Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
High Hopes: Experimental Evidence on Saving and the Transition to High School in Kenya
James HabyarimanaWilliam Jack
Working Paper Series | No. 4 | January 2018
Georgetown University Initiative on Innovation, Development and Evaluation Department of Economics and the McCourt School of Public Policy 306 Reiss Building | Washington, DC 20057 https://gui2de.georgetown.edu/
HighHopes:ExperimentalEvidenceonSavingandtheTransitiontoHighSchoolinKenya1
JamesHabyarimana2andWilliamJack3
January,2018
Abstract
Wereportresultsofarandomizedcontroltrialinwhichparentsofprimaryschoolleaverswereencouragedtoopenaconvenientbankaccountoperatedoveramobilemoneyplatform.Alocksavingsaccount(LSA)wasrandomlypromotedtohalfthetreatmentgroup.Treatmentboostedaccounttake-upby25percentagepoints.Intent-to-treatestimatesshowthatbeingofferedeitheraccountincreasedsavingsonthemobilephone.Totalfinancialsavingsincreasedby3-4times,suggestingaccesstothemobilebankaccountcrowdedinotherformsofsavings.Highschoolenrollmentwas5-6percentagepointshigher–representingaonethirdincreaseforcompliers.
1ThankstoEmilyKayser,AlexBerg,SarahBaran,AlexWendo,andMarthaMutuaforprojectimplementationinKenya,andPitchayaIndravudh,AmirJilani,LayaneElHor,andAliHamzaforresearchassistance.2McCourtSchoolofPublicPolicy,GeorgetownUniversity3DepartmentofEconomics,GeorgetownUniversity
2
1.Introduction
Despitetherecentlargeexpansioninsecondaryschoolenrollment,stillmorethanonein
fivechildrenofsecondaryschoolagearenotenrolled(UNESCO2015).Suchlowenrolment
ratescouldbeduetolowperceivedbenefits(Jensen2010),howeverhighcostshavebeen
foundtobebindinginanumberofcontexts(seeGarlick(2013),Blimpo,Gajigoand
Pugatch(2015),MuralidharanandPrakash(2016),Brudevold-Newman(2017)).Asin
otherdomains(Tarozzi,etal.(2014)),inwhichliquidityconstraintslimitdemand,without
accesstofinancethehopeofgoingtohighschoolisjustthat,ahighhope.4Inthispaper,
weexaminetheextenttowhichfinancialinclusioncanhelpparentsandthechildren
achievesuchgoals.
Accesstoversatileandaffordablefinancialservices,includingpayments,saving,credit,and
insurance,iswidelyseenasanimportantcomponentofstrategiestolifthouseholdsoutof
poverty.5Indeed,theWorldBank’sGlobalFindexreport,basedondatacollectedin
interviewswith150,000adultsacross140countries,concludesthat“Financialinclusionis
criticalinreducingpovertyandachievinginclusiveeconomicgrowth.Whenpeoplecan
participateinthefinancialsystem,theyarebetterabletostartandexpandbusinesses,
investintheirchildren’seducation,andabsorbfinancialshocks.”(Demirguc-Kuntetal
2015).Thisstudyaddstoagrowingbodyofexperimentalevidenceinsupportofalink
betweenaccesstofinancialservicesandproductivityenhancinginvestments.6
Despitearecentexpansioninfinancialinclusion,accesstofinancialservicesinthe
developingworldremainslow,andutilizationconditionalonaccessisoftenlimited.Dupas
etal.(2012)documentbothsupplysideanddemandsidereasons,includinglackof
infrastructure,unreliableservice,hightransactionfees,andlowlevelsoftrustinfinancial
institutions.ThemobilemoneyrevolutioninKenyaandotherpartsofthedeveloping
worldhasallowedsomegeographicalbarrierstoaccesstobeovercome.
4Ozier(forthcoming)usesaregressiondiscontinuitydesigntoshowthatattendanceofsecondaryschoolinKenyaincreasescognitiveachievement.5SeeDupasandRobinson(2013a),BruhnandLove(2014),Prina(2015)andBurgessandPande(2005)forrecentworkonthemicroeconomicimpactsoffinancialaccessondownstreamoutcomes.6Seeforexample,Coleetal(2017),Coleetal(2013),GineandYang(2009),Karlanetal(2014)andMobarakandRozenzweig(2013)ontheroleoffinancialaccessandagriculturalproductiondecisions.
3
Inthispaperwestudytheimpactofaccesstoaconvenientsavingstechnologythatuses
Kenya’smostsuccessfulanddominantmobilemoneyplatform,M-PESA,7operatedbythe
country’slargestmobilenetworkoperator,Safaricom.Launchedin2007primarilyasa
mechanismtoexecutedomesticremittances,M-PESAisnowusedbyover90percentof
Kenyanhouseholds,and,consistentwiththeconclusionsoftheFindexreport,hasbeen
showntohavehadimportantimpactsonrisksharing,poverty,andlaborallocation(Jack
andSuri,2014,andSuriandJack,2016),simplyasaresultoftheremittanceandpayments
functionality.
InNovember2012,themobilenetworkoperatorandalocalbankintroducedamobile
bankaccount(MBA)knownas“M-Shwari”,8thatprovidesmoresophisticatedbanking
services,includingsavingsandcredit,overtheM-PESAplatform.Savingsbalancesearn
interestonaslidingscaleofbetween2%and5%APR,whileone-monthloansattracta
7.5%fee.9Acommitmentsavingsdevice,intheformofalocksavingsaccount(LSA),was
addedtotheMBAsuiteofservicesinJune,2014.10BalancesontheLSAearnabonusof1%
additionalinterest,whichisforfeitediffundsarewithdrawnpriortoanagreedtermof
betweenoneandsixmonths,andearlywithdrawalsareonlybeavailableaftera48-hour
waitingperiod.
WeexaminetheimpactofaccesstotheMBA,andtotheLSA,inaspecificcontextinwhich
userscanbeexpectedtohavesimilarsavingsobjectives.Inparticular,wepromotetheuse
ofthefinancialserviceswithparentsofchildrenhalfwaythroughtheirfinalyearof
primaryschool,andwithafocusontheimportanceofsavingforthetransitiontohigh
schoolsixmonthslater.
Whileschoolfeesinpublicprimaryschools(upto8thgrade)wereofficiallyabolishedin
2003andinpublicsecondaryschools(9thto12thgrades)in2008,parentsstillfacecostsof
supplies,uniforms,andtransportassociatedwiththetransitiontosecondaryschool,as
7“M”isformobile,and“pesa”ismoneyorcashinSwahili.Individualaccountsareheldbythemobilephoneoperator,whichinturndepositscustomers’accountbalancesinasmallnumberofaccounts,underthecompany’sname,withcommercialbanks.8“Shwari”means“calm”inSwahili.9Annualinflationin2014inKenyawasabout7percent.10ToopenanLSA,acustomermusthavethemorebasicMBA,throughwhichtheLSAisaccessed.
4
wellasfeesforextratutoringandacademicsupport.Also,attendanceatprivateboarding
anddayschoolsiscommoninKenya,andindeedincreasedsignificantlyafterfeesfor
publicschoolswerereduced(LucasandMbiti(2012);WorldBank,2009).
Werandomizedpromotionofthetwosavingsinterventions(theMBAandtheLSA)to
parents,plusacontrol,attheschoollevel.Parentsinallgroupswereinformedaboutthe
importanceofcontinuededucation,assupportedbytheMinistryofEducation,andof
savingforthetransitiontosecondaryschool.Weuseadministrativedatafromthemobile
networkoperatorandthebank,alongwithsurveydata,tomeasurechangesinsavings
behaviorandschoolingdecisionsacrossexperimentalgroups.
Karlanetal.(2011)andKastandPomeranz(2014)pointtoproblemsofattentionand
focus,motivatingtheideathatreminderscanbeeffectivetoolstoincreasesavingrates.In
lightofsuchfindings,orthogonaltothesavingsaccounttreatments,werandomizedthe
opportunitytoreceiveSMSmessagesthatwouldremindparticipantstosavein
anticipationofthecostsofhighschool.However,wefindnoeffectoftheremindersonany
savingorbehavioralindicators,suggestingthatattentionchallengesdidnotrepresenta
bindingconstraint,oratleastthattheSMSinterventionasdelivereddidnotrelaxit.11
NearlyallparentshadanM-PESAaccount,oraccesstooneintheirhouseholds,atbaseline,
andwhileroughly25%hadanMBA,virtuallynonehadanLSAatthetime.Afterthe
encouragement,take-upoftheMBAwascloserto60%inthetwotreatmentgroups,while
take-upoftheLSAremainedclosetozerointhecontrolandMBAgroups,butwas28%in
thegrouptowhichitwaspromoted.
Anumberofempiricalstudieshavehighlightedtheimportanceofcommitmentdevicesin
boostingsavingsinthecontextofhyperbolicdiscounting.12Howeverinthecurrentstudy,
whileLSAswereopenedandusedbyanon-trivialshareoftheLSAtreatmentgroup,we
findnosystematicdifferencesinimpactbetweenthetwotreatmentgroups.13Thus,like
11TheSMSreminderresultsareavailableonrequest,butwedonotdiscussthemfurtherinthispaper.12See,forexample,Ashraf,KarlanandYin(2006),Bernheim,RayandYeltekin(2011)andBrune,Giné,GoldbergandYang(2012).13ThismighthavebeenatleastpartlyduetoalessthanseamlessLSAinterface.LSAholders,mostofwhomusedfeaturephones,hadtorememberandenteraUSSDshortcode(*234#6),insteadofbeingabletoaccesstheaccountfromtheM-PESAmenu.
5
attentiondeficitchallenges,commitmentproblemsmightnotbeespeciallybindinginthe
currentcontext.Ontheotherhand,accesstotheMBA,withorwithouttheLSA,had
importanteffectscomparedtothecontrolgroup,eventhoughvirtuallyallcontrolgroup
membersusedM-PESA,themobilemoneytransfer(butnotbanking)service.
First,usingtheadministrativedata,wefindthatsavingsheldonmobilephonesbyparents
inthetwotreatmentgroupsshowsmallbutstatisticallysignificantintent-to-treat
increasesoverthesix-monthperiodrelativetothecontrolgroup.However,thisaverage
effectmasksthefactthatthebalancesofabout66%ofindividualsinthetwotreatment
groupschangedbylessthan50KSh(50UScents)overtheperiod,14comparedwithabout
62%ofthoseinthecontrol.Theshareofdissaverswassimilaracrossallgroups(14-15%),
buttheshareofindividualswhosavedmorethan50KShwas5-6percentagepointshigher
inthetreatmentgroups.
Wealsoestimateeffectsonabroadermeasureoffinancialsavings,includingtraditional
bankaccounts,informalsavingsgroups,andcash,whichshowlargerITTimpacts.Once
again,alargeshareofrespondents–about50%-reportzerochangeinfinancialassets.
However,ofthosewithnon-zerochanges,mostreportdeclinesinfinancialholdingsover
theperiod,whichincludedthewidelycelebratedChristmasfestiveseason,duringwhich
certainunusualexpenses(travelandotherfestivity-relatedconsumption)mightbe
incurred.While5-10%ofindividualsineachgroupreportpositivesavings,30-40%report
dis-savingovertheperiod.Butdis-savingbythedis-saversistwiceashighforthoseinthe
controlgroupasforthoseinthetreatmentgroups.Accesstofinancialservicesmightthus
militateagainstcertaintemptations,consistentwiththefindingsofDupasandRobinson
(2013b),Banerjeeetal.(2015)andforasurveyarticle,EvansandPopova(2014).
Localaveragetreatmenteffectestimatesofchangesinfinancialsavingsareeconomically
meaningful,suggestingthatthosewhoopenedmobilesavingsaccountsinresponsetothe
experimentsaved$US40-50morethantheyotherwisewouldhave.Assignmenttoeither
treatmentarmalsoincreasedaccesstoandutilizationofcredit–weobservea50percent
increase,attheextensivemargin,ofutilizationofcredit.Thesekindsofsavingsrates,and14Fiftyshillingsisabout3%oftheaveragelevelofsavingsbythosewhosaved,and5%oftheaveragedissavingbythosewhodissaved.
6
creditresponses,couldconceivablymakeadifferencetoschoolenrollmentdecisions,in
lightofthemediancostoftransitiontohighschoolreportedatbaselineofaround$US380
acrossallgroups.
Indeed,ourendlinesurveyshowslargeimpactsofassignmenttothesavingsarmson
secondaryschoolenrollmentinJanuary2015.Seventy-twopercentofcontrolgroup
respondentsreportedthattheirchildrenenteredhighschoolin2015,immediately
followingtheintervention,whileinthetreatmentgroups,about78percentwereenrolled.
Correspondingly,theTOTestimatessuggestthatamongstthe25percentofthesample
whowouldnototherwisehaveopenedanMBAaccount,doingsoincreasedthelikelihood
ofenrollmentfromabout62%tonearly83%-anincreaseinenrollmentofone-third.The
TOTimpactoftheLSAwassimilar,increasingtheenrollmentrateofchildrenofparents
whowouldotherwisenothaveopenedacommitmentsavingsaccountfromabout65%to
83%.
Toexplorethemechanismsthatcouldunderlietheselargeenrollmenteffects,wefirst
modeltheimpactofactualsavingsonthelikelihoodofenrollinginsecondaryschool,
instrumentingwithtreatmentassignment.Wefindweakevidencethathigherlevelsof
mobilesavingareassociatedwiththetransitiontohighschool,andstrongerevidencethat
increasesinbalancesintheMBAitselfboostenrollment.Puttingmoneyinthebank,not
justonthephone,appearstobeeffective.
Next,weinvestigateimpactsoftreatmentassignmentonanumberofintermediary
outcomes,includingtestscoresintheprimaryschoolexitexamination,thesourceof
financeusedtocoverthecostsofhighschool,andthetypeofhighschoolattended.
However,wefindlittleevidenceoftreatmenteffectsonthesevariables.
Ourresultssuggestthataccesstoaconvenientfinancialservicewitharelativelysmall
depositinterestratecanbeinstrumentalingeneratinghigherlevelsofsavings,andatleast
inthisexperimentalcontext,inpromotingenrollmentinhighschool.Ourresultsare
consistentwitharecentrandomizedstudybyLipscombandSchechter(2017)where
individualsofferedmobilesavingsaccountsaremorelikelytopurchasesubsidized
desludgingservicesinSenegal.Thenextsectiondocumentsourrecruitmentactivitiesand
7
baselinedata.Section3reportsoveralltreatmenteffectsonsavings,whilesection4
documentssavingspatternsandheterogeneousresponses,andsection5reportsimpacts
onschoolingoutcomes.Section6concludes.
2.Recruitmentanddata
InJuneandJulyof2014,halfwaythroughtheKenyanschoolyear,wevisited337primary
schoolsinthreecounties,15andconductedmeetingsattendedbyatotalof4,802parentsof
eighth-gradestudents,ofwhom4,673consentedtotakepartinthestudy.Ofthosewho
consentedtotakepart4,082,or85percent,hadamobilephonewithanM-PESAaccount.
Ofthese,some4,020(98.5%)gaveAdministrativeDataSharingconsent(ADSconsent)–
thatis,theyconsentedtoallowthemobilenetworkoperatorandthebanktosharetheir
administrativeaccountdatawiththeresearchteam.
Randomizingattheschoollevel,wepromotedadoptionanduseofthemobilebankaccount
(MBA)amongstonegroupofparents,andoftheLockSavingsAccount(LSA)amongst
another.Thenumberofparentsandschools(inparentheses)ineachofthethreeprimary
experimentalcellsareshowninTable1,includingincolumn(1)allthosewhoconsentedto
takepart,incolumn(2)thosewhoalsogaveADSconsent,andincolumn(3)thosein
column(2)whowereinterviewedatendline.
Parentmeetingswereheldattheschools,duringwhichmembersofallexperimental
groups,includingthecontrol,werepresentedwiththesameinformationaboutthe
importanceofcontinuededucation,andofsavingforthetransitiontosecondaryschool.
Aspartofthebaselinesurvey,weaskedparentstorecordtheexpectedcostofsending
theirchildtosecondaryschool,andtosetanassociatedsavingsgoalwiththiscostinmind.
ThoseintheMBAtreatmentgroupwereshownhowtoopenbothanM-PESAaccountand
15ThecountieswereKisumu,Nyeri,andKilifi.AnotherRCTwasbeingadministeredindependentlybythesamePIsinthesamesetofprimaryschools.ThatprojectexaminedtheimpactofvariousinterventionsaimedatimprovingthedeliveryofschoolWASH(water,sanitationandhygiene)services,andinvolvedstudentsingradesbeloweighthgrade.Therandomizationsinthetwostudieswereorthogonal.
8
anMBA(iftheydidn’talreadyhavethem),andhowtheycouldbeused.ThoseintheLSA
groupwereinstructedadditionallyonhowtoopenandoperatetheLSA.16
Table2,panelsA,B,andC,reportbalancetestsforbaselinerespondent,child,andschool
level,characteristicsrespectivelyacrosstheexperimentalgroups.Forthoseself-reported
characteristicsinthebaselinesurvey,weusethesampleof4,673whoconsentedtotake
part.Forinformationonvariablesthatcouldonlybediscernedfromtheadministrative
phonedata,weusethesmallersampleof4,020whograntedADSconsent.17Table3
presentsbaselinecharacteristicsbyeachofthetwosavingstreatmentgroupsandthe
control,andpairwiseandthree-waybalancetests.
Sixty-eightpercentoftheparentsatthemeetingswerewomen,andtheaverageageof
respondentswas44.Sixty-twopercenthadcompletedprimaryschool,butonly19percent
hadfinishedsecondaryschool.Ninety-threepercentownedamobilephone,only2percent
didn’thaveaccesstoone,andnearlyallofthe90percentofthesamplewithaSafaricom
SIMcardalsohadanM-PESAaccount.Therewerenearly5childrenperhousehold,halfof
whomweregirls.Everyonereportedexpectingtheirchildrentogoontosecondaryschool,
andtheaverageexpectedcostofdoingsowasaboutUSD400,themedianvalueofwhich
wasUSD380.Schoollevelattributesarealsoreported,andarebalanced.
Werevisitedtheprimaryschoolsandsuccessfullyconductedface-to-faceendlinesurveys
of3,994ofthe4,673originalsampleofrecruitedparents.Wefollowedthisupwithphone
surveysof854oftheremainingparticipants,foratotalre-contactrateof85.5percent.The
re-contactrateamongstthosewhogaveADSconsentwas87.5percent.
16UponopeningaLockSavingsAccount,acustomerchoosesthematurityoftheaccount,anintegernumberofmonths,afterwhichthefundscanbeaccessedwithoutpenalty.However,inourstudy,weworkedwiththebanktoallowaparticulardatetobespecified,andactivelyencouragedrespondentsintheLSAtreatmentgrouptochooseamaturitydateofJanuary5th,2015,fortworeasons.First,studentstypicallybeginsecondaryschoolneartheendofJanuaryeachyear,sohavingthefundsavailableafewweeksaheadofthestartofclassescouldhavebeenvaluableinfinancingthetransition.Andsecond,amaturitydateafterChristmas–apopularholidayandgift-givingtimeinmuchofKenya(althoughlesssoinKilifi,whichhasasizeableMuslimpopulation)–wouldensurethatfundswouldbeavailableforschoolingexpenses,unlesstheyhadbeenwithdrawnwithpenalty.Ofthe399respondentsintheLSAarmwhoopenedLSAs,306choseamaturitydateofJanuary5th,2015.
17Similarly,inourregressionsinvolvingbothbaselineandendlinedata,weusethesampleof4,673individualswhoconsentedtotakethebaselinesurvey,withattritedvaluestreatedasmissing.Inanyregressionsinwhichweuseadministrativedata(withorwithoutbaselineorendlinesurveydata),weuseinformationonthesmallersampleof4,020,againtreatingattritedindividualsasmissing.
9
Table3reportsattritionbyexperimentalarmamongstboththefullbaselinesample,and
forthosewhogaveADSconsent.Inbothcases,attritionrateswere3-5percentagepoints
higherinthetreatmentgroupsthaninthecontrol.InouranalysisbelowwereportManski
boundsontreatmenteffectstoaccountforthedifferentialratesofre-contact.
Take-upratesofmobilefinancialservicesacrossthethreesavingsarmsarereportedin
Table4usingadministrativedatafromthe4,020respondentswhogaveADSconsent.We
reportratesatwhichMBAsandLSAswereopenedineachofthethreeexperimental
groupsasofJanuary5th,2015,sixmonthsaftertheintervention.About34%ofthecontrol
grouphadanMBAbythattime,whilethesharesofthetwotreatmentgroupswithan
accountwere24-25percentagepointshigher.Virtuallyno-oneinthecontrolorMBA
treatmentgrouphadanLSAbyearly2015,while27%oftheLSAtreatmentgroupdid.
Thus,ourencouragementdesignhadnon-negligibleimpactsonadoptionofthetwobank
products,asintended.
Belowwereportimpactsoftheexperimentalinterventionsonanumberofoutcomes.In
thissection,wediscusstheseoutcomesandtheirmeasurementinmoredetail,
distinguishingbetweensavingsoutcomesontheonehand,andthoserelatedtoschooling
decisionsandbehaviorontheotherhand.
Wemeasure“mobilesavings”usingadministrativedataasthenetchangeintotalbalances
heldonindividuals’M-PESAaccounts,regularMBAaccounts,andLSAaccounts.Alongwith
theLSA,theMBAincludedbothatransactionalsavingsaccountaswellasaloanaccount.
Wethusreportbothgrosssavings(thetotalchangeinbalancesonallmobilesavings
accounts)andnetsavings(grosssavingsminusnetincreaseinoutstandingdebtinthe
MBAloanaccounts).18
Intheendlinesurveysweaskedindividualstoestimatethechangeinvalueofanyother
financialassetstheyheld,includingmobilemoneyaccountsheldwithothermobile
operators,otherbankaccounts,depositswithSACCOs,ROSCAs,etc.,andcashheldathome
18EligibilityforloansontheMBAisbasedonacreditscoregeneratedfromdataonusageofmobilephone,mobilemoney,andMBAactivitylevels.Loanproceedsaretransferredtothecustomer’sM-PESAaccount,andtheliabilityrecordedinherloanaccountwiththebank.
10
orwithfriends.Addingthesesavingstomobilesavingsgivesusavalueofwhatwereferto
as“financialsavings”.19
Ourprimaryschoolingoutcomeofinterestisenrollmentinhighschool,whichweelicitin
theendlinesurvey.Weaskparentsiftheirchildhadbeenenrolledinhighschoolbythe
timeofthesurvey(whichtookplaceatleastonemonthafterthebeginningoftheschool
year),whatkindofschool(private/public,co-ed/singlesex,andwhetheritisanational,
county,district,orlocalschool),thecostofattendance,andotherattributes.Wealso
collectinformationonresultsoftheend-of-primaryschoolexaminations,whichdetermine
eligibilityforentranceintothedifferenttypesofsecondaryschool,alongwithparents’
expectationsabouttheirchild’sperformance,andparentaldecision-makingroles.
Theparentsofchildrenwhoenrolledinhighschoolwereaskedhowtheyfinancedthe
costsofthetransition,withoptionsincludingdrawingonfinancialsavings,thesaleof
assets,andborrowingfromfriendsandrelativesorfrommoreformalsources.
3.Impactsonsavings
Inthissection,wepresentourmainresultsonparentalsavingsoverthecourseofthe
secondhalfoftheirchild’sfinalyearofprimaryschool.Wheninvestigatingtheimpactof
havingaccesstotheMBA,weadopttwoapproaches:firstweuseassignmenttoeitherthe
MBAorLSAtreatmentgroup,comparedwithassignmenttothecontrolgroup;andsecond,
wecomparethoseassignedjusttotheMBAtreatmentgrouponlywiththecontrol.When
estimatingtheimpactofthelocksavingsaccount,wecomparethoseassignedtotheLSA
treatmentwiththecontrolgroupontheonehand,andwiththeMBAtreatmentgroupon
theother.Thelastcomparisonallowsustodeterminewhatmarginaleffect,ifany,access
tothecommitmentdevicehadonsavingbehavior.
19Wealsoaskedrespondentstoreporttheirownership,purchase,andsalesofrealassetsoverthesix-monthperiodbetweenrecruitmentandthebeginningoftheschoolyear,includingitemssuchaslivestock,householddurables,andotheritems.Thismeasurehasveryhighvarianceandweexpectsuffersfromconsiderablemeasurementerror,dueinparttothedifficultyrespondentslikelyhadinestimatingthevalueofassetsheldatbaseline.
11
Threedimensionsofsavingaredistinguishedinthetables:grossvsnet(toaccountfor
loanstakenoutontheMBAplatform);mobilevsfinancial;andshort-term(thatis,until
January5th,2014,justbeforethebeginningofthenewschoolyear)and“long-term”(until
January31st,bywhichtimehighschooltransitionalexpensesareexpectedtohavebeen
incurred).20Withinthecategoryofmobilesavings,wealsoestimateimpactsonsavings
balancesheldatthebank,thatisintheMBAorLSA,butnotontheM-PESAaccount,asit
reflectsamoveintointerestbearingdepositswithinthemoreformalbankingsector.
(a) Impactsonmobilesavings
Table5presentsestimatesoftreatmenteffectsonshort-termsavingsheldonmobile
accounts,ITTestimatesintheupperpanel,andtheTOTestimatesinstrumentingwith
treatmentassignmentinthelowerpartofthetable.Thefirstfourcolumnsmeasure
impactsongrosssavings,andthemiddlefouronnetsavings.Thefinalfourcolumnsshow
changesingrossbalancesheldatthebank,aggregatedacrossboththeMBAandLSA
accounts.
InColumn(1),theaveragetreatmenteffectsofaccesstotheMBAacrossbothgroupsare
reported.TheITTandTOTestimatesarebothpositive,butimprecise.InColumn(2),
thoseassignedtotheLSAarmareexcluded,andtheimpactofassignmenttotheMBA
treatmentgroupcomparedwiththecontrolisestimated.Beingencouragedtoopen,and
actuallyopening,anMBAhaveeffectsonmobilesavingsthataresignificantatthe10
percentlevel.Thosewhocomplywiththetreatmentaccumulate1,093KSh(aboutUSD11)
morethattheywouldhaveotherwise.Forcomparison,thecorrespondinglevelofsavings
ofcomparablecompliersinthecontrolgroup,reportedinthelowerpartofthetable,is253
KSh.Treatmentgroupcompliersthussaveaboutfourtimesasmuchontheirphonesas
similarindividualsinthecontrolgroup.Theinterest-bearingbankaccountitselfseemsto
20Ourdistinctionbetweenshort-andlong-termfinancialsavingsisimperfect.Wecanusetheadministrativedatatocalculatemobilesavingsuptoanyenddate;butthenon-mobilecomponentoffinancialsavingsisreportedasofthetimeoftheendlinesurvey,onetotwomonthslater,anddoesnotdifferentiatebetweenresourcesaccumulateduptoJanuary5thandJanuary31st.Thusshort-termtotalfinancialsavingsareinfactamixofchangesinshort-termmobileandlongertermnon-mobilebalances.
12
beboostsavings,eventhoughtheinterestrateislow,andsavingonM-PESAisjustas
convenient.21
ExcludingtheMBAgroup,impactsofassignmenttotheLSAgroupvis-à-visthecontrol
(Column(3))aresmallerandinsignificant.ThefourthcolumncomparestheMBAandLSA
groupsdirectly,excludingthecontrol.AlthoughthepointestimatesonLSAassignment
andtreatmentarebothnegative,theyareinsignificant,anditisnotpossibletodistinguish
betweentheeffectsoftheLSAandtheMBA.
NetmobilesavingsinColumns(5)to(8)exhibitverysimilareffectsizeswiththesame
statisticalproperties,althoughtheTOTestimateusingbothtreatmentgroups,ofthe
impactofopeninganaccount,shownincolumn(5),cannowbedistinguishedfromzeroat
the10percentlevel.
Finally,Columns(9)to(12)showaclearshiftofsavingsintotheformalbankaccount.
ComparingthepointestimatesinColumns(1)to(3),whichareabouthalfthesizeofthose
inColumns(9)-(11),wecaninferthatroughlyhalftheincreaseinbankbalancesisdueto
“new”savings,whilehalfisduetoashiftfromthemobilemoneyaccounttothebank
account.
InAppendixTable1wereportLeeBoundsontheestimatesinTable5,accountingfor
attritionfromtheendlinesurvey.Notsurprisingly,therangeofcoefficientsoftenincludes
zero,butingeneralthemagnitudessuggestattritionisunlikelytohavebiasedourresults
toaquantitativelylargedegree.
Table6reportsthesamesetofregressionsfor“long-term”mobile-phonesavings,upto
January31.AllcoefficientsinColumns(1)through(8)aremuchsmallerandhighly
insignificant.ItthusappearsthattotheextentexposuretotheMBAand/orLSA
treatmentsincreasedmobilesavings,theseincrementsweredepletedinthe2to3weeks
beforethebeginningofthenewschoolyear.Evenso,columns(9)through(12)suggest
thatamongstthosebalancesthatweremaintainedonmobileaccounts,ashiftfromthe
21Thiscouldbedueatleastinparttothedesiretobuildupahistoryofbanksaving,soastobecomeeligibleforalargerloan,althoughthecreditscoretakesintoconsiderationM-PESAbalancesaswellasthoseontheMBA.
13
mobilemoneyplatform,M-PESA,totheformalbankaccountscanbeobserved.TheTOT
estimatessuggestthatbeinginducedtoopenanMBAincreasedbankaccountsavingby
between1,000and1,500KSh(USD10-15),andbeinginducedtoopenanLSAincreased
thembyabout500KSh(USD5).
Beingsubjecttothemobilebankaccountencouragementthrougheitherarmalso
increasedthelikelihoodthatparentswouldtakeoutaloanontheplatform.Thisis
importantasitprovidesanadditionalsourceoffinancingforeducationexpenditures,
althoughunderstandingwhetheraccesstocreditcrowdsoutsavingisofcoursealsoof
interest.Table7showsthatbetweenrecruitmentandJanuary5th,2015,parentsineachof
thetreatmentgroupswerebetween3and5percentagepointsmorelikelytodrawonan
MBAloan,comparedwithan8percentrateinthecontrolgroup(seeColumns(1)-(4)).
TheTOTestimatessuggestthatbetween12and18percentofuserswhoopenedan
accountasaresultoftheencouragementtookadvantageofthecreditoption.Similarly,the
averagenumberofloanswashigherinthetreatmentgroups(Columns(5)-(8)),andthe
amountsborrowedwerehigher,butestimatedwithlessprecision(Columns(9)-(12)).
(b) Financialsavings
Itisimportanttoassesstheimpactofaccesstobankaccountsoverthemobileplatformon
abroadermeasureofsavings,soastodistinguishbetweensimpleshiftsinassetholdings
andincreasesinaccumulatedresources.Tothisend,wecombineourdataonmobile
savingswithinformationprovidedintheendlinesurveyonotherfinancialsavings,suchas
holdingsinotherbankaccounts,“underthemattress”savings,SACCO(savingsandcredit
cooperative)andChama(micro-savingsgroups)accountbalances,advancepurchases,and
savingswithfamilyorotherentities.
Asmentionedabove,oneshort-comingofourdataisthatthesenon-mobilefinancial
savingsamountswereelicitedatthetimeoftheendlinesurvey,whichtookplacein
February/March,afterthebeginningoftheschoolyear.Weknowalreadyfromthemobile
savingsthatsignificantactivityoccurredduringthemonthofJanuary,asbalanceson
mobileaccountsweredepleted.Thesamecouldofcoursebetrueforotherfinancialassets.
14
Wethusconstructtwooutcomemeasures,inadditiontothenon-mobilefinancialsavings
variable.Inthefirst,weaddnon-mobilesavingsreportedintheendlinesurveytoshort-
termmobilesavingsaccruedthroughJanuary5th,drawnfromadministrativedata;andin
thesecond,weaddreportednon-mobilesavingstolong-termmobilesavings,accrued
throughJanuary31st.
IftheobservedmovementsoutofmobilesavingsinJanuarywerematchedone-for-oneby
increasesinnon-mobilefinancialbalances,thenthecombinationofthelatterwithmobile
savingsthroughJanuary31stwouldprovideareliableindicatorofbothshort-andlong-
termfinancialsavings.Ontheotherhand,itseemsunlikelythatfamilieswouldnecessarily
engageinthiskindofrebalancingbehaviorspecificallyinthemonthofJanuary.
Ifinsteadtheevolutionofusers’non-mobilesavingsbalancesfollowedasimilarpatternto
thatoftheirmobilesavingsaccounts,thenthesumofnon-mobilesavingsandmobile
savingsonJanuary31stwouldprovideareasonableestimateoflongtermsavings,while
thesumofnon-mobilebalancesandmobilesavingsonJanuary5thwouldunder-estimate
short-termsavings.Wemightfurtherspeculatethat,undertheassumptionthatthetwo
kindsofsavingsmoveintandem,thattotalshort-termfinancialsavingscouldbecalculated
asthesumofmobilesavingsasofJanuary5thandnon-financialsavingsmultipliedbythe
ratioofshort-tolong-termmobilesavings.
Table8reportsITTandTOTestimatesoftheimpactoftreatmentonnon-mobilesavings
(Columns(1)through(4))andfinancialsavingsasmeasuredbythesumofnon-mobile
savingsandmobilesavingsthroughJanuary5th(Columns(5)through(8))andmobile
savingsthroughJanuary31st(Columns(9)through(12)).Imprecisionhampersourability
todiscernstatisticallysignificanteffects,althoughthepointestimatesinallcasesare
positiveandeconomicallymeaningful.
FocusingfirstonColumns(5)through(8),whichreportwhatarelikelyunder-estimatesof
totalshort-termfinancialsavingsgrossofloanproceeds,wedocumentmeaningful
responsestoassignmenttobothtreatmentgroups,withITTestimatesofbetween940and
over1,400KSh.ThecorrespondingTOTestimatesareintherangeof3,600tonearly4,800
KSh(aboutUSD50).Comparedwiththe(negative)estimatedsavingsofcompliersinthe
15
referencegroup,theseestimatessuggestthatforthosewhoopenanaccountinresponseto
eithertheMBAorLSAencouragement,theincentivetodissavethroughJanuary5this
effectivelyeliminated.
FromTables5and6,theratioofthepointestimatesofshort-tolong-termmobilesavings
impactsofthetwotreatmentgroupscombined,forexample,isabout15.Ifnon-mobile
financialsavingsweretofollowasimilartemporalpatterntomobilesavings,theshort-
termimpactoftreatmentonthetreatedwouldbeseveralhundreddollars–enoughto
covermostifnotallofthecostoftransitionofhighschool.
4.Savingspatternsandheterogeneousresponses
Tobetterunderstandthepatternofsavingsacrossindividuals,Figures1and2illustrate
thepatternsofshort-termgrossmobilesavingsbehaviorandshort-termfinancialsavings
behavior(asdefinedabove)ofmembersofthecontrolgroupandthecombinedMBAand
LSAtreatmentgroups.22About15percentofindividualsineachgroupexhibitmobile
dissaving,whilefully40percentshowreductionsinfinancialsavings.Ontheotherhand,
about20percentshowpositivemobilesavings,andasimilarshareshowpositivefinancial
savings.
Thetreatmenteffectsarereflectedindifferencesbetweenthecumulativesavings
behaviors,althoughpoint-wiseinterpretationdependsonastrongmonotonicity
assumption.Nonetheless,undersuchanassumption,wewouldinferthatindividualsinthe
combinedtreatmentgroupswhowouldotherwisehavesavedapositiveamountontheir
phones,savemore;andthosewhootherwisewouldhavereducedtheirmobilebalances,
reducethemless.
Whenitcomestothebroaderconceptoffinancialsavingshowever,alltheactionappears
tobeamongstthedis-savers.Againassumingmonotonicity,exposuretotheMBAand/or
22Inthecontrolgroup,thegraphisconstructedbyfirstorderingindividualsbythenetchangeinbalancesintherelevantaccounts.Foralargeshare,thischangeiszero.Forthosewithnegativechanges,wecalculatethecumulativereductionmovingtotheleft,andforthosewithpositivechanges,thecumulativeincreaseiscalculatedmovingtotheright.Theresultingvaluesarenormalizedbythenumberofindividualsinthecontrolgroup.AsimilarexerciseisperformedforthoseinthecombinedCBAandLSAtreatmentgroups.
16
LSAappearstoinducenochangeinpositivefinancialsavingbehavior,butmeaningful
reductionsindissavingfromfinancialassets.
TheseobservationsarequantifiedinthequantileregressionsreportedinTable9.Inthe
firstthreecolumnsweobservepositiveimpactsofassignmentatthe10thand90th
percentilesofthedistributions,withstatisticalsignificanceatthetopend.Theeffectatthe
medianiszero.Forfinancialsavings(columns4-6),theestimatesatthemedianandthe
90thpercentileareallsmall,buttheimpactsatthebottomendarelarge,andinthecaseof
LSAtreatmentarm,highlysignificant.
Table10reportsmarginalresultsoforderedlogitregressionsexploringtheimpactof
treatmentassignmentonthelikelihoodofanindividualsavinganegative,zero,orpositive
amountoverthesix-monthperiod.Usingthepredictedprobabilityestimatesforeach
outcome,wefindthatassignmenttotheMBApromotionreducesthelikelihoodof
dissavingbyjustunder4percentagepointsandincreasesthelikelihoodofpositivesavings
by4percentagepoints.AssignmenttotheLSApromotionreducesthelikelihoodof
dissavingbyjustover5percentagepointsandincreasesthelikelihoodofpositivesavings
bynearly6percentagepoints.
Tables11through13reportestimatesofheterogeneoustreatmenteffectsalonganumber
ofdimensions.First,weaskiftheidentityofthesurveyrespondent,inparticularwhether
s/heistheprimaryfinancialdecision-makerinthehousehold,isassociatedwithany
differenceinimpact.Whenconsideringgrossornetmobilesavings(Columns(1)and(2)),
theinteractioneffectbetweendecision-makerstatusandtreatmentassignmentisvery
smallandinsignificant.Theeffectsonfinancialsavings(Column(4))aresimilarly
imprecise.However,thereappearstobeasignificantlylargerimpactoftreatmenton
savingsinthebankaccountitselfwhentherespondentisthedecision-maker.Thiscould
reflecttheimportanceofin-personexplanationaboutthefeaturesoftheMBAandLSA.
Table12askswhethertheimpactofaccesstoformalbankingservicesdiffersfor
householdsinwhichwomenaretheprimarydecisionmakersregardingfinancialmatters.
WereportITTestimatesoftheimpactofassignmenttoeithertreatmentarmonbothgross
andnetmobilesavingsusingadministrativedata.WhileaccesstoM-PESAitselfhasbeen
17
showntobeespeciallybeneficialforpoorruralwomeninKenya(seeSuriandJack,2016),
inthecurrentexperimentalsettingwefindreasonablystrongevidencetoindicatethatthe
savingsofhouseholdsinwhichthemalemadefinancialdecisionsweremoreresponsiveto
improvedaccesstodigitalfinancialservicesthanwerethoseofhouseholdsinwhich
womenheldtheelectronicpursestrings.
Finally,Table13asksifhouseholdsinwhichtheparentsassessedtheirchildrentohave
above-averageschoolperformancerespondeddifferentlytoothers,perhapsbecausethe
prospectofsecondaryschoolenrollmentwasmoresalient,andthereturnstofurther
schoolinglarger.Infacthowever,thereislittleevidenceofheterogeneityonthis
dimension.
5.Impactsonschooling
Intheendlinesurvey,whichtookplaceafterthebeginningofthefollowingschoolyear,we
collectedinformationfromrespondentsregardingtheirchildren’senrollmentinsecondary
school.Columns(1)through(3)ofTable14reportITTandTOTestimatesoftheimpactof
thetwotreatmentscombined,andseparately.Seventy-twopercentofparentsinthe
controlgroupreportedthattheirchildrenwereenrolledinsecondaryschool.PanelA
reportsITTimpactsof5-6percentagepoints,whichtranslateintoTOTeffectsofbetween
18and24points.Comparedwithcompliersinthecontrolgroup,about60percentof
whomweestimateenrolledinsecondaryschool,theseTOTestimatessuggestthatopening
abankaccountisassociatedwitha27to40percentboosttoenrollment,whichreached83
percentforbothgroups.
Toinvestigatetheselargeimpactsonthetransitiontohighschool,weaskifthe
inducementtosavemorecausallyincreasestheprobabilityofenrolling.Inparticular,in
Table15weregressenrollmentoutcomesonactualmobilesavings,instrumentingwith
assignmenttotreatment.Columns(1)through(4)followthesamepatternasinprevious
tables,thefirstthreecomparingoutcomesineitherbothoroneofthetreatmentgroups
withthecontrol,andthefourthcomparingoutcomesintheLSAgroupwiththoseinthe
18
MBAtreatment.SavingsaremeasuredinthousandsofKenyanshillings(tensofUS
dollars).
ThecoefficientsongrossornetmobilesavingsinColumns(1)-(3)ofPanelsAandBare
uniformlypositive,althoughmostlyimpreciselyestimated,butpointtowardsa20-30
percentagepointincreaseinthelikelihoodofenrollinginsecondaryschoolforevery
thousandshillingincreaseinbalances.Thereisagainnoapparentdifferencebetweenthe
MBAandLSAeffects(Column(4)).PanelCshowsamorerobustrelationshipbetween
increasesinbankaccountbalancesandschoolenrollment,suggestingoncemorea
substantiveimpactoftheuseofformalbankingservicesonparentalbehavior.
InPanelD,theimpactofhighertotalfinancialsavingsonenrollmentisassessed,usingthe
definitionofendlinenon-mobilefinancialsavingsplusshort-termmobilesavingsasof
January5th.Theseregressionssufferfromweakinstrumentation,andyieldsmallerbutstill
positive,ifimpreciseestimatesoftheimpactofadditionalsavingonhighschool
enrollment.
Whatmechanismliesbehindtheselargeeffectsonschoolenrollment?Weinvestigatea
numberoffinancialindicatorstoprobethisquestion.InTable16wereportITTandTOT
estimatesoftreatmentonthecostofhighschoolreportedatendline(thecostfornon-
enrolleesistreatedaszero).IntheITT,thoseintheLSAgroupappeartohavechosen
schoolsthatwereonaverageabout10percentlesscostlythanthoseinthecontroland
MBAgroups,significantatthe10percentlevel(Columns(3)and(4)).TheTOTestimates
oncompliersaresomefourtimesaslarge,andare25percentofthecostofschools
attendedbycomparisongroupcompliers–thatis,similarindividualsinthecontroland
MBAgroups.TheimpactofassignmenttotheMBAgroupismuchsmallerandstatistically
insignificant.However,thefactthatourenrollmenteffectswerelargerforassignmentto
theMBAgroupsuggestscautioninattributingthoseenrollmenteffectstoselective
targetingoflowercostschools.
Wenextaskedrespondentshowtheyfinancedthecostsofhighschool,asreportedinTable
17.Ofcourse,responsesareonlyrecordedforrespondentswithenrolledchildren,sothe
reportedmeansinTable17reflectbothselectionandtreatmenteffects.Onlyoneofthe12
19
coefficientsissignificant(andthenatthe10percentlevel),andallareeconomically
irrelevant,sotreatmentassignmentappearsnottohaveaffectedhowparentsfinance
school.
Finally,weaskiftreatmentassignment,ortake-upofMBAandLSAaccounts,couldbe
associatedwithstudentsfocusingmoreonafuturethatincludedattendanceathighschool,
higheraspirations,andgreatereffortintheirfinalexamsatthecompletionofprimary
school.23Self-reportedexamscores,howevershownosignificantrelationshiptotreatment
assignment(Table18).
6.Conclusions
Financialinclusionthroughmobiletechnologycouldholdpromiseforexpandingthe
opportunitiesofthepoortosaveandinvest.Inthispaper,wehavefoundthatatleastfor
parentsoffinalyearprimaryschoolchildreninKenya,promotingmobilephone-based
savingsaccountsappearstohaveledtoincreasesinsavingsheldonthemobilephone,and
tolargerincreasesinfinancialsavingsaggregatedacrossinstruments.
Incontrasttosomeearlierstudies,wefindlittleevidencethatincreasesinsavingaredue
toattentionproblemsontheonehand,asourSMSreminderinterventionhadlittleeffect,
ortocommitmentissuesontheother,sincethemarginalimpactofalockedsavings
accountoverandaboveasimplebankaccountwasgenerallyeconomicallyandstatistically
insignificant.Instead,the(admittedlylow)interest-bearingfeatureofthebankaccount,
andtheaccesstoshort-termcreditthatitprovided,bothdeliveredoveraneasilyaccessible
platform,werelikelyenoughtonudgepeopletowardssavingmore,oratanyratedissaving
less.
Amongsttheroughlyone-quarterofoursamplewhoopenedaccountsinresponsetothe
encouragementtreatment,mobilesavingsoverasix-monthperiodincreasedbyabout
USD10.Undersomereasonableimputationassumptionsregardingnon-mobilefinancial
assets,ourresultssuggestthattotalfinancialsavingsbythosewhorespondedtothe
23AllschoolstudentsinKenyatakeanationalexaminationattheendofprimaryschool,resultsonwhichdeterminethekindsofsecondaryschoolstheywillbeeligibletoattend.
20
encouragementcouldhaveincreasedenoughtocovermostofthecostsoftransitioning
fromprimaryschooltohighschool.
Indeed,ourmoststrikingresultisthattreatmentassignmentisassociatedwithhigher
ratesofhighschoolenrollment,ontheorderof5-6percentagepointsonanITTbasis,and
18-24pointsfortheTOT.Theprecisemechanismsbywhichtheseimpactsaremediated
arelesseasytopindown.Accountingforendogeneity,highersavings,especiallyinthe
mobilebankaccount,appeartoincreaseenrollmentinhighschool,but,notwithstanding
issuesoffungibility,parentsdon’tnecessarilyreportusingthesavingstopaytuitionand
othercosts.Thereislimitedevidencethattreatmentgrouphouseholdschoseless
expensiveschoolsfortheirchildren,andnoevidencethatfinancialinclusionaffects
aspirationsinawaythatmightinducestudentstoachievehigherend-of-primaryschool
testscores.
Nonetheless,accesstoamobilebankaccount,bymovingtheneedleofsecondaryschool
enrollment,appearstobeaninnovationthatcouldbrightentheprospectsoflargenumbers
ofchildren,givingthemhighhopesforthefuture.
21
References
Ashraf,Nava,DeanKarlanandWesleyYin(2006):“TyingOdysseustothemast:evidence
fromacommitmentsavingsproductinthePhilippines,”QuarterlyJournalofEconomics,pp.
635-672.
Banerjee,Abhijit,EstherDuflo,RachelGlennersterandCynthiaKinnan(2015):“The
miracleofmicrofinance?Evidencefromarandomizedevaluation,”AmericanEconomic
Journal:AppliedEconomics,7(1):22-53.
Blimpo,M.P.,O.Gajigo,andT.Pugatch(2015):“FinancialConstraintsandGirls’Sec-
ondaryEducation:EvidencefromSchoolFeeEliminationintheGambia,”IZADiscussion
Paper,(No.9129).
Berheim,B.Douglas,DebrajRayandSevinYeltekin(2015):“Povertyandself-control,”
Econometrica,83(5),pp.1877-1911.
Brudevold-Newman,Andrew(2017):“TheImpactsofFreeSecondaryEducation:Evidence
fromKenya.”WorkingPaper.
Bruhn,MiriamandInessaLove(2014):“TheRealImpactofImprovedAccesstoFi-nance:
EvidencefromMexico.”JournalofFinance59(3):2807–2833.
Brune,Lasse,XavierGiné,JessicaGoldbergandDeanYang(2016):“FacilitatingSavingsfor
Agriculture:FieldExperimentalEvidencefromMalawi,”EconomicDevelopmentand
CulturalChange,UniversityofChicagoPress,vol.64(2),pp.187-220.
Burgess,RobinandRohiniPande(2005):“DoRuralBanksMatter?Evidencefromthe
IndianSocialBankingExperiment.”AmericanEconomicReview95(3),780–795.
ColeS.,Giné,X.Tobacman,J.Topalova,P.Townsend,R.andVickery.J(2013):“Barriersto
householdriskmanagement:EvidencefromIndia.”AmericanEconomicJournal:Applied
Economics5:104–35.
ShawnCole,XavierGiné,JamesVickery(2017):“HowDoesRiskManagementInfluence
ProductionDecisions?EvidencefromaFieldExperiment.”TheReviewofFinancialStudies,
Volume30,Issue6,Pages1935–1970
22
DeanKarlan,MargaretMcConnell,SendhilMullainathan,JonathanZinman(2016):“Getting
totheTopofMind:HowRemindersIncreaseSaving,”ManagementScience62(12),pp.
3393-3411.https://doi.org/10.1287/mnsc.2015.2296
Demirguç–Kunt,Asli,LeoraKlapper,andDorotheSinger(2015).“TheGlobalFindex
Database2014:MeasuringFinancialInclusionaroundtheWorld.”WorldBankPol-icy
ResearchWorkingPaper#7255.
Dupas,Pascaline,SarahGreen,AnthonyKeats,andJonathanRobinson(2012):“Challenges
tobankingtheruralpoor:EvidencefromKenya’sWesternProvince,”
http://www.nber.org/chapters/c13363.pdf
Dupas,PascalineandJonathanRobinson(2013a):“Savingsconstraintsand
microenterprisedevelopment:EvidencefromafieldexperimentinKenya,”American
EconomicJournal:AppliedEconomics,5(1),pp.163-192.
Dupas,PascalineandJonathanRobinson(2013b):“WhyDon’tthePoorSaveMore?
EvidencefromHealthSavingsExperiments.”AmericanEconomicReview103(4):1138–
1171.
Evans,DavidK.;Popova,Anna(2014):“Cashtransfersandtemptationgoods:areviewof
globalevidence,”PolicyResearchworkingpaper;no.WPS6886;ImpactEvaluationseries;
no.IE127.Washington,DC:WorldBankGroup.
http://documents.worldbank.org/curated/en/617631468001808739/Cash-transfers-and-temptation-
goods-a-review-of-global-evidence
Garlick,R.(2013):“HowPriceSensitiveisPrimaryandSecondarySchoolEnrollment?
EvidencefromNationwideTuitionFeeReformsinSouthAfrica,”WorkingPaper.
Giné,Xavier,DeanKarlanandJonathanZinman(2010):“Putyourmoneywhereyoubuttis:
Acommitmentcontractforsmokingcessation,”AmericanEconomicJournal:Applied
Economics,2(4),pp.213-235.
Giné,X.,andYang.D.(2009):“Insurance,credit,andtechnologyadoption:Field
experimentalevidencefromMalawi.”JournalofDevelopmentEconomics89:1–11.
23
Jensen,R.(2010):“The(Perceived)ReturnstoEducationandtheDemandforSchooling.”
QuarterlyJournalofEconomics,125(2),515–48.
Karlan,Dean,AishwaryaLakshmiRatan,andJonathanZinman(2014):“Savingsbyandfor
thePoor:AResearchReviewandAgenda.”ReviewofIncomeandWealth60(1):36–78.
Karlan,D.,Osei,R.Osei-Akoto,I.andUdryC.(2014):“Agriculturaldecisionsafterrelaxing
creditandriskconstraints.”QuarterlyJournalofEconomics129:597–652
Kast,FelipeandDinaPomeranz(2014)."SavingMoretoBorrowLess:Experimental
EvidencefromAccesstoFormalSavingsAccountsinChile."NBERW.P.20239.
Lipscomb,MollyandLauraSchecter(2017):“MobilePaymentSystems:TheImpactof
EarmarkedSavingsonSanitationPurchases.”WorkingPaper,UniversityofVirginia
Lucas,AdrienneandIsaacMbiti(2012):“Access,sorting,andachievement:Theshort-run
effectsoffreeprimaryeducationinKenya,”AmericanEconomicJournal:AppliedEconomics,
4(4),pp.226-253.
Mobarak,A.M.,andRosenzweig.M.(2013):“Sellingformalinsurancetotheinformally
insured.”WorkingPaper,YaleUniversity.
Muralidharan,Karthik,&Prakash,Nishith(2016):“CyclingtoSchool:IncreasingSecondary
SchoolEnrollmentforGirlsinIndia.”WorkingPaper19305
Ohba,Asayo(2011):“TheabolitionofsecondaryschoolfeesinKenya:Responsesbythe
poor,”InternationalJournalofEducationalDevelopment,Vol31(4),pp.402-408.
Ozier,O.(Forthcoming):“TheImpactofSecondarySchoolinginKenya:ARegression
DiscontinuityAnalysis,”JournalofHumanResources.
Prina,Silvia(2015):“BankingthePoorviaSavingsAccounts:EvidencefromaField
Experiment.”JournalofDevelopmentEconomics115:16–31.
Tarozzi, Alessandro, Aprajit Mahajan, Brian Blackburn, Dan Kopf, Lakshmi Krishnan, and Joanne 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.
24
UNESCOInstituteforStatisticsglobaldatabases.(2015).
WorldBank(2009):“AbolishingschoolfeesinAfrica:LessonsfromEthiopia,Ghana,Kenya,
Malawi,andMozambique,”WorldBank,incollaborationwiththeUnitedNationsChildren’s
Fund(UNICEF),WashingtonDC.
25
Table1:Design
Baseline
BaselinewithADSconsent
EndlinewithADSconsent
Control 1,688 1,441 1,300
(111) (110) (110)
MBA 1,431 1,227 1,056
(110) (108) (108)
LSA 1,554 1,352 1,163
(116) (113) (113)
Total 4,673 4,020 3,519
(337) (331) (331)
Cellentriesarenumberofparentsinterviewed.Numberofschoolspercellinbrackets.
26
Table2A:Balancetest(Respondent)
MeansbyTreatmentArms DifferenceinMeans
Control MBA LSA Overall
CvsMBA
CvsLSA
MBAvsLSA
Gender-Female 0.70 0.70 0.66 0.68 0.00 0.04* 0.04
(0.01) (0.02) (0.02) (0.01) (0.02) (0.02) (0.02)
Age 43.57 43.04 44.33 43.66 0.53 -0.76 -1.29***
(0.34) (0.34) (0.32) (0.20) (0.48) (0.47) (0.47)
IsprimarydecisionmakerinHH 0.41 0.41 0.41 0.41 -0.01 0.00 0.00
(0.02) (0.02) (0.02) (0.01) (0.03) (0.02) (0.03)
Hascompletedprimaryschooling 0.61 0.65 0.60 0.62 -0.04 0.01 0.05
(0.03) (0.03) (0.03) (0.02) (0.04) (0.04) (0.04)
Hascompletedsecondaryschooling
0.20 0.19 0.19 0.19 0.00 0.01 0.00
(0.01) (0.01) (0.01) (0.01) (0.02) (0.02) (0.02)
Hasaformaljob 0.11 0.09 0.10 0.10 0.01 0.01 -0.01
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Ownsamobilephone 0.91 0.93 0.95 0.93 -0.02 -0.04* -0.01
(0.02) (0.01) (0.01) (0.01) (0.02) (0.02) (0.02)
Canaccessamobilephone 0.98 0.99 0.98 0.98 -0.01* 0.00 0.01
(0.00) (0.00) (0.00) (0.00) (0.01) (0.01) (0.01)
HasaSafaricomSIM(allsample) 0.89 0.91 0.92 0.90 -0.02 -0.04* -0.01
(0.02) (0.02) (0.01) (0.01) (0.02) (0.02) (0.02)
GaveAdmin.DataSharing(ADS)consent(amongSafaricomSIMowners)
0.93 0.92 0.92 0.92 0.01 0.01 0.01
(0.01) (0.02) (0.03) (0.01) (0.02) (0.03) (0.03)
HasaSafaricomSIM(sampleexcl.Kilifi)
0.97 0.97 0.97 0.97 0.00 0.00 0.00
(0.01) (0.01) (0.01) (0.00) (0.01) (0.01) (0.01)
HasaSafaricomSIM(Kilifisample) 0.75 0.76 0.85 0.79 -0.01 -0.10** -0.08*
(0.03) (0.04) (0.03) (0.02) (0.05) (0.04) (0.05)
HasM-PESA 0.85 0.88 0.89 0.87 -0.03 -0.04* -0.01
(0.02) (0.02) (0.02) (0.01) (0.03) (0.02) (0.02)
HasM-PESA(amongSafaricomSIMowners)
0.96 0.97 0.97 0.96 -0.01 -0.01 0.00
(0.01) (0.01) (0.01) (0.00) (0.01) (0.01) (0.01)
N 1688 1431 1554 4673 3119 3242 2985
27
Table2B:Balancetest(ChildrenandSchooling)
MeansbyTreatmentArms DifferenceinMeans
Control MBA LSA Overall Cvs
MBACvsLSA
MBAvsLSA
Nb.ofchildreninHH 4.85 4.72 5.05 4.88 0.13 -0.20 -0.33*
(0.14) (0.12) (0.14) (0.08) (0.18) (0.19) (0.18)
Shareofgirlsamongchildren 0.51 0.51 0.49 0.51 0.00 0.02 0.02
(0.01) (0.01) (0.02) (0.01) (0.02) (0.02) (0.02)
Shareexpectedtoattendsecondaryschool
1.00 1.00 1.00 1.00 0.00 0.00** 0.00
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Sharewithaboveaveragegrades 0.31 0.30 0.29 0.30 0.01 0.02 0.01
(0.02) (0.02) (0.02) (0.01) (0.03) (0.03) (0.02)
Logexpectationofsecondaryschoolexpenses
10.47 10.48 10.44 10.46 -0.02 0.03 0.05
(0.04) (0.03) (0.03) (0.02) (0.05) (0.05) (0.05)
N 1688 1431 1554 4673 3119 3242 2985
Table2C:Balancetest(Schools)
MeansbyTreatmentArms DifferenceinMeans
Control MBA LSA Overall
CvsMBA
CvsLSA
MBAvsLSA
No.ofPermanentClassrooms,Usable
10.80 11.16 11.22 11.05 -0.36 -0.42 -0.06
(0.49) (0.61) (0.51) (0.31) (0.78) (0.71) (0.79)
AgeofSchool 47.16 45.02 46.24 46.20 2.14 0.92 -1.22
(2.28) (1.56) (2.14) (1.19) (2.76) (3.12) (2.64)
EnrollmentTotal(AllStudents) 539.56 483.91 546.97 524.95 55.66 -7.41 -63.06
(34.96) (34.55) (35.61) (20.35) (49.05) (49.80) (49.51)
PupilTeacherRatio 37.00 35.22 38.01 36.79 1.78 -1.02 -2.79
(1.42) (1.67) (1.44) (0.87) (2.19) (2.02) (2.20)
%studentswhoattendedsecondaryschool(amonglastyeargraduates)
69.32 71.58 68.43 69.72 -2.26 0.89 3.15
(2.74) (2.66) (3.10) (1.65) (3.81) (4.13) (4.07)
N 1688 1431 1554 4673 3119 3242 2985
28
Table3:Attrition
FullSample FullSamplewithADSConsent
MBATreatment 0.035** 0.049***
(0.014) (0.014)
LSATreatment 0.033** 0.042***
(0.016) (0.015)
AdjustedR-squared 0.02 0.01
ControlGroupMean 0.127 0.098
Nb.NotFound 679 501
Observations 4673 4020
Column(1):Attritionforbaselinesample
Column(2):AttritionforbaselinewithADSconsent
Standarderrorsclusteredbyschool.
Estimationstratifiesoncountyandwealth
*p<0.10,**p<.05,***p<.01
Table4:TakeupofMBAandLSAbytreatmentgroup
MBATake-up LSATake-up
MBATreatment 0.246*** -0.008
(0.025) (0.007)
LSATreatment 0.241*** 0.268***
(0.023) (0.022)
ControlGroupMean 0.336 0.015
Observations 4020 4020
Take-upratesreportedfromadministrativedataonJan.5th2015.
SamplerestrictedtothosewithADSconsent
Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth
*p<0.1,**p<0.05,***p<0.01
29
Table5:MobileSavings(January5th)
GrossMobileSavings NetMobileSavings GrossCBASavings (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)PanelA:ITTEstimates
MBA/LSAtreat 200 205 357***
(125) (125) (78)
MBATreatment 272* 278* 435***
(164) (164) (112)
LSATreatment 132 -116 134 -119 282*** -115 (144) (181) (145) (182) (97) (137)
AdjustedR-Squared 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.01 0.01 0.00PanelB:TOTEstimates
MBATake-up 824 1093* 841* 1120* 1468*** 1752***
(508) (650) (508) (651) (321) (439)
LSAtake-up 490 -413 498 -425 1045*** -410 (535) (642) (537) (646) (356) (488)
F-Statisticforweakidentification 149 102 154 155 149 102 154 155 149 102 154 155AdjustedR-squared 0.00 -0.00 0.00 -0.00 0.00 -0.00 0.00 -0.00 -0.02 -0.04 0.01 -0.00Ref.GroupCompliers'mean 399 354 225 1161 399 354 275 1270 0 0 -147 1407ReferenceGroupMean 253 253 253 543 294 294 294 590 21 21 21 467ReferenceGroup Control Control Control MBA Control Control Control MBA Control Control Control MBA
TreatmentInstrumentUsed Both MBA LSA LSA Both MBA LSA LSA Both MBA LSA LSA
ArmsExcludedfromSample None LSA MBA Control None LSA MBA Control None LSA MBA Control
Observations 4020 2668 2793 2579 4020 2668 2793 2579 4020 2668 2793 2579
GrossMobileSavings=M-PESAsavings+MBAsavings+LSAsavings NetMobileSavings=M-PESAsavings+MBAsavings+LSAsavings-MBALoans GrossCBASavings=MBAsavings+LSAsavings SamplerestrictedtothosewithADSconsent Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth *p<0.1,**p<0.05,***p<0.01
30
Table6:MobileSavings(January31st)
GrossMobileSavings NetMobileSavings GrossCBASavings (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)PanelA:ITTEstimates
MBA/LSAtreat 13 17 256***
(147) (147) (68)
MBATreatment 57 64 374***
(185) (186) (108)
LSATreatment -19 -75 -17 -78 144** -205* (159) (177) (159) (178) (66) (116)
AdjustedR-Squared 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00PanelB:TOTEstimates
MBATake-up 53 231 70 257 1053*** 1506***
(602) (739) (602) (742) (282) (437)
LSAtake-up -71 -266 -63 -278 533** -730* (588) (626) (587) (633) (252) (410)
F-Statisticforweakidentification 149 102 154 155 149 102 154 155 149 102 154 155AdjustedR-squared 0.00 0.00 0.00 -0.00 0.00 0.00 0.00 -0.00 -0.01 -0.03 -0.00 -0.01ReferenceGroupCompliers'mean 780 819 552 920 780 819 602 1029 0 0 -284 1117ReferenceGroupMean 476 476 476 552 517 517 517 598 24 24 24 409ReferenceGroup Control Control Control MBA Control Control Control MBA Control Control Control MBA
TreatmentInstrumentUsed Both MBA LSA LSA Both MBA LSA LSA Both MBA LSA LSA
ArmsExcludedfromSample None LSA MBA Control None LSA MBA Control None LSA MBA Control
Observations 4020 2668 2793 2579 4020 2668 2793 2579 4020 2668 2793 2579
GrossMobileSavings=M-PESAsavings+MBAsavings+LSAsavings NetMobileSavings=M-PESAsavings+MBAsavings+LSAsavings-MBALoans GrossCBASavings=MBAsavings+LSAsavings SamplerestrictedtothosewithADSconsent Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth *p<0.1,**p<0.05,***p<0.01
31
Table7:Credit AtleastoneMBALoan NumberofMBALoans TotalAmountofMBALoans (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
PanelA:ITTEstimates
MBA/LSAtreat 0.04*** 0.40*** 187.02
(0.01) (0.12) (114.58)
MBATreatment 0.05*** 0.45*** 190.29
(0.01) (0.15) (129.43)
LSATreatment 0.03*** -0.01 0.34** -0.11 180.88 -14.47 (0.01) (0.01) (0.14) (0.16) (144.18) (153.21)
AdjustedR-Squared 0.01 0.02 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.00 0.00 0.00PanelB:TOTEstimates
MBATake-up 0.16*** 0.18*** 1.64*** 1.82*** 768.64* 765.73
(0.04) (0.04) (0.45) (0.55) (461.85) (508.44)
LSAtake-up 0.12*** -0.05 1.26** -0.37 670.27 -51.52 (0.04) (0.05) (0.51) (0.59) (531.34) (544.08)
F-Statisticforweakidentification 149 102 154 155 149 102 154 155 149 102 154 155AdjustedR-squared 0.12 0.13 0.06 -0.02 0.07 0.08 0.05 -0.01 0.03 0.03 0.02 0.00Ref.GroupCompliers'mean 0 0 0.13 0.30 0 0 0.99 2.62 0 0 380 1375ReferenceGroupMean 0.08 0.08 0.08 0.12 0.75 0.75 0.75 1.19 415 415 415 604ReferenceGroup Control Control Control MBA Control Control Control MBA Control Control Control MBA
TreatmentInstrumentUsed Both MBA LSA LSA Both MBA LSA LSA Both MBA LSA LSA
ArmsExcludedfromSample None LSA MBA Control None LSA MBA Control None LSA MBA Control
Observations 4020 2668 2793 2579 4020 2668 2793 2579 4020 2668 2793 2579
Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth SamplerestrictedtothoseADSconsent LoanvariablescalculatedtillJan5th *p<0.1,**p<0.05,***p<0.01
32
Table8:FinancialSavings
NonMobileGrossFinancialSavings GrossFinancialSavings(Jan5) GrossFinancialSavings(Jan31) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)PanelA:ITTEstimates
MBA/LSAtreat 950 1196* 1017 (613) (624) (631)
MBATreatment 604 940 728 (647) (659) (661)
LSATreatment 1253 712 1408* 555 1271 605 (779) (751) (796) (771) (806) (768)
AdjustedR-Squared 0.00 0.00 0.00 0.00 0.00 0.00 0.00 -0.00 0.00 0.00 0.00 -0.00PanelB:TOTEstimates MBATake-up 3642 2327 4586* 3623 3898 2807
(2365) (2512) (2408) (2564) (2426) (2563)
LSAtake-up 4308 2365 4841* 1843 4370 2008 (2705) (2504) (2769) (2567) (2800) (2559)
F-Statisticforweakidentification 154 100 162 160 154 100 162 160 154 100 162 160
AdjustedR-squared -0.01 -0.00 0.00 0.00 -0.01 -0.01 -0.00 0.00 -0.01 -0.00 -0.00 0.00Ref.GroupCompliers'mean -4422 -3982 -6124 -4562 -3999 -3592 -5937 -3283 -3656 -3143 -5635 -3557ReferenceGroupMean -3752 -3752 -3752 -3262 -3470 -3470 -3470 -2619 -3257 -3257 -3257 -2617ReferenceGroup Control Control Control MBA Control Control Control MBA Control Control Control MBATreatmentInstrumentUsed Both MBA LSA LSA Both MBA LSA LSA Both MBA LSA LSAArmsExcludedfromSample None LSA MBA Control None LSA MBA Control None LSA MBA ControlObservations 3519 2356 2463 2219 3519 2356 2463 2219 3519 2356 2463 2219Sample:RestrictedtothosewithADSconsentandfoundatendline.Allmissingvaluesaretreatedaszeros Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth UsingGrossMobileSavingsfromAdmindatatillJan5forColumn5-8,andtillJan31tillforColumn9-12 NonMobileGrossFinancialSavings=BankAccount+MattressesSavings+SACCO+CHAMA+Advancedpurchases+Family+otherGrossFinancialSavings(Jan5)=NonMobileGrossFinancialSavings+GrossMobileSavingsAdmin(Jan5th) GrossFinancialSavings(Jan31)=NonMobileGrossFinancialSavings+GrossMobileSavingsAdmin(Jan31st)
33
Table9:MobileSavings-QuantileRegressions
GrossMobileSavings GrossFinancialSavings
(1) (2) (3) (4) (5) (6)
MBATreatment 131 0 383* 2907 3 -317
(201) (0) (218) (2291) (35) (628)
LSATreatment 236 0 397** 4984*** 3 -60
(156) (0) (202) (1570) (40) (565)
ControlGroupMean 253 253 253 -3470 -3470 -3470
Quantile 0.1 0.5 0.9 0.1 0.5 0.9
Observations 4020 4020 4020 3519 3519 3519
GrossMobileSavings(Jan5)=M-PESAsavings+MBAsavings+LSAsavingsNetMobileSavings(Jan5)=M-PESAsavings+MBAsavings+LSAsavings-MBALoansGrossCBASavings(Jan5)=MBAsavings+LSAsavings
SamplerestrictedtothosewithADSconsent
Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth
*p<0.1,**p<0.05,***p<0.01
Table10:OrdinalLogitMargins-Gross&NetMobileSavings
OrdinalVar:GrossMobileSavings OrdinalVar:NetMobileSavings
MBA
DependentVariable=1 -0.035** -0.032**
(0.016) (0.016)
DependentVariable=2 -0.004* -0.005*
(0.002) (0.003)
DependentVariable=3 0.039** 0.037**
(0.018) (0.019)
LSA
DependentVariable=1 -0.052*** -0.053***
(0.016) (0.016)
DependentVariable=2 -0.006** -0.008***
(0.002) (0.003)
DependentVariable=3 0.058*** 0.061***
(0.018) (0.018)
PseudoR-squared 0.002 0.0023
Observations 4020 4020
*p<0.1,**p<0.05,***p<0.01OrdinalDep.Variable:=1ifsavingsnegative,=2ifsavingszeroand=3ifsavingspositive
34
Table11:MobileSavingsbydecisionmakingrespondent(ITT)
GrossMobileSavings NetMobileSavings GrossCBASavings
GrossFinancialSavings(Jan5)
MBA/LSAtreat 229 228 125** 550
(236) (237) (51) (1365)Respondentisthedecision-maker 173 192 -15 -546
(178) (179) (41) (1131)MBA/LSAxRespondentisthedecision-maker -36 -30 270*** 661
(262) (263) (88) (1466)R-Squared 0 0 0.01 0
AdjustedR-Squared 0 0 0.01 0ControlGroupMean 256 297 21 -3406
Observations 4000 4000 4000 3508
GrossMobileSavings(Jan5)=M-PESAsavings+MBAsavings+LSAsavings
NetMobileSavings(Jan5)=M-PESAsavings+MBAsavings+LSAsavings-MBALoans GrossCBASavings(Jan5)=MBAsavings+LSAsavings
GrossFinancialSavings(Jan5)=NonMobileGrossFinancialSavings+GrossMobileSavings(Jan5th)
SamplerestrictedtothosewithADSconsentfor(1),(2)and(3),samplerestrictedtothosewithADSconsentandfoundatendlinefor(4)Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth
*p<0.1,**p<0.05,***p<0.01
35
Table12:MobileSavingsandFemaleDecision-maker
GrossMobileSavings NetMobileSavings GrossCBASavings GrossFinancialSavings(Jan5)
MBA/LSAtreat 333** 337** 388*** 1307*
(151) (152) (93) (750)
Femaledecision-maker 202 193 59 982
(209) (209) (64) (822)
MBA/LSAxFemaledecision-maker -527* -526* -122 -346
(268) (268) (147) (1018)
R-Squared 0.00 0.00 0.01 0.00
AdjustedR-Squared 0.00 0.00 0.01 0.00
ControlGroupMean 253 294 21 -3470
Observations 4020 4020 4020 3519
GrossMobileSavings(Jan5)=M-PESAsavings+MBAsavings+LSAsavings
NetMobileSavings(Jan5)=M-PESAsavings+MBAsavings+LSAsavings-MBALoans GrossCBASavings(Jan5)=MBAsavings+LSAsavings
GrossFinancialSavings(Jan5)=NonMobileGrossFinancialSavings+GrossMobileSavings(Jan5th) SamplerestrictedtothosewithADSconsentfor(1),(2)and(3),samplerestrictedtothosewithADSconsentandfoundatendlinefor(4)Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth
*p<0.1,**p<0.05,***p<0.01
36
Table13:MobileSavingsandexpectationsofaboveaverageschoolperformance
GrossMobileSavings NetMobileSavings GrossCBASavings GrossFinancialSavings
MBA/LSAtreat 196 205 321*** 1557**
(132) (133) (81) (682)
Expectsaboveavg.schoolperformance 51 92 5 591
(205) (205) (73) (794)MBA/LSAxExpectsaboveavg.schoolperformance
-34 -48 122 -1414
(294) (295) (186) (1277)
R-Squared 0.00 0.00 0.01 0.00
AdjustedR-Squared 0.00 0.00 0.01 0.00
ControlGroupMean 256 297 21 -3406
Observations 3976 3976 3976 3487
GrossMobileSavings(Jan5)=M-PESAsavings+MBAsavings+LSAsavings
NetMobileSavings(Jan5)=M-PESAsavings+MBAsavings+LSAsavings-MBALoans GrossCBASavings(Jan5)=MBAsavings+LSAsavings
GrossFinancialSavings(Jan5)=NonMobileGrossFinancialSavings+GrossMobileSavings(Jan5th) SamplerestrictedtothosewithADSconsentfor(1),(2)and(3),samplerestrictedtothosewithADSconsentandfoundatendlinefor(4)Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth
*p<0.1,**p<0.05,***p<0.01
37
Table14:Take-upofMBA/LSAandSecondarySchoolEnrollment
(1) (2) (3) (4)
PanelA:ITTEstimates
MBA/LSAtreat 0.06***
(0.02)
MBATreatment 0.06***
(0.02)
LSATreatment 0.05** -0.01
(0.02) (0.02)
AdjustedR-Squared 0.10 0.09 0.10 0.09
PanelB:TOTEstimates
MBATake-up 0.21*** 0.24***
(0.07) (0.08)
LSAtake-up 0.18** -0.04
(0.08) (0.07)
F-Statisticforweakidentification 167 108 169 166
AdjustedR-squared 0.06 0.04 0.09 0.09
ReferenceGroupCompliers'mean 0.62 0.59 0.65 0.94
ReferenceGroupMean 0.72 0.72 0.72 0.81
ReferenceGroup Control Control Control MBA
TreatmentInstrumentUsed Both MBA LSA LSA
ArmsExcludedfromSample None LSA MBA Control
Observations 3761 2521 2630 2371
Standarderrorsclusteredbyschool.EstimationstratifiesoncountyandwealthSamplerestrictedtothosewithADSconsentandfoundattheendline*p<0.1,**p<0.05,***p<0.01
38
Table15:SavingsandSecondarySchoolEnrollment(TOT)
(1) (2) (3) (4)
PanelA
GrossMobileSavings 0.24 0.20 0.37 0.07
(0.16) (0.12) (0.47) (0.15)
F-Statisticforweakidentification 3 4 1 1
AdjustedR-squared -5.99 -3.90 -11.09 -0.54
PanelB
NetMobileSavings 0.24 0.19* 0.36 0.07
(0.16) (0.12) (0.44) (0.15)F-Statisticforweakidentification 3 4 1 1
AdjustedR-squared -5.79 -3.81 -10.44 -0.54
PanelC
GrossCBASavings 0.14** 0.13** 0.16* 0.08
(0.06) (0.05) (0.09) (0.17)
F-Statisticforweakidentification 24 17 10 1
AdjustedR-squared -0.79 -0.69 -0.48 -0.42
PanelD GrossFinancialSavings(Jan5) 0.06 0.09 0.04 -0.02
(0.05) (0.10) (0.04) (0.04)
F-Statisticforweakidentification 2 1 2 0
AdjustedR-squared -5.35 -10.08 -3.11 -0.55
Observations 3761 2521 2630 2371
ReferenceGroup Control Control Control MBA
TreatmentInstrumentUsed Both MBA LSA LSA
ArmsExcludedfromSample None LSA MBA Control
GrossMobileSavings(Jan5th)=M-PESAsavings+MBAsavings+LSAsavings NetMobileSavings(Jan5th)=M-PESAsavings+MBAsavings+LSAsavings-MBALoansGrossCBASavings(Jan5th)=MBAsavings+LSAsavings
GrossFinancialSavings(Jan5)=NonMobileGrossFinancialSavings+GrossMobileSavings(Jan5th)SamplerestrictedtothosewithADSconsentandfoundattheendlineStandarderrorsclusteredbyschool.EstimationstratifiesoncountyandwealthGross,Net,CBAandFinancialsavingsaremeasuredinunitsofthousandsofKSh
*p<0.1,**p<0.05,***p<0.01
39
Table16:RealizedCostofSchool
TotalSchoolCost(incurred+expected)
(1) (2) (3) (4)
PanelA:ITTEstimates
MBA/LSAtreat -1953
(1830)
MBATreat 330
(2194)
LSATreat -3692* -3736*
(1968) (2093)
AdjustedR-Squared 0.01 0.01 0.02 0.01
PanelB:TOTEstimates
MBAtake-up -7488 1272
(7114) (8415)
LSAtake-up -12691*-
12409* (6941) (7141)
F-Statisticforweakidentification 154 100 162 160
AdjustedR-squared -0.02 0.01 -0.00 -0.01Ref.GroupCompliers'mean 33906 31340 44976 45160
ReferenceGroup Control Control Control MBAReferenceGroupMean 31729 31729 31729 32103
TreatmentInstrumentUsed Both MBA LSA LSAArmsExcludedfromSample None LSA MBA Control
Observations 3519 2356 2463 2219
Standarderrorsclusteredbyschool.EstimationstratifiesoncountyandwealthSample:EveryonewhogaveBaselineandAdminDataSharing(ADS)consentandwasfoundattheendline
*p<0.1**p<0.05***p<0.01
40
Table17:FinancingSchool
SavedMoney
BorrowingMoney
SoldFarmprod.orlivestock
SoldHHitems
Giftitemsormoney
other
MBATreat -0.02 0.01 0.01 -0.00 0.01 -0.00
(0.02) (0.02) (0.02) (0.00) (0.01) (0.01)
LSATreat -0.04* 0.01 0.03 -0.00 0.02 -0.01
(0.02) (0.02) (0.02) (0.00) (0.01) (0.01)
AdjustedR-Squared 0.03 0.03 0.03 -0.00 0.01 0.00
ReferenceGroupMean 0.48 0.24 0.24 0.00 0.09 0.09
Observations 3519 3519 3519 3519 3519 3519
Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth Sample:EveryonewhogaveBaselineandAdminDataSharing(ADS)consentandwasfoundattheendline
Table18:StandardizedTestScores
(1) (2) (3) (4)PanelA:ITTEstimates MBA/LSAtreat 0.01
(0.06) MBATreatment -0.01
(0.07) LSATreatment 0.05 0.05
(0.07) (0.07)AdjustedR-Squared 0.03 0.02 0.03 0.04PanelB:TOTEstimates MBATake-up 0.05 -0.03
(0.22) (0.24) LSAtake-up 0.15 0.17
(0.24) (0.21)F-Statisticforweakidentification 178 120 148 144AdjustedR-squared 0.03 0.02 0.04 0.04ReferenceGroupCompliers'mean 0.00 -0.03 0.01 0.00ReferenceGroupMean 0.00 0.00 0.00 0.00ReferenceGroup Control Control Control MBATreatmentInstrumentUsed Both MBA LSA LSAArmsExcludedfromSample None LSA MBA ControlObservations 3249 2218 2255 2025
Standarderrorsclusteredbyschool.EstimationstratifiesoncountyandwealthStandardizedtestscoresusingmeanandstandarddev.ofcontrolgroupSamplerestrictedtothosewithADSconsentandfoundattheendline*p<0.1,**p<0.05,***p<0.01
41
TableA2:LeeboundsforTable8(FinancialSavings)
NonMobileFinancialSavings GrossFinancialSavings(Jan5) GrossFinancialSavings(Jan31)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)lower
-1325**-
1759***-931 714 -1071* -1389** -782 493 -1217** -1567** -899 554
(577) (665) (757) (4149) (596) (695) (772) (4163) (602) (698) (782) (4163)upper 2076*** 1632*** 2479*** 880 2469*** 2139*** 2768*** 673 2411*** 2049*** 2739*** 732
(536) (589) (698) (1405) (560) (634) (721) (1801) (559) (626) (725) (1756)ArmsExcludedfromSample None LSA MBA Control None LSA MBA Control None LSA MBA ControlObservations 4020 2668 2793 2579 4020 2668 2793 2579 4020 2668 2793 2579
Sample:RestrictedtothosewithbaselineandADSconsent. Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth
UsingGrossMobileSavingsfromAdmindatatillJan5forColumn5-8,andtillJan31tillforColumn9-12 NonMobileGrossFinancialSavings=BankAccount+MattressesSavings+SACCO+CHAMA+Advancedpurchases+Family+other
GrossFinancialSavings(Jan5)=NonMobileGrossFinancialSavings+GrossMobileSavingsAdmin(Jan5th) GrossFinancialSavings(Jan31)=NonMobileGrossFinancialSavings+GrossMobileSavingsAdmin(Jan31st)
TableA1:LeeboundsforTable5(MobileSavings)
GrossMobileSavings NetMobileSavings GrossCBASavings (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
lower -136 66 -236 -485** -135 73 -240 -492** 70 247 15 -410*** (217) (465) (167) (221) (218) (465) (167) (235) (129) (299) (66) (143)
upper 405*** 410 341** 31 408*** 416 343** 27 397*** 480*** 298*** -151 (156) (326) (140) (188) (156) (324) (141) (190) (76) (141) (79) (138)
ArmsExcludedfromSample None LSA MBA Control None LSA MBA Control None LSA MBA ControlObservations 4673 3119 3242 2985 4673 3119 3242 2985 4673 3119 3242 2985
GrossMobileSavings=M-PESAsavings+MBAsavings+LSAsavings NetMobileSavings=M-PESAsavings+MBAsavings+LSAsavings-MBALoans
GrossCBASavings=MBAsavings+LSAsavings
Samplerestrictedtothosewithbaselineconsent
Standarderrorsclusteredbyschool.Estimationstratifiesoncountyandwealth *p<0.1,**p<0.05,***p<0.01
42
Figure1:Cumulativeshort-termgrossmobilesavings
Figure2:Cumulativeshort-termfinancialsavings