0.353 0.140 0 0.873
Age25–44 9638 0.133 0.077 0 0.987
Age45–64 9638 0.209 0.090 0 0.662
Age65plus 9638 0.110 0.094 0 0.780
EduPrimary 9638 0.128 0.109 0 0.722
EduSecondary 9638 0.344 0.105 0 1
EduTechnical 9638 0.183 0.063 0 0.5
EduDegreeplus 9638 0.296 0.179 0 1
EduRefused 9638 0.049 0.054 0 1
TenOwnmortgage 9638 0.347 0.192 0 0.953
TenOwnnomortgage 9638 0.275 0.196 0 0.808
TenRentland 9638 0.249 0.215 0 0.985
TenRentlocal 9638 0.091 0.165 0 0.987
TenRenvol 9638 0.011 0.043 0 0.688
TenRentfree 9638 0.011 0.020 0 0.890
Dwelling DwellBungalow 9638 0.798 0.292 0 1
DwellFlat 9638 0.176 0.280 0 1
DwellBedsit 9638 0.006 0.025 0 0.746
DwellOther 9638 0.021 0.031 0 0.982
ProportionEFG 9638 0.326 0.292 0 1
AgePre1945 9638 0.138 0.221 0 0.988
Age1945–60 9638 0.088 0.171 0 0.954
Age1960–80 9638 0.214 0.275 0 1
Age1980–2000 9638 0.247 0.268 0 0.990
A.2. Resultsfromfirststagesupplyequation
TableB1
Firststagesupplyequation.
DepVar:maxlengthinyearssincenetworkinplace
Variable Linear Quadratic
Coefficient RobustSE Coefficient RobustSE
Householdcount 0.102*** (0.028) 6.332*** (1.790)
Householdcountsq −0.000** (0.000) −0.022*** (0.008)
Areakm 0.162*** (0.016) 2.773** (1.079)
Areakmsq −0.000*** (0.000) −0.000** (0.000)
log(distancetocutbog) 0.972*** (0.087) 44.660*** (6.365)
log(distancetobktbog) 0.399*** (0.140) 8.285 (10.992)
log(meandistancetogasnetwork) −3.899*** (0.129) −91.128*** (9.847)
Coalbandummy 2.319*** (0.216) 98.124*** (15.353)
EconWorking [REF] [REF]
EconLookingforfirstjob −6.700 (16.052) −181.729 (1166.800)
EconUnemployed −7.628** (3.704) −542.257* (301.803)
EconStudent −6.209 (4.497) −423.807 (359.568)
EconHome 9.683* (5.359) 382.041 (442.248)
EconRetired −4.277 (6.972) −708.001 (560.839)
EconDisabled 2.635 (5.257) −614.494 (413.031)
EconOther 9.977 (8.273) 241.815 (686.109)
Age25–44 [REF] [REF]
Age0–14 3.929 (3.868) 23.942 (331.530)
Age15–24 16.346*** (5.366) 706.930 (430.658)
Age45–64 6.820** (3.219) −2.660 (269.857)
Age65plus 14.968** (7.520) 1291.967** (639.143)
EduSecondary [REF] [REF]
EduPrimary −2.259 (2.866) −147.458 (237.319)
EduTechnical −5.784* (3.271) −331.153 (264.163)
EduDegreeplus 5.392*** (1.917) 283.807* (158.999)
EduRefused 9.252** (3.887) 589.154* (338.634)
TenOwnmortgage [REF] [REF]
TenOwnnomortgage −0.600 (2.191) −45.547 (184.580)
TenRentland 2.843** (1.381) 90.053 (114.931)
TenRentlocal 3.347** (1.415) 279.572** (113.928)
TenRenvol 0.160 (2.666) 93.094 (202.774)
TenRentfree 11.902* (6.710) 321.387 (517.277)
DwellHouse [REF] [REF]
DwellFlat 0.513 (0.859) 21.788 (72.837)
DwellBedsit 8.401 (10.856) 1315.792 (1011.618)
DwellOther 8.870* (4.635) 151.917 (369.691)
ProportionEFG 0.160 (1.163) 66.561 (95.703)
AgePost2000 [REF] [REF]
AgePre1945 9.851*** (1.721) 652.140*** (145.094)
Age1945–60 4.592*** (1.577) 359.113*** (125.876)
Age1960–80 5.355*** (0.943) 227.464*** (76.441)
Age1980–2000 6.254*** (0.615) 235.929*** (48.380)
Constant −5.484* (3.266) −662.089** (262.435)
N 9638 9638
Weakid(Kleibergen–PaaprkWaldF)a 63.2 7.72
Weakid(Kleibergen–PaaprkWaldF)b 95.18 20.73
Underid(Kleibergen–PaaprkLM)a 0.001 0.001
Underid(Kleibergen–PaaprkLM)b 203.642 30.82
Notes:ResultsfromIV-GMMspecification.Cluster-robuststandarderrorsinparenthesis.
* p<0.1. ** p<0.05. ***p<0.01.
A.3. Robustnessandsensitivityanalysis
TableC1
Resultsofalternativespecifications.
Variable (1) (2) (3) (4)
Maxlengthyearshat 0.030*** 0.028*** 0.030*** 0.031***
Maxlengthyearssquaredhat −0.000*** −0.000*** −0.001*** −0.001***
log(distancetocutbog) 0.014*** 0.019*** 0.014 0.018***
log(distancetobktbog) −0.007* −0.010*** −0.013 −0.007
log(meandistancetogasnetwork) −0.115*** −0.132*** −0.119*** −0.120***
Coalbandummy 0.051*** 0.066*** 0.055** 0.062***
EconWorking [REF] [REF] [REF]
EconLookingforfirstjob −0.258 −0.312 −0.333
EconUnemployed −0.342*** −0.380* −0.417***
EconStudent −0.344*** −0.359** −0.413***
EconHome −0.334*** −0.282 −0.326*
EconRetired −0.665*** −0.671*** −0.758***
EconDisabled −0.598*** −0.512*** −0.700***
EconOther −0.173 −0.071 −0.147
Age25–44
Age0–14 0.568*** 0.493* 0.585***
Age15–24 0.122 0.176 0.203
Age45–64 −0.269*** −0.283* −0.273**
Age65plus 0.878*** 1.002*** 1.026***
EduSecondary [REF] [REF] [REF]
EduPrimary 0.207*** 0.099 0.198*
EduTechnical −0.076 −0.130 −0.117
EduDegreeplus 0.224*** 0.216** 0.243***
EduRefused −0.090 −0.020 −0.046
TenOwnmortgage
TenOwnnomortgage −0.350*** −0.304*** −0.347***
TenRentland −0.156*** −0.086 −0.151***
TenRentlocal 0.079** 0.164** 0.110**
TenRenvol −0.038 −0.001 −0.009
TenRentfree −0.263* −0.220 −0.251
DwellBungalow [REF] [REF] [REF] [REF]
DwellFlat −0.223*** −0.275*** −0.277*** −0.219***
DwellBedsit −0.415** −0.550*** −0.290 −0.259
DwellOther −0.322*** −0.443*** −0.426*** −0.327**
ProportionEFG −0.505*** −0.569*** −0.477*** −0.496***
AgePost2006 [REF] [REF] [REF] [REF]
AgePre1945 0.360*** 0.177*** 0.399*** 0.422***
Age1945–60 0.312*** 0.116*** 0.307*** 0.348***
Age1960–80 −0.150*** −0.425*** −0.175*** −0.134***
Age1980–2000 −0.149*** −0.271*** −0.148*** −0.133***
Deprivationindex 0.009**
Averageemploymentcompensation 0.359**
Constant 0.880*** 0.984*** 0.608** 0.858***
N 9638 9642 9638 9638
Notes:ResultsfromIV-GMMspecification.Cluster-robuststandarderrorsinparenthesis.