Policy Research Working Paper 7859
A First Step up the Energy Ladder?
Low Cost Solar Kits and Household’s Welfare in Rural Rwanda
Michael GrimmAnicet Munyehirwe
Jörg PetersMaximiliane Sievert
Development Economics Vice PresidencyOperations and Strategy TeamOctober 2016
WPS7859P
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
edP
ublic
Dis
clos
ure
Aut
horiz
ed
Produced by the Research Support Team
Abstract
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Policy Research Working Paper 7859
This paper is a product of the Operations and Strategy Team, Development Economics Vice Presidency. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at [emailprotected], [emailprotected], and [emailprotected].
More than 1.1 billion people in developing countries are lacking access to electricity. Based on the assumption that electricity is a prerequisite for human development, the United Nations has proclaimed the goal of providing elec-tricity to all by 2030. In recent years, Pico-Photovoltaic kits have become a low-cost alternative to investment intensive grid electrification. Using a randomized controlled trial, the paper examines uptake and impacts of a simple Pico-Photovoltaic kit that barely exceeds the modern energy
benchmark defined by the United Nations. The authors find significant positive effects on household energy expen-ditures and some indication for effects on health, domestic productivity, and on the environment. Since only parts of these effects are internalized, underinvestment into the technology is likely. In addition, our data show that adop-tion will be impeded by affordability, suggesting that policy would have to consider more direct promotion strategies such as subsidies or financing schemes to reach the UN goal.
AFirstStepuptheEnergyLadder?LowCostSolarKitsand
Household’sWelfareinRuralRwanda
MichaelGrimm,AnicetMunyehirwe,JörgPeters,andMaximilianeSievert
JELcodes:D13,H23, H43, I31,O13,O18,Q41.Keywords:SustainableEnergyforAll(SE4All),householdwelfare,householdtechnology
adoption,Sub‐SaharanAfrica,RandomizedControlledTrial.
Michael Grimm is a Professor of Development Economics at University of Passau. He is also affiliated with Erasmus University Rotterdam and IZA, Bonn; [emailprotected]. Anicet Munyehirwe is director of IB&C Rwanda; [emailprotected]. Jörg Peters is heading the research group “Climate Change in Developing Countries” at RWI, Germany. He is Visiting Associate Professor at the University of the Witwatersrand, Johannesburg, South Africa; [emailprotected]. Maximiliane Sievert (corresponding author) is Research Fellow at RWI, Germany; [emailprotected]. We thank two anonymous referees and the editor for very valuable comments. We also thank conference participants in Münster (VfS 2015), in Oxford (CSAE 2015), in Kiel (VfS Development Economics 2015), Copenhagen (PEGNET 2013) and Düsseldorf (IAEE 2013) as well as participants at research seminars at the University of Bonn (ZEF), the Kiel Institute for the World Economy, and the University of Groningen. The data underlying this research was collected for an impact evaluation commissioned by the Policy and Evaluation Department of the Ministry of Foreign Affairs of the Netherlands (IOB). This work was supported by the German Federal Ministry for Economic Affairs and Energy and the Ministry of Innovation, Science, and Research of the State of North Rhine-Westphalia [Sondertatbestand – special grant] to JP and MS. Please cite the version of this paper published in the World Bank Economic Review (http://wber.oxfordjournals.org/).
Morethan1.1billionpeopleindevelopingcountrieslackaccesstoelectricity.Some
590millionofthemliveinAfrica,wheretheruralelectrificationrateisonly14percent
(SE4All 2015). Providing access to electricity is an explicit goal of the sustainable
developmentgoals(SDGs)andfrequentlyconsideredapreconditionforeconomicand
socialdevelopment(UN2005).Basedonsuchassumptions,theUnitedNationsaims
foruniversalaccesstoelectricityby2030viaitsinitiativeSustainableEnergyforAll
(SE4All; seealsoUN2010).The investment requirements toachieve this targetare
enormous,estimatedbytheInternationalEnergyAgency(IEA)(2011)tobeabout640
billionUSDollars.
In recent years, so‐calledPico‐Photovoltaic (Pico‐PV) kits have become a low‐cost
alternativetoexistingelectrificationtechnologiesthankstoasubstantialcostdecrease
ofphotovoltaicandbatterysystemsaswellasenergysavingLED lamps.Different
Pico‐PV kits exist that provide basic energy services like lighting,mobile phone
charging,andradiousage. In theSE4All initiative’smulti‐tierdefinitionofwhat is
consideredasmodernenergy,thePico‐PVtechnologyconstitutestheTier1andthus
thefirststeponthemetaphoricenergyladder.InvestmentcostsforPico‐PVkitsare
farlowerthanfortheprovisionofon‐gridelectricityorhighertierPVsystems.
Thispaperinvestigatesusagebehaviorandthechangesinpeople’slivingconditions
whenhouseholdsmakethisfirststeptowardmodernenergybasedonarandomized
controlled trial (RCT) thatwe implemented in rural Rwanda. The kit,whichwe
randomlyassignedfreeofchargeto150outof300householdsin15remotevillages,
consistsofa1Wattsolarpanel,a40lumenlamp,atelephonecharger,andaradio—
and thereby justbarelyreaches thebenchmarkofwhatqualifiesasmodernenergy
accessintheSE4Allframework.ThemarketpriceofthefullPico‐PVkitisataround
30USD.OurstudypopulationisthemaintargetgroupofthePico‐PVtechnology,that
is,thebottom‐of‐the‐pyramidlivinginacountry’speripherywhowillnotbereached
bytheelectricitygridintheyearstocomeandwhowillhaveproblemstoaffordhigher
tierPVsystems.
WeinvestigatetheadoptionofthePico‐PVkitatboththeextensiveandtheintensive
margin.At theextensivemargin,weexaminewhetherhouseholdsactuallyuse the
Pico‐PV kit. This is not obvious given thatwe distribute the kit for free and the
technologyisnewforthehouseholds.Thereisanintensedebateinthedevelopment
communityaboutusageintensityoffreelydistributedgoods(see,e.g.,Dupas2014).
Attheintensivemargin,soconditionalonhouseholdsusingthekit,weexaminethe
effectsofPico‐PVusageonthreetypesofoutcomes:energyexpenditures,healthand
environment,andproductivity indomesticwork.Theamplitudeofeffectsheavily
dependsonusagebehavior:isthekitusedinadditionorasasubstitutetotraditional
lightingsourceslikekerosene?Whichhouseholdmemberusesthekitandforwhich
purposes?Dohouseholdsexpand theiractivities that require lighting into evening
hours,ordotheyjustshiftactivitiesfromdaytimetonighttime?Doesthetotaltime
awakeofhouseholdmemberschange?NonelectrifiedruralhouseholdsinAfricaare
increasinglyusingLEDlampsthatrunondry‐cellbatteries.Sincethesebatteriesare
notdisposedofappropriatelyandpotentiallyharm the localenvironment,Pico‐PV
usagemightalsoinduceenvironmentalbenefits.Wealsoanalyzewhetherpotential
productivitygainsindomesticworkreleasetimethatcannowbededicatedtoincome
generatingactivities.
OurpapercomplementstheseminalworkofFurukawa(2014)whostudiestheeffects
ofPico‐PV lampson children’s learningoutcomes in ruralUganda.Weextend the
scopebyexamining theeffectsofPico‐PVkitsonvarious in‐houseactivitiesofall
householdmembers,notonlythoseofschoolchildren.Wefindthathouseholdsuse
the kits intensively in spite of the zero price and the novelty of the product.
Furthermore,thekitconsiderablyreducesconsumptionofkerosene,candles,anddry‐
cellbatteriesand, in consequence,energy expenditures.The reductionofkerosene
improveshousehold airquality and the reductionofdry‐cellbattery consumption
plausiblyleadstoenvironmentalbenefits.Moreover,wefindthatchildrenshiftpart
of theirhomework into theeveninghours.Primaryschoolboyseven increase their
totalstudy time.Whilepartsof theseeffectsareclearly internalizedbenefits,other
partsareimportantexternalities,whichmayprovidethecauseforpublicsubsidies,in
particularifitturnsoutthathouseholdsaresimplytoopoortoraisetheupfrontcosts
alone.
TheroleofpublicpolicyinthepromotionofPico‐PVtechnologyisnotdefinedsofar.
TheexpectationoftheWorldBank’sLightingGlobalprogram,forexample,aswellas
otherdonorsisthatPico‐PVkitsmakeinroadstoAfricanhouseholdsviacommercial
markets,implyingthatenduserspaycost‐coveringprices(seeLightingGlobal2016).
Thismightinfactworkoutfortherelativelywell‐offstratainruralareasbutismuch
moreuncertainfortheruralpoor.Infact,themajortargetgroupofPico‐PVkitswithin
theSE4Allendeavorislocatedbeyondthereachofthegridinremoterareas.These
households are short on cash, credit constrained, andmight havemore essential
prioritiestospendtheirmoneyon.Ifthesegroupsintheperipheryofthedeveloping
world shall be reached by the SE4All initiative, direct subsidies or even a free
distributionmightberequired.Thisisindeedthepolicyinterventionwemimicinour
study.Fromawelfareeconomicspointofviewthiswouldbejustifiediftheusageof
Pico‐PVkitsgeneratesprivateandsocialreturnsthatoutweightheinvestmentcost.
Ourpaperprovidesempiricalsubstancetothisdebate.
Sofar,onlyverylittleevidenceexistsonthetake‐upandimpactsofPico‐PVlamps.To
ourknowledge, theonlypublishedstudy isFurukawa (2014),whoconcentrateson
educational outcomes alone. Furukawa randomized Pico‐PV lamps among 155
primaryschoolstudentsinUgandawhoatbaselineusedkerosenewicklampsasthe
mainlightingsourceathome.AlthoughFurukawa(2014)findsthatchildren’sstudy
hours clearly increased among Pico‐PV lamp owners, he curiously observes
decreasing test scores. Furukawa tests different explanations of this “unexpected
result”.Withouthavingthedataathandtoobtainarobustanswer,hehypothesizes
thatthelowpowerofthelampsandtheinadequaterechargingbehaviorcouldhave
ledtoflickeringlight,whicheventuallyworsenedstudyingconditions.Basedonthis
experience,wewillthereforecarefullycheckthelightingqualityandusers’satisfaction
inourexperiment.
Much more evidence exists on the socioeconomic effects of classical rural
electrificationprogramsusinghigher tier technologies,mostly theextensionof the
electricitygrid.Theseinterventionsdifferfromourrandomizedsolarkittotheextent
thatmuchhighereffectsizescanbeexpected,butalsomuchhighercostsareincurred.
Nonetheless, this literature constitutes an important background of ourwork, in
particular those studies that explore the effects of electricity usage on similar
outcomes.VandeWalleetal.(2016)forinstancefindthatinruralIndiaelectrification
led to a significant increase in households’ expenditures. For the case of a grid
extensionprograminElSalvador,BarronandTorero(2014,2015)findreductionsin
kerosene consumption, in particulate matter exposure, and respiratory disease
prevalenceaswellasanincreaseinstudyhoursamongchildren.Thelatterfindingis
confirmedinagridextensionprograminBangladesh(Khandkeretal.2012)butnotin
apreviousstudy inRwandaon theeffectsofmini‐gridelectrification (Benschetal.
2011).ForSouthAfricaandNicaragua,respectively,Dinkelman (2011)andGrogan
andSadanand(2013)provideevidencethattheuseofelectricitysaveswomen’stime
inhouseholdchoresandleadstoincreasedlaborsupplyofwomen.1
In SE4All’s multi‐tier framework solar home systems are the intermediate step
betweenPico‐PVandgridelectricity.Samadetal.(2013)evaluateasolarhomesystem
programinBangladeshandfindincreasesineveningstudyhoursofschoolchildren,
TV usage, and female decision‐making power. They also find reduced kerosene
consumptionand somemoderateevidence forpositivehealtheffects.Benschetal.
1Furtherstudiesexistthatexaminewhetheron‐gridruralelectrificationprogramscanspurincomegenerationand
economicgrowth(see,e.g.,Benschetal.2011;Dinkelman2011;Bernard2012;Khandkeretal.2012,2013;Grogan
andSadanand2013;Lipscometal.2013;BarronandTorero2014;Lenzetal.2016;PetersandSievert2016).As
discussedabove,wedonotexpectthePico‐PVsystemstoaffectsuchoutcomes.
(2013) confirmpositive effects of solar home systemusage on children’s studying
hoursinSenegal.
Itistheaimofourpapertoextendthescopeofthisliteraturetothebottomstepofthe
energy ladder. Hence, these findings are important to classify our observations,
although of course the cost‐related and technological differences between on‐grid
electricity,50Wattsolarhomesystemsandour1WattPico‐PVkithavetobebornein
mind.
The remainderof thepaper isorganizedas follows:Section Igives thepolicyand
countrybackground.SectionIIprovidestheoreticalconsiderationsthatwillguideour
empiricalanalysis.SectionIIIpresentsourexperimentaldesign.SectionIVdiscusses
allresults,andSectionVconcludes.
Background
PolicyBackground
Intheabsenceofelectricity,peopleinruralSub‐SaharanAfricalighttheirhomesusing
traditional lighting sources—candlesorkerosenedrivenwick lampsandhurricane
lamps.Inrecentyears,dry‐cellbatterydrivenLED‐lampshavebecomeavailable in
almosteveryruralshopandareincreasinglyused(seeBenschetal.2015).Themost
common ones are small LED‐torches andmobile LED‐lamps that exist in various
versions (see Figure 1). In addition,many rural households use hand‐crafted LED
lamps,thatis,LED‐lampsthatareremovedfromtorchesandinstalledsomewherein
thehouseorona stick thatcanbecarriedaround.For ruralhouseholds inAfrica,
expenditures forboth traditional lightingsourcesanddry‐cellbatteriesconstitutea
considerablepartoftheirtotalexpenditures.Inveryremoteandpoorareas,people
whoarecashconstrainedgenerallyusevery littleartificial lightingand sometimes
evenonlyresorttothelightingthatthecookingfireemits.Forthisstratum,theday
inevitablyendsaftersunset.
Figure1:Traditionallightingdevices
Hurricane
lamp
Traditionaltin
lamp
Ready‐
madetorch
Hand‐crafted
LEDlamp
MobileLEDlamp
Source:Ownillustration
Obviously, this lightingconstraintrestrictspeople inmanyregards.Activitiesafter
nightfallareliterallyexpensivebutalsodifficultandtiringbecauseofthelowquality
ofthelighting(seeSectionIIformoreinformationonlightingquality).Atthesame
time,itbecomesevidentthatmodernenergyisnotabinarysituation.Rather,thereare
severalstepsbetweenacandleandanincandescentlightbulb.
Thiscontinuumhassometimesbeenreferredtoastheenergyladder.Infact,SE4Allhas
defineddifferenttiersofmodernenergyaccesswithinitsGlobalTrackingFramework
(SE4All2013)accordingtotheelectricitysupplythatismadeavailable.Forexample,
aregularconnectiontothenationalgridqualifiesasTier3,becauseitallowsforusing
generallighting,atelevision,andafanthewholeday.Asolarhomesystemwould
qualifyforTier1or2(dependingonitscapacity).Tier1requireshavingaccesstoa
peakcapacityofatleast1Wattandbasicenergyservicescomprisingatasklightand
aradiooraphonecharger for fourhoursperday.Thespreadbetween theservice
qualitiesof thedifferent tiers isalso reflected in the required investmentcosts: the
retailpriceofthePico‐PVkitusedinthisstudyisataround30USD.TheWorldBank
(2009)estimatesacostrangeforon‐gridelectrificationinruralareasof730to1450USD
perconnection.2
The promotion of Pico‐PV kits ismost prominently pursued by theWorld Bank
programLightingGlobal.BasedontheassumptionthatthemarketforPico‐PVsystems
is threatened by a lack of information and information asymmetries, it provides
technical assistance togovernments, conductsmarket research, facilitates access to
financetomarketplayers,andhasintroducedaqualitycertificateforPico‐PVsystems.
TheobjectiveofLightingGlobal’sinitiativeintheregion,LightingAfrica,istoprovide
accesstocertificatedPico‐PVkitsto250millionpeopleby2030.ThePico‐PVlantern
andthepanelusedforthepresentstudywerecertifiedbyLightingAfrica.3
2TheinvestmentrequirementscalculatedbyIEA(2011)ofadditional640billionUSDtoachieveuniversalaccess
toelectricityarebasedonelectricityconnectionsthatprovideaminimumlevelofelectricityof250kWhperyear.
ThisroughlycorrespondstoaTier2electricitysource.3At thepointof thePico‐PVkit’s certification,LightingAfricadidnotyet issue certificates formobilephone
chargingandotherservices.
CountryBackground
Rwanda’senergysectorisundergoinganextensivetransitionwithaccesstoelectricity
playing a dominating role. The Government of Rwanda’s goal is to increase the
electrificationrateto70percentofthepopulationby2017/2018andtofullcoverageby
2020. The key policy instrument clearly is the huge Electricity Access Roll‐Out
Program (EARP) that since 2009 quintuplicated the national connection rate to 24
percentcountrywide.Threefurtherprogramsexistthathavenotbeenimplemented
sofar,though.First,theGovernmentplanstoestablishamechanismtoprovidethe
pooresthouseholds(categorizedasUbudehe1accordingtothenationalpovertyscale)
with abasic solar system corresponding toTier 1 electricity access. Second, a risk
mitigationfacilityshallbeestablishedtoencouragetheprivatesectortoincreasesales
ofsolarproductsand services.Third,mini‐gridsshallbedevelopedby theprivate
sector(MININFRA2016).Theseprogramsarecomplementedbytheso‐calledByeBye
Agatadowa initiative that aims at eliminating kerosene lamps completely from the
country.
Intheabsenceofpublicpromotionschemes,fewprivatefirmsthatsellLightingAfrica
verified Pico‐PV kits were active in the country at the time of the study
implementation.TheyoperatemostlyintheRwandancapitalKigaliandothercities.
In rural areas,Pico‐PV kits are sometimes available,but their retailprice ismuch
higher compared to lower quality dry‐cell battery driven LED‐lamps that can be
boughtinruralshopsalloverthecountry.Thesedevicesarenotqualityverified,but
costonlybetween500FRW(0.82USD4)forhand‐craftedLEDlampsand3000FRW
(4.95USD)foranLEDhurricanelamp.ThebatterycoststorunanLEDhurricanelamp
foronehourarearound0.01USD.Thisischeaperthanrunningakerosenedrivenwick
lamp(around0.03USDperhour)andthelightingqualityisslightlybetter,whichis
whymanyhouseholdsarenowusingsuchready‐madeorhand‐craftedLED‐lamps.
Compared to both battery‐driven LED lamps and kerosene lamps, Pico‐PV kits
providehigherquality lighting (dependingon thenumberofLEDdiodes)atzero
operatingcosts.Assumingthatahouseholdusesthelampforfourhoursperday,the
investmentintothePico‐PVlampusedforthisstudyamortizesafter10monthsifa
ready‐madeLEDlampisreplacedandafterlessthan5monthsifitreplacesakerosene
wicklamp.
TheoreticalConsiderations
Based on the literature on rural electrification presented in the Introduction,we
assumethatthePico‐PVtreatmentaffectsthreedimensionsoflivingconditions:First,
the budget effect which arises because households with access to a Pico‐PV kit
experienceachange in thepriceofenergy,whileno (substantial) investmentcosts
occuraslongasweassumethatthePico‐PVtreatmentissubsidizedordistributedfor
free. Second, health and environmental effects occurwhenever Pico‐PV kits replace
kerosene lamps,candles,anddry‐cellbatteries.Adecrease inkeroseneandcandles
4ExchangerateasofNovember2011:1USD=607FRW.
consumptionreduceshouseholdairpollutionwithpotentialeffectsonhealth(seeLam
etal.2012;WHO2016).Environmentalbenefitsarisedue to inappropriatedry‐cell
batterydisposal(seeBenschet.al2015)thatisreducedifdry‐cellbatteryconsumption
goes down. Third, we analyze the productivity of domestic production, that is,
productionnot intended tobe tradedon competitivemarkets.Thiswe refer to as
domesticproductivityeffectinwhatfollows.Thereasonforonlyfocusingondomestic
production is that income insuchremoteruralareas isvirtuallyonlygeneratedby
subsistence agriculture.ThePico‐PV kit, in turn, is too small to affect agricultural
production. For non‐agricultural products, access tomarkets is very limited and,
hence, localnonagricultural labormarketsarenonexistent.Atbaseline,only seven
percentofheadofhousehold’smainoccupationandonepercentof spouse’smain
occupationwasanon‐agriculturalactivity.Yet,sinceintheorythePico‐PVkitcould
liberatetimefromdomesticlaborandextendthetimeawakeofhouseholdmembers,
weexamineatleastthetimededicatedtoanyincomegeneratingactivity(agricultural
andnonagriculturalactivities).Labordemandinsuchruralregionsistoolow,though,
to absorb increases in labor supply, and therefore measurable effects on
nonagriculturalincomecannotbeexpected.
Themechanismleadingtothebudgetandhealthandenvironmentaleffectsarequite
intuitive,whereasthetransmissionchannelforthedomesticproductivityeffectmight
belessobvious.Productiveactivitiesathomeincludecooking,cleaning,andmaking
andrepairingofhouseholdgoodsaswellasstudyingandchargingacellphone.Since
thevisualperformanceofhumansstronglyincreaseswiththelightinglevel(Brainard
etal.2001),weassumethattheproductivityinperformingtheseactivitiesincreases
withthequantityandqualityoflight.Productivityinfineassemblyworkforinstance
hasbeenshowntoincreaseby28percentasthelightinglevelincreasesfrom500to
1500lumen(lm)(Lange1999).Butevenincreasingthelightinglevelfrommuchlower
levelscomeswithsignificantproductivityeffects.Evidencecomesforinstancefrom
weavingmills(Lange1999).5Theliteratureattributesgoodqualitylightingtodevices
that provide sufficient, nonglaring, nonflickering and uniform light, balanced
luminousdistribution throughout the room,good color renderingandappropriate
lightcolor(Lange1999).Alongallthesecriteria,thePico‐PVkitsperformbetterthan
other traditional lighting devices such as kerosene lamps and candles, but also
comparedtosmallerhand‐craftedLEDlamps.OurPico‐PVlampemits40lm,whilea
candleonlyemitsaround12lm,ahurricanelampusedatfullcapacityaround32lm
andlargemobileLEDlampscanreachlevelsaround100lm(O’SullivanandBarnes
2006). The LED lamps used in poor and remote areas are less luminous, though.
Lumenlevelsemittedbyhand‐craftedLEDlampsvarysubstantiallydependingonthe
numberandqualityofdiodesandbatteriesused.Twotothreediode‐lampsconnected
toabatterypackageemitabout10lm.6
5Moreevidenceexistsalsoonsofterimpactssuchasapositivelinkagebetweenlightingandworkmood(Kuller
andWetterberg1993;Boyceetal.1997;PartonenandLönnqvist2000),fatigue(Dauratetal.1993;Grunbergeret
al.1993;Begemannetal.1997),andeyestrainandheadache(Wilkinsetal.1989,KullerandLaike1998)thatcan
beassumedtoimproveworkingperformance.Foradetailedpresentationoftheevidenceforproductivityeffects
associatedwithlight,seethesupplementalappendix.
6Sincelumennumbersforthesehand‐craftedlampsdonotexist,wetestedthetwomostwidelyusedstructures(a
twodiode‐lampandathreediode‐lampstructure)inalaboratoryatUniversityofUlm,Germany,usingstandard
lumenemissiontestprocedures.Accordingtothesetests,thelevelofemittedlumensbyhand‐craftedLEDlamps
isataround10lm.
Oneadditionaleffectassociatedwithapossibleincreaseinradiousageisbetteraccess
toinformation,whichinturnmayhaveproductivityeffectsiftheinformationrelates
tomarketdataorcanaffectnorms,suchasgendernormsforinstance,andpreferences
(Bertrand et al. 2006; Jensen andOster 2009; La Ferrara et al. 2012; Sievert 2015).
AlthoughweanalyzewhetherradiosareusedwiththePico‐PVkitanddisplayradio
ownershipandusage in the supplementalappendix (seeAppendixS5),wedonot
furtherinvestigateanyoftheseeffectsasmosthouseholdsusethePico‐PVkitonlyfor
lighting.
ResearchApproachandData
OuridentificationstrategyreliesontherandomizedassignmentofPico‐PVkitsafter
thebaselinesurvey.Theintention‐to‐treateffect(ITT)inourcaseisalmostidenticalto
theaveragetreatmenteffectonthetreated(ATT)becauseofthehighcompliancerate
inthetreatmentgroupandnotreatmentcontaminationinthecontrolgroup.Sinceall
resultsarerobustwithregardtobothwaysofestimatingimpacts,weshowonlythe
moreconservativeITTresults.
Treatment
Therandomizedkits includea1Wattpanel,arechargeable4‐LED‐diodes lamp(40
lumenmaximum) including an installed battery, amobile phone charger, a radio
including a charger, and a back‐up batterypackage (see Figure 2)Error!Reference
sourcenotfound..7Therearedifferentoptionstousethepanel.First,itcanbeusedto
directlychargethelamp’sbattery.Afteronedayofsolarchargingitisfullycharged.
Thelampcanbeusedinthreedimminglevelsand—fullycharged—provideslighting
forbetween6and30hoursdependingonthechosenintensitylevel.Second,thekit
canbeconnecteddirectlytothemobilephoneconnectorplugandtheradioconnector
tochargemobilephonesortheradio.Third,thekitcanbeusedtochargetheback‐up
batterypackagethatcanthenbeusedtochargetheotherdeviceswithoutsunlight.
Thecompletekitcostsaround30USD,thesmallestversionwithonlythesolarpanel
andanLEDlampincludinganinstalledbatterycostsaround16.50USD.
Figure2:ThePico‐PVkit
Source:Ownillustration
ImpactIndicators
As a precondition for the three effects on budget, health, and environment, and
domesticproductivitythehouseholds’usagebehaviorisourfirstmatterofinterest.
WelookatusageandchargingpatternsofthePico‐PVkitandanalyzewhichofthe
7ThekitusedinourexperimentprovidesmoreenergyservicesthanthesolarlanternusedbyFurukawa(2014),
butthepanelisalsotwiceaslarge(1Wattcomparedto0.5Watt).
different energy services—lighting, radio operation, andmobilephone charging—
householdsusemost.Sincethekitismostlyusedforlighting(seebelow),wefocusin
particularonthisservice.
Forbudget effects,we first look at changes in thepriceof the energy service.We
calculatethepriceperlightinghourandpriceperlumenhourthehouseholdseffectively
pay. Second,we analyzewhether price effects translate into a change in lighting
consumption.Here,welookattheaverageamountoflightinghoursconsumedperday
andlumenhoursconsumedperday.Lightinghoursarecalculatedasthesumofusage
time of all lamps used during a typical day (including candles and ready‐made
torches).Thepriceperlightinghouriscalculatedbydividingexpendituresonlighting
fuelsby thenumberof lightinghours consumed.For calculating lumenhours,we
multiplythelampspecificlightinghourswiththeamountoflumenemittedperlamp.
Finally,welookatchangesintotalenergyexpendituresandintheexpendituresforthe
differentenergysourceskerosene,batteries,andcandles.
For health and environmental effects,we first explore reductions in kerosene and
candleconsumptionandtowhatextentthisleadstoaperceivedimprovementofair
quality,measuredbythesubjectiveassessmentoftherespondents.Alsoformeasuring
the household members’ health status, we rely on self‐reported information on
whetheranyhouseholdmembersuffersfromrespiratorydiseasesandeyeproblems.We
distinguishbetweenmaleandfemaleadultsaswellasprimary,andsecondaryschool
children.We did not measure air quality or undertake any medical exams. For
environmentaleffects,weanalyze reductions indry‐cellbatteryconsumptionand the
wayhowhouseholdsdisposeofdry‐cellbatteries.
In order to investigate domestic productivity effects,we look at themain users’
domesticlaboractivitiesexercisedwhenusingthePico‐PVlamp.Themaindomestic
laboractivityforadultsishousework;childrenusethelampmainlyforstudying.We
assesstheincreaseofdomesticproductivitybyanalyzingthelightingsourceusedfor
these respective activities. Based on the evidence from the literature presented in
SectionIIandthesupplementalappendix,weassumethathouseholdsbecomemore
productivewhen they switch from a lower quality lighting source or no artificial
lightingtothePico‐PVlamp.Thisseemsreasonablesinceevenatdaytime,thetypical
dwellinginruralRwandaisquitedark.Windowsaresmallinordertokeeptherain
and the heat out of the inner of the dwelling. To analyze lamp switching, we
enumeratedalllampsineachhouseholdinterviewandaskedrespondentstonameall
usersforeachlampandtherespectivepurposeofusingit.Theinformationontime
spentondifferentactivitieswaselicitedintheinterviewsthroughanactivityprofile
for eachhouseholdmember. If a certain activitypursuedby thehousehold isnot
associatedwithoneoftheemployedlamps,weassumethatnospecificlightingdevice
isused for thisactivity, and it is either exercisedusingdaylight,orusing indirect
lightingfromthefireplaceorlampsusedforotherhouseholdtasks.
Inordertoanalyzewhetherthehigherproductivityalsoleadstoanincreaseintotal
domesticlaborinput,weanalyzethetotalamountoftimededicatedtodomesticlaborper
day.Wefurthermoreexaminewhethertotaltimehouseholdmembersareawakechanges
dueto increased lightingavailabilityandwhethertimededicatedto incomegenerating
activitieschangesasaresultoftimesavingsindomesticproduction.
RCTImplementation
ThekeyfactsoftheimplementationarepresentedinTable1.Adetaileddescriptionof
theimplementation includingamapofthesurveyareaandafigureillustratingthe
participant flow can be found in the supplemental appendix.A discussion of the
externalvalidityofourresultsisalsopresentedinthesupplementalappendix.
Table1.KeyFactsonRCTImplementation
Baselinesurvey November2011
DeliveryofPico‐PVkits December2011
Follow‐upsurvey June2012
Studypopulation15nonadjoinedcommunitiesinfourruraldistrictsofRwanda
locatedintheNorthern,WesternandSouthernProvince.
NoPico‐PVkitsavailableonthemarket
~5.5hoursofsunlightperday(whichissimilartocountryaverage)
Sample 300randomlysampledhouseholds
Randomization
Stratifiedrandomizationandadditionalre‐randomizationusing
minmaxt‐statmethodatthehouseholdlevel;randomassignmentof
150Pico‐PVkits
StratificationcriteriaConsumedlightinghoursperday,usageofmobilephones(binary),
radiousage(binary),anddistrict
Re‐randomization
BalancingcriteriaaremarkedinThesurveyedhouseholdsare
mainlysubsistencefarmersthatliveinvery
modestconditions.Theeducationallevelofthe
headofhouseholdislowandhouseholdsown
onlyafewdurableconsumptiongoods.The
householdsinoursamplehavecashexpenditures
ofonaverage0.45USD(1.12USDPPP)adayper
personwiththelower25%‐stratumhavingonly
0.07USD(0.18USDPPP).Eventheupperquartile
hascashexpendituresof1.14USD(2.86USDPPP)
only.Byanystandard,thesampledhouseholds
qualifyasextremelypoor.
Also energy consumption patterns illustrate the
precarioussituationofmosthouseholds(see
Table 3). They consume on average only around
threehours of artificial lightingperdaywhich is
mainlyprovided throughkerosenewick lampsor
battery‐driven small hand‐crafted LED lamps.
Around11percentofhouseholdsevendonotuse
any artificial lighting devices and rely only on
lighting from the fireplace after nightfall. For the
baselinevalues,wecalculate lightinghoursas the
sumoflightingusageperdayacrossallusedlamps,
excludingcandlesandtorches,forwhichwedidnot
elicitusagehoursat thebaselinestage.Almost65
percentof thehouseholdown a radio, around 40
percenthaveacellphone.
Table2and
Table3
Compensationforcontrolhouseholds Onebottleofpalmoilanda5kgsackofricewortharound7USD
Attritionrate <1%
Compliancerate 87%(18householdsdeclaredtheirPico‐PVkittobesold,lostor
stolen;Onehouseholdreceivedkitonlyduringfollow‐up)
Source:Householddataset2011/2012.
Results
BalanceofSocioeconomicCharacteristicsofParticipatingHouseholds
Thissectionexaminesthebalancingbetweentreatmentandcontrolgroupand,atthe
sametime,portraysthesocioeconomicconditionsinthestudyareas.Baselinevalues
ofthehouseholds’socioeconomiccharacteristicsshowthattherandomizationprocess
wassuccessfulinproducingtwobalancedgroups(seeThesurveyedhouseholdsare
mainlysubsistencefarmersthatliveinverymodestconditions.Theeducationallevel
oftheheadofhouseholdislowandhouseholdsownonlyafewdurableconsumption
goods.Thehouseholdsinoursamplehavecashexpendituresofonaverage0.45USD
(1.12USDPPP)adayperpersonwiththelower25%‐stratumhavingonly0.07USD
(0.18USDPPP).Eventheupperquartilehascashexpendituresof1.14USD(2.86USD
PPP)only.Byanystandard,thesampledhouseholdsqualifyasextremelypoor.
Also energy consumption patterns illustrate the precarious situation of most
households(see
Table3).Theyconsumeonaverageonlyaroundthreehoursofartificiallightingper
daywhichismainlyprovidedthroughkerosenewicklampsorbattery‐drivensmall
hand‐crafted LED lamps.Around 11 percent of households even do not use any
artificiallightingdevicesandrelyonlyonlightingfromthefireplaceafternightfall.
Forthebaselinevalues,wecalculatelightinghoursasthesumoflightingusageper
dayacrossallusedlamps,excludingcandlesandtorches,forwhichwedidnotelicit
usagehoursatthebaselinestage.Almost65percentofthehouseholdownaradio,
around40percenthaveacellphone.
Table2).
The surveyedhouseholds aremainly subsistence farmers that live inverymodest
conditions.Theeducationalleveloftheheadofhouseholdislowandhouseholdsown
only a fewdurable consumption goods.Thehouseholds in our samplehave cash
expendituresofonaverage0.45USD(1.12USDPPP)adayperpersonwiththelower
25%‐stratumhavingonly0.07USD(0.18USDPPP).Eventheupperquartilehascash
expenditures of 1.14 USD (2.86 USD PPP) only. By any standard, the sampled
householdsqualifyasextremelypoor.
Also energy consumption patterns illustrate the precarious situation of most
households(see
Table3).Theyconsumeonaverageonlyaroundthreehoursofartificiallightingper
daywhichismainlyprovidedthroughkerosenewicklampsorbattery‐drivensmall
hand‐crafted LED lamps.Around 11 percent of households even do not use any
artificiallightingdevicesandrelyonlyonlightingfromthefireplaceafternightfall.
Forthebaselinevalues,wecalculatelightinghoursasthesumoflightingusageper
dayacrossallusedlamps,excludingcandlesandtorches,forwhichwedidnotelicit
usagehoursatthebaselinestage.Almost65percentofthehouseholdownaradio,
around40percenthaveacellphone.
Table2.BalanceofSocioeconomicCharacteristicsbetweenTreatmentandControlGroup
(BaselineValues)
Treatment Control
t‐test/chi‐2‐test
(totaltreatedvs.control
p‐values)
total
(SD)
noncompliant
(SD)
total
(SD)
Householdsize1 4.85(2.0) 5.5(1.5) 5.0(2.0) .491
HH’scomposition(%)
Sharechildren0–15years 39(24) 51(16) 38(23) .680
Shareelderly65+ 7(20) 2(6) 5(16) .389
HH’sheadmale(%) 76 84 76 .892
AgeoftheHH’shead 47(15) 45(17) 48(15) .795
EducationofHHhead(%)1
None 35 53 35 .857
Primaryeducation 61 42 60
Secondaryeducationandmore 4 5 5
Cultivationofarableland(%)1 99 100 98 .314
Ownershipofarableland(%)1 95 90 95 .791
Ownershipofcows(%)1
Nocow 63 84 69 .542
Onecow 22 11 19
Morethanonecow 15 5 12
Ownershipofgoats(%)1
Nogoat 68 79 74 .476
Onegoat 16 5 12
Morethanonegoat 16 16 11
Materialofthewalls(%)1
Highervaluethanwood,mud,orclay 14 11 14 1.000
Materialofthefloor(%)1
Highervaluethanearthordung 12 5 11 .854
District(%)2
Gicumbi 19 16 20 .997
Gisagara 26 32 27
Huye 28 26 27
Rusizi 27 26 26
Numberofobservations 148 19 148
Note:1Usedforre‐randomization;2usedforstratification.
Source:Householddataset2011.
Ifwelookatthesmallgroupofnon‐compliers,whodeclaredtheirkittobesold,lost
orstolen,weseethattheyaregenerallypoorerthancomplyinghouseholds:Theyhave
morechildren,ownlessland,havelesscowsandgoats,andhavelessradiosandcell
phones.
Table3.BalanceofOutcomeRelatedCharacteristicbetweenTreatmentandControlGroup
(BaselineValues)
Treatment Controlt‐test/chi‐2‐
test
(totaltreated
vs.control
p‐values)
total
(SD)
non‐
compliant
(SD)
total
(SD)
Lightinghours,categorized(%)2
Nolampsorcandles 19 26 19
Lessorequal3h/day 51 42 51
Morethan3h/day 30 32 30 1.000
Lightinghoursperday,continuous1 3.1 2.7 3.2 .910
Usageofhand‐craftedLED1(%) 37 26 35 .628
UsageofmobileLED1(%) 4 5 3 .520
Consumptionofcandles1(piecespermonth) 1.25 2.32 1.76 .356
Usageofwicklamps(%) 49 47 47 .727
Usageofnoartificiallighting(%) 12 16 11 .715
Consumptionofkeroseneforlighting1(inliterpermonth) .46 .35 .54 .372
Radioownership2(%) 64 32 64 1.000
Mobilephoneownership2(%) 36 32 36 1.000
Numberofmobilephones1 .49 .21 .47 .876
Numberofobservations 148 19 148
Note:1usedforre‐randomization;2usedforstratification.
Source:Householddataset2011.
ImpactAssessment
Take‐UpandLightingUsage
Among the 131 households that still have a Pico‐PV kitwhen interviewed in the
follow‐upsurvey,usageratesareveryhigh(seeTable4).Insum,86percentusethe
kit at least once per day, primarily for lighting. Radio and especially cell phone
chargingusageratesareratherlow.Mosthouseholdsreportthatboththeradioand
thecellphonechargerwereverydifficulttousewiththekit,whichwasconfirmedby
technicalinspectorsinvolvedintestingthekitforLightingAfrica.Themajorreason
forthisseemstobethelowcapacityofthepanel,whichonlyallowsforchargingall
devicescompletelywithinonedayifthedailysunlightisexploitedatamaximum.In
practice,householdsusedthechargingcapacitiesmainlyforthelightingdevice.Given
thispreferenceforlighting,toolittlecapacityisleftfortheothertwoservices.Forcell
phonecharging,noncompatibilityofthesolarchargerwithsomeofthewidelyused
cell phone types in rural Rwanda posed additional problems. In linewith these
technicaldeficienciesandthehouseholds’expressedprioritiesforlighting,charging
patternsaredominatedbythelamp:mostofthetime,thekitisusedtochargethelamp
(26hoursperweek),followedbyoperatingtheradio(20hours).Itishardlyusedto
chargeacellphone(onlytwohours8).
DuetothetechnicaldrawbacksofthePico‐PVkit,wewillconcentrateinthefollowing
oneffectsrelatedtotheusageofimprovedlightingservice.Virtuallyallkitowning
8Theshareofhouseholdsusingthekitforcellphonechargingisverylowatlessthantenpercent.Thosehouseholds
thatdochargetheirphonewiththekitchargeit19hoursperweek.
householdspredominantlyuseitforlighting.9Somedetailsonradiousage,preferred
programsandotherinformationsourcesareshowninthesupplementalappendix.10
ThePico‐PV lampsaremainlyusedby femaleadults, followedbymaleadults (see
Table4).Childrenusethelampslessfrequently.
Traditional lampusagegoesdown substantially,with 47percent of the treatment
groupusingexclusively thePico‐PV lamp for lightingpurposes.11While treatment
grouphouseholdsuseonaverage0.8traditionallamps(anytype,includingcandles),
controlgrouphouseholdsuse1.4traditionallampsimplyingthatthePico‐PVlamps
havereplacedhalfofthetraditionallightingsources.Treatmenthouseholdsuseabove
allsignificantlylesswicklampsandhand‐craftedLEDlamps,butalsolessready‐made
torches,hurricanelamps,andmobileLEDlamps.Theshareofhouseholdsthatdonot
useanyartificiallightingsource,amountingtoninepercentinthecontrolgroup,still
reachesfivepercentamongtreatmenthouseholds.Theyeitherbelongtothegroupof
non‐compliersortothehouseholdswithtechnicalproblemswiththePico‐PVlamp.
Table4.UsageofPico‐PVKits(ShareofTreatmentHouseholdsinPercent)
9Theonlyexceptionsarefourhouseholdsthatreportedtohavetechnicalproblemswiththelampandcannotuse
itforthisreason.10Itcanbeseenthatradiousagesignificantlyincreasedinthetreatmentgroup,onaverageandacrossalltypesof
householdmembers.Adultslistenabovealltonewsontheradio,whilechildrenlistentomusic.Consequently,
radioissubstantiallymoreoftenthemainsourceofinformationfortreatmenthouseholds.Inthecontrolgroup
communitygatheringsconstitutetypicallyamoreimportantsourceofinformation.
11TableS6.1inthesupplementalappendixshowsacomprehensivepresentationoflampusageinthetreatment
andthecontrolgroup.
Shareoftreatmenthouseholds…
(inparentheses:onlycomplianthouseholds) %
Pico‐PVlampismainlyusedby… %
usingthekitatleastonceaday 86(95) Femaleadult>17yearsold 49
…usingthekitforlighting 85(97) Maleadult>17yearsold 23
…usingthekitforlisteningtotheradio 68(79) Femalebetween12and17yearsold 10
…usingthekitforchargingmobilephones 10(11) Malebetween12and17yearsold 7
…usethebatterypack 65(71) Collectivelyusedbywholefamily 6
Childrenbetweensixand11yearsold 5
Source:Householddataset2012.
Most lampusers are satisfiedwith the lightingqualityof the lamp.More than 70
percentofalllampusersreporttheyare“always”or“often”satisfiedwiththelighting
quality.Only22percentreporttobesatisfiedonlyseldomandsixpercentarenever
satisfied.Satisfaction levelswithtraditional lampsaresubstantially lower.Forwick
lampsandhand‐craftedLEDlamps,94percentand91percent,respectively,reportto
besatisfiedseldomornever.
Sincebothtreatmentandcontrolhouseholdsarelocatedwithinthesamecommunities,
spill‐overeffectsmightoccur.Especiallychildrenoftenmeetandplaywithfriendsand
theremightbepositivespill‐overeffectsonotherhouseholds’children.Ifamongthese
‘other’householdsarehouseholds fromourcontrolgroup, itmayevendownward
bias our impact estimates. Yetwe did not find any evidence for spill‐overs. For
instance,inthecontrolgrouptheshareofchildrenstudyingoutsidetheirhomedid
notincreaseandisnegligibleatlessthanonepercent.Moregenerally,thequalitative
interviewsweconducteddidnotprovideanyindicationforjointactivitiesusingthe
kitsandhencespill‐oversofthatsort.
BudgetEffectsandKeroseneConsumption
Lookingatthepriceperconsumed lightinghourandthepriceperconsumed lumenhour
(Table5),householdsinthecontrolgrouppayapproximatelyfivetimesasmuchper
lightinghourashouseholdsinthetreatmentgroup(950FRWvs.180FRW;1.56USD
vs.0.30USD).Thedifference isobviouslyevenmorepronounced for thepriceper
lumenhour:Ahouseholdinthecontrolgrouppaysseventimesmoreperlumenhour
thanahouseholdinthetreatmentgroup(70FRWvs.9FRW;0.12USDvs.0.02USD).
This reduction in lighting costs effectively translates into a strong increase in the
amountoflumenhoursconsumedperdayintreatedhouseholds,whichismorethantwo
timesashighasincontrol‐grouphouseholds(seeTable5)—reflectingtheverypoor
lightingqualityof traditional lightingsources.Yet,alsowithoutaccounting for the
improved quality of lighting, the Pico‐PV kit leads to an increase in lighting
consumption.Theamountoflightinghoursconsumedperdayissignificantlyhigherin
thetreatmentgroupafterhavingreceivedthePico‐PVlamp.
Table5.PriceandConsumptionofLightingEnergy
Treatment Control ITT p‐value
Costperlightinghour(inFRWper100hours) 176 950 ‐702 .000
Costperlumenhour(inFRWper100hours) 9 70 ‐57 .000
Lightinghoursconsumedperday 4.43 3.85 0.59 .074
Lumenhoursconsumedperday 142 61 78 .000
Note:TheITTdepictsthedifferenceinmeansatthefollow‐upstagebetweenthewholetreatmentandcontrolgroup,
includingalsonon‐complyinghouseholds.Wecontrolforallstratificationandre‐randomizationcharacteristics.
Detailedestimationresultscanbefoundinthesupplementalappendix.ExchangerateasofNovember2011:1USD
=607FRW.
Source:Householddataset2011/2012.
Lookingattotalenergyexpenditure(Table6.ExpendituresperMonthperCategory(inFRW)
Treatment Control ITT
p‐
value
Candles 42 109 ‐20 .339
Keroseneforlighting 155 609 ‐418 .000
Bigbatteries(TypeD) 358 352 ‐9 .750
Smallbatteries(TypeAA) 30 72 ‐43 .003
Mobilephonecharging 407 520 ‐68 .407
Totaltraditionalenergysources(withoutcookingenergy) 993 1,662 ‐557 .000
Totalexpenditures 37,971 31,334 7,249 .276
Shareofenergyexpenditureontotalexpenditures 0.04 0.07 ‐0.03 .001
Note:TheITTdepictsthedifferenceinmeansatthefollow‐upstagebetweenthewholetreatmentandcontrolgroup,
includingalsonon‐complyinghouseholds.Wecontrolforallstratificationandre‐randomizationcharacteristics.
Detailedestimationresultscanbefoundinthesupplementalappendix.ExchangerateasofNovember2011:1USD
=607FRW).
Source:Householddataset2011/2012.
As a consequence, we observe a significant reduction in expenditures for small
batteries, but not for the larger batteries since households use their Pico‐PV kit
predominantlyforlightingbutonlyveryseldomtoruntheirradio.Theconsumption
ofcandlesisalsosignificantlyreduced.Inaddition,wefindamoderatereductionin
expenditures on cell phone charging, although the difference is not significant.
Estimating an ATT only among mobile phone users by employing the random
treatmentassignmentasan instrumentshowsastatisticallysignificantreductionof
costsforphonechargingof1,662FRW(2.74USD).Theaveragehouseholdthatpays
forchargingthemobilephonepays1,400FRWpermonth(2.31USD).
In total,energyexpenditureswithoutcookingenergyare557FRW (0.92USDPPP)
lowerinthetreatmentgroup.Thisdifferenceisstatisticallysignificant.Ifwecompare
this to the total household expenditures it shows the importance of energy
expenditures for thehouseholdbudget:The shareof energy expenditureswithout
cookingdecreasesbythreepercentagepointsfromsevenpercenttofourpercent.
),weobservethathouseholdsspendaroundfivepercentoftheiroverallexpenditures
on kerosene, candles, and dry‐cell batteries. In treated households we expect a
significantdecreaseofexpenditures forkerosene,candlesanddry‐cellbatteries. In
fact,we observe a significant and considerable drop of kerosene expenditures by
almost70percent.Twotypesofdry‐cellbatteriesareusedinoursample,big(TypeD)
andsmall(TypeAA)batteries.Whilemorethan90percentofsmallbatteriesareused
forlighting,morethanthree‐fourthsofbigbatteriesareusedforradios.
Table6.ExpendituresperMonthperCategory(inFRW)
Treatment Control ITT
p‐
value
Candles 42 109 ‐20 .339
Keroseneforlighting 155 609 ‐418 .000
Bigbatteries(TypeD) 358 352 ‐9 .750
Smallbatteries(TypeAA) 30 72 ‐43 .003
Mobilephonecharging 407 520 ‐68 .407
Totaltraditionalenergysources(withoutcookingenergy) 993 1,662 ‐557 .000
Totalexpenditures12 37,971 31,334 7,249 .276
Shareofenergyexpenditureontotalexpenditures 0.04 0.07 ‐0.03 .001
Note:TheITTdepictsthedifferenceinmeansatthefollow‐upstagebetweenthewholetreatmentandcontrolgroup,
includingalsonon‐complyinghouseholds.Wecontrolforallstratificationandre‐randomizationcharacteristics.
Detailedestimationresultscanbefoundinthesupplementalappendix.ExchangerateasofNovember2011:1USD
=607FRW).
Source:Householddataset2011/2012.
As a consequence, we observe a significant reduction in expenditures for small
batteries, but not for the larger batteries since households use their Pico‐PV kit
predominantlyforlightingbutonlyveryseldomtoruntheirradio.Theconsumption
ofcandlesisalsosignificantlyreduced.Inaddition,wefindamoderatereductionin
expenditures on cell phone charging, although the difference is not significant.
Estimating an ATT only among mobile phone users by employing the random
treatmentassignmentasan instrumentshowsastatisticallysignificantreductionof
costsforphonechargingof1,662FRW(2.74USD).Theaveragehouseholdthatpays
forchargingthemobilephonepays1,400FRWpermonth(2.31USD).
In total,energyexpenditureswithoutcookingenergyare557FRW (0.92USDPPP)
lowerinthetreatmentgroup.Thisdifferenceisstatisticallysignificant.Ifwecompare
this to the total household expenditures it shows the importance of energy
expenditures for thehouseholdbudget:The shareof energy expenditureswithout
cookingdecreasesbythreepercentagepointsfromsevenpercenttofourpercent.
12 This difference seems not to be driven by the treatment. The (nonsignificant) difference in total expenditures
had already existed at baseline. Moreover, the different subcategories of expenditures do not show any significant changes over time neither.
HealthandEnvironmentalEffects
The combustionofkerosene is associatedwithharmful emissions that can lead to
respiratory diseases (WHO 2016). Although the relative contribution of kerosene
lamps tohouseholdairpollution is rather lowcompared to firewoodandcharcoal
usageforcookingpurposes,itistheimmediateexposureofpeoplesittingnexttoa
wicklampforaspecifictask(e.g.,studying),thatmakeskeroseneasubstantialhealth
threat(Lametal.2012).
Indeed,inoursamplekerosenelampsareaboveallusedbychildrenforstudyingand
bywomen for cooking. In qualitative in‐depth interviews preceding the baseline
surveymanyhouseholdscomplainedaboutsootykerosenelampsleadingtorecurring
eyeproblemsandkidshavingblacknasalmucus.Wethereforeexaminedtheextent
towhichthedecreaseinkerosenelampusagetranslatesintoaperceivedimprovement
ofairqualityand,potentially, intoadecrease inrespiratorydiseasesymptomsand eye
problems.Atthebaselinestagethejudgementofmosthouseholds(around67percent
inbothgroups)wasthatairqualityintheirhouseswasgood,inthefollow‐upsurvey
45percentoftreatedhouseholdsandonlythreepercentofcontrolhouseholdssaythat
theairqualityintheirhomeshasimprovedincomparisontothebaselineperiod.In
anopenquestion,virtuallyall treatedhouseholdsascribe this improvement to the
Pico‐PVlamp.Lookingatself‐reportedhealthindicators,though,wecannotconfirm
thatthisimprovedairqualityleadstoabetterhealthstatusofthehouseholdmembers,
which isnot surprisinggiven that cooking fuelsare still thedominating sourceof
householdairpollution.13
HouseholdsinnonelectrifiedareasinAfricaareincreasinglyusingdry‐cellbatteries
and LED lamps to light their homes. Therefore, a potential reduction in dry‐cell
batteriesdeservesspecialattentionbecausetheymightcontainharmfulmaterialsand
a proper collection system does not exist. In fact, in our sample 95 percent of
households throwdischargedbatteries into theirpit latrines, that is,nonsealed3–4
meterholesintheirbackyard.Twopercentofthehouseholdscollectthemwiththeir
garbage,and threepercent throw themawaysomewhere in theirbackyard.Hence,
potentiallytoxicsubstancescanbeexpectedtoenterthegroundwater.Theextentto
which thisposesa threat topeople’shealth isunclear,as little isknownabout this
process,neitherinRwandanorelsewhere(seealsoBenschetal.2015).
DomesticProductivityEffects
BuildingonthesubstantialusageofthePico‐PVlampweexaminetheextenttowhich
thisinducesapotentialgainindomesticproductivity.Forthispurpose,welookatthe
mainusers’activitiesexercisedwhenusingthePico‐PVlampand—inordertoassess
the extent of the quality improvement—which lighting sources are used among
householdsinthecontrolgroupfortherespectiveactivity.
ThemostfrequentusersofthePico‐PVlamparefemaleadults,ofwhich87percent
usethelampmainlyforhousework(seeTable7).Houseworkdonebywomenrefers
13Seesupplementalappendix,TableS6.2,formoredetailedresults.
abovealltocookingbutalsoincludes,forexample,childcaring,preparingthebeds
beforegoingtosleep,andrepairingclothes.ThePico‐PVlampreplacesaboveallwick
lampsand isusedbyfemaleadultsthathadnotbeenusinganyparticular lighting
devicebefore.14Maleadultsalsousethelampmostlyforhousework,althoughthese
aremorediverseactivitiesthanforwomen.Formaleadults,thePico‐PVlampreplaces
wicklamps,ready‐madetorches,andhand‐craftedLEDsandisalsousedbymenwho
hadnotusedanyartificiallightingdevicebeforeforhouseworkactivities.
Table7.ActivityUsingPico‐PVLamp,AdultsandChildreninTreatmentHouseholds(%)
FirstActivity SecondActivity ThirdActivity
Femaleadult>17yearsold N=149 Housework 87 Study 5 Eat 4
Maleadult>17yearsold N=60 Housework 71 Recreation 10 Study 10
Children6to17yearsold N=56 Study 75 Housework 16 Recreation 4
Note:Informationonactivitiesstemfromanopenquestionamongtreatmenthouseholdsatfollow‐up,askingfor
themainactivitiesthedifferentlampusersareexercisingwhileusingthelamp.
Source:Householddataset2011/2012.
Table8showsthathouseworkisdoneprimarilyduringdaytime,alsointhetreatment
group, and the total time dedicated to domestic work per day does not change
significantly. The total time household members are awake per day does not change
significantly,either.ThisrevealsthatthePico‐PVlampisalsousedduringdaytimefor
housework activities,which is in linewith observationsmade during qualitative
14Weanalyselampswitchingbycomparinglampsusedforthecorrespondingactivitiesbytreatmentandcontrol
households.Detailedresultsoftheanalysiscanbefoundinthesupplementalappendix,TableS6.3.
interviews:thetypicalRwandandwellingisquitedarkevenduringthedaytimeand
peoplesometimesuseartificiallightingintheirhomes.TotheextentthePico‐PVlamp
replacesa traditional lightingsource for theirdaytimehouseworkactivity, lighting
qualityclearlyimproves.Yet,peoplemightalsorelocateoutsideactivitiesindoorsand
replacenaturaldaylightby thePico‐PV lamp. In this case, lightingqualitywould
probablynot improve,butstill itdemonstratesthehigherflexibilitypeoplehave in
organizingtheirdailytasks.
Table8.DailyTimeAwake,TimeSpentonDomesticLaborandAnyIncomeGenerating
Activity
Treatment Control ITT p‐value
Timeawake
Headofhousehold 14h28 14h27 0h05 .739
Spouse 14h46 14h36 0h11 .378
Domesticlabor
Headofhousehold,total 2h08 2h10 ‐0h01 .950
Headofhousehold,afternightfall 0h16 0h12 0h04 .542
Spouse,total 2h48 2h30 0h16 .333
Spouse,afternightfall 0h32 0h31 0h02 .779
Any income generating activity of subsistence
farmers
Headofhousehold,total 5h37 5h29 0h21 .215
Headofhousehold,afternightfall 0h01 0h01 0h00 .823
Spouse,total 5h37 5h25 0h10 .354
Spouse,afternightfall 0h00 0h01 0h00 .462
Note:TheITTdepictsthedifferenceinmeansatthefollow‐upstagebetweenthewholetreatmentandcontrolgroup,
includingalsonon‐complyinghouseholds.Wecontrolforstratumdummiesandre‐randomizationcharacteristics.
Detailedestimationresultscanbefoundinthesupplementalappendix.
Source:Householddataset2011/2012.
Moreover, Table 8 probes into the questionwhether time dedicated to any income
generating activity increases, which might happen because the higher domestic
productivitycouldsetfreetimeforotherpurposes.Weconcentrateouranalysison
subsistencefarmersthatconstitute86percentofhouseholdheadsand85percentof
spousesatbaseline.15Boththeheadofhouseholdandthespouseslightlyincreasethe
timetheydedicatetoincomegeneration(bysixandthreepercent,respectively),but
thisdifferenceisnotstatisticallysignificant.
Thethirdmostimportantusergrouparechildrenbetweensixand17years.Theyuse
thePico‐PVlampmainlyforstudying(seeTable7).Inordertounderstandchangesin
the productivity of studying at home, we first need to analyze children’s study
patternsandhowtheydividetheirstudytimebetweendaylighttimeandevening.
AscanbeseeninTable9,inaroundone‐thirdofthehouseholdswithchildrenatschool
age, childrendonot study after school.There isno significantdifferencebetween
householdsinthecontrolandtreatmentgroups.Theshareofchildrenstudyingafter
15 We distinguish as income generating activities between subsistence farmers, governmental employees,
independentoccupations,andotherdependentoccupations.Thegroupsizesofthelatterthreearesmallatn=2,
n=1, and n=2 for spouses andn=4,n=11, and n=9 forhead ofhouseholds.Therefore, these groups are very
unbalancedacrosstreatmentandcontrolhouseholdsatbaseline(seeTableS6.4inthesupplementalappendix).
Whenestimatingeffectsontimededicatedtoincomegenerationincludingtheseoccupationgroups,wefinda
significantpostiveeffectforoverallincomegenerationtimeforspouses.Thisdifferenceisdriven,however,by
thesenon‐balancedsub‐groupsandcanthusnotbeinterpretedasaneffect.
nightfall,though,issignificantlyhigherinthetreatedgroup.Thetotalstudytime,that
is,afternightfallandduringdaytime,increasesonlyformaleprimaryschoolchildren.
Femaleprimaryschoolchildrenjustshifttheirstudytimefromafternoonhourstothe
evening leading to an increase in study time after nightfall. For secondary school
childrenwedonotobserveanysignificantchanges.Hence,thePico‐PVkitsbenefit
primarilyyoungerchildren.Besidestheincreaseoftotalstudytimeforprimaryschool
boys,itseemstoincreasetheflexibilityingirl’stimeallocation,althoughwedonot
detectwhethertheyusethefreedtimeduringthedayfordomesticworkorrecreation.
Inanycase,atleastforthosechildrenwhousedwicklampsbefore,thelightingand
airqualityincreases.
Table9.StudyPattern(OnlyHHwithChildrenatSchoolAge;6–17years)
N Treatment Control ITT p‐value
Share of HH with children studying after
school
209 67 61 5 .369
Share of HH with children studying after
nightfall
209 26 14 14 .006
Dailystudytimeafterschool(inminutes)
malechildren6‐11,total 100 0h37 0h26 0h13 .009
malechildren6‐11,afternightfall 100 0h29 0h12 0h12 .045
femalechildren6‐11,total 92 0h51 0h30 0h12 .533
femalechildren6‐11,afternightfall 92 0h27 0h11 0h11 .090
malechildren12‐17,total 89 1h01 0h54 0h21 .191
malechildren12‐17,afternightfall 89 0h50 0h32 0h14 .382
femalechildren12‐17,total 94 1h02 0h58 0h10 .327
femalechildren12‐17,afternightfall 94 0h44 0h37 0h10 .191
Note:TheITTdepictsthedifferenceinmeansatthefollow‐upstagebetweenthewholetreatmentandcontrolgroup,
includingalsonon‐complyinghouseholds.Wecontrolforallstratificationandre‐randomizationcharacteristics.
Detailedestimationresultscanbefoundinthesupplementalappendix.
Source:Householddataset2011/2012.
Altogether,we observe an effect of Pico‐PV kit ownership on time dedicated to
domesticactivitiesonly in thecaseofstudy timeofprimaryschoolboys.Forother
Pico‐PVusers,weonlyobserveahigherflexibilityinorganizingtheirdomesticduties.
Moreover, it is plausible to expect that the improved lighting also increases the
effectivenessofthetasksitisusedfor.Quantifyingthiseffectisbeyondthescopeof
ourstudy,though.Atleastforthecaseofstudents’schooltest‐scorestheFurukawa
(2014) results advise some caution inmaking hasty statements about improving
learningoutcomesbasedonlongerstudytimesandbetterlight.16
Conclusion
Our resultsshow that simplebutqualityverifiedPico‐PVkits in factconstitutean
improvementcomparedtothebaselineenergysources,mostlydry‐cellbatteriesand
kerosene.Given the small sizeof thepanel, thechargingcapacity isobviouslynot
abundantly available, andmany households did notmanage to use the panel for
chargingtheradioandmobilephones;lightingturnedouttobethemostoftenused
16RememberthatFurukawa(2014)observesevenadeclineofschoolchildren’stestscoresinspiteofanincreasein
studytimeinanRCTusingsolarlanternsinUganda.Oneexplanationforthispuzzlingoutcomeheprovidesis
thepotentiallybadlightingqualityofthesolarlamp.Inourcase,wehavenoindicationforsuchabadlighting
qualityandwebelievethiswouldhavebeendisclosedinthevariousqualitativeinterviewsweconducted(in
whichmanyotherproblemswerediscussedprettyopenly,seeSectionIV).
service.Intheseremoteandpoorareas, lighting isascarcegoodandthe lampwas
indeed intensively used by virtually all treatment group households leading to
increasesinboththequalityandthequantityoflightingusage.
The most important finding of our study is that total energy expenditures and
expenditures fordry‐cellbatteriesandkerosenegodownconsiderably.Thisshows
thatbeneficiariessubstitutetraditionalenergysourcesinsteadofjustincreasingtheir
energyconsumption.Beyondthedirecteffectthishasonhouseholdwelfare,theusage
of the lamp also implies social returns. It induces advantages for people’s health
because kerosene usage is associated with harmful smoke emissions and the
environmentbecausedry‐cellbatteriesareusuallydisposedofinunprotectedlatrines
or in the landscape. Since households in rural Sub‐Saharan Africa are rapidly
switchingfromkeroseneorcandlestoLED‐lampsthatrunondry‐cellbatteriesthis
findingdeservesparticularattention.
In addition,we find thatbeneficiariesuse thekit forvariousdomesticproduction
activitieslikecookingorstudying.Althoughwecannotquantifythebenefits,evidence
fromtheliteraturestronglysuggeststhatthesolarlampallowsdoingtheseactivities
betterandfasterthanwithtraditionallightingsources,whichplausiblyresultsinan
overallincreaseindomesticoutput.Thesolarlampalsoenableshouseholdstoallocate
theirtimemorefreelyandtoshiftactivitiestowardtheeveninghours.Schoolchildren,
forexample,findbetterandmoreflexiblestudyingconditionsthankstotheimproved
lighting source.Even if thisdoesnot lead toan immediatemeasurable increase in
domesticoutput,pursuingtheactivitieswithbetterlightandinamoreflexibleway
willatleastreducetheeffortthatisneededtoundertakehouseholdchores.Thiswould
stillbeanimportantimprovementofhousehold’slivingconditions.
Whileultimatepovertyimpactson,forexample,incomeoreducationalinvestments
mightbesmallcomparedtoproductivitygainsassociatedwithbiggerinfrastructure
interventions, these effects are still considerable from the poor’s perspective,
particularlyhavinginmindthelowinvestmentcostsoftheinterventionat30USDper
kit.
Our resultshence substantiate theTier‐1‐thresholdofmodernenergyaccess in the
SE4AllGlobalTrackingFramework.ThePico‐PVkitscan in factmeet theneed for
basicenergyservices inpoorareaswhereenergyconsumption isstillatavery low
level.Yet,comparingourfindingstomoreadvancedregionsandlargerinterventions,
suchasgridextension, italsobecomesevident thatPico‐PVkitscannot satisfy the
wholeportfolioofenergydemand(Lenzetal.2016).Hence,inmanynotsoremote
areasPico‐PVkitscanbeconsideredaseitheracomplementtoagridconnectionfor
backuppurposesorasabridgingtechnologytowardagridconnectionatalaterpoint
intime.Forverypoorareasintheperipheryofacountryasstudiedinthispaper,in
contrast,Pico‐PVisinmanycasestheonlyoptiontoobtainmodernenergybecause,
first,theseregionsarebeyondthereachoftheelectricitygridformanyyearstocome
and, second, other off‐grid solutions such as larger solar home systems are too
expensive.We thereforeargue thathouseholds in such remoteareasare themajor
targetgroupofTier1energysystemswithintheSE4Allinitiative.
Whatiscrucialfortheacceptanceofthisnewtechnologyistheproperfunctioningand
easeinusageofthekit—inparticulariftheobjectiveistosetupamarketaspursued
byprogramslikeLightingAfrica.Ithasturnedoutthatarelativelymatureproduct
suchasthePico‐PVkitusedinthisstudy,ofwhichtheprincipalcomponentshadbeen
testedandcertifiedbyLightingAfricaaswellasmassivelysold inothercountries,
mightstillexhibittechnicalproblemsunderrealusageconditions.Thisisinlinewith
findingsofFurukawa(2014)whoobservesthatinsufficientchargingunderrealusage
conditionsledtoflickeringlightquality.Testingandcertificationproceduresaswell
asthedevelopmentofcomprehensibleusageguidelinesshouldthereforeencompass
a strong component of field tests and not only laboratory examinations. This is
particularlyimportantinthelightoftherapidpenetrationofruralAfricawithnon‐
brandedLEDlampsthathasoccurredinrecentyears(seeBenschetal.2015).Interms
oflightingquality,thesedry‐cellbattery‐runlampsareonaparwithPico‐PVkits.
Nonetheless,Pico‐PVkitsthatmeetqualitystandardsintermsofusabilityandlifetime
areaworthwhileinvestment.Ifkeroseneordry‐cellbatteriesarereplaced,households
withconsumptionpatternsasobserved inour researcheconomizeonaverage0.95
USD PPP per month, which is around two percent of monthly household
expenditures.TheinvestmentintothePico‐PVkitthenpaysoffafter18months,which
islessthanitslife‐spanof2–3years,buttheinterplayofcashandcreditconstraints,
thelackofinformation,andhighdiscountrateswillmakemosthouseholdsforegothis
investment.
ThisclaimpointsatadilemmaofLightingAfricaandotherdonorandgovernmental
interventions,whichintendtodisseminatePico‐PVkitsviasustainablemarketsasa
contributiontoSE4All:Themajortargetpopulationwillhardlybeabletobringupthe
requiredinvestment.Financingschemesmightinsomeregionsbeanobvioussolution.
But given the long pay‐off period for the bottom of the income distribution and
noninternalizedadvantages,suchfinancingschemesareprobablynoteffective.Atthe
sametime,ifitisclearlythepoliticalwillbothinnationalgovernmentsandamongthe
international community to provide electricity also to the very poor, one should
considermoredirectpromotionoptions.Subsidizedorevenfreedistributionofkits
mightthenbeanalternativetoreachthepoorestofthepoor.Whilemanydevelopment
practitionersareopposedtoafreedistributionpolicyanditwouldbeinstarkcontrast
tothestrategiespursuedbyongoingdisseminationprograms,theempiricalliterature
provides evidence from other field experiments that supports such an approach
(KremerandMiguel2007;CohenandDupas2010;Tarozzietal.2014;Benschand
Peters2015).Asamatterofcourse,a subsidizeddistributionpolicywould require
establishing institutions thatmaintain the subsidy scheme including an effective
systemformaintenanceandreplacementofbrokenkitsinordertoensurelong‐term
sustainability.Moreover,sincesubsidieswouldrequirepublicfunds,thepriorityof
the SE4Allgoalwould obviouslyneed tobepondered againstotherdevelopment
objectives.
Havingsaidthis,itisalsoclearthatfurtherexperimentalstudiesthatcanexaminethe
mechanismsbehindtake‐upbehavior,suchasthehouseholds’willingness‐to‐payfor
electricenergy,theroleofcreditconstraints,andinformationarecertainlyuseful.Such
researcheffortswouldhelptodesignappropriateleast‐coststrategiestoachievethe
modern“energy‐for‐all‐goals”oftheinternationalcommunity.
REFERENCES
Barron,M.,andM.Torero.2014.“ElectrificationandTimeAllocation:Experimental
EvidencefromNorthernElSalvador.ʺMPRAPaper63782,UniversityLibraryof
Munich,Germany.
———. 2015. ʺHousehold Electrification and Indoor Air Pollution.ʺMPRA Paper
61424,UniversityLibraryofMunich,Germany.
Begemann,S.H.A.,G.J.vandenBeld,andA.D.Tenner.1997.“Daylight,ArtificialLight
and People in an Office Environment, Overview of Visual and Biological
Responses.”InternationalJournalofIndustrialErgonomics3:231–39.
Bensch, G., and J. Peters. 2015. ʺThe IntensiveMargin of Technology Adoption:
ExperimentalEvidenceonImprovedCookingStovesinRuralSenegal.ʺJournalof
HealthEconomics42:44–63.
Bensch,G.,J.Peters,andM.Sievert.2013.“FearoftheDark?HowAaccesstoElectric
LightingAffects SecurityAttitudes andNighttimeActivities inRural Senegal.”
JournalofRuralandCommunityDevelopment8(1):1–19.
———. 2015. “The Lighting Transition inAfrica: FromKerosene to LED and the
EmergingDry‐CellBatteryProblem.”(No.579).Essen:RuhrEconomicPapers.
Bensch,G.,J.Kluve,andJ.Peters.2011.“ImpactsofRuralElectrificationinRwanda.”
JournalofDevelopmentEffectiveness3(4):567–88.
Bernard, Tanguy. 2012. “Impact Analysis of Rural Electrification Projects in Sub‐
SaharanAfrica.”WorldBankResearchObserver27(1):33–51.
Bertrand, J.T.,K.OʹReilly, J.Denison,R.Anhang,andM.Sweat.2006.“Systematic
Review of the Effectiveness of Mass Communication Programs to Change
HIV/AIDS‐relatedBehaviors inDevelopingCountries.”HealthEducationResearch
21(4):567–97.
Boyce,P.R.,J.W.Beckstead,N.H.Eklund,R.W.Strobel,andM.S.Rea.1997.“Lighting
theGraveyardShift:TheInfluenceofaDaylight‐SimulatingSkylightontheTask
PperformanceandMoodofNightShiftWorkers.”LightingResearchandTechnology
29(3):105–34.
Brainard,G.C.,J.P.Hanifin,J.M.Greeson,B.Byrne,G.Glickman,E.Gerner,andM.D.
Rollag.2001.“ActionSpectrumforMelatoninRegulationinHumans:Evidencefor
aNovelCircadianPhotoreceptor.”JournalofNeuroscience21(16):6405–6412.
Cohen, J.,andP.Dupas.2010.“FreeDistributionorCost‐Sharing?Evidence froma
RandomizedMalariaPreventionExperiment.”Quarterly JournalofEconomics125
(1):1–45.
DauratA,AguirreA,ForetJ,GonnetP,KeromesA,andBenoitO.1993.“Bright
LightAffectsAlertnessandPerformanceRhythmsduringa24‐hourConstant
Routine.”PhysicsandBehaviour53(5):929–36.
Dinkelman, T. 2011. “The Effects of Rural Electrification on Employment: New
EvidencefromSouthAfrica.”AmericanEconomicReview101(7):3078–108.
Dupas, P 2014. ʺShort‐Run Subsidies and Long‐Run Adoption of New Health
Products:EvidencefromaFieldExperiment.ʺEconometrica82(1):197–228.
Furukawa,C.2014.“DoSolarLampsHelpChildrenStudy?ContraryEvidencefroma
PilotStudyinUganda.”JournalofDevelopmentStudies50(2):319–41.
Grogan,L.,andA.Sadanand.2013.“RuralElectrificationandEmployment inPoor
Countries:EvidencefromNicaragua.”WorldDevelopment43:252–65.
Grunberger,J.,L.Linzmayer,M.Dietzel,andB.Saletu.1993.“TheEffectof
BiologicallyActiveLightontheNoo‐andThymopsycheonPsycho‐physiological
VariablesinHealthyVolunteers.”InternationalJournalofPsychophysiology15(1):
27–37.
IEA.2011.WorldEnergyOutlook2011.Paris.
Jensen,R.,andE.Oster.2009.“ThePowerofTV:CableTelevisionandWomen’sStatus
inIndia.”QuarterlyJournalofEconomics124(3):1057–94.
Khandker,ShahidurR.,DouglasF.Barnes,andHussainA.Samad.2012.“TheWelfare
ImpactsofRuralElectrificationinBangladesh.”EnergyJournal33(1):187–206.
Khandker,ShahidurR.,DouglasF.Barnes,andHussainA.Samad.2013.“Welfare
ImpactsofRuralElectrification:APanelDataAnalysisfromVietnam.”Economic
DevelopmentandCulturalChange61(3):659–92.
Kremer,M.,andE/Miguel.2007.“TheIllusionofSustainability.”QuarterlyJournalof
Economics122(3):1007–65.
Kuller,R.,andT.Laike.1998.ʺTheImpactofFlickerfromFluorescentLightingon
Well‐Being,PerformanceandPhysiologicalArousal.ʺJournalofErgonomics41(4):
433–47.
Kuller,R.,andL.Wetterberg.1993.ʺMelatonin,Cortisol,EEG,ECGandSubjective
ComfortinHealthyHumans:ImpactofTwoFluorescentLampTypesatTwo
LightIntensities.”LightingResearchandTechnology25(2):71–80.
Lenz, L., A. Munyehirwe, J. Peters, and M. Sievert. 2016. “Does Large Scale
Infrastructure Investment Alleviate Poverty? Impacts of Rwanda’s Electricity
AccessRoll‐OutProgram.”WorldDevelopment,forthcoming.
LaFerrara,E.,A.Chong,andS.Duryea.2012.“SoapOperasandFertility:Evidence
fromBrazil.”AmericanEconomicJournal:AppliedEconomics4(4):1–31.
Lam,N.L.,K.R.Smith,A.Gauthier,andM.N.Bates.2012.“Kerosene:AReviewof
Household Uses and Their Hazards in Low‐ andMiddle‐Income Countries.”
JournalofToxicologyandEnvironmentalHealth,PartB15(6):396–432.
Lange,H.1999.HandbuchfürBeleuchtung,SLG,LiTG,LTG,NSVV,5.Auflage,ecomed
SICHERHEIT.LandsbergamLech.
Lighting Global. 2016. Off‐Grid SolarMarket Trends Report 2016, Bloomberg New
Energy Finance and Lighting Global in cooperation with the Global Off‐Grid
LightingAssociation(GOGLA).
Lipscom M., M. Mobarak, and T. Barham. 2013. “Development Effects of
Electrification:EvidencefromtheTopographicPlacementofHydropowerPlants
inBrazil.”AmericanEconomicJournal:AppliedEconomics5(2):200–231.
MININFRA.2016.RuralElectrificationStrategy.April2016.MinistryofInfrastructure,
RepublicofRwanda.Kigali.
O’Sullivan,K.,andD.F.Barnes.2006.“EnergyPoliciesandMultitopicHousehold
Surveys.GuidelinesforQuestionnaireDesign inLivingStandardsMeasurement
Studies.”EnergyandMiningSectorBoardDiscussionPaper(17).
Partonen,T.,andJ.Lönnqvist.2000.“BrightLightImprovesVitalityandAlleviates
DistressinHealthyPeople.”JournalofAffectiveDisorders57(1–3):55–61.
Peters,J.,andM.Sievert.2016. “ImpactsofRuralElectrificationRevisited:The
AfricanContext.”JournalofDevelopmentEffectiveness,forthcoming.
Samad,HussainA.,ShahidurR.Khandker,M.Asaduzzaman,andM.Yunus.2013.
“TheBenefitsofSolarHomeSystems:AnAnalysisfromBangladesh.”WorldBank
PolicyResearchWorkingPaperNo.6724,WorldBank,WashingtonD.C.
Sievert,M.2015.“RuralElectrificationandDomesticViolenceinSub‐SaharanAfrica.”
RWIWorkingPaperNo.570.
SustainableEnergyforAll(SE4All).2013.SustainableEnergyforAllGlobalTracking
Framework Consultation Document. http://www.se4all.org/tracking‐progress/
[AccessedSeptember10,2014].
SustainableEnergyforAll(SE4All).2015.SustainableEnergyforAllGlobalTracking
Framework Consultation Document. Retrieved from:
http://www.se4all.org/tracking‐progress/.
Tarozzi,A.,A.Mahajan, B. Blackburn,D.Kopf, L.Krishnan, and J.Yoong. 2014.
“Micro‐loans, Insecticide‐Treated Bednets, and Malaria: Evidence from a
Randomized Controlled Trial in Orissa, India.” American Economic Review 104
(7):1909–41.
United Nations. 2005. The Energy Challenge for Achieving the Millennium
DevelopmentGoals.NewYork.
———.2010.Energy foraSustainableFuture.ReportandRecommendations.New
York.
vandeWalle,D.,M.Ravallion,V.Mendiratta,andG.Koolwal.2016.“Long‐Term
Gains from Electrification in Rural India.” World Bank Economic Review,
forthcoming.
WHO.2016.BurningOpportunity:CleanHouseholdEnergyforHealth,Sustainable
Development,andWellbeingofWomenandChildren.Washington,D.C.:World
HealthOrganization.
Wilkins,A.J.,I.Nimmo‐Smith,A.Slater,L.Bedocs.1989.ʺFluorescentLighting,
HeadachesandEyestrain.ʺLightingResearchandTechnology21(1):11–18.
WorldBank.2009.“UnitCostsofInfrastructureProjectsinSub‐SaharanAfrica.”Africa
InfrastructureCountryDiagnostic,BackgroundPaperNo.11,WorldBank,Washington
D.C.
A First Step up the Energy Ladder? Low Cost Solar Kits and
Household’sWelfareinRuralRwanda
SupplementalAppendix
AppendixS1:Productivityeffectsassociatedwithlighting 46
AppendixS2:RCTImplementation 48
AppendixS3:Contractforlotterywinners 52
AppendixS4:ExternalValidityofresults 53
AppendixS5:RadioUsage 56
AppendixS6:Additionaltables 59
AppendixS7:Fullregressionresultsoftablesinmaindocument 66
AppendixS1:Productivityeffectsassociatedwithlighting
Sincethevisualperformanceofhumansstronglyincreaseswiththelightinglevel(Brainardet
al.2001),weassumethatthelaborproductivityinperformingtheseactivitiesincreaseswith
thequantityandqualityoflight.Withhighquantityandqualityoflight,activitiescanbedone
fasterandwithmoreprecision,andhenceoutput increases.Productivity in fineassembly
workforinstancehasbeenshowntoincreaseby28%asthelightinglevelincreasesfrom500
to1500lm(Lange1999).Butevenincreasingthelightinglevelfrommuchlowerlevelscomes
withsignificantproductivityeffects.Evidencecomesforinstancefromweavingmills(Lange
1999).Theliteraturealsoshowsthatlighthasastimulatingeffectontheworkmood(Kuller
andWetterberg1993;Boyceetal.1997;PartonenandLönnqvist2000).Italsohelpstoavoid
accidentsasalertnessincreaseswithlight(Dauratetal.1993).Studieshavealsoshownthat
theuseofhigherlightinglevelshelpstocopewithfatigue(Dauratetal.1993;Grunbergeret
al.1993;Begemannetal.1997).Moreover,Wilkinsetal.(1989)showthatworkinginpooror
lowqualitylighting,peoplecansuffereyestrainwhichagainresultsinpoorerperformance
andisoftenaccompaniedbyheadaches.Headachesandstressinpeoplearealsocausedby
lampflicker(KullerandLaike1998).Theliteratureattributesgoodqualitylightingtodevices
thatprovidesufficientlightatthevisualtask,gooduniformityofthelightingoverthewhole
taskarea,balancedluminousdistributionthroughouttheroom,alightinginstallationwithout
glare,goodcolorrenderingandappropriatelightcolor,andlightingwithoutflicker(Lange
1999).
References
Begemann,S.H.A.,G.J.vandenBeld,andA.D.Tenner.1997.“Daylight,artificiallightandpeopleinan
office environment, overview of visual and biological responses”, International Journal of Industrial
Ergonomics,3,231–239.
Boyce,P.R.,J.W.Beckstead,N.H.Eklund,R.W.Strobel,M.S.Rea.1997.“Lightingthegraveyard‐shift:
the influence of a daylight‐simulating skylight on the task performance andmood of night shift
workers”,LightingResearchandTechnology,29(3),105‐134.
Brainard,GeorgeC., John P.Hanifin, JeffreyM.Greeson, BrendaByrne,GenaGlickman, Edward
Gerner,andMarkD.Rollag.2001.“Actionspectrumformelatoninregulationinhumans:Evidencefor
anovelcircadianphotoreceptor”,JournalofNeuroscience,21(16):6405‐6412.
DauratA,AguirreA,ForetJ,GonnetP,KeromesA,BenoitO.1993.“Brightlightaffectsalertnessand
performancerhythmsduringa24‐hourconstantroutine”,Physicsandbehaviour,53(5):929‐36.
Grunberger,J.,L.Linzmayer,M.Dietzel,B.Saletu.1993.“Theeffectofbiologicallyactivelightonthe
noo‐ and thymopsyche on psycho‐physiological variables in healthy volunteers”, Int. J. of
Psychophysiology,15(1):27‐37.
Kuller,R.andT.Laike.1998.̋ Theimpactofflickerfromfluorescentlightingonwell‐being,performance
andphysiologicalarousal.ʺJournalofErgonomics,41(4):433‐47.
Kuller,R.andL.Wetterberg.1993.ʺMelatonin,cortisol,EEG,ECGandsubjectivecomfortinhealthy
humans:impactoftwofluorescentlamptypesattwolightintensities.”LightingResearchandTechnology,
25(2)71‐80.
Lange, Horst. 1999. Handbuch für Beleuchtung, SLG, LiTG, LTG, NSVV, 5. Auflage, ecomed
SICHERHEIT.LandsbergamLech.
Partonen,T.andJ.Lönnqvist.2000.“Bright light improvesvitalityandalleviatesdistress inhealthy
people.”JournalofAffectivedisorders,57(1‐3):55‐61.
Wilkins,A.J.,I.Nimmo‐Smith,A.Slater,L.Bedocs, ʺFluorescentlighting,headachesandeyestrain.ʺ,
LightingResearchandTechnology,21(1):11‐18.
AppendixS2:RCTImplementation
The RCT for this studywas conducted betweenNovember 2011 and July 2012 in close
cooperationwiththeRwandansurveycompanyIB&CandtheRwandanEnergyWaterand
SanitationAuthority(EWSA).IB&CteammembersandEWSAstaffwereincludedatallstages
oftheplanningandimplementationprocess.InNovember2011,wedidapreparationmission
toselecttheregionsinwhichtheRCTshouldbeimplemented.Inordertomimictheeffects
Pico‐PVkitswouldhaveontheirultimatetargetpopulation–householdsbeyondthereachof
theelectricitygridanditsextensions–weselected15remotecommunitiesequallydistributed
acrossfourdistrictsintheperipheryofthecountry(seeFigureS2.1).Thecommunitiesdonot
bordereachother.AccordingtoRwandansolarexperts,theseregionsshowamediumsolar
radiationlevelwithayearlyaverageof5.5hoursofsunlightperday.Alsointhe(cloudier)
rainyseasontheradiationlevelistypicallyenoughforthePico‐PVkittoproducesufficient
electricity.Inordertoavoidtreatmentcontamination,noneofthefewregionswereselected
inwhichPico‐PVkitswerealreadyavailable.17
FigureS2.1:Mapofsurveyregions
Source:OwnrepresentationbasedonmapprovidedbyREG.
Together with IB&C we conducted a baseline survey among 300 randomly sampled
households inDecember 2011. The baseline datawas used to build strata of comparable
householdswith regards to theconsumed lightinghoursperday,usageofmobilephones
17Foradiscussionoftherepresentativenessoftheseruralcommunities,pleaserefertothesectiononexternal
validityinthesupplementalappendix,SectionS4.
(binary),radiousage(binary),anddistrict.Wethenrandomizedthetreatmentwithinthe48
strataresultingfromthisstratificationandadditionallyappliedaminmaxt‐statmethodfor
furtherimportantbaselinecriteria(seeBruhnandMcKenzie2009).18Fortheimpactanalysis,
we include stratum dummies according to our stratification process and control for all
householdcharacteristicsusedforre‐randomization.
A fewdays after the baseline survey, thePico‐PV lampsweredelivered to the randomly
selectedhouseholds.Thosehouseholdsassignedtothecontrolgroupreceivedacompensation
(onebottleofpalmoilanda5kgsackofricewortharound7USD)inordertoavoidresentment
amongthevillagers.ThePico‐PV“winners”furthermorewereinstructedonhowtousethe
kit.Thisinstructionwasconductedbystaffmembersoftheorganizationthatmarketedthe
Pico‐PVkitinotherregionsandwhoarehencealsoresponsibleforinstructingrealcustomers
thatbuyakitataregularsalesman.
Since the surveywas embedded into abroader setof evaluation studies in theRwandan
energysectoronotherongoinginterventionsindifferentareasofthecountry,itwaspresented
asageneralsurveyonenergyusageandnotasastudyonPico‐PVorlightingusage.Neither
treatmentnorcontrolgroupmemberswereinformedabouttheexperiment.Anofficialsurvey
permissionissuedbytheRwandanenergyauthoritywasshowntobothlocalauthoritiesand
theinterviewedhouseholds.BoththePico‐PVkitandthecontrolgroupcompensationwere
presented toparticipantsnotasagift,butasa reward forparticipation in the survey.We
conductedtherandomizationinourofficeusingthedigitalizedbaselinedata.Localauthorities
aswellasthefieldstaffofIB&Cwereonlyinformedonthefinalrandomizationresults.
FigureS2.2:Participantsflow
18 SeeAshrafetal.(2010)foranapplicationofthiscombinedstratifiedre‐randomizationapproach. All
balancingcriteriaarehighlightedinTable2and3ofSection5.1.
Source:ownillustrationinaccordancewithguidelinesprovidedinBOSE(2010)
Giventhehighpovertyratesintheregion,ourlocalpartnersassessedtheriskofhouseholds
selling the Pico‐PV kit to be fairly high. Since it was our ambition to mimic a policy
interventioninwhichbasicenergyservicesareprovidedforfreetoallhouseholds(andthus
potentials tosell thekitswouldbereducedconsiderably)we tried toavoid this.Our local
researchpartnersaddressedthisriskbypreparingashortcontracttobesignedbythedistrict
mayorsandthewinnersthatobligedthewinnersnottosellthePico‐PVsystem(seeOnline
Appendix).Thegovernmentalauthorityiswellrespectedalsoinremoteareasofthecountry
andRwandansgenerallytendtocomplywithformalagreements.Atthesametimewewere
assuredthatsuchaprocedurewouldnotinduceirritationsinthevillages.Amonitoringvisit
amongallwinnerseachtwomonthswasconductedtoensuretheproperfunctioningofthe
Pico‐PVsystemsandmayremindthewinnersoftheircommitmentnottosellthesystems.
Sixmonthsaftertherandomizationwerevisitedthe300householdsforthefollow‐upsurvey.
Exceptfortwo,allhouseholdsinterviewedduringthebaselinecouldberetrievedgivingusa
fairlylowattritionrateofonly1percent.Alsocomplianceturnedouttobehighwithonly18
householdsthatdeclaredtheirPico‐PVkittobesold,lostorstolen(itcanbesuspectedthat
also the lostandstolenonesweresold in fact).Onehouseholdgot thekitonlyduring the
follow‐up, since the household had been absent duringmultiple delivery attempts after
baseline.TheparticipantflowisvisualizedinFigureS2.2.
References
Ashraf,Nava, JamesBerry,and JesseM.Shapiro.2010.“CanHigherPricesStimulateProductUse?
EvidencefromaFieldExperimentinZambia.”AmericanEconomicReview,100(5):2383–413.
Bose,Ron.2010.“AChecklistfortheReportingofRandomizedControlTrialsofSocialandEconomic
Policy Interventions inDeveloping countries:CEDEVersion 1.0.”WorkingPaperNo. 6.NewDelhi:
InternationalInitiativeforImpactEvaluation.
Bruhn,Miriam and DavidMcKenzie. 2009. “In Pursuit of Balance: Randomization in Practice in
DevelopmentFieldExperiments.”AmericanEconomicJournal:AppliedEconomics,1(4):200–32.
AppendixS3:Contractforlotterywinners
AGREEMENT OF COOPERATION (translated from Kinyarwanda) Between……………………………………..Representative of RWI/ISS And the beneficiary of solar kits:
-Name: ………………………..... -Phone number: ……………………….... -Code of household: ………………………....
-Village ……………………….... -Cell: ……………………….... -Sector: ……………………….... -District: ……………………….... -Province: ……………………….... Article 1: This agreement concerns the cooperation between RWI/ISS and beneficiaries of solar kits during research on impact of electricity on living conditions of beneficiaries. Article 2: The Agreement is valid for one year from the date of signature. Article 3: RWI/ISS’s responsibilities:
To offer beneficiaries solar kits freely (solar kits consist of 1. solar panel, 2. lamp, 3. battery power pack, 4. active and passive radio connectors, 5. radio, and 6. phone connector)
To conduct survey on impact of electricity on living conditions of beneficiaries Assist beneficiaries in collaboration with Though Stuff in any case of technical problems of
solar kits Article 4: Responsibilities of beneficiaries of solar kits:
To follow rules given by Though Stuff about how to keep well solar kits To give all required information on the impact of electrification on the living conditions To communicate Though Stuff on the encountered problems about the use of solar kits Don’t sell or give freely solar kits to someone else Turn back to RWI/ISS solar kits when beneficiaries are not able to keep them
Done at ….., the….December 2011 Signature Beneficiary’s name:……………………………………
Signature Name………………………………………………………. Local Authorities representative………………………………….
Signature Name…………………………………………………. Representative of RWI-ISS
AppendixS4:ExternalValidityofresults
Externalvalidity refers to thequestionwhether resultsobserved ina certain study canbe
expectedtobetransferabletootherregionsorwhethertheywouldalsoapplyiftheprogram
underevaluationwasupscaled(outsidetherandomisedexperiment).Itisfrequentlyargued
that RCTs aremore prone to external validity problems than observational studies.We
thereforediscussthehazardstoexternalvalidityastheyareidentifiedbyDuflo,Glennerster,
andKremer(2008).
Representativenessofthestudypopulationforadifferentpolicypopulation
Onpurpose,weselectedregionsthatareveryremoteforRwandancomparisonandhavevery
limitedaccesstomodernenergysources.Inaddition,theseregionswerenotscheduledtobe
connectedtotheelectricitygridinthesubsequentyears.Thecommunitiesarelocatedinfour
differentdistricts,coveringthreeoutoffiveRwandanprovinces.Thismightnotrepresentthe
typicaltargetareaofcommercialPico‐PVdisseminationapproaches.Forpurchasingpower
reasonssuchapproachesmightratherfocusontheperipheryofthegridcoveredareasoreven
ongrid‐connectedandurbanareas,inwhichPico‐PVdevicesareusedasback‐upintimesof
outages.Incontrast,itwastheaimofthisevaluationtoassesstheextenttowhichPico‐PVcan
generallycontributetothecombatagainstenergypoverty.Thiscontributionwouldhappen
inregions thatare farbeyond theoutreachof thegrid (orothermoreexpensiveelectricity
sourceswithhighercapacities).Hence,theresultsobtainedinthisstudyaretransferableto
other set‐ups inwhich Pico‐PV is not onlymarketed for commercial reasons, but as an
instrumenttoprovidemodernenergytothepoor.ThisistheexplicitgoalofSE4Allandalso
therolethatisassignedtoPico‐PVwithinthisSE4Allinitiative.
Moreover,intheevaluationreportunderlyingthepresentstudy(seeGrimmetal.2013)we
comparetheRCTsampleandusagebehaviourintheRCTtorealusersofthePico‐PVkitin
otherregionsofthecountry,thisis,customerswhodeliberatelydecidedtobuythekit.Inline
withexpectations,itcanbeseenthatthesereal‐worldusersaresomewhatwealthierthanthe
averageRCTuser,butusagepatterns forthekit’slamparequitesimilar(seeGrimmetal.
2013,Section4.4.).
GeneralEquilibriumeffects
GeneralEquilibriumeffectsareeffectsthatonlyoccurorbecomeperceivableifthetreatment
isprovided toa largerpopulationor fora longerperiod.Hence, these effectsmighthave
repercussionsontheRCTsampleandcanonlybecapturedbytheRCTiftheperiodbetween
randomizationandtheimpactassessmentislongenoughandthestudypopulationislarge
enough.Inthepresentcase,onecouldtheoreticallythinkofdecreasingpricesoftraditional
lightingsourcesbecauseofadecreasingdemand,whichinturncouldincreaseconsumption
oftraditionallightingsourcesbyhouseholdsbothintheRCTsampleandbeyond.Thiseffect
canbeexpectedtobeverysmall,though,sinceenergypricesaremostlydrivenbyregional
marketsifnotworldmarkets.Ontheotherhand,incaseonewouldupscaletheprogramto
thewholecountry(i.e.distributePico‐PVkitsonalargescale),forexample,sucheffectscould
occurforproductswitharegionalvaluechain(i.e.thatdonotonlydependonworldmarket
prices).
As for the timehorizon,ourevaluationexaminesshort‐termusageand impacts.Whilewe
believe that the sixmonthsperiodbetweendeliveryof thePico‐PVkitandour follow‐up
surveyislongenoughtoallowtheusers’adaptationtothenewtechnology,wecannotrule
outthatusagebehaviourwouldchangeovertime.
Hawthorne‐andJohn‐Henry‐Effects
Hawthorne‐ and John‐Henry‐effects occur if participants in an experiment change their
behaviourbecausetheyknowthattheyareparticipatinginanexperiment.Totheextentthat
the fieldwork teamhas to interactwith the studypopulation, these effects canhardlybe
excludedcompletelyinmostRCTs.However,therearewaystokeepthemassmallaspossible,
mostlybyreducingtheattentionthatisevokedintheparticipatinghouseholdandthevillages.
Thesurveysusedforthisstudywerepresentedaspartofa*generalsurveyonenergyusagein
relation to on‐going and well‐known energy interventions. Respondents were asked to
consent toparticipating inthesurvey,buttherandomizationor theexperimentalcharacter
were notmentioned.A permission letter issued by the Rwandan EnergyAuthoritywas
presented to the local authorities aswell as theparticipatinghouseholds bothduring the
baselinesurveyandthefollow‐upsurvey.Infact,ourstudywaspartofabroaderevaluation
engagement inthecountry,whichalsocovered50different targetvillagesoftheRwandan
gridroll‐outprogrammeEARP(seeLenzetal.2016).Althoughthestudyregionsofthepresent
paperwerenotscheduledtobeconnectedbyEARPinthenearfuture,mostoftheresidents
areawareoftheelectrificationprogrammethatpossiblywillalsoreachtheircommunities.The
survey work was implemented as unobtrusive as possible. Each household was visited
individually.
Furthermore, the randomly assigned Pico‐PV systemwas not labelled as a gift, but as a
compensationforparticipationinthesurvey.Also,householdsassignedtothecontrolgroup
receivedacompensationconsistingofasackof5kgofriceandonelitreofcookingoil.19Asa
sideeffect,thiscompensationforthecontrolgroupaddressesapotentialethicalconcernthat
issometimesbroughtforwardagainstRCTs:Randomlyassigningatreatmenttoonegroup
mayinduceuncomfortablefeelingsintheothergroup.
Insum,whilesomesortofsurveyeffectisunavoidable,thereisnoreasontoexpectstrong
Hawthorne and John Henry‐effects, since the participants were not informed about an
experiment,both treatmentandcontrolgroups receiveda reimbursem*nt forparticipation
andthesurveysandinterviewswereimplementedasunobtrusiveaspossible.
SpecialCare
Theway inwhichweprovided thePico‐PVkit(most importantly the training)was in line
19ThisimplementationdesignfollowstheapproachpresentedinDeMel,McKenzie,andWoodruff(2008)andwas
alsoappliedinaRCTwithimprovedcookingstovesinSenegal(BenschandPeters2015).
withwhatthecompanythatmarketedtheproducthadforeseenforthemarketingprocess.
AlthoughthelevelofcarewasprobablyhighercomparedtosomeotherPico‐PVkitsthatare
justsoldinshopsanddonotcompriseatraining,manyregularmarketvendorsalsooffersuch
trainings.
References
Bensch,GuntherandJörgPeters.2015.“Theintensivemarginoftechnologyadoption–Experimental
EvidenceonimprovedcookingstovesinruralSenegal.”JournalofHealthEconomics42:44‐63.
Duflo,E.,R.Glennerster, andM.Kremer. 2008.UsingRandomization inDevelopmentEconomics
Research:AToolkit,Chapter61,HandbookofDevelopmentEconomics.
Grimm,Michael,JörgPeters,andMaximilianeSievert.2013.ImpactsofPico‐PVSystemsUsageusinga
Randomized Controlled Trial andQualitativeMethods. Final Report on behalf of the Policy and
OperationsEvaluationDepartment(IOB)oftheNetherlandsMinistryofForeignAffairs.
Lenz, Luciane,AnicetMunyehirwe, Jörg Peters andMaximiliane Sievert (2016)Does Large Scale
InfrastructureInvestmentAlleviatePoverty?ImpactsofRwanda’sElectricityAccessRoll‐OutProgram.
WorldDevelopment,forthcoming.
Mel, Suresh de, David McKenzie, and Christopher Woodruff. 2008. “Returns to Capital in
Microenterprises:EvidencefromaFieldExperiment.”QuarterlyJournalofEconomics,123(4):1329–72.
AppendixS5:RadioUsage
TableS5.1portrays theeffectsof thePico‐PVkit treatmenton radioownershipandusage
patterns.SincetherandomizedPico‐PVkitencompassesaradio,theshareofradioownersis
closeto100percentinthetreatmentgroup,whileslightlymorethan50percentofthecontrol
grouphouseholdsownaradio.Itisabovealltheheadofthehouseholdwhousestheradio,
buttheincreasednumberofradiosinthehouseholdsalsoleadstosignificantlyincreasedradio
usageofotherhouseholdmembers.Theshareofmemberslisteningtotheradioissignificantly
higherinthetreatmentgroupforallhouseholdmembers.Thelisteninghoursforthosewho
usearadioonlyincreasessignificantlyfortheheadofhouseholds.
TableS5.1:Radioownershipandusage
Treatment Control ITT p‐value
RadioOwnership 95% 52% 41% 0.000
HHmemberlistensregularlytoradio
HeadofHH 86% 43% 41% 0.000 ‐ Listeninghoursperday(onlyuser) 4.3 3.2 1.1 0.012
Spouse 78% 42% 34% 0.000 ‐ Listeninghoursperday(onlyuser) 3.3 2.7 0.4 0.289
Boys12‐17years 76% 47% 33% 0.048
‐ Listeninghoursperday(onlyuser) 2.2 2.0 ‐0.1 0.907
Girls12‐17years 80% 42% 26% 0.151
‐ Listeninghoursperday(onlyuser) 1.7 2.1 ‐0.3 0.639
Children6‐11years 63% 33% 16% 0.090 ‐ Listeninghoursperday(onlyuser) 1.9 2.2 0.4 0.631
Radiosaremostlyusedtolistentoprogramsthattransmitinformation(seeTableS5.2).We
askedallradiouserstonametheirtwofavouriteradioprograms:thebyfarmostpreferred
programamongadultsarenews,followedbymusic.Thethirdpreferenceareprogramsthat
peoplerefertoas“théatre”.Theseareradioplaysthattrytoraiseawarenessondifferenttopics
likereconciliation,workingattitude,orjustice.Thelastcategorysubsumesspecialbroadcasts
topicslikepoliticsorhealth(‘broadcasts’inTableS5.2).
TableS5.2:Preferredradioprogramsperhouseholdmember
(inpercent)Treatment Control ITT p‐value
HeadofHH Music 46 62 ‐17 0.062
(N=193) News 84 86 ‐4 0.408
Theatre 13 5 10 0.056
Broadcast 13 8 1 0.847
Other 14 15 3 0.697
Spouse Music 43 67 ‐29 0.006
(N=158) News 71 68 5 0.426
Theatre 24 14 8 0.245
Broadcast 10 9 ‐1 0.893
Other 16 12 12 0.114
Children Music 76 71 ‐2 0.907
6‐11yearsold News 31 39 ‐1 0.963
(N=77) Theatre 22 14 ‐7 0.498
Broadcast 8 11 4 0.709
Other 4 11 ‐2 0.836
Male Music 70 83 8 0.824
12‐17yearsold News 47 63 ‐24 0.425
(N=54) Theatre 10 17 ‐20 0.247
Broadcast 13 8 ‐21 0.238
Other 13 8 12 0.562
Female Music 89 84 16 0.309
12‐17yearsold News 44 58 ‐11 0.613
(N=64) Theatre 18 16 ‐2 0.940
Broadcast 9 5 ‐3 0.856
Other 16 11 5 0.777
Inlinewiththeseusagepatterns,radioisbyfarthemostimportantsourceofinformationin
thesurveyedvillages.Morethan90percentoftreatmenthouseholdsand40percentofcontrol
households answer to an open question on their main source of information that they
exclusivelyortogetherwithothersourcesreceive informationthrough theradio(seeTable
S5.3). Apart from this, direct conversations with other people (community gatherings,
neighboursandfriends)arethemostimportantsourcesofinformation.TVsandnewspaper
are only used by a negligible number of households.While for treated households the
importance of radio is substantially higher, control households relymore on community
gathering and information exchangewith neighbours and friends. This obviously creates
potentialsforspilloversofincreasedaccesstoinformationforsomehouseholdstothewhole
village.
TableS5.3:Mainsourceofinformation(multipleanswerspossible)
Treatment Control ITT p‐value
Radio 91 40 51 0.000
Communitygathering 70 82 ‐11 0.029
Neighbours/friends 28 28 0 0.929
BoxS5.1:RadiostationsinRwanda
InRwanda, thebiggest radiostation is thestate‐financedRadioRwanda that reachesmore than90
percentofthepopulation.Itbroadcasts24hoursinKinyarwanda,French,EnglishandSwahili.Radio
Rwandamaintainsadditionally5communityradiostationsthatpartlybroadcastcontributionsfrom
RadioRwandaandadditionallycover localnewsandregional information.Since2002,alsoprivate
radiostationshavereceivedlicencestobroadcast.Theirreceptionarea,though,ismainlyrestrictedto
Kigaliandbigger towns.Withinoursurveyarea,people frequentlyreported thatbesidesRwandan
radiostationstheycanalsoreceiveradiofromBurundiorCongo.
RadioRwandaanditscommunityradiostationscoverbothentertainmentandinformationbroadcasts.
ForexampleatRadioRusizi,thecommunityradioofRusizi,onaregulardaywheretheyofferprogram
from5amto23pm,musiccovers7hours,newssumuptoalmost4hours,3.5hoursofentertainment
showslikesoapoperas(“théatre”),quizzesorsportsevents,andalmost3hoursonbroadcastwithsome
educationalbackground.Theseeducationalbroadcastsdiffuseforexampleinformationonhygieneand
cleanliness,onagriculturalactivities,onanimalhusbandry,ongoodgovernance,ortoraiseworkers’
motivation.
AppendixS6:Additionaltables
LampUsage
TableS6.1:Numberoflightingdevicesandconsumption
Shareofhouseholdsusing[lamp] Operationhoursper
dayandlamp
Treatment Control ITT p‐
value
Treatment Control
Pico‐PVlamp 0.86 0.00 0.86* 0.00 2.89 ‐
Hand‐craftedLEDlamps 0.28 0.45 ‐0.18 0 3.45 3.40
Ready‐madetorch 0.14 0.22 ‐0.10 0.01 2.23 2.12
Wicklamp 0.12 0.43 ‐0.22 0 2.47 2.98
Candles 0.07 0.15 0.04 0.935 2.05 1.58
Hurricanelamp 0.04 0.10 ‐0.07 0.002 3.2 2.43
MobileLEDlamp 0.03 0.05 ‐0.03** 0.014 2 2.57
Nolamp 0.05 0.09 0.03 0.16 ‐ ‐
SUM 1.62 1.43 0.16 0.004 4.3 3.8
Note:TheITTdepictsthedifferenceinmeansatthefollow‐upstagebetweenthewholetreatmentandcontrolgroup,
alsoincludingnon‐complyinghouseholds.Wecontrolforallstratificationandre‐randomizationcharacteristics.
Rechargeable lamps and gas lamps are not included in the table, since only one control household uses a
rechargeablelampandonlyonetreatmenthouseholdusesagaslamp.
*Probitestimationisnotapplicable,sincecontrolgrouphouseholdsdonotusethelampleadingtoconvergence
problems;wedisplaysimpledifferencesinmeansinstead.**Controllingforrandomizationstratumdummiesleads
toconvergenceproblems.Weincludethestratificationcriteriainstead.
Health
TableS6.2:Shareofhouseholdswithhouseholdmemberssufferingdiseases(inpercent)
Treatment Control ITT p‐value
Maleadult Respiratorydiseases 7 7 ‐2 0.596
Eyeproblem 7 8 0 0.972
Femaleadult Respiratorydiseases 5 6 0 0.904
Eyeproblem 9 13 ‐6 0.128
Male Respiratorydiseases 2 2 1 0.785
6‐11yearsold Eyeproblem 2 6 ‐3 0.440
Female Respiratorydiseases 0 0 0 ‐‐‐
6‐11yearsold Eyeproblem 9 10 5 0.472
Male Respiratorydiseases 0 0 0 ‐‐‐
12‐17yearsold Eyeproblem 2 0 1 0.672
Female Respiratorydiseases 4 0 3 0.452
12‐17yearsold Eyeproblem 6 3 8 0.292
Note:Thedataisself‐reportedinformationofwhetheranyhouseholdmembersuffersfromanyofthediseases.The
ITT depicts the difference inmeans at the follow‐up stage between thewhole treatment and control group,
includingalsonon‐complyinghouseholds.Wecontrolforallstratificationandre‐randomizationcharacteristics.
Lampsusedfordomesticlabor
TableS6.3:Mostfrequentlyusedlampsforhouseworkbymaleandfemaleadult(percent
ofallhouseholds)
Femaleadultsdoing
housework
Maleadultsdoing
housework
Children(6‐17years)
studying
Lamp
Treat. Ctrl. ITT p‐
value
Treat. Ctrl. ITT p‐
value
Treat. Ctrl. ITT p‐value
Wicklamp 732 ‐23 0.000
3 9 ‐7 0.001
2 12
‐
12***0.000
Ready‐madetorch 8 12 ‐7 0.056 3 7 ‐8 0.000
Hand‐craftedLED 79 ‐3 0.182
1 5
‐
6**
0.003
Pico‐PVlamp 72 0 72* 0.000 26 0 26* 0.000 30 0 30* 0.000
None 15 42 ‐25 0.000 68 78 ‐9 0.006 32 41 ‐19 0.000
Noneandstudyingatdaytimeonly 9 18 ‐19 0.000
Noneandstudyingafternightfall 23 22 ‐2 0.633
Note:TheITTdepictsthedifferenceinmeansatthefollow‐upstagebetweenthewholetreatmentandcontrolgroup,
includingalsonon‐complyinghouseholds.Wecontrolforallstratificationandre‐randomizationcharacteristics.
Detailedestimationresultscanbefoundinthefollowing.
*Probitestimationisnotapplicable,sincecontrolgrouphouseholdsdonotusethelampleadingtoconvergence
problems;wedisplaysimpledifferencesinmeansinstead.**Controllingforrandomizationstratumdummiesleads
to convergence problems.We include the stratification criteria instead. *** Controlling for baseline kerosene
consumption (continuous) causes convergence problems.We include a dummy indicating baseline kerosene
consumptionyes/noinstead.
TableS6.3a:Mostfrequentlyusedlampsforhouseworkbyfemaleadult(%ofall
households)
Wicklamp Ready‐made
torch
Hand‐
crafted
LED
None
Treatment ‐0.228 ‐0.069 ‐0.034 ‐0.249
(0.000)*** (0.056)* (0.182) (0.000)***
Consumptionofcandles ‐0.001 ‐0.016 0.000 ‐0.001
(0.612) (0.001)*** (0.829) (0.854)
Consumptionofkerosene 0.013 ‐0.022 ‐0.017 0.017
(0.229) (0.427) (0.652) (0.151)
Numberofhouseholdmembers ‐0.005 0.011 ‐0.006 0.017
(0.676) (0.106) (0.347) (0.091)
Numberofmobilephones 0.047 0.005 0.013 ‐0.025
(0.270) (0.836) (0.484) (0.734)
Plastereddwelling ‐0.056 0.071 ‐0.000 0.147
(0.481) (0.024)** (0.994) (0.200)
Modernwall ‐0.073 ‐0.106 0.042 0.075
(0.217) (0.119) (0.390) (0.468)
Modernfloor 0.078 0.090 ‐0.058 ‐0.136
(0.094)* (0.076)* (0.246) (0.161)
Hand‐craftedLED ‐0.082 ‐0.095 0.071 0.042
(0.134) (0.022)** (0.012)** (0.507)
MobileLED ‐0.040 0.058 ‐0.008 0.007
(0.763) (0.379) (0.916) (0.962)
Householdownsland ‐0.238 0.104 ‐0.060 0.147
(0.003)*** (0.434) (0.176) (0.218)
Householdownsonegoat ‐0.064 0.072 ‐0.005 ‐0.019
(0.126) (0.109) (0.915) (0.808)
Householdownsseveralgoats ‐0.005 0.034 0.060 ‐0.005
(0.897) (0.430) (0.148) (0.932)
Householdownsonecow ‐0.033 0.001 0.025 ‐0.065
(0.543) (0.978) (0.521) (0.321)
Householdownsseveralcows 0.026 ‐0.056 ‐0.086 0.036
(0.657) (0.287) (0.055)* (0.722)
Headofhouseholdcompletedprimaryschool 0.046 0.085 ‐0.002 ‐0.095
(0.385) (0.045)** (0.937) (0.083)
Headofhouseholdcompletedsecondaryschool ‐0.100 0.185 n.i. ‐0.038
(0.136) (0.021)** (0.816)
Redistributedbetweenstrataforrandomization ‐0.029 0.078 n.i. 0.112
(0.853) (0.314) (0.409)
PseudoR‐Squared 0.36 0.32 0.29 0.20
NumberofObservations 294 294 294 294
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;Standard
errorsareclusteredatthevillagelevel.
n.i.:notincludedsincevariablepredictssuccessorfailureperfectly.
Table S6.3b: Most frequently used lamps for housework by male adult (% of all
households)
Wicklamp Ready‐
madetorch
Hand‐crafted
LED
None
Treatment ‐0.070 ‐0.078 ‐0.055 ‐0.085
(0.001)*** (0.000)*** (0.003)*** (0.006)***
Consumptionofcandles 0.004 ‐0.007 ‐0.013 0.002
(0.062)* (0.006)*** (0.113) (0.728)
Consumptionofkerosene 0.011 ‐0.043 ‐0.057 0.007
(0.006)*** (0.046)** (0.000)*** (0.545)
Numberofhouseholdmembers 0.007 0.009 0.002 0.003
(0.195) (0.025)** (0.427) (0.856)
Numberofmobilephones 0.006 ‐0.005 0.000 ‐0.048
(0.794) (0.587) (0.979) (0.391)
Modernfloor ‐0.010 0.050 n.i. ‐0.031
(0.874) (0.056)* (0.781)
Hand‐craftedLED ‐0.003 ‐0.053 0.037 ‐0.090
(0.900) (0.047)** (0.140) (0.089)*
Householdownsland ‐0.029 n.i. ‐0.022 0.009
(0.581) (0.447) (0.942)
Householdownsonegoat 0.039 ‐0.051 n.i. ‐0.023
(0.168) (0.216) (0.723)
Householdownsseveralgoats 0.077 0.082 ‐0.044 ‐0.107
(0.004)*** (0.000)*** (0.036)** (0.038)**
Householdownsonecow ‐0.007 ‐0.003 n.i. 0.030
(0.866) (0.896) (0.614)
Householdownsseveralcows ‐0.016 ‐0.045 0.013 0.130
(0.656) (0.244) (0.470) (0.134)
Headofhouseholdcompletedprimaryschool 0.036 0.048 ‐0.025 ‐0.036
(0.073)* (0.138) (0.040)** (0.485)
Redistributedbetweenstrataforrandomization 0.137 n.i. n.i. 0.109
(0.039)** (0.646)
Plastereddwelling n.i. 0.009 n.i. 0.191
(0.526) (0.079)*
Modernwall n.i. ‐0.059 ‐0.047 0.072
(0.073)* (0.049)** (0.488)
MobileLED n.i. ‐0.050 0.121 ‐0.110
(0.317) (0.000)*** (0.359)
Head of household completed secondary
school
n.i. 0.075 0.020 ‐0.065
(0.147) (0.568) (0.528)
Householdownsaradio n.i. n.i. ‐0.004 n.i.
(0.811)
Householdownsamobilephone n.i. n.i. 0.038 n.i.
(0.000)***
Consumptionoflightinghours n.i. n.i. ‐0.001 n.i.
(0.883)
PseudoR‐Squared 0.36 0.51 0.51 0.17
NumberofObservations 294 294 295 294
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
n.i.:notincludedsincevariablepredictssuccessorfailureperfectly.
TableS6.3c:Mostfrequentlyusedlampsforstudyingbychildren(%ofhouseholdswith
childrenatschoolage;N=208)
Wicklamp Nolamp Noneand
studyingat
daytime
only
Noneand
studying
after
nightfall
Treatment ‐0.118 ‐0.194 ‐0.185 ‐0.020
(0.000)*** (0.000)*** (0.000)*** (0.633)
Consumptionofcandles 0.002 ‐0.004 ‐0.009 ‐0.001
(0.327) (0.235) (0.116) (0.780)
Numberofhouseholdmembers 0.021 0.009 0.003 0.006
(0.049)** (0.623) (0.748) (0.669)
Numberofmobilephones ‐0.027 0.089 0.059 0.015
(0.211) (0.319) (0.272) (0.827)
Plastereddwelling ‐0.031 ‐0.053 ‐0.061 ‐0.083
(0.526) (0.616) (0.540) (0.284)
Modernwall ‐0.137 ‐0.069 ‐0.041 ‐0.082
(0.008)*** (0.378) (0.530) (0.232)
Modernfloor 0.025 ‐0.088 n.i. 0.049
(0.692) (0.525) (0.453)
Hand‐craftedLED ‐0.049 ‐0.092 ‐0.065 ‐0.020
(0.312) (0.360) (0.005)*** (0.773)
Householdownsonegoat ‐0.041 0.109 0.083 0.054
(0.415) (0.257) (0.005)*** (0.628)
Householdownsseveralgoats ‐0.138 0.178 0.027 0.137
(0.014)** (0.012)** (0.644) (0.015)**
Householdownsonecow ‐0.022 0.100 0.125 ‐0.039
(0.652) (0.292) (0.040)** (0.621)
Householdownsseveralcows 0.118 0.217 ‐0.051 0.142
(0.021)** (0.033)** (0.674) (0.086)*
Headofhouseholdcompletedprimaryschool ‐0.013 ‐0.058 0.056 ‐0.106
(0.797) (0.543) (0.269) (0.179)
Headofhouseholdcompletedsecondaryschool 0.033 ‐0.123 ‐0.027 ‐0.229
(0.574) (0.543) (0.842) (0.216)
Consumptionofkerosene n.i. ‐0.144 ‐0.111 ‐0.060
(0.015)** (0.005)*** (0.110)
Householdownsland n.i. 0.055 0.014 0.064
(0.751) (0.891) (0.663)
Redistributedbetweenstrataforrandomization n.i. 0.177 ‐0.655 0.204
(0.379) (0.000)*** (0.216)
MobileLED n.i. 0.116 n.i. 0.186
(0.558) (0.147)
PseudoR‐Squared 0.50 0.27 0.40 0.33
NumberofObservations 207 207 207 207
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
n.i.:notincludedsincevariablepredictssuccessorfailureperfectly.
TableS6.4:MainOccupationatbaseline(inpercent)
Treatment Control
Headofhousehold SubsistenceFarmer 90 83
GovernmentEmployee 1 1
Otherindependentoccupation 3 5
Otherdependentoccupation 1 5
Housewife,Retired 4 6
Unemployed 1 0
Spouse SubsistenceFarmer 98 93
GovernmentEmployee 2 0
Otherindependentoccupation 0 1
Otherdependentoccupation 0 2
Housewife,Retired 0 3
Studies 0 1
AppendixS7:Fullregressionresultsoftablesinmaindocument
Table5:Priceandconsumptionoflightingenergy
VARIABLES Costper
lightinghour
Costper
lumenhour
Lightinghours
consumed
perday
Lumenhours
consumed
perday
Treatment ‐7.022 ‐0.566 0.585 77.736
(0.000)*** (0.000)*** (0.074)* (0.000)***
Consumptionofcandles 0.182 0.012 0.021 0.026
(0.013)** (0.009)*** (0.267) (0.983)
Consumptionofkerosene 1.260 0.109 0.054 ‐3.696
(0.001)*** (0.000)*** (0.718) (0.163)
Numberofhouseholdmembers ‐0.017 0.001 0.013 ‐4.296
(0.911) (0.950) (0.851) (0.374)
Numberofmobilephones ‐0.392 ‐0.034 1.296 21.644
(0.528) (0.355) (0.007)*** (0.019)**
Plastereddwelling ‐2.112 ‐0.208 0.709 ‐14.458
(0.120) (0.012)** (0.372) (0.607)
Modernwall 1.529 0.147 ‐1.249 ‐27.763
(0.310) (0.141) (0.039)** (0.077)*
Modernfloor 3.310 0.132 0.461 ‐40.420
(0.064)* (0.120) (0.549) (0.243)
HandcraftedLED ‐1.494 ‐0.059 0.592 ‐10.708
(0.136) (0.414) (0.295) (0.452)
MobileLED ‐1.607 ‐0.167 0.873 47.231
(0.300) (0.108) (0.405) (0.199)
Householdownsland ‐1.447 ‐0.085 ‐0.181 19.863
(0.139) (0.285) (0.868) (0.430)
Householdownsonegoat ‐2.448 ‐0.154 ‐0.296 ‐38.858
(0.119) (0.090)* (0.563) (0.307)
Householdownsseveralgoats 0.757 0.024 0.021 ‐16.033
(0.591) (0.789) (0.967) (0.608)
Householdownsonecow ‐0.004 ‐0.063 0.879 19.648
(0.998) (0.409) (0.078)* (0.225)
Householdownsseveralcows 1.784 0.213 0.205 75.429
(0.090)* (0.044)** (0.670) (0.239)
Headofhouseholdcompletedprimaryschool 0.919 0.049 0.492 13.202
(0.165) (0.466) (0.311) (0.305)
Headofhouseholdcompletedsecondaryschool 3.560 0.276 ‐0.980 ‐33.814
(0.085)* (0.076)* (0.543) (0.348)
Redistributedbetweenstrataforrandomization ‐0.018 0.006 1.319 20.259
(0.990) (0.967) (0.202) (0.530)
Constant 9.486 0.661 2.138 ‐17.590
(0.012)** (0.023)** (0.329) (0.603)
Observations 265 265 288 288
AdjustedR‐squared 0.397 0.404 0.121 0.165
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
Table6a:Expenditurespermonthpercategory(inFRW)
VARIABLES Candles Kerosene Char
coal
Big
batteries
Small
batteries
Mobile
phone
charging
Treatment ‐19.927 ‐418.007 1.917 ‐9.344 ‐43.352 ‐67.916
(0.339) (0.000)*** (0.447) (0.750) (0.003)*** (0.407)
Consumptionofcandles 29.621 8.273 ‐0.019 0.405 ‐1.165 19.456
(0.000)*** (0.165) (0.780) (0.911) (0.537) (0.039)**
Consumptionofkerosene 1.833 169.929 ‐0.096 9.460 ‐0.074 7.410
(0.753) (0.007)*** (0.594) (0.133) (0.978) (0.686)
Numberofhouseholdmembers 8.228 8.295 ‐0.934 ‐2.416 1.281 29.491
(0.267) (0.567) (0.382) (0.829) (0.678) (0.383)
Numberofmobilephones ‐9.833 195.483 7.513 60.265 17.418 1,100.513
(0.631) (0.067)* (0.367) (0.149) (0.232) (0.009)***
Plastereddwelling 16.502 ‐63.128 ‐0.548 ‐70.298 ‐12.142 205.632
(0.581) (0.576) (0.726) (0.330) (0.686) (0.198)
Modernwall ‐44.866 ‐120.093 ‐0.820 64.199 ‐58.167 63.933
(0.322) (0.267) (0.617) (0.504) (0.058)* (0.686)
Modernfloor 56.056 281.553 ‐2.738 ‐65.299 22.367 ‐434.782
(0.246) (0.050)* (0.360) (0.282) (0.599) (0.060)*
Hand‐craftedLED 18.316 9.387 ‐4.571 20.710 ‐15.293 86.275
(0.392) (0.925) (0.334) (0.686) (0.440) (0.552)
MobileLED ‐2.633 255.134 ‐3.422 139.388 21.636 ‐84.409
(0.965) (0.234) (0.403) (0.361) (0.735) (0.867)
Householdownsland ‐0.584 ‐117.541 5.549 179.692 ‐19.392 ‐14.158
(0.986) (0.287) (0.185) (0.036)** (0.544) (0.942)
Householdownsonegoat 73.643 ‐134.849 ‐1.542 26.665 ‐16.252 ‐11.673
(0.000)*** (0.225) (0.500) (0.621) (0.202) (0.912)
Householdownsseveralgoats 10.973 ‐111.840 ‐1.314 164.868 33.091 287.823
(0.609) (0.363) (0.510) (0.042)** (0.219) (0.036)**
Householdownsonecow ‐11.154 ‐151.915 5.046 134.052 14.472 74.082
(0.533) (0.076)* (0.290) (0.019)** (0.707) (0.600)
Householdownsseveralcows 93.441 91.004 ‐1.129 68.589 ‐19.832 ‐136.453
(0.027)** (0.486) (0.673) (0.315) (0.233) (0.568)
Headofhouseholdcompletedprimaryschool ‐27.737 89.954 1.490 115.139 ‐2.168 26.694
(0.095)* (0.332) (0.353) (0.004)*** (0.885) (0.799)
Headofhouseholdcompletedsecondaryschool 83.751 26.200 ‐0.821 326.077 110.878 ‐523.982
(0.277) (0.880) (0.717) (0.047)** (0.408) (0.153)
Redistributedbetweenstrataforrandomization 44.463 56.283 1.826 126.782 ‐80.459 185.076
(0.488) (0.769) (0.426) (0.403) (0.231) (0.292)
Constant ‐227.965 ‐283.081 ‐
18.863
‐419.144 261.650 ‐320.098
(0.049)** (0.463) (0.259) (0.173) (0.130) (0.723)
Observations 296 296 295 296 296 296
AdjustedR‐squared 0.622 0.356 ‐0.019 0.275 0.046 0.480
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations.Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
Table6b:Expenditurespermonthpercategory(inFRW)
VARIABLES Totaltraditional
energyexpenditures
(withoutcooking)
Total
expenditures
Shareofenergy
expenditureon
totalexpenditures
Treatment ‐556.653 7,249.033 ‐0.030
(0.000)*** (0.276) (0.001)***
Consumptionofcandles 56.573 ‐603.375 0.001
(0.002)*** (0.074)* (0.017)**
Consumptionofkerosene 188.463 2,594.154 0.003
(0.010)** (0.002)*** (0.144)
Numberofhouseholdmembers 43.955 3,700.456 0.000
(0.315) (0.135) (0.922)
Numberofmobilephones 1,371.322 22,314.462 0.013
(0.004)*** (0.051)* (0.164)
Plastereddwelling 75.918 21,254.017 ‐0.023
(0.724) (0.165) (0.066)*
Modernwall ‐95.781 23,581.834 ‐0.002
(0.672) (0.137) (0.893)
Modernfloor ‐142.721 ‐7,699.614 0.016
(0.550) (0.719) (0.143)
Hand‐craftedLED 114.774 1,974.949 ‐0.004
(0.525) (0.762) (0.724)
MobileLED 325.633 ‐1,101.144 ‐0.003
(0.561) (0.930) (0.854)
Householdownsland 33.578 2,184.018 0.010
(0.881) (0.485) (0.364)
Householdownsonegoat ‐64.025 ‐7,456.327 0.006
(0.721) (0.112) (0.610)
Householdownsseveralgoats 383.684 8,874.944 ‐0.005
(0.082)* (0.524) (0.643)
Householdownsonecow 64.631 ‐4,661.881 0.001
(0.693) (0.381) (0.924)
Householdownsseveralcows 95.617 14,374.902 ‐0.015
(0.777) (0.171) (0.302)
Headofhouseholdcompletedprimaryschool 203.383 5,767.114 ‐0.003
(0.217) (0.329) (0.791)
Headofhouseholdcompletedsecondaryschool 22.096 ‐8,550.415 ‐0.004
(0.965) (0.701) (0.836)
Redistributedbetweenstrataforrandomization 333.947 ‐9,091.174 0.013
(0.137) (0.214) (0.729)
Constant ‐1,007.517 ‐45,103.310 0.071
(0.342) (0.298) (0.150)
Observations 296 296 295
AdjustedR‐squared 0.582 0.250 0.136
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations.Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
Table8a:Dailytimespentondomesticlabour
VARIABLES Totaltimeof
headof
household
Timeduring
nightofhead
ofhousehold
Totaltime
ofspouse
Timeduring
nightof
spouse
Treatment ‐1.674 6.150 26.953 2.752
(0.950) (0.542) (0.333) (0.779)
Consumptionofcandles ‐0.431 ‐0.858 3.755 0.320
(0.871) (0.099)* (0.182) (0.705)
Consumptionofkerosene ‐6.370 ‐1.981 ‐4.528 0.123
(0.274) (0.129) (0.339) (0.954)
Numberofhouseholdmembers ‐3.233 ‐0.591 3.293 5.514
(0.660) (0.793) (0.735) (0.248)
Numberofmobilephones 39.520 7.932 32.397 28.709
(0.103) (0.367) (0.306) (0.029)**
Plastereddwelling ‐2.937 8.097 ‐15.112 2.584
(0.952) (0.556) (0.764) (0.890)
Modernwall ‐42.319 ‐10.732 29.055 6.767
(0.353) (0.389) (0.627) (0.768)
Modernfloor ‐0.894 15.761 ‐50.724 ‐20.211
(0.986) (0.211) (0.184) (0.430)
HandcraftedLED ‐9.829 0.960 ‐64.496 ‐19.938
(0.726) (0.904) (0.108) (0.354)
MobileLED ‐104.698 ‐13.120 72.072 99.724
(0.061)* (0.552) (0.331) (0.008)***
Householdownsland ‐57.488 ‐18.418 ‐180.938 ‐67.553
(0.314) (0.378) (0.135) (0.113)
Householdownsonegoat 44.618 11.146 ‐60.298 ‐0.667
(0.359) (0.438) (0.186) (0.965)
Householdownsseveralgoats ‐10.393 ‐0.429 ‐42.010 ‐28.114
(0.820) (0.967) (0.478) (0.170)
Householdownsonecow ‐14.606 ‐13.052 36.974 8.110
(0.748) (0.290) (0.281) (0.565)
Householdownsseveralcows ‐0.574 ‐18.584 ‐38.480 2.483
(0.989) (0.065)* (0.509) (0.897)
Headofhouseholdcompletedprimaryschool ‐10.490 ‐6.117 ‐25.867 2.283
(0.755) (0.488) (0.427) (0.793)
Headofhouseholdcompletedsecondaryschool ‐22.247 ‐20.982 12.455 23.273
(0.748) (0.459) (0.902) (0.601)
Redistributedbetweenstrataforrandomization 99.878 ‐28.758 0.900 55.525
(0.347) (0.139) (0.991) (0.187)
Constant 463.535 86.662 80.674 ‐55.506
(0.054)* (0.060)* (0.603) (0.313)
Observations 287 287 257 257
AdjustedR‐squared 0.000 ‐0.042 ‐0.006 0.092
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
Outcomevariableshavebeentransformedtoadecimalsystem.Forretransformationmultiplywith0.6.
Table8b:Dailytimespentondomesticlabourandincomegeneration
VARIABLES Totaltimeof
headof
household
Timeduring
nightofhead
ofhousehold
Totaltime
ofspouse
Timeduring
nightof
spouse
Treatment 34.898 ‐0.633 16.890 ‐0.000
(0.215) (0.823) (0.354) (0.462)
Consumptionofcandles 0.115 ‐0.030 ‐0.038 ‐0.000
(0.953) (0.828) (0.983) (0.178)
Consumptionofkerosene 17.951 0.640 ‐44.361 0.000
(0.022)** (0.292) (0.107) (0.070)*
Numberofhouseholdmembers 14.813 0.366 12.094 0.000
(0.072)* (0.647) (0.208) (0.063)*
Numberofmobilephones 27.912 ‐2.668 ‐47.178 ‐0.000
(0.374) (0.169) (0.088)* (0.029)**
Plastereddwelling 5.573 15.244 ‐77.895 0.000
(0.935) (0.138) (0.131) (0.708)
Modernwall ‐45.806 ‐9.566 4.021 ‐0.000
(0.471) (0.071)* (0.955) (0.084)*
Modernfloor 24.199 6.033 111.740 0.000
(0.701) (0.377) (0.058)* (0.364)
HandcraftedLED ‐13.573 0.767 ‐90.818 0.000
(0.729) (0.637) (0.073)* (0.035)**
MobileLED 90.041 32.162 41.571 0.000
(0.510) (0.297) (0.597) (0.040)**
Householdownsland 34.368 0.377 ‐113.986 ‐0.000
(0.734) (0.924) (0.001)*** (0.042)**
Householdownsonegoat ‐61.557 ‐3.939 ‐3.032 0.000
(0.229) (0.144) (0.860) (0.694)
Householdownsseveralgoats ‐67.281 ‐4.119 ‐3.641 0.000
(0.354) (0.156) (0.940) (0.508)
Householdownsonecow 9.356 ‐3.255 ‐35.505 ‐0.000
(0.850) (0.116) (0.366) (0.898)
Householdownsseveralcows 37.263 4.393 ‐51.729 0.000
(0.159) (0.361) (0.371) (0.065)*
Headofhouseholdcompletedprimaryschool 30.875 2.090 13.715 ‐0.000
(0.457) (0.156) (0.657) (0.355)
Headofhouseholdcompletedsecondaryschool 62.866 ‐3.830 ‐45.834 0.000
(0.530) (0.642) (0.588) (0.198)
Redistributedbetweenstrataforrandomization ‐58.066 ‐8.810 ‐100.196 0.000
(0.712) (0.321) (0.499) (0.708)
Constant 268.389 ‐21.343 1,324.191 150.000
(0.204) (0.195) (0.000)*** (0.000)***
Observations 218 218 219 219
AdjustedR‐squared ‐0.003 0.051 0.189 1.000
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
Outcomevariableshavebeentransformedtoadecimalsystem.Forretransformationmultiplywith0.6.
Table8c:Dailytimeawake
VARIABLES Totaltimeof
headof
household
Timeduring
nightofhead
ofhousehold
Treatment 9.019 17.002
(0.739) (0.378)
Consumptionofcandles ‐0.913 0.304
(0.685) (0.673)
Consumptionofkerosene 1.820 ‐5.699
(0.689) (0.133)
Numberofhouseholdmembers 13.346 6.049
(0.046)** (0.104)
Numberofmobilephones 32.596 15.604
(0.234) (0.189)
Plastereddwelling 50.196 ‐30.972
(0.293) (0.676)
Modernwall ‐39.148 10.260
(0.222) (0.751)
Modernfloor ‐20.131 22.464
(0.727) (0.508)
HandcraftedLED 23.839 ‐52.063
(0.551) (0.081)*
MobileLED 27.487 7.605
(0.741) (0.841)
Householdownsland 124.099 ‐46.603
(0.325) (0.149)
Householdownsonegoat ‐66.221 ‐3.483
(0.068)* (0.916)
Householdownsseveralgoats ‐52.641 6.490
(0.094)* (0.770)
Householdownsonecow 24.014 ‐20.545
(0.308) (0.268)
Householdownsseveralcows 67.650 13.639
(0.078)* (0.540)
Headofhouseholdcompletedprimaryschool 59.966 4.630
(0.083)* (0.745)
Headofhouseholdcompletedsecondaryschool 101.419 10.572
(0.160) (0.781)
Redistributedbetweenstrataforrandomization 47.874 ‐15.860
(0.504) (0.762)
Constant 1,132.298 1,563.754
(0.000)*** (0.000)***
Observations 287 256
AdjustedR‐squared 0.001 0.040
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
Outcomevariableshavebeentransformedtoadecimalsystem.Forretransformationmultiplywith0.6.
Table9a:Studypattern(onlyhouseholdswithchildrenatschoolage;6‐17years)
Shareofhouseholdswith
childrenstudying
afterschool
Shareofhouseholdswith
childrenstudying
athomeafternightfall
Treatment 0.053 0.142
(0.389) (0.006)***
Consumptionofcandles ‐0.008 ‐0.006
(0.188) (0.088)*
Consumptionofkerosene 0.200 0.109
(0.005)*** (0.028)**
Numberofhouseholdmembers 0.013 0.039
(0.522) (0.025)**
Numberofmobilephones 0.152 ‐0.062
(0.077)* (0.115)
Plastereddwelling ‐0.093 ‐0.029
(0.405) (0.768)
Modernwall 0.071 ‐0.074
(0.344) (0.181)
Modernfloor ‐0.105 0.001
(0.407) (0.992)
Hand‐craftedLED 0.170 0.111
(0.093)* (0.122)
Householdownsland 0.084
(0.379)
Householdownsonegoat 0.160 0.049
(0.164) (0.494)
Householdownsseveralgoats 0.163 0.009
(0.081)* (0.889)
Householdownsonecow 0.179 ‐0.035
(0.099)* (0.579)
Householdownsseveralcows 0.139 0.043
(0.211) (0.659)
Headofhouseholdcompletedprimaryschool ‐0.051 ‐0.086
(0.469) (0.076)*
Head of household completed secondary
school
0.174 0.047
(0.353) (0.747)
Redistributed between strata for
randomization
1.404 0.284
(0.000)*** (0.021)**
PseudoR‐Squared 0.22 0.29
NumberofObservations 209 209
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
77
Table9b:Studypattern(onlyhouseholdswithchildrenatschoolage;6‐17years) (1) (2) (3) (4)
Time children study m611
Time children study m611
Time children study f611
Time children study f611
VARIABLES total night total night
Treatment 21.677 20.717 19.396 17.896
(0.093)* (0.045)** (0.533) (0.090)*
Consumptionofcandles
-0.056 -0.818 -2.145 -2.005
Consumptionofkerosene (0.968) (0.525) (0.354) (0.142)
27.913 36.604 35.199 42.368
#ofhouseholdmembers (0.141) (0.021)** (0.190) (0.000)***
-0.394 -1.548 0.586 -4.349
Numberofmobilephones (0.926) (0.728) (0.952) (0.132)
17.281 -8.491 -36.963 -13.436
Plastereddwelling (0.450) (0.503) (0.218) (0.144)
-14.352 -38.274 -87.895 -35.751
Modernwall (0.639) (0.052)* (0.183) (0.072)*
3.435 26.824 91.559 -6.578
Modernfloor (0.861) (0.159) (0.145) (0.726)
-48.736 -36.603 8.429 -29.381
Hand‐craftedLED (0.397) (0.131) (0.889) (0.071)*
23.893 25.171 47.443 57.773
MobileLED (0.285) (0.293) (0.279) (0.001)***
47.961 78.711 162.397 139.533
Householdownsland (0.275) (0.071)* (0.023)** (0.000)***
49.601 41.979 -2.465 0.857
Householdownsonegoat (0.117) (0.136) (0.941) (0.966)
36.085 6.248 -65.871 -2.252
Hhownsseveralgoats (0.218) (0.828) (0.325) (0.934)
18.801 14.272 -67.680 29.021
Householdownsonecow (0.497) (0.581) (0.397) (0.162)
42.575 32.030 88.579 -13.953
Hhownsseveralcows (0.091)* (0.232) (0.244) (0.430)
39.019 69.939 85.677 22.076
Headofhouseholdcompletedprimary
school
(0.286) (0.012)** (0.376) (0.427)
-10.078 3.784 -19.635 9.669
Head of household completed
secondaryschool
(0.611) (0.748) (0.345) (0.528)
8.449 -13.685 -105.612 -18.220
Redistributed between strata for
randomization
(0.879) (0.798) (0.526) (0.662)
-22.938 -37.743 242.863 58.804
Constant (0.498) (0.129) (0.049)** (0.104)
-55.092 1.256 -4.198 -33.329
(0.231) (0.985) (0.959) (0.346)
Observations 100 101 92 92
AdjustedR‐squared 0.204 0.258 -0.040 0.329
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
Outcomevariableshavebeentransformedtoadecimalsystem.Forretransformationmultiplywith0.6.
78
Table9c:Studypattern(onlyhouseholdswithchildrenatschoolage;6‐17years) (5) (6) (7) (8)
Time children study m1217
Time children study m1217
Time children study f1217
Time children study f1217
VARIABLES total night total night
Treatment 35.387 23.497 17.404 16.837
(0.191) (0.382) (0.327) (0.191)
Consumptionofcandles
-4.250 -4.168 -1.684 -0.854
Consumptionofkerosene (0.256) (0.207) (0.130) (0.486)
23.674 24.780 19.667 34.258
#ofhouseholdmembers (0.349) (0.245) (0.384) (0.166)
7.720 3.868 -0.121 -9.756
Numberofmobilephones (0.453) (0.473) (0.990) (0.224)
53.754 48.771 16.059 30.158
Plastereddwelling (0.096)* (0.098)* (0.540) (0.115)
60.881 16.034 24.574 -23.757
Modernwall (0.128) (0.685) (0.702) (0.715)
-84.749 -45.013 -7.744 27.478
Modernfloor (0.113) (0.411) (0.876) (0.356)
-130.294 -90.490 -40.608 -48.511
Hand‐craftedLED (0.015)** (0.134) (0.482) (0.372)
-12.113 -5.273 29.156 9.080
MobileLED (0.775) (0.894) (0.356) (0.796)
-98.814 -19.806 75.930 134.141
Householdownsland (0.163) (0.761) (0.459) (0.104)
30.819 21.946 -40.240 -54.034
Householdownsonegoat (0.398) (0.572) (0.566) (0.431)
69.840 21.248 29.489 4.769
Hhownsseveralgoats (0.114) (0.575) (0.404) (0.908)
28.914 26.683 7.878 28.657
Householdownsonecow (0.545) (0.522) (0.767) (0.105)
-0.046 -20.610 -15.382 -14.186
Hhownsseveralcows (0.999) (0.555) (0.538) (0.712)
-58.250 -81.834 40.908 13.216
Headofhouseholdcompletedprimary
school
(0.210) (0.041)** (0.474) (0.791)
25.100 0.865 -25.229 -26.903
Head of household completed
secondaryschool
(0.357) (0.971) (0.281) (0.294)
21.427 25.836 -96.871 -92.160
Redistributed between strata for
randomization
(0.740) (0.703) (0.003)*** (0.002)***
246.270 221.981 16.778 32.953
Constant (0.008)*** (0.019)** (0.773) (0.562)
-469.789 -337.756 108.898 177.112
(0.003)*** (0.033)** (0.487) (0.294)
Observations 89 89 94 94
AdjustedR‐squared 0.131 0.080 -0.152 0.017
Note:Robustpvalinparentheses;***p<0.01,**p<0.05,*p<0.1;
Randomizationstratadummiesareincludedinallestimations;Controlvariablesrefertobaselinevalues;
Standarderrorsareclusteredatthevillagelevel.
Outcomevariableshavebeentransformedtoadecimalsystem.Forretransformationmultiplywith0.6.