documento de trabcollaborated with us on an evaluati on of the yo elijo mi pc program that is the...
TRANSCRIPT
D O C U M E N T O
D E T R A B A J O
Instituto de EconomíaD
OC
UM
EN
TO d
e TR
AB
AJO
I N S T I T U T O D E E C O N O M Í A
www.economia.puc.cl • ISSN (edición impresa) 0716-7334 • ISSN (edición electrónica) 0717-7593
Parental Monitoring ande Children´s Internet Use: The Role of Information,Control, and Cues
Francisco Gallego, Ofer Malamud, Crsitian Pop-Eleches.
4962017
1
ParentalMonitoringandChildren'sInternetUse:TheRoleofInformation,Control,andCues
FranciscoGallegoPontificiaUniversidadCatólicadeChile
OferMalamudNorthwesternUniversity
CristianPop‐ElechesColumbiaUniversity
October2017
Abstract*Thispaperexploreshowasymmetricinformationbetweenparentsandchildrenanddirectparentalcontrolscaninfluencechildren’sinternetuseinChile.Wedesignedandimplementedasetofrandomizedinterventionswherebyapproximately7700parentsweresentweeklySMSsmessageswith(i)specificinformationabouttheirchildren’sinternetuse,and/or(ii)encouragementandassistancewiththeinstallationofparentalcontrolsoftware.WeseparatetheinformationalcontentfromthecueassociatedwithSMSmessagesandvarythestrengthofthecuesbyrandomlyassigningwhetherparentsreceivedmessagesinapredictableorunpredictablefashion.Ouranalysisyieldsthreemainfindings.First,wefindthatmessagesprovidingparentswithspecificinformationreducechildren’sinternetuseby6‐10percentandhelpparentsmitigatetheproblemofasymmetricinformationinthehousehold.Second,wedonotfindsignificantimpactsfromhelpingparentsdirectlycontroltheirchildren’sInternetaccesswithparentalcontrolsoftware.Third,thestrengthorsalienceofthecueassociatedwithreceivingamessagehasanindependentimpactoninternetuse.
* We would like thank Ramon Rodriguez and his staff at the Ministry of Education for providing the data and technical assistance necessary to conduct this study. We are especially grateful to Jaime Bellolio who collaborated with us on an evaluation of the “Yo Elijo mi PC” program that is the setting of the present study. Cristian Larroulet, Jose Ignacio Cuesta, Antonia Asenjo, Magdalena Bennet, Ana Mendoza, Dario Romero, Sebastian Otero, and Alejandro Saenz provided excellent research assistance. We would like to thank Paloma Acevedo, Peter Bergman, Samuel Berlinski, Marianne Bertrand, Lucas Coffman, Stefano Dellavigna, Jeanne Lafortune, Philip Oreopoulos, as well as seminar participants at Columbia Teacher’s College, the Inter-American Development Bank, Princeton University, and PUC-Chile for comments and suggestions. We would like to thank FONDECYT (Project 1141111), J-PAL, the Columbia-Chile fund, and the Population Research Center, grant # R24 HD051152-05 from the National Institute of Child Health and Human Development for financial support. All errors are our own.
2
1. Introduction
Economistshavelongbeeninterestedinhowparentscanaffecttheirchildren’sactions.
Becker(1974,1981)introducedhisfamous“RottenKidTheorem”toshowthat,under
certainassumptions,altruisticparentscancontroltheirchildren’sactionsindirectly
throughex-posttransfers.However,asBergstrom(1989)pointedout,theRottenKid
theoremdoesnotholdinthepresenceofmoralhazardduetoasymmetricinformation
—i.e.whenparentscannotobservetheirchildren’sactions.BursztynandCoffman
(2012)provideevidenceforsuchasymmetricinformationamongparentsinBrazil.
Nevertheless,evenintheabsenceofasymmetricinformation,parentsmaynotbeable
tocontroltheirchildren’sactionsbecausetheyareunableorunwillingtomake
negativetransfersthatimposelargecostsonthechild(Weinberg,2001;Berry,2015).
Inthesecases,parentsmaywishforthepossibilityofcontrollingtheirchildren’sactions
directly.
Themainmotivationforthispaperistoexploretheroleofdirectcontrolsand
asymmetricinformationbetweenparentsandchildreninthecontextofhome
computersandinternetuse.Tothisend,wedesignedandimplementedasetof
randomizedinterventionstotesttheimpactofsendingparentsweeklySMSmessages
containingspecificinformationabouttheirchildren’srecentinternetuseand/or
encouragementandassistancewithinstallingparentalcontrolsoftware.Providing
parentswithinformationabouttheirchildren’sinternetuseshouldhelpalleviate
informationalasymmetriesandthereforereducethepotentialformoralhazard.
Encouragingparentstoinstallparentalcontrolsoftwarecanhelpparentsbypassthe
needtoincentivizetheirchildrenorenforcerulesrelatedtocomputeruse,assuming
thatparentsareabletoinstallandoperatesuchparentalcontrolsoftware.
3
Exploringparent-childinteractionswhennavigatingtechnologyisespecially
instructivebecausetheinformationalasymmetriesarelikelytobepronouncedand
implementingdirectcontrolscanbedifficult;childrenareoftenquickertoadapttonew
technologiesandparentsmayencounterchallengesinunderstandinghowchildrenuse
technology.ArecentstudybyMalamudandPop-Eleches(2011)foundthatparental
rulesabouthomeworkandcomputeruseattenuatedthenegativeeffectsofcomputer
ownership,suggestingthatparentalmonitoringandsupervisionmaybeimportant
mediatingfactors.Buttheseresultswereonlysuggestivebecausetheruleswerenot
randomlyassignedandmayhavereflectedotherparentalcharacteristics.
Asecondmotivationforthispaperistounderstandhowandwhytheprovision
ofinformationaffectsbehavior.Tothisend,wedesignedourinterventionstoseparate
theinformationalcontentfromthecueassociatedwiththeSMSmessages,and
attemptedtovarythestrengthorsalienceofthecuesbyrandomlyassigningwhether
parentsreceivedmessagesinapredictableorunpredictablefashion.Thiswasbasedon
researchinneurosciencewhichsuggeststhathumanresponsesmayberelatedtothe
predictabilityornoveltyofthestimuli(Parkin,1997;Bernsetal.,2001;Fenkeretal.,
2008).Itisalsocloselyrelatedtoresearchinpsychologyonhowdifferentschedulesof
reinforcementaffectbehavior(FersterandSkinner,1957).
Thereisalargeliteratureexaminingtheimpactofprovidinginformationto
consumers,voters,students,parents,etc.(AlcottandRogers,2014;FinanandFerraz,
2011;Jensen,2010;Dizon-Ross,2016).Moreover,acoupleofrecentpapersstudythe
effectofsendingSMSmessagestoparentswithinformationabouttheirmissed-
assignment,attendance,andgrades(Bergman,2016;Berlinskietal.,2017).However,
thesestudieshavenotseparatedtheeffectonbehaviorduetotheinformationalcontent
ofthemessagevs.thecuethatisassociatedwithgettingthemessageitself.Thereisalso
4
researchexamininghowSMSmessagescanserveasreminderstopromotebeneficial
changestobehaviorineducation(Castleman,2015),health(Pop-Elechesetal.,2011)
andsavings(Karlanetal.,2014)).Someofthesestudiesalsoconsidertheroleoflimited
attentioninexplainingtheeffectivenessofSMSmessagesandreminders.1Nevertheless,
ourunderstandingofthesalienceorstrengthofthesetypeofmessagesremains
limited.2
Wefocusonasampleofchildrenin7thand8thgradewhoreceivedfree
computersand12monthsoffreeinternetsubscriptionsthroughChile’s“YoElijomiPC”
(YEMPC)programin2013.Wehavedataontheintensityofinternetuseatthedaily
levelfromtheinternetserviceprovider(ISP)whichservedallofthecomputers
providedtothechildreninoursample.Accordingtothisdata,childrendownloaded
approximately150MBofinternetcontentdailywhichtranslatestoabout3hoursof
internetuseonadailybasis.GiventhattheAmericanAcademyofPediatrics(AAP)
recommendsnomorethan2hoursofscreentimeforchildren(AAP,2016),this
suggestsaroleforparentalsupervisionofInternetuseregardlessofthetypeofuse.
Ouruniquedataenabledustointroduceatreatmentthatprovidedparentswith
informationabouttheirchildren’sinternetuse.Forthis“information”treatment,we
sentparentsweeklySMSmessagesprovidingspecificinformationfromtheISPabout
theintensityofinternetuse,intermsofMBsuploaded/downloaded,overtheprevious
week.Forthe“parentalcontrol”treatment,wesentparentsweeklySMSsoffering
assistancewiththeinstallationofWindows8(W8)parentalcontrolsoftware.Wealso
incorporatedatreatmentarmthatincludedbothISPinformationandassistancewith
W8parentalcontrolstotestforpossibleinteractionsbetweenthesetreatments.To
1Taubinsky(2014)andEricson(2017)providetheoreticalanalysesofhowlimitedattentionandpresentbiascanaffecttheeffectivenessofmessagesandreminders.2Bertrandetal.(2010)considertheeffectofdifferentcuesinthecontextofamarketingfieldexperiment.
5
disentangletheinformationalcontentandtheofferofassistancefromthecue
associatedwithSMSmessages,wecomparethesetreatmentstoacontrolgroupin
whichparentsreceivedweeklySMSsremindingthemthatchildrenshouldmakegood
useoftheircomputers,amessagethatwasincludedineverytreatment.Inaddition,we
attemptedtovarythestrengthofthecuewithineachofourtreatmentarmsby
randomlyassignparentstoeitherreceivetheSMSsonthesamedayoftheweek(the
“fixed”subgroups)oraonrandomdayoftheweek(the“random”subgroups).Allof
theseinterventionslastedfor14weeks.
Wehavethreemainsetsofresults.First,wefindthathouseholdsinwhich
parentsreceivedISPinformationaboutinternetusehada6to10percentlower
intensityofinternetuseduringthetreatmentperiodrelativetohouseholdsinthe
controlgroup.Theseeffectspersistintheweeksandmonthsaftertreatmentended.
Thissuggeststhatourtemporaryinterventionprovidinginformationoninternetuse
mayhavealteredthepermanentintra-householdequilibriumandhelpedparentssolve
theproblemofasymmetricinformationinamorepermanentway.Wealsoshowthat
therearestatisticallysignificantreductionsinusepreciselyonthedaysimmediately
afterreceivingtheISPinformationandthiseffectismorerelevantintheearlyweeksof
theexperiment.Moreover,itistheSMSmessageswhichconveythe“badnews”,i.e.that
childrenusedmoreinternetthanthereferencegroupinaspecificweek,whichproduce
amuchlargerdeclineininternetuse.Thus,thesefindingsconfirmthatitisthespecific
informationprovidedtoparentsabouttheirchildren’sinternetusewhichleadstoa
significantreductionofinternetuse.
Second,wedonotfindsignificantimpactsfromhelpingparentsdirectlycontrol
theirchildren’sinternetaccess.Inparticular,wedonotfindadifferenceininternetuse
betweenparentswhowereencouragedandprovidedassistancetoinstallparental
6
controlsoftwareascomparedwiththoseinthecontrolgroupwhoonlyreceiveda
genericcue.Take-upofthisinterventionwasonly15percentwhenmeasuredinterms
oftheparentswhoactuallyrespondedtoourmessages.However,forthosetakingup
theintervention,wedonotfindchangesinInternetuseevenonthedaysimmediately
afterinstallingtheparentalcontrolsettings.Webelievethisfindingreflectsthe
considerableobstaclesfacedbylow-incomeparentsinimplementingtechnological
solutionsformonitoringandsupervisingtheirchildren.
Third,wehaveseveralresultsthathelpusopenthe“blackbox”ofhowmessages
thatcontaininformationaffectbehavior.Asmentionedabove,bysendingmessagesthat
varytheamountofinformationwecanshowtheimportantroleofamessage’s
informationalcontentonreducingInternetuse.Butouranalysisalsoshowstwo
additionalfindingsthatsuggesttheimportanceofsalience.Whenweexperimentally
variedthestrengthofthecue,wefindthathouseholdswhoreceivedSMSsonarandom
scheduleexperiencedsignificantlygreaterreductionsininternetusethanthoseon
fixedschedules,aneffectsimilarinmagnitudetothemaineffectassociatedwith
receivingtheISPinformation.Furthermore,wefindthateventheSMSmessagessentto
thecontrolgrouphadshort-termimpactsonInternetuseinthefirstweeksofthe
experiment,perhapsduetothenoveltyofthemessage.
Ourpapermakesseveralcontributions:First,weidentifyreal-timeimpactsof
theprovisionofinformationoninternetuse,animportantdimensionofchildren’s
behaviorathomethatisoftenimperfectlyobservedtoparents.Second,weuse
experimentalvariationtoisolatethecausaleffectofprovidingparentswithspecific
informationabouttheirchildren’sbehaviorandhelpingparentsexercisedirectcontrol
overtheirchildren’sbehavior.Thisdealswiththepotentialendogeneityassociatedwith
observationalcomparisonsofparentswhohaveaccesstomoreorlessinformationand
7
controlovertheirchildren’sbehavior.Third,weisolatetheimpactofprovidingparents
withspecificinformationfromtheeffectofacuefromreceivingamessagebyusinga
controlgroupinwhichparentsreceivedmessageswithoutthisinformation.Suchcues
arelikelytobeanimportantpartofanyattempttoprovideparents(orothereconomic
actors)withinformation.Fourth,weattempttolearnmoreabouttheroleofthesecues
byintroducingexperimentalvariationinhowandwhenparentsreceivedthemessages.
Thus,thisstudyprovidesauniqueopportunitytobetterunderstandthenatureofintra-
familydynamicsbetweenparentsandchildrenandtobetterunderstandhowandwhy
messagescontaininginformationaffectbehavior.
Thepaperisorganizedasfollows:Section2providessomebackgroundonthe
YoElijomiPCprogram.Section3introducestheexperimentaldesign.Section4
describesthedatausedforthestudy.Section5explainstheempiricalstrategy
underlyingtheanalysis.Section6presentsthemainfindings.Section7discussesand
interpretsthesefindings.Finally,Section8concludes.
2. Background
Wedesignedandimplementedourexperimentinthecontextofthe2013cohortofthe
YEMPCprogram.YEMPCisaChileangovernmentprogramthatprovidescomputersto
7thgraderswithhighacademicachievementfromdisadvantagedhouseholds.Students
areeligiblefortheprogramiftheyattainedasufficientlyhighgradepointaverage
(GPA)in4th,5th,andthefirstsemesterof6thgradeandiftheirhouseholdscoredbelow
acertainlevelonafocalizationinstrumentcalledFichadeProtecciónSocial(FPS).3
Figure1presentsthetimelineoftheprocessesassociatedwiththe2013cohort
oftheYEMPCandwiththeexperimentweimplementedwithpartofthestudentsofthis
3Carneiroetal.(2014)presentadetaileddescriptionoftheFPSinstrument.
8
cohort.ThetimelineofYEMPCforeachroundisasfollows:Eligiblestudentsare
identifiedbasedontheirFPSandGPAscoresinSeptember-Octoberoftheyearpriorto
receiptofthecomputer(2012inourcase).Inprinciple,eachstudentwhomeetsthe
FPSandGPArequirementsiseligibletoreceiveacomputer;thereisnoapplication
process.However,studentsarethenrequiredtoselectacomputerinNovember-
December.Anumberofdifferentoptionsareavailableeachyear,includinglaptopsand
desktopswithdifferentfeaturesandspecifications.Eachcomputerisequippedwith
Windows8andMicrosoftOffice.Computersaredistributedtostudentsduringthe
monthsofAprilandMay.Thecomputersaregiveninschools,inaceremonyorganized
bythemunicipalitywherethestudentisenrolled.Beginningwiththe2011cohort,
studentswithsometypesofcomputersalsoreceived12monthsoffreeInternetservice.
Ourexperimentconsidersstudentswhoenterto7thgradein2013andwere
selectedtotheprograminNovember2012.Wefocusinthebeneficiariesofthiscohort
whoreceivedfreeinternetaccesswiththeircomputersstartingfrommid-2013.
Internetisprovidedthroughaprivateprovider.32,270beneficiariesareinthis
category(outofthe52,122beneficiariesoftheprograminthe2013cohort).
Afteraprocessthatinvolvedcontactingfamiliesbyphone(usingadministrative
data)andaskingthemtoparticipateintheexperimentandtocompleteatelephone
baselinesurvey,weendedupwithasampleof9,636parentswithavalidcellphone(to
receivetheSMSs)andwhoconsentedtoparticipateintheexperiment.4
In terms of access to computers and Internet at home before receiving the
YEMPCprogram,Gallegoetal.(2017)documentthat40%ofbeneficiarieshadaPCat
home,23%hadInternetaccessathome,and6%haveacellphonewithInternetaccess.
4AppendixTable1comparesthestudentsinourexperimentalsamplewithabroadersampleofthosewhoappliedforacomputerwithaninternetconnectionunderthe“YoElijoMiPC”program.Wedonotidentifyeconomicallyrelevantdifferencesbetweentheexperimentalsampleandthebroadersample.
9
Intermsofwheretheyhaveaccesstocomputer(differentfromhome):87%hadaccess
toacomputeratschool,30%inacyber-café,and68%canaccessacomputer´sfriend.
ThisimpliesthatmostofthekidsinoursamplehadaccesstoacomputerandInternet
before theprogramstartedbut that amuch smaller fractionhadpersonal computers
and personal Internet access. However, in terms of total Internet use, the median
student reported that timeusewas "access to Internet some times in theweek".This
contrasts with an average time of about three hours of Internet use per day in the
baseline of our experiment that followed the distribution of computers to recipients.
ThissuggeststheYEMPCprogramincreaseddramaticallyInternetuse.
3. ExperimentalDesign
Werandomized7,707familiesintothedifferentexperimentalgroupspresentedin
Table1.5Inadditiontothedatacomingfromthetelephonebaselinesurveywehave
administrativedataoninternetusestartingfromSeptember,2013.Usingthis
information,weimplementedastratifiedrandomizationconsideringthefollowing
strata:(i)guardian'seducation(NoHigh-School,HighSchool,College),(ii)parent
perceptionofwhetherthestudentstaystoolonginfrontofthecomputer(YesorNo,
usinginformationfromabaselinesurvey),and(iii)InternetUseastotalMBs
downloadedbetweenSeptemberandDecember15,2013(usingadministrativedata
fromtheISPprovider).
Theexperimentconsistsofthedeliveryofweeklytextmessagestotheparents
includedinthesample.TheSMSsdifferedintermsofcontentandthedayoftheweekin
5Wealsohavea“referencegroup”of1,929familieswhichweusetocomputeInternetuseforthepurposesofcomparisontooneoftheexperimentalarms.
10
whichtheyweredelivered.Intermsofcontents,wesentthreetypesofSMSsusingthe
followingtexts:
• SMS-only:“WehopeyourchildmakesgooduseoftheYoElijoMiPClaptopthat
he/shewon”.
• ISP:“Wehopeyourchildmakesgooduseofthelaptopthathe/shewon.Yourchild
downloadedXXMBstheweekoftheDD-MMM,{“morethan",or“similarto",or“less
than"}whatatypicalchilddownloaded:YYMBs.”6
• W8:“Wehopeyourchildmakesgooduseofthelaptopthathe/shewon.The
ParentalControlprogramofWindows8canhelpyousuperviseyourchild's
computeruse.CallusatXXX-XXXXforassistance.”
GroupT0receivedtheSMS-onlymessage,T1receivedtheISPmessage,T2receivedthe
W8message,andT3receivedboththeISPandW8messages(inthatorder).Ineach
groupabouthalfofthefamiliesreceivedthetreatmentsinfixed-daysoftheweek(and
werandomizethedayinwhichtheyreceivedthemessage)andabouthalfofthe
familiesreceivedthemessagesonrandomdaysofeachweek.
Themessagesweresentweeklybetween4pmand5pmondifferentweekdays
betweenDecember23,2013andApril6,2014.Thisperiodcoversthesummervacation
(fromDecember,232013toMarch6,2014)andtheschoolperiod(fromMarch7,
2014).WediscusstheactualshareofSMSsreceivedineachtreatmentarminSection
6.1.
4. Data
ThemainsourceofadministrativeinformationforourstudyisinformationonInternet
useforeachbeneficiarycollectedbytheInternetprovider.Thisincludesdaily
6The“referencegroup”isusedtocomputetheweeklyaverageofMBsdownloadedbyatypicalchild.
11
informationonMBsdownloadedanduploaded.Wereceivedinformationforeach
beneficiaryfortheperiodcoveredbetweenSeptember22,2014andJune,17,2015.
ThisimpliesthatwehaveinformationonInternetusefortheperiodbeforetheSMS
treatmentsstarted,fortheperiodinwhichtheSMSsweredelivered,andalsofor12
weeksafterthetreatmentwasdiscontinued.
Inaddition,weuseinformationfromthebaselinesurveyforthestratified
randomizationandtocontrolforseveralbaselinecharacteristicsinourmain
specifications.Theseincludehouseholddatawithinformationaboutthestudentsand
parentssuchas,studentgender,guardianage,familycomposition,numberofsiblings,
parents’education,parents’workingconditions,andguardian´sperceptionsofinternet
andcomputeruse.Wehaveinformationforalltheindividualsincludedinthesample,
asthiswaspartoftheenrollmentprocess.
Table2summarizesstudentandparentalcharacteristicsforourmainanalytical
sample.ThedailymeanMBsdownloadinthe3monthsofthepre-treatmentperiodwas
approximately150MBwhichcorrespondedto186minutesofpredictedinternetuse.
Almostallofthechildreninoursamplelivewiththeirmothers,andoversixtypercent
alsolivewiththeirfathers.Moreover,approximatelythree-quartershaveasiblingliving
withthemwhilefifteenpercentalsolivewithagrandparent.Oursampleofstudentsis
43percentfemaleandtheyhaveanaverageof1.7siblings.Theaverageageofthe
guardianis40yearsold.Mostoftheguardianshavesecondaryeducation,withalmost
halfhavingcompleteditandanotherfifteenpercenthavesomesecondaryeducation.
Theremainderhaveonlyanelementaryeducationwithjustfourpercenthavingsome
highereducation,whichisnotsurprisinggiventhetargetpopulationoftheYoElijomi
PCprogram.
12
Duringthetreatmentperiod,wewereabletogatherinformationaboutwhether
theSMSssentwerereceivedonthecellphonesoftreatedparents.Thisservesto
measurethetechnicalpartofthetake-up,asrelatedtotheactualdeliveryofthe
messages.WealsocollecteddataontheinstallationofW8parentalcontrolsetting
throughourcallcenter.Thus,thismeasure,capturesthetake-upoftheW8treatment
directlyfromus,thoughparentscouldinstallparentalcontrolsoftwarebyothermeans.
WediscussandusethisinformationinTable4.
Then,afterthedeliveryofthetreatmentwasended,weappliedabriefphone
interviewbetweenAprilandearlyMay2014toexaminesomepotentialmechanisms
underlyingourestimatedimpactsoninternetuse.Wewereabletocontact5,001
parentswhoconsentedtoparticipateinthesurvey.Thisisequivalentto57%ofthe
originalsample.Thisismostlyaconsequenceofthedifficultytoreachparentsonthe
phone,astherejectionrateofthesurveywasjustabout14%.Thesurveyincludesa
seriesofquestionsaboutparentrecollectionsofreceivingSMSs,theusefulnessofthe
SMSs,andthedecisiontoinstalltheparentalcontrolsoftware.7Weusethisinformation
inTable5.
Finally,inordertoeasetheinterpretationoftheresults,weconstructedaproxy
fortimeuse(inseconds)usinginformationfromstudentsofthe2012YEMPCcohort.
WecollectedinformationonbothMBs(downloadedanduploaded)andtimeofinternet
connectionforasampleof48,920studentsfor125days(frommid-Apriltoearly
December,2012).UsingthisinformationweestimatedOLSregressionsmodelsinwhich
timeofInternetuseisanon-linearfunctions(includinginteractions)ofMb
downloaded,Mbuploaded,dummiesforthedayoftheweek,dummiesforholidays,and
7AppendixTable1presentsacomparisonofbaselinecharacteristicsbetweenthemainsampleandthesurveysample.
13
dummiesfordiscretelevelsofuse(fourcategoriesthatreflecthigher,high,normal,low
use).WeusetheestimationwiththehighestR2(0.621)toimputetimeofInternetuse
foroursample.Wekeeptheseestimatesjustasreferencefortheinterpretationofthe
mainresultsinthepaperandpresentimpactestimatesintheAppendixasrobustness
checks.
Table3showsbalanceinthemaindemographiccharacteristicsforoursample
acrosseachofourtreatmentarms,T1,T2,andT3relativetothecontrolgroupT0.The
F-testpresentedinthelastcolumnrejectsbalanceacrossthetreatmentsandthe
controlgroupatthe10%levelforjustonevariable.Thiscorrespondstowhetherthe
childliveswiththeirmother.8Still,thedifferencesinaveragesforthisvariableacross
groupsdonotseemtobeeconomicallyrelevant.Wecontrolforthisvectorofcovariates
insomeofourregressionspecificationsand,notsurprisinglygiventhebalanceacross
treatmentarms,ourcoefficientsremainlargelyunchanged.
5. Empiricalstrategy
Wepresenttwoalternativeregressionapproachesforestimatingtheimpactofour
maininterventionsonInternetuse.First,weanalyzetheimpactofthedifferent
interventionsbycomparingaverageInternetuseacrosshouseholdsallocatedtothe
differentgroups.ThisallowsustoidentifytheaverageeffectonInternetuseoverthe
entiretreatmentperiod.Second,wealsoanalyzetheeffectsoftheactualreceptionof
theSMSmessagesusinganeventstudyanalysisinwhichweexploitwithin-event
8AppendixTable2presentsbalancetestsfortherandom/fixedschedulesub-treatmentsandAppendixTable3presentsbalancetestsforthesurveysample.Weonlyobservetwounbalancedvariablesfortherandom/fixedschedulecomparisons(outof15),andwithsmalldifferences:therandom-schedulehouseholdshaveslightlylessparentswithcompletehighereducation(1p.p.)andwerepresentwithahigherprobabilityintheothercategorygroupforcurrentemploymentstatus(1.p.p.).
14
variationinInternetuseonadailybasis.Thisallowstobetterunderstandthe
mechanismsbehindthechangesinbehavior.
Inthefirstapproach,weadoptthestandardspecificationusedtoanalyze
randomizedexperimentsbyseparatelyidentifyingtheimpactsofeachtreatmentarm,
T1,T2,orT3relativetoourcontrolgroupT0:
!! = !′!! + !!!1! + !!!2! + !!!3! + !!
where!! isameasureofinternetuseforhousehold!,.Forsomespecificationswealso
includeasetofcontrolvariables(!!).ThecoefficientonT1capturestheeffectof
receivingISPinformationwithrespecttothecontrolgroupT0,thecoefficientonT2
capturestheeffectofreceivinginformationonhowtoinstallparentalcontrolswith
respecttothecontrolgroup,andT3capturestheeffectofreceivinginformationonboth
ISPandhowtoinstallparentalcontrols.TotheextentthatnotalloftheSMSssentare
actuallyreceived,thesecoefficientswillreflectsintention-to-treat(ITT)parameters.To
theextentthatthereisimperfectcomplianceonthismargin,thesecoefficientscanbe
scaledupbythefractionofmessagesreceived(althoughasshownbelow,thevast
majorityofSMSssentwereactuallyreceived).
Tofurtherimproveprecision,wealsoconsideranalternativeregressionmodel
thataccountsforthefactthatgroupT3effectivelyreceivesbothofthetreatments
providedtogroupsT1andT2:
!! = !′!! + !!!"#_!"#$%&'()$"! + !!!"#$%&"'()%&#)'*! + !!
where!! and!! aredefinedasbefore,!"#_!"#$%&'()$"! isanindicatorforhouseholds
thatareeitheringroupT1orT3,and!"#$%&"'()%&#)'*! isanindicatorforhouseholds
thatareeitheringroupT1orT3.Thus,weestimatetheimpactofparentsreceivingISP
informationregardinginternetusewhethertheyareinT1orT3andtheimpactof
receivinginformationaboutW8parentalcontrolswhetherornottheyareinT2orT3.
15
Notethatthisspecificationdoesrequireustoassumethatthereareno
complementaritiesbetweenthetwoseparatetreatments.Aswillbeseenbelow,wedo
notobserveanysignificanteffectsforT2relativetothecontrolgroupandtheestimated
impactsforT3aresimilartothoseestimatedforT1.Therefore,wethinkthatthe
assumptionunderlyingthisalternativemodelisfairlyinnocuous.
Second,wetakeadvantageofthefactthatwehavedailyinformationoninternet
usetoestimatethedynamiceffectofeachSMS“event”ondailyInternetuseonthedays
immediatelyprecedingandfollowingthedayonwhichthemessagewassent.Westack
alltheeventsforeachsub-treatmentandestimatethefollowingmodel:
!!"! = !!!!!!
!!!!+ !!!!
!
!!!+ !!!"#!! ∗ !"#!
!!
!!!!+ !!!"#!! ∗ !"#!
!
!!!
+ !!!"!! ∗ !"!!!
!!!!+ !!!"!! ∗ !"!
!
!!!+ !! + !!"!
or,withabuseofnotation(becausewearenotexplicitlyexcludingd=-1)
!!"# = !!!!!
!!!!+ !!!"#!! ∗ !"#!
!
!!!!+ !!!"!! ∗ !"!
!
!!!!+ !! + !!"#
whereYandiaredefinedasbefore,dreferstotheday,ereferstotheevent,ISPisa
dummythattakesavalueof1forhouseholdsintheISPinformationgroup,PCisa
dummythattakesavalueof1forhouseholdsintheparentalcontrolsgroup,Drefersto
daydummyvariables,andµe denotetheeventfixedeffects.Thisapproachallowsusto
estimateavectorofcoefficientsthatcapturedifferencesinInternetusewithrespectto
day-1(i.e.,thedaybeforeactuallyreceivingthetreatment)forallthetreatmentgroups.
Forinstanceθ-3measuresthedifferenceinInternetthreedaysbeforereceivingthe
messagewithrespecttothedaybeforethemessagewasreceivedforthecontrolgroup,
(θ-3+θ-3ISP)isthesameeffectforhouseholdsintheISPinformationgroup,and(θ-3+θ-
3PC)fortheparentalcontrolgroup.
16
Wealsoestimatetheimpactofoursub-treatmentsinwhichwevarywhetherthe
SMSsaresentinapredictableorunpredictablefashion.Todothis,weestimatethe
followingregressionmodel:
!! = !′!! + !!"#$%&! + !!
whereRandomiequals1iftheSMSsweresentonarandomdayoftheweekand0ifthe
SMSsweresentonthesamedayofeachweek.Thecoefficient!capturestheimpactof
receivingthemessageonarandomdayrelativetoafixeddayoftheweek.
Finally,weconsideraspecificationthatallowsfortheinteractionofourmain
treatmentsthatprovideISPinformationorparentalcontrolswithoursub-treatments
whichvarywhethertheSMSsaresentinapredictableorunpredictablefashion:
!! = !!!! + !!!"!_!"#$! + !!!"#$%&"'()%&#)'*! + !!!"#$%&!
+ ! !"!_!"#$! ∗ !"#$%&! + ! !"#$%&"'()%&#)'*! ∗ !"#$%&! + !!
Thecoefficientsηandθindicatewhetherprovidinginformationandparentalcontrols
arecomplements(orsubstitutes)withthestrengthorsalienceofthecue.
6. Take-Up
Webeginbyshowingthepatternsoftake-upusingouradministrativedatainTable4.
Columns(1)and(2)confirmthathouseholdswerecorrectlytargetedtoreceiveSMSs
providinginformationaboutinternetusefromtheISPprovider.FromPanelA,thosein
groupsT1andT3receivedapproximately82percentoftheseSMSswhereasthosein
groupT2andthecontrolgroupdidnotreceivethem.Thisisalsoapparentwhenusing
ouralternativeregressionmodelinPanelBtoestimatethecombinedimpactof
providingISPInformationfromT1andT3.Similarly,columns(3)and(4)confirmthat
householdswerecorrectlytargetedtoreceiveSMSsregardingtheWindows8(W8)
parentalcontrolsoftware.ThoseingroupsT2andT3received83percentand81
17
percentoftheseSMSswhilethoseingroupT1andthecontrolgroupdidnotreceive
thematall.9Theimperfectcomplianceintheadministrativedatarepresentscasesin
whichtheSMSmessageswerenotdeliveredduetotechnicalissues(i.e.server
problems,lackofreception,etc.).However,asshowninAppendixTable4,thevast
majorityofparentsreceivedatleastonemessage(98%inthecaseofT1andT2and
97%inthecaseT3).
Finally,columns(5)and(6)ofTable4showthatabout14percentofhouseholds
intreatmentgroupT2and16percentofhouseholdsingroupT3receivedassistance
fromuswithinstallingtheW8parentalcontrolsoftware;again,asexpected,theserates
werezerointreatmentgroupT1andthecontrolgroup.10
WealsoaskedparentsabouttheirrecollectionsofreceivingSMSs,theusefulness
oftheSMSs,andtheirdecisiontoinstallparentalcontrolsoftware.Column(1)ofPanel
AinTable5indicatesnosignificantdifferencesinwhetherparentsrecalledever
receivinganSMSacrossthedifferenttreatmentarmsT1,T2andT3,relativetoabaseof
86percentinthecontrolgroup.Thisisnotsurprisinggiventhatallhouseholdswere
sentaweeklySMS(thoughPanelBdoessuggestthatslightlyfewerparentswho
receivedtheParentalControlinterventionsreporttohaveeverreceivedanSMS).
However,column(2)indicatesthat,amongparentsingroupT2whoonlyreceivedan
SMSregardingW8parentalcontrols,significantlyfewerrememberedwhattheSMS
actuallysaidascomparedtothecontrolgroup.Incontrast,amongparentsingroupsT1
andT3whoreceivedanSMSregardingtheISPinternetuse,significantlymore
rememberedwhattheSMSactuallysaidascomparedtothecontrolgroup.This
9PanelBdoesshowasmallbutsignificanteffectofthecombinedimpactofISPInformationfromT1andT3onthelikelihoodofreceivingSMSsregardingW8parentalcontrolsoftware.Thisisaresultofthesmalldifferencesintake-upbetweenT2andT3.10Again,PanelBshowsasmallbutsignificanteffectofthecombinedimpactofISPInformationfromT1andT3onthelikelihoodofinstallingtheW8parentalcontrolsoftwareasaresultofthesmalldifferencesintake-upbetweenT2andT3.
18
differentialrateofrecallmayalsoexplainsomeofthedifferencesintheimpacts
betweentheIPSandW8interventions.
Columns(3)-(7)showwhether,conditionalonreportingthereceiptofanSMS,
parentsfoundtheSMSsusefulandforspecificpurposesrelativetothecontrolgroup.
Forthemostpart,thepatternsinPanelAandBarenotsurprisingandpresentevidence
thathelpustounderstandthepotentialimpactsofthetreatments.Parentsingroups
T1andT3whoreceivedSMSsregardingtheISPinternetuseweresignificantlymore
likelytofindthesemessagesusefulforbeinginformedaboutinternetuse.Important
forourstudy,whileabout20%ofparentsinthecontrolgroupdiscussedtheSMSswith
theirkids,thispercentagemorethandoublesforparentsingroupsthatreceived
information.Incontrast,parentswhoreceivedinformationabouttheparentalcontrols
SMSswereslightlylesslikelytodiscussthemessageswiththeirkid.
ParentsingroupsT2andT3whoreceivedSMSsabouttheW8parentalcontrols
weresignificantlymorelikelytofindthemusefulforlearningabouttoolsthatwouldbe
helpfultomonitoruse.ButmoreparentsingroupT2foundtheSMSstobeofnouseat
all.Thus,forsomereason,informationabouttheW8softwarewasnotveryusefulto
parents,evenincontrasttothegeneralmessagereceivedbyparentsinthecontrol
group.Wealsofindthat,despitethefactthatalltreatmentarmsalsocontaineda
sentenceremindingparentstoensuretheirchildrenmadegooduseofthecomputer,
fewerhouseholdsintreatmentgroupsT1,T2,andT3reportedthattheirmessageswere
usefulforthispurposeascomparedtothecontrolgroupwhichcontainedonlythis
sentence.Thismaypresentafirstindicationforthelimitedattentionofparents;
includingadditionalcontentintheSMSmayhaveledparentstopaylessattentiontothe
firstpartoftheSMS.
19
Finally,column(8)indicatesthatparentsingroupsT2andT3whoreceived
informationabouttheW8parentalcontrolssoftwareweremorelikelytoinstallit.Yet
thereisalsoevidencethatsomeparentsingroupT1andthecontrolgroupsucceededin
installingparentalcontrolsoftwaredespitenotreceivinganyassistancefromus.
Overall,ouradministrativeandsurveydatasuggestthattheinterventionsworkedas
intendedandthattheactualcontentoftheSMSsdidmatter.Nevertheless,take-up
associatedwithinstallingtheWindows8(W8)parentalcontrolsoftwarewasquitelow.
7. MainResults
Thissectiondescribesourmainresultsontheroleofinformation,parentalcontrols,
andcuesforparentalmonitoringandsupervision.Theanalysisofinformationand
parentalcontrolismainlybasedonTable6andFigure2;thatofcuesisbasedonTable
7andFigure5.Bothtablesarestructuredinasimilarfashiontothetablesshowing
take-up:PanelAdisplaystheimpactseparatelyforeachtreatmentarmT1,T2,andT3
relativetothecontrolgroup;PanelBdisplaysthecombinedimpactofprovidingISP
InformationfromT1andT3aswellasthecombinedeffectofprovidingParental
ControlsoftwarefromT2andT3.Ineachpanel,weshowtheoverallimpactandthe
separateimpactsforweekdaysandweekends.
7.1 Information
7.1.1 Aggregateimpacts
WebeginwithadiscussionoftheaggregateimpactsofprovidingparentswithISP
informationontheintensityofinternetuse.AcrossthedifferentspecificationsinPanel
AofTable6,thereisevidencethathouseholdsingroupT1inwhichparentsreceived
theISPinformationaboutinternetusehadlowerintensityofinternetuseoverthe
20
treatmentperiod.Thedailyreductionof11-16megabytesdownloadedrepresentsa6-
10percentdecreaserelativetothecontrolgroup.Theestimatesaremoreprecisewith
theinclusionofthecontrolvariables,andonlyslightlylargerinmagnitudeforthe
weekendascomparedtoweekdays.TheimpactsforhouseholdsingroupT3,inwhich
parentswereprovidedwithbothinformationaboutinternetuseandhelpinstalling
parentalcontrols,arenegativebutsomewhatsmallerinmagnitudeandlesssignificant
thanthoseforT1.AbroadlysimilarpatternisobservedinPanelBwheretheincreased
precisionyieldsaconsistentlysignificant(combined)impactofprovidingISP
InformationfromT1andT3.
Theseresultssuggestthatprovidingparentswithspecificinformationabout
theirchildren’sinternetusedoesleadtoasignificantreductionof6-10percentin
contemporaneousinternetuse.Thistranslatestoadailyreductionofabout7minutes
ofinternetuseinhouseholdswhichreceivedtheISPinformationintervention(see
AppendixTable5forestimatesintermsofpredictedminutesofuse).
Forthemostpart,theimpactofourinterventionsareverysimilaracross
weekdaysandweekends.Thismaybebecausethepatternsofinternetuseandparental
monitoringdonotvarybetweenweekdaysandweekends.Indeed,wedonotseelarge
differencesininternetuseforthecontrolgroupbetweenweekdaysandweekends;
averageinternetuseis169.4and167.6forweekdaysandweekendsrespectively.
However,itisalsopossiblethattherearecountervailingforcesatplay.Forexample,
children’sdemandforinternetmaybehigherduringweekendsbuttheabilityof
parentstomonitortheirchildren’sinternetusemayalsobecorrespondinglygreater.As
aresult,theequilibriumlevelofusebothwithandwithoutourinterventionsmayend
upbeingquitesimilaracrossweekendsandweekdays.Alternatively,thesimilarityin
treatmenteffectsacrossweekdaysandweekendscanbeaconsequencethatour
21
interventionsaffectedbehaviorthatchangeInternetuseinapermanentwayacrossall
daysoftheweek.
7.1.2 Eventstudy
Wealsoexplorethehigh-frequencydynamicsofourinterventionsbyimplementingan
eventstudyanalysisthatexploitsthetimingofthemessageswithineachweek.The
resultsarepresentedinPanelAofFigure2whichplotscoefficientestimatesforthe
controlgroup(SMS-only),theinformationtreatment(ISPinfo)group,andtheparental
controltreatments(PC).Day0marksthedayonwhichtheSMSsarereceivedeach
week,althoughthemessageswerereceivedintheafternoonsowemayexpectlarger
impactsonthefollowingday(day1).Foreaseofcomparison,wenormalizeallof
coefficientstoequalzeroonthedaypriortoreceiptoftheSMS(day-1).These
coefficientsarealsopresentedinAppendixTable6alongwiththeirstatistical
significance.
Wedonotobserveatrendforanyofthegroupsinthedaysprecedingreceiptof
theSMS(days-3to-1).However,wedoseesignificantdifferenceemergeastheSMSs
arereceivedbythehouseholds.InternetusestartsdecliningfortheISPInfogroupon
day0,declinesfurtheronday1,andremainsbelowtheinternetuseinthepre-
treatmentperiod.Theplotsforthecontrolgroupdoesnotfollowthesamepattern.
Whilethereisasmalldecreaseinday1,thisquicklyrevertstothelevelofthedays
beforethemessagesweresent.Thissupportstheresultsoftheprevioussectionby
confirmingthatitisthereceiptoftheSMSmessagesthemselvesthatleadstoa
discernibleeffectoninternetuse.
Theimpactassociatedwiththeactualcontentofthemessagereceivedbythe
ISP-infogroupisshowninPanelsBandCofFigure2.Inparticular,wesplitthesample
22
betweenthosereceivingamessagestatingthatInternetuseinthepreviousweekwas
“above”themeanofthereferencegroupandthosestatingthatInternetuseinthe
previousweekwasthe“sameorbelow”themean.Theseresultsindicatethatthe
observedeffectsinPanelAaredrivenbythoseSMSmessagescontaining“badnews”for
theparents.Wedonotseeasimilarpatternfortheothertwogroups,whichsuggests
thisisnotexplainedbymean-reversioninInternetuse.Thisfurtherconfirmsthatitis
theactualcontentofthemessagemattersandnotsimplybecauseofthereceiptofthe
SMSmessages.
Tosummarize,theseresultssuggestthatprovidingparentswithinformation
abouttheirchildren’sInternetusehelpstoalleviatetheinformationalasymmetries
withinthefamily.ByhavingacontrolgroupthatalsoreceivesanSMSmessage,wecan
isolatetheimpactofinformationfromthecueassociatedwiththemessageitself.
Moreover,theevidencefromoureventstudyanalysisthatthemessagecontentdrives
theimpactsonlyservestoreinforcetheviewthatitistheinformationitselfwhich
generatesthecausalimpactonInternetuse.
7.2 ParentalControls
Inthissection,weuseTable6andtheeventstudyframeworkinFigure2toconsider
theimpactofofferingassistancewithinstallingparentalcontrolsoftwareonthe
intensityofinternetuse.
LookingatPanelAofTable6,weseenosignificanteffectsforhouseholdsin
groupT2inwhichparentswereprovidedwithinformationaboutinstallingparental
controls;ifanythingthecoefficientsareslightlypositive.Therearealsonosignificant
effectsinPanelBwhereweestimatethe(combined)impactofofferingparentalcontrol
softwarefromT2andT3;thepointestimatesareallclusteredclosetozero.Thus,the
23
aggregatedataindicatesthatofferingparentswithassistancetoinstallparentalcontrol
softwareisnotaneffectivewayofchangingbehavior.Thisconclusionisalsoconfirmed
bytheresultsfromtheeventstudyanalysis.Incontrasttothepatternsobservedforthe
informationaltreatment,Figure2showsnodiscernableimpactofparentalcontrol
interventiononintensityofinternetuseinthedaysimmediatelyfollowingreceiptof
theSMSmessage.
Asafurtherexercise,weconsideranalternativeevent-studyanalysisinwhich
weestimatetheshort-termimpactofactuallyinstallingW8parentalcontrolsoftware.
Sinceweprovidedassistancewithinstallingparentalcontrolsoftwaretofamiliesin
treatmentgroupsT2andT3,weknowtheprecisedateonwhicheachofthe564
parentswhocalledreceivedthisassistance.ThesedatesarestaggeredthroughJanuary
(afterwhichnomorecallswerereceived)whichallowsustoestimateaneventstudy
thatcontrolsforseasonality,similartothoseusedinestimatingtheimpactofreceiving
anSMS.TheresultsofthisanalysisareshowninFigure3andAppendixTable7which
indicatenosignificantshort-termimpactsinthedaysimmediatelyafterinstallationof
theW8parentalcontrolsoftware.Giventhatthedecisiontoinstallparentalcontrol
softwarecouldbeendogenoustointernetuse,thesefindingsneedtobeinterpreted
withcare.However,theyareconsistentwiththelong-termestimatesfromcomparisons
acrossourmaintreatmentgroups.
Theabsenceofsignificantimpactsfromprovidingassistanceforinstalling
parentalcontrolsoftwarecouldindicatethatparentsalreadyhaveaccesstoother
meansofcontrollingtheirchildren’scomputeruse.Thismayalsoexplainthelowrateof
take-upforthisintervention.Alternatively,thelowrateoftake-upcouldreflectthe
considerableobstaclesfacedbylow-incomeparentsinimplementingtechnological
solutionsformonitoringandsupervisingtheirchildren.Asnotedpreviously,parentsin
24
thistreatmentarmweremorelikelytoreportlearningabouttoolsthatcanbehelpfulin
monitoringtheirchildren.Butperhapssuchparentsneedmorehands-onassistanceto
actuallyinstallanduseparentalcontrolsoftwareontheirchildren’scomputers.11
Moreover,installingandoperatingparentalcontrolsoftwarecanimposesubstantial
timecostswhichmayleadtoprocrastination,status-quobias,andotherbiasesthat
arisewiththedemandforcommitmentdevices(seeBryanetal.,2010forareview).
7.3 Cues
Asexplainedabove,ourinterventionsweredesignedtoseparatetheinformational
contentandtheofferofassistancewithparentalcontrolsoftwarefromthecue
associatedwiththeSMSmessages.Thissectionpresentsadditionalevidencesuggesting
thatthecuesthemselvesalsoplayanimportantroleinaffectedparentalbehavior.
First,weuseoureventstudyframeworktoshowthatSMSmessagessenttothe
controlgrouphadshort-termimpactsinthefirstweeksoftheexperiment.Appendix
Table8presentstheimpactsfromtheeventstudyforeachtreatmentgroupduringthe
firstandsecondhalfofthetreatmentperiod.Wewilldiscusstheimplicationsofthese
patternsforthepersistenceofourmaininterventionsinthesubsequentsection.
However,itisnotablethatthereisanegativeandstatisticallysignificantdecreaseof
about10MBsonedayaftertheSMSmessagewasreceivedforthecontrolgroupduring
thefirsthalfofthetreatmentperiod.Thissuggeststhatthesalienceofthemessagealso
matterssincetheSMSmessageswithoutspecificinformationoninternetusearelikely
tobemoresalientatthebeginningoftheexperiment.
11Wehaveexaminedwhichparentalcharacteristicspredicttake-upoftheW8parentalcontrolsoftware.Thestrongestpredictorsarethegenderofthestudent(lesslikelytoinstallforfemales)andthestatedintentiontoinstallparentalcontrolsoftwareinthebaselinesurvey.
25
Second,weconsidertheeffectofvaryingthestrengthofthecueassociatedwith
messagesbysendingthemineitherapredictableorunpredictablefashion.Foreach
treatmentarm,arandomsubsetofhouseholdsreceivedSMSsonthesamedayofthe
week(thisfixeddaywasrandomlydrawnamonghouseholds)whiletheremainderof
householdsreceivedSMSsonarandomdayoftheweek.12Table7examinestheeffectof
receivingSMSsonarandomversusafixedscheduleforallthetreatmentarms
combined.ThereisstrongevidencethathouseholdswhoreceivedSMSsonarandom
schedulehadgreaterreductionsof10-15MBsindailyinternetusethanhouseholdson
fixedschedules.Thisissimilarinmagnitudetothemaineffectassociatedwithreceiving
ISPinformationrelativetothecontrolgroup,andsuggeststhatthestrengthofthecue
associatedtothemessageisasimportantasthemessageitself.13
Webelievethesefindingsareconsistentwithresearchinneuroscienceand
psychologyfindingthatunpredictableandnoveltystimulihavelargerimpacts(e.g.
Parkin,1997;Bernsetal.,2001;Fenkeretal.,2008).Theyarealsorelatedtoresearchin
behavioraleconomicsthatemphasizestheroleofinattentioninthecontextof
reminders(e.g.,Karlan,etal.2014,Taubinsky2014,Ericsson,2017).Onealternative
explanationforthesepatternsisthatrandomschedulesallowformoreflexible
responsesbyparentswhenreceivingamessageisnotasconvenientonsomedays(and
theimpactofrepeatedmessagesisnon-linear).Inthiscase,wewouldexpecttofind
heterogeneoustreatmenteffectsbythedayoftheweekinwhichthemessagewas
12AppendixTable10presentsregressionsoftake-upfortherandom/fixedschedulesubtreatments.Resultsimplysmalldifferencesintake-upusingadministrativedataininstallationofW8inrandomschedulegroupwithintheT2treatmentgroup,andnegativedifferencesinthepercentageofmessagesreceivedagainsttherandomschedulehouseholdsintheT3group.13InAppendixTable11,weestimateaspecificationthatestimatestheeffectofrandomvs.fixedschedulesseparatelyforeachtypeoftreatment.ReceivingSMSsonarandomscheduleleadstonegativeimpactsineverytreatmentarm,althoughitissmallerandinsignificantamonghouseholdsthatreceivedinformationonparentalcontrols.Thefactthatthecontrolgroupalsoshowsdifferentialeffectsbyrandomvs.fixedschedulessuggeststhatthestrengthofthecueprovidedbytherandomschedulealsoaffectsmessageswithoutspecificinformationalcontent.
26
delivered.However,wedonotfindstatisticallysignificantdifferencesacrossdays.A
secondalternativeexplanationisthatparentspaymoreattentiontoSMSssentona
randomschedulebecausethey(mistakenly)interpretthemasbeingsentactivelyby
someonemonitoringtheirinternetuse,whereasSMSmessagessentonafixedschedule
aremorelikelysentthroughanautomatedprocess;inreality,allofthemessageswere
automated.Whilewecannotcompletelyruleoutthispossibility,itisworthnotingthat
thedifferencesbetweenrandomandfixedschedulesareevenlargerinmagnitudefor
thecontrolgroup(asdiscussedinSection8.3),wherethemessagesdonotcontainany
specificinformationabouttheirchildren’sinternetuse.
8. FurtherResults
8.1 Dynamics
Didtheimpactsassociatedwithourinterventionsdisplaydifferentdynamicsduringthe
treatmentperiod?WebeginwithabroadlookatthepatternsovertimeinFigure4
whichshowsthetreatmenteffectsforeachweekinthepre-treatmentperiod,treatment
period,andpost-treatmentperiod(relativetothecontrolgroup).Weobservetheinitial
impactsoftheinformationtreatmentbuildupduringthefirst4weeksoftreatmentand
thenappeartostabilizethroughtherestofthetreatmentperiod.Theseresultsare
confirmedinTable8,whichpresentsthecoefficientestimatesfortheimpactofeach
treatmentforthefirst-halfofthetreatment(fromweeks1to7)andforthesecondhalf
ofthetreatment(weeks8to14).
Furthermore,Figure5presentstheimpactsfromtheeventstudyforthefirst
andsecondhalfofthetreatmentperiod(andthepost-treatmentperiod)forthefixed
27
schedulesub-treatment.14Resultsinbothpanelsindicatethatthedynamiceffectsof
providingISPinformationwerestrongerinthefirst-halfofthetreatmentperiod.
AppendixTable8presentstheeconometricresultscorrespondingtothesetables.They
showsizeabledecreasesininternetuseofapproximately13megabytesonthedaythe
SMSwasreceivedand20megabytesonedayafterreceiptoftheSMS.Incontrast,the
effectsinthesecondhalfofthesample,whilestillnegative,arenotstatistically
significant.
Theseresultsprovidecomplementaryevidenceonwhythecuesassociatedwith
anSMSmessagehaveanimpactinouranalysis.Thedifferentdynamicsofrandom
versusfixedmessagesduringthefirstandsecondpartoftheinterventionareconsistent
withtheviewthattheincreasedstrengthofthecueforrandom-schedulemessages
shouldbemorerelevantinthesecondpartoftheinterventionafterrecipientsonthe
fixed-schedulehavelikelybecomeaccustomedtoreceivingtheirmessagesonthesame
dayeveryweek.
8.2 Persistence
Ifourinterventionsprovidedparentswithnewtoolstoaddressthechallengeof
monitoringandsupervisingtheirchildren,wewouldexpecttheseimpactstopersist.On
theotherhand,ifparentsdependontheSMSsthemselvestohelpthemmonitorand
supervisetheirchildren,theseeffectswouldlikelydisappearwhentheystopreceiving
theirSMSs.Inordertoanswerthisimportantquestion,weanalyzetheimpactofour
treatmentsduringtheperiodaftertheinterventionshadended(i.e.the“post-treatment
period”).
14Wejustuseinformationfortheindividualswhoreceivedmessagesonafixeddayoftheweekbecauseitisnotobvioushowtoshow“placebo”impactsinthepost-treatmentperiodforthesubsampleofindividualswhoreceivedmessagesonarandomdayoftheweek.
28
InFigure4,whichalsoshowsthetreatmenteffectsforeachweekinthe
posttreatmentperiod(relativetothecontrolgroup),weobservethatthetreatment
effectsremainatasimilarlevelevenaftertheinterventionsconcludeinweek14.This
evidencesuggeststhatourmainimpactsdidpersistfollowingthetreatmentperiod.A
similarpictureemergesintheregressionresultspresentedinTable8,whichconfirm
thatintheposttreatmentperiod(weeks15to26)therearesignificantimpactseven
afterthetreatmentends,atroughlythesameorderofmagnitudeasthetreatment
impactsduringthelastweeksofthetreatmentperiod.15Notethatwedonotfindany
significanteffectintheposttreatmentperiodwhentheSMSswerenotactuallyreceived.
Theseresultsshedlightonthemechanismsunderlyingourmainresults.Thefact
thatourmainimpactsdidpersistfollowingthetreatmentperiod,suggeststhatthe
interventionaffectedinternetusenotsimplythroughadirecteffectoftheSMSsthat
mighthelpparentstobettermonitorandsupervisetheirchildren.Rather,thetools
providedbythisinterventionmayhavechangedtheequilibriumuseovertheinitial
interventionperiodand,therefore,ledthetreatmenteffectstopersistevenafterthe
SMSsstopped.
8.3 Interactionsbetweentreatments
WealsoconsidertheinteractionbetweenourmaintreatmentsthatsentSMSs
providingISPInformationaboutinternetuseandreminders/assistanceforinstalling
ParentalControlssoftwarewithoursub-treatmentsthatvariedwhetherthoseSMSs
15Theexperimenttookplaceduringboththevacationperiod(fromDecember,2013toearlyMarch2014)andtheschoolperiod(fromearlyMarchonwards).Thishasanimportantoverlapwiththeanalysesweperforminthissection.AppendixTableA13estimatestreatmenteffectsforthelasttwoweeksofthevacationperiodandthefirsttwoweeksoftheschoolperiodinordertocomparetheeffectofthetreatmentwhileinvacationandwhileinschool,Theseresultssuggestthetreatmenteffectsarenotsubstantiallydifferentforthevacationandschoolperiodand,therefore,weconcludethatthedynamiceffectswepresentinthissectionareprobablyunrelatedtothisalternativeexplanation.
29
werereceivedonarandomorafixedschedule.Theseinteractionseffectsaredisplayed
inTable9forourcombinedtreatmentsandinAppendixTable10fortheseparate
treatments.
Ineithercase,weobservemaineffectsthataresimilartotheonesestimatedin
previoustables:receivingSMSswithISPInformationaboutinternetuse(forthoseona
fixedschedule)leadstosignificantlylowerinternetuse;receivingreminders/assistance
forinstallingParentalControlssoftware(forthosewhoreceivedthemonafixed
schedule)hasanegativebutstatisticallyinsignificantimpactoninternetuse;and
receivingSMSsonarandomscheduleleadstoverylargeandsignificantreductionsin
internetuse.
TheinteractioneffectsbetweenISPinformationandindicatorsforarandom
scheduleareconsistentlypositive,albeitnotsignificant(asimilarpatternholdswith
respecttotheinteractionbetweentherandomscheduleandtheparentalcontrols
treatment).ThissuggeststhatISPinformationandanycueassociatedwitharandom
schedulearesubstitutes,notcomplements.Inotherwords,providingspecific
informationappearstocrowdouttheeffectofthecueassociatedwiththemessage.
9. Conclusion
Parentsareoftenconfrontedwiththechallengeofmonitoringandsupervisingtheir
children’sactions.Thischallengehasbecomeevenmorepressingwiththeincreasing
availabilityofinternetaccessathomewhichmaydisplaceproductiveactivities(and
exposestudentstoinappropriatecontent).Inthepaper,weexaminedtheimportant
roleofasymmetricinformationandthepotentialfordirectparentalcontrolsinaffecting
children’sinternetuse.Theinformationalasymmetriesinthiscontextarelikelytobe
30
especiallypronouncedbecausechildrenareoftenquickertoadapttonewtechnologies
andparentsmaynotbeawareofhowchildrenareusingtechnology.
Wedesignedandimplementedasetofrandomizedexperimentstotestwhether
theintensityofchildren’sinternetuserespondstotheprovisionofspecificinformation
aboutchildren’sinternetuseandtotheofferofassistancewiththeinstallationof
parentalcontrolsoftware.Thesampleincludeschildrenin7thand8thgradewho
receivedfreecomputersand12monthsoffreeinternetsubscriptionsthroughChile’s
“YoElijomiPC”(YEMPC)programin2013,andwetakeadvantageofdetailed
informationontheintensityofinternetuseatthedailylevelfromtheinternetservice
provider(ISP)whichservedallofthecomputersprovidedtothechildreninoursample.
Ourresultsshowthatprovidingparentswithinformationabouttheirchildren’s
internetuseleadstosubstantialreductionsinuse:householdsinwhichparents
receivedISPinformationaboutinternetusehadsignificantlylowerintensityofinternet
useduringthetreatmentperiodascomparedtohouseholdsinthecontrolgroup.We
observestatisticallysignificantreductionsinusepreciselyonthedaysimmediately
afterreceivingtheISPinformation.Furthermore,itisthoseSMSmessagesindicating
thatchildrenusedmoreinternetthanthereferencegroupinaspecificweek,which
producethelargestdeclinesininternetuse.Wefindnoimpactofreceivingassistance
withtheinstallationofparentalcontrolsoftwareontheintensityofinternetuse.
Moreover,wedonotobserveshort-termimpactsofactuallyinstallingparentalcontrol
softwareamongthefamiliesthatreceivedassistance.
Takentogether,thesefindingsindicatethatprovidingparentswithspecific
informationabouttheirchildren’sinternetuseaffectbehaviorwhileprovidingparents
withparentalcontrolsoftwaredoesnot.Thefactthattheimpactsofinformationeffects
persistaftertreatmentendssuggeststhatourtemporaryinterventionmayhavealtered
31
theequilibriumofinternetuseandhelpedparentssolvetheproblemofasymmetric
informationinamorepermanentway.
WealsofindstrongevidencethathouseholdswhoreceivedSMSswithan
unpredictablescheduleexperiencedsignificantlygreaterreductionsininternetuse
thanthoseonpredictableschedules,aneffectsimilarinmagnitudetothemaineffect
associatedwithreceivingtheISPinformation.Inaddition,wefindthattheSMS
messagessenttothecontrolgrouphadshort-termimpactsonInternetuseinthefirst
weeksoftheexperiment,perhapsduetothenoveltyofthemessage.Thesefindings
suggestthatthecuesassociatedwithmessageshaveanindependenteffectonbehavior
andthatthestrengthofsuchcuesisanimportantdeterminantofouroutcomes.Thus,
ourstudyshedslightontheroleofinformationandcuesinaffectingbehavior.
32
References
Alcott,HuntandToddRogers(2014)TheShort-RunandLong-RunEffectsofBehavioralInterventions:ExperimentalEvidencefromEnergyConservation”AmericanEconomicReview104(10):3003–3037
AmericanAcademyofPediatrics(2016)“PolicyStatement:MediaUseinSchool-AgedChildrenandAdolescents”,Pediatrics138(5).
Becker,GaryS.(1974)"ATheoryofSocialInteractions."JournalofPoliticalEconomy82:1063-94.
Becker,GaryS.(1981).ATreatiseontheFamily.Cambridge,Mass.:HarvardUniv.Press.
Bergman,Peter(2016)“Parent-ChildInformationFrictionsandHumanCapitalInvestment:EvidencefromaFieldExperimentInvestment”.Workingpaper.
Bergstrom,TheodoreC.1989.“AFreshLookattheRottenKidTheorem—andOtherHouseholdMysteries.”JournalofPoliticalEconomy97(5):1138–59.
Berlinski,Samuel,Busso,Matias,Dinkelman,TarynandClaudiaMartinez(2016)“PapásalDía:Usinghighfrequencytextmessagingtoreduceparent-schoolinformationgapsandimproveschooloutcomes”Workingpaper.
Berry,James(2015)ChildControlinEducationDecisions:AnEvaluationofTargetedIncentivestoLearninIndia.JournalofHumanResources50(4):1051-1080
Berns,Gregoryetal.(2001)“PredictabilityModulatesHumanBrainResponsetoReward”TheJournalofNeuroscience21(8):2793–2798
Beuermann,Diether,JulianCristia,SantiagoCueto,OferMalamud,andYyannuCruz-Aguayo.2015.“OneLaptopperChildatHome:Short-TermImpactsfromaRandomizedExperimentinPeru.”AmericanEconomicJournal:AppliedEconomics7(2):1-29.
Bryan,G.,D.Karlan,S.Nelson(2010)“CommitmentDevices”AnnualReviewofEconomics2:671-698
Bursztyn,LeonardoandLucasCoffman(2012).“TheSchoolingDecision:FamilyPreferences,IntergenerationalConflict,andMoralHazardintheBrazilianFavelas,”JournalofPoliticalEconomy120(3):359-397
Castleman,BenjaminL.(2015)The160-CharacterSolution:HowTextMessagingandOtherBehavioralStrategiesCanImproveEducation.JohnsHopkinsUniversityPress
Dizon-Ross,Rebecca(2016),“Parents’BeliefsandCHildren’sEducation:ExperimentalEvidencefromMalwai,”Workingpaper.
Ericson,K.M.(2017)“OntheInteractionofMemoryandProcrastination:ImplicationsforReminders,Deadlines,andEmpiricalEstimation”JournaloftheEuropeanEconomicAssociation,15(3):692–719
33
Fairlie,Robert,andJonathanRobinson.2013.“ExperimentalEvidenceontheEffectsofHomeComputersonAcademicAchievementamongSchoolchildren.”AmericanEconomicJournal:AppliedEconomics5(3):211–240.
Fenker,DanielaB.JuliettaU.Frey,HartmutSchuetze,DorotheeHeipertz,Hans-JochenHeinze,andEmrahDuzel(2008)“NovelScenesImproveRecollectionandRecallofWords”JournalofCognitiveNeuroscience20(7):1250–1265
FersterCBandBF.Skinner.(1957)SchedulesofReinforcement.EnglewoodCliffs,NJ:Prentice-Hall
Finan,FredericoandClaudioFerraz(2011)“ElectoralAccountabilityandCorruptioninLocalGovernments:EvidencefromAuditReports”.AmericanEconomicReview101:1274-1311.
Gallego,Francisco,DavidPreiss,andAlejandroSaenz(2017)“Fromaccesstouse:DeterminantsofInternetuseinastatesponsoredfreeInternetprogramforchildren”.Manuscript.
Jensen,Robert.2010.“The(Perceived)ReturnstoEducationandtheDemandforSchooling,”QuarterlyJournalofEconomics125(2):515–48.
Karlan,Dean,MargaretMcConnell,SendhilMullainathan,JonathanZinman(2014)“GettingtotheTopofMind:HowRemindersIncreaseSaving”ManagementScience62(12):3393-3411
Malamud,O.andC.Pop-Eleches,C.2011.“HomeComputerUseandtheDevelopmentofHumanCapital.”QuarterlyJournalofEconomics126:987–1027.
Parkin,AlanJ.(1997)“Humanmemory:Novelty,associationandthebrain”CurrentBiology7:768–769
Pop-ElechesC,ThirumurthyH,HabyarminaJ,GraffZivinJ,GoldsteinM,DeWalqueD,MacKeenL,HabererJ,SidleJ,NgareD,BangsbergD(2011)“Mobilephonetechnologiesimproveadherencetoantiretroviraltreatmentinresource-limitedsettings:arandomizedcontrolledtrialoftextmessagereminders”.AIDS25(6)
Taubinsky,D.(2014)“FromIntentionstoActions:AModelandExperimentalEvidenceofInattentiveChoice”Mimeo.
Vigdor,J.,andH.Ladd.,ErikaMartinez2014.“ScalingtheDigitalDivide:HomeComputerTechnologyandStudentAchievement.”EconomicInquiry52(3):1103-1119.
Weinberg,Bruce(2001),“AnIncentiveModeloftheEffectofParentalIncomeonChildren,”JournalofPoliticalEconomy109(2):266-280
Table 1: Sample Size by Treatment and Subtreatment
Treatment Fixed RandomGroup Day Day Total
T1 ISP 963 964 1927T2 W8 965 963 1928T3 ISP + W8 962 962 1924T0 SMS-only 964 964 1928
Total 3853 3854 7707
Notes: The sample was stratified by Guardian’s education (No High-School, High School, College), Parent perception of whether the stu-dent stays too long in front of the computer (Yes or No) and InternetUse as the total MB downloaded between September and December(the 15th).
Table 2: Summary Statistics
Mean S.D.
Panel A: Student CharacteristicsDaily mean download, pre 153.51 179.77Average daily use in minutes, predicted, pre 186.37 162.81Live with mother 0.96 0.20Live with father 0.62 0.49Live with Brother/Sister 0.76 0.42Live with Grandfather/Grandmother 0.15 0.36Female 0.43 0.49Number of siblings 1.72 1.28
Panel B: Guardian CharacteristicsGuardian Age 40.42 7.78What is your education level?
Elementary incomplete 0.10 0.30Elementary complete 0.14 0.35Secondary incomplete 0.15 0.36Secondary complete 0.47 0.50High incomplete 0.04 0.20High complete 0.09 0.29
What is your current employment status?Working full time 0.33 0.47Working part-time 0.13 0.33Not working looking for a job 0.06 0.23Not working not looking for a job 0.47 0.50Other 0.02 0.14
Notes: This table presents estimated means (Column 1) and stan-dard deviations (Column 2) for students included in the experimentalsample.
Table 3: Balance by Treatment
(1) (2) (3) (4) (5)T1 T2 T3 SMS-Only P-Value(F-Test)
Panel A: Student CharacteristicsDaily mean download, pre 153.75 153.96 153.12 153.36 0.999
( 180.83) ( 177.29) ( 174.63) ( 183.42)Average daily use in minutes, predicted, pre 185.26 188.04 188.17 184.60 0.862
( 161.91) ( 164.42) ( 164.60) ( 160.53)Live with mother 0.95 0.96 0.97 0.95 0.085
( 0.21) ( 0.19) ( 0.18) ( 0.21)Live with father 0.62 0.62 0.61 0.62 0.931
( 0.49) ( 0.49) ( 0.49) ( 0.49)Live with Brother/Sister 0.76 0.78 0.78 0.75 0.112
( 0.43) ( 0.42) ( 0.41) ( 0.43)Live with Grandfather/Grandmother 0.16 0.15 0.15 0.15 0.766
( 0.37) ( 0.36) ( 0.36) ( 0.35)Female 0.42 0.42 0.42 0.45 0.361
( 0.49) ( 0.49) ( 0.49) ( 0.50)Number of siblings 1.69 1.73 1.73 1.74 0.625
( 1.25) ( 1.30) ( 1.31) ( 1.28)
Panel B: Guardian CharacteristicsGuardian Age 40.29 40.64 40.49 40.50 0.587
( 7.82) ( 7.98) ( 7.92) ( 7.67)What is your education level?
Elementary incomplete 0.09 0.10 0.10 0.10 0.961( 0.29) ( 0.30) ( 0.30) ( 0.30)
Elementary complete 0.13 0.14 0.14 0.15 0.773( 0.34) ( 0.35) ( 0.35) ( 0.35)
Secondary incomplete 0.16 0.15 0.15 0.15 0.691( 0.37) ( 0.36) ( 0.36) ( 0.36)
Secondary complete 0.47 0.47 0.47 0.47 0.994( 0.50) ( 0.50) ( 0.50) ( 0.50)
High incomplete 0.04 0.04 0.05 0.04 0.329( 0.19) ( 0.21) ( 0.22) ( 0.20)
High complete 0.10 0.09 0.09 0.09 0.727( 0.30) ( 0.29) ( 0.28) ( 0.29)
What is your current employment status?Working full time 0.33 0.33 0.34 0.32 0.707
( 0.47) ( 0.47) ( 0.47) ( 0.47)Working part-time 0.13 0.14 0.13 0.13 0.758
( 0.33) ( 0.34) ( 0.33) ( 0.34)Not working looking for a job 0.05 0.05 0.06 0.06 0.570
( 0.22) ( 0.22) ( 0.23) ( 0.24)Not working not looking for a job 0.47 0.46 0.46 0.47 0.790
( 0.50) ( 0.50) ( 0.50) ( 0.50)Other 0.02 0.02 0.02 0.02 0.575
( 0.14) ( 0.15) ( 0.14) ( 0.13)
Note: This table presents estimated di↵erences between students in the di↵erent experimental groups. Columns1 to 4 present means and stadard deviations in parentheses. Column 5 presents the p-value of a of joint testfor di↵erences between the T1, T2 and T3 and SMS-only groups.
Table 4: Take-up: using Administrative Data
(1) (2) (3) (4) (5) (6)SMS ISP SMS ISP SMS W8 SMS W8 W8 installed W8 installed
Panel A: T1, T2, T3T1 0.821*** 0.820*** 0.000 -0.000 -0.001 -0.000
(0.006) (0.006) (0.005) (0.005) (0.008) (0.008)T2 -0.000 -0.001 0.832*** 0.832*** 0.135*** 0.135***
(0.006) (0.006) (0.005) (0.005) (0.008) (0.008)T3 0.816*** 0.816*** 0.815*** 0.815*** 0.156*** 0.157***
(0.006) (0.006) (0.005) (0.006) (0.008) (0.008)
Observations 7,707 7,707 7,707 7,707 7,707 7,707Control Mean 0 0 0 0 0 0BL Controls X X X
Panel B: ISP Info and Parental ControlsISP Information 0.818*** 0.818*** -0.008** -0.008** 0.011* 0.011*
(0.004) (0.004) (0.004) (0.004) (0.006) (0.006)Parental Controls -0.002 -0.003 0.824*** 0.823*** 0.146*** 0.146***
(0.004) (0.004) (0.004) (0.004) (0.006) (0.006)
Observations 7,707 7,707 7,707 7,707 7,707 7,707Control Mean 0 0 0 0 0 0BL Controls X X X
Note: This table presents estimated e↵ects on take-up for di↵erent treatment groups with respect to the control group. Columns 1 and 2 presentestimates on the reception of SMSs including ISP information. Columns 3 and 4 present estimates on the reception of SMSs including an o↵erof help to install parental control settings. Columnd 5 and 6 present estimates on the installation of parental control settings through the callcenter of the experiment. Odd-numbered columns present estimates without controls and even-numbered columns present estimates includingbaseline control variables. Control variables include the baseline values of mean of MBs of Internet use; guardian gender, age, educationlevel and employment status; number of siblings; and dummies for family composition (indicating whether the child lives with mother, father,step-mother or father’s partner, step-father or mother’s partner, uncle/aunt, brother/sister, grandfather/grandmother, other relatives, andother non-relatives). Robust estimated standard errors are reported in parentheses. *** Significant at the 1 percent level. ** Significant at the5 percent level. * Significant at the 10 percent level.
Tab
le5:
Tak
e-up:
usi
ng
Surv
eyD
ata
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Hav
eyou
ever
What
are
you
usi
ng
the
info
rmat
ion
inth
eSM
Sfo
r?(c
ond.
on
rece
ivin
g)re
ceiv
edan
SM
Sin
Do
you
rem
ember
Dis
cuss
wit
hB
ein
form
edG
etin
form
edabout
Be
rem
inded
toN
ouse
Hav
eyou
ever
your
Phone
from
what
the
your
child
about
the
fam
ily
tools
that
can
be
ensu
rea
good
use
inst
alled
pare
nta
l
YE
MP
Cpro
gra
m?
SM
Ssa
id?
inte
rnet
use
hel
pfu
lto
monit
or
ofth
eco
mpute
rco
ntr
olso
ftw
are
Panel
A:T
1,T
2,T
3T
10.0
16
0.1
15**
*0.
268*
**0.
073*
**-0
.010
-0.3
32*
**0.0
12
0.02
8**
(0.0
16)
(0.0
20)
(0.0
22)
(0.0
16)
(0.0
16)
(0.0
23)
(0.0
21)
(0.0
12)
T2
-0.0
26-0
.158
***
-0.0
17-0
.028
*0.
077*
**-0
.216*
**0.1
74**
*0.
096
***
(0.0
16)
(0.0
20)
(0.0
22)
(0.0
17)
(0.0
17)
(0.0
23)
(0.0
21)
(0.0
12)
T3
-0.0
140.0
54**
*0.
220*
**0.
054*
**0.
030*
-0.2
91**
*0.
036*
0.1
09*
**(0
.016)
(0.0
20)
(0.0
22)
(0.0
16)
(0.0
17)
(0.0
23)
(0.0
21)
(0.0
12)
Obse
rvation
s3,
959
3,36
33,
275
3,27
53,
275
3,27
53,
275
3,8
49C
ontr
olM
ean
0.864
0.76
10.
196
0.10
40.
110
0.5
870.
190
0.0
258
Panel
B:IS
PIn
foand
Pare
nta
lC
ontrols
ISP
Info
rmat
ion
0.014
0.16
2***
0.25
3***
0.07
7***
-0.0
28**
-0.2
08**
*-0
.060
***
0.021
**(0
.011)
(0.0
14)
(0.0
16)
(0.0
12)
(0.0
12)
(0.0
16)
(0.0
15)
(0.0
09)
Par
enta
lC
ontr
ols
-0.0
28*
*-0
.108*
**-0
.033
**-0
.023
**0.
058*
**-0
.085*
**0.
098**
*0.0
89**
*(0
.011)
(0.0
14)
(0.0
16)
(0.0
12)
(0.0
12)
(0.0
16)
(0.0
15)
(0.0
09)
Obse
rvation
s3,
959
3,36
33,
275
3,27
53,
275
3,27
53,
275
3,8
49C
ontr
olM
ean
0.864
0.76
10.
196
0.10
40.
110
0.5
870.
190
0.0
258
Note
:T
his
table
pre
sents
estim
ate
de↵
ects
on
take-
up
for
di↵
eren
ttr
eatm
ent
gro
ups
wit
hre
spec
tto
the
contr
olgro
up.
Robust
esti
mate
dst
andard
erro
rsare
report
edin
pare
nth
eses
.***
Sig
nifi
cant
at
the
1per
cent
level
.**
Sig
nifi
cant
at
the
5per
cent
level
.*
Sig
nifi
cant
at
the
10
per
cent
level
.
Table 6: Impact of Treatments on Intensity of Internet Use
(1) (2) (3) (4) (5) (6)Period: All All Weekdays Weekdays Weekend Weekend
Panel A: T1, T2, T3T1 -12.422 -13.178* -11.154 -11.982* -15.658* -16.230**
(7.978) (7.086) (8.115) (7.215) (8.279) (7.471)T2 0.058 -0.578 1.281 0.569 -3.062 -3.504
(8.218) (7.512) (8.232) (7.562) (8.800) (8.060)T3 -9.594 -11.744 -9.865 -12.143* -8.902 -10.723
(7.782) (7.148) (7.785) (7.168) (8.450) (7.824)
Observations 7,707 7,707 7,707 7,707 7,707 7,707Control Mean 168.9 168.9 169.4 169.4 167.6 167.6BL Controls X X X
Panel B: ISP Information and Parental ControlISP Information -11.038** -12.172** -11.150** -12.348** -10.751* -11.724**
(5.439) (4.848) (5.498) (4.893) (5.716) (5.200)Parental Controls 1.443 0.429 1.285 0.204 1.844 1.001
(5.469) (4.855) (5.527) (4.906) (5.748) (5.191)
Observations 7,707 7,707 7,707 7,707 7,707 7,707Control Mean 168.9 168.9 169.4 169.4 167.6 167.6BL Controls X X X
Note: This table presents estimated e↵ects on Internet use measured as daily average MBs uploaded and down-loaded. Odd-numbered columns present estimates without controls and even-numbered columns present estimatesincluding baseline control variables. Control variables include the baseline values of mean of MBs of Internetuse; guardian gender, age, education level and employment status; number of siblings; and dummies for familycomposition (indicating whether the child lives with mother, father, step-mother or father’s partner, step-father ormother’s partner, uncle/aunt, brother/sister, grandfather/grandmother, other relatives, and other non-relatives).Robust estimated standard errors are reported in parentheses. *** Significant at the 1 percent level. ** Significantat the 5 percent level. * Significant at the 10 percent level.
Table 7: Impact of Random on Intensity of Internet Use
(1) (2) (3) (4) (5) (6)Period: All All Weekdays Weekdays Weekend Weekend
Random -14.530*** -12.401** -15.294*** -13.123*** -12.582** -10.557**(5.458) (4.873) (5.516) (4.921) (5.735) (5.216)
Observations 7,707 7,707 7,707 7,707 7,707 7,707BL Controls X X X
Note: This table presents estimated e↵ects of the random sub-treatment (with respect to the fixed sub-treatmentgroup) on Internet use measured as daily average MBs uploaded and downloaded. Odd-numbered columns presentestimates without controls and even-numbered columns present estimates including baseline control variables.Control variables include the baseline values of mean of MBs of Internet use; guardian gender, age, education leveland employment status; number of siblings; and dummies for family composition (indicating whether the child liveswith mother, father, step-mother or father’s partner, step-father or mother’s partner, uncle/aunt, brother/sister,grandfather/grandmother, other relatives, and other non-relatives). Robust estimated standard errors are reportedin parentheses. *** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10percent level.
Table 8: Impact of Treatments on Intensity of Internet Use Across Periods
(1) (2) (3) (4) (5) (6)Period: All All Weekdays Weekdays Weekend Weekend
Panel A: 1st Half of TreatmentISP Information -11.096** -12.286*** -9.724* -10.957** -14.822*** -15.894***
(4.947) (4.386) (5.016) (4.447) (5.335) (4.853)Parental Controls 4.594 3.667 4.110 3.101 5.908 5.202
(4.974) (4.405) (5.044) (4.470) (5.357) (4.865)
Observations 7,707 7,707 7,707 7,707 7,707 7,707Control Mean 163.8 163.8 166.2 166.2 157.3 157.3BL Controls X X X
Panel B: 2nd Half of TreatmentISP Information -10.978 -12.055* -12.655* -13.816** -6.952 -7.831
(6.912) (6.375) (7.057) (6.505) (7.287) (6.827)Parental Controls -1.771 -2.873 -1.696 -2.854 -1.949 -2.920
(6.946) (6.383) (7.091) (6.523) (7.326) (6.819)
Observations 7,707 7,707 7,707 7,707 7,707 7,707Control Mean 174 174 172.7 172.7 177.1 177.1BL Controls X X X
Panel C: Post-TreatmentISP Information -11.273** -12.126*** -11.101** -12.007*** -11.710** -12.428**
(4.905) (4.452) (4.847) (4.383) (5.400) (4.999)Parental Controls 0.032 -0.717 -0.621 -1.436 1.694 1.112
(4.921) (4.455) (4.863) (4.387) (5.418) (5.002)
Observations 7,707 7,707 7,707 7,707 7,707 7,707Control Mean 149.5 149.5 147.2 147.2 155.3 155.3BL Controls X X X
Note: This table presents estimated e↵ects on Internet use measured as daily average MBs uploaded and down-loaded. Odd-numbered columns present estimates without controls and even-numbered columns present estimatesincluding baseline control variables. Control variables include the baseline values of mean of MBs of Internetuse; guardian gender, age, education level and employment status; number of siblings; and dummies for familycomposition (indicating whether the child lives with mother, father, step-mother or father’s partner, step-father ormother’s partner, uncle/aunt, brother/sister, grandfather/grandmother, other relatives, and other non-relatives).Robust estimated standard errors are reported in parentheses. *** Significant at the 1 percent level. ** Significantat the 5 percent level. * Significant at the 10 percent level.
Table 9: Interactions of Treatments with Random
(1) (2) (3) (4) (5) (6)Period: All All Weekdays Weekdays Weekend Weekend
ISP Info -12.256 -15.980** -11.470 -15.291** -14.263* -17.739**(8.266) (7.286) (8.472) (7.442) (8.369) (7.588)
PC -7.149 -5.672 -7.584 -6.117 -6.039 -4.536(8.279) (7.278) (8.488) (7.444) (8.376) (7.554)
Random -24.341** -22.340** -24.484** -22.419** -23.976** -22.140**(10.114) (9.293) (10.141) (9.367) (10.893) (10.023)
ISP Info ⇥ Random 2.428 7.608 0.631 5.878 7.014 12.020(10.911) (9.677) (11.030) (9.773) (11.460) (10.373)
PC ⇥ Random 17.193 12.259 17.749 12.701 15.775 11.133(10.908) (9.613) (11.028) (9.709) (11.456) (10.312)
Observations 7,707 7,707 7,707 7,707 7,707 7,707BL Controls X X X
Note: This table presents estimated e↵ects on Internet use measured as daily average MBs uploaded and down-loaded. Odd-numbered columns present estimates without controls and even-numbered columns present estimatesincluding baseline control variables. Control variables include the baseline values of mean of MBs of Internetuse; guardian gender, age, education level and employment status; number of siblings; and dummies for familycomposition (indicating whether the child lives with mother, father, step-mother or father’s partner, step-father ormother’s partner, uncle/aunt, brother/sister, grandfather/grandmother, other relatives, and other non-relatives).Robust estimated standard errors are reported in parentheses. *** Significant at the 1 percent level. ** Significantat the 5 percent level. * Significant at the 10 percent level.
Figure 1: Timeline
Figure 2: Event Study
-30
-25
-20
-15
-10
-50
510
15
-3 -2 -1 0 1 2 3day
All Events
-30
-25
-20
-15
-10
-50
510
15
-3 -2 -1 0 1 2 3day
Above Events
-30
-25
-20
-15
-10
-50
510
15
-3 -2 -1 0 1 2 3day
Not-Above Events
ISP Info PCSMS-Only
Notes:This figure presents estimated e↵ects for each treatment group in day around the reception of an SMS message.Day 0 marks the day on which the SMSs are received each week. See Appendix Table 6 for more details.
Figure 3: W8 Install Event Study, controlling for seasonality
-70
-40
-10
2050
80M
B U
se-3 -2 -1 0 1 2 3
Day
Parental Controls
-70
-40
-10
2050
80M
B U
se
-3 -2 -1 0 1 2 3Day
T2
-70
-40
-10
2050
80M
B U
se
-3 -2 -1 0 1 2 3Day
T3
MB Use
Notes:This figure presents estimated e↵ects for the installation of the W8 parental control settings on Internet use,measured using MBs downloaded and uploaded. Day 0 marks the day on which the program was installed. SeeAppendix Table 7 for more details.
Figure 4: Impact on MB Use Across Weeks, with controls
-60
-40
-20
020
Impa
ct
-20 -10 0 10 20 30Week
ISP Information
-40
-20
020
40Im
pact
-20 -10 0 10 20 30Week
Parental Controls
Notes: This table presents estimated e↵ects (and confidence intervals) for each treatment group (with respect to thecontrol group) for each week of the experiment on Internet use measured as daily average Control variables includethe baseline values of mean of MBs of Internet use; guardian gender, age, education level and employment status;number of siblings; and dummies for family composition (indicating whether the child lives with mother, father, step-mother or father’s partner, step-father or mother’s partner, uncle/aunt, brother/sister, grandfather/grandmother,other relatives, and other non-relatives).
Figure 5: Event Study, Post-Treatment for Fixed Subgroup
-30
-25
-20
-15
-10
-50
510
15
-3 -2 -1 0 1 2 3day
1st Half of Treatment
-30
-25
-20
-15
-10
-50
510
15
-3 -2 -1 0 1 2 3day
2nd Half of Treatment
-30
-25
-20
-15
-10
-50
510
15
-3 -2 -1 0 1 2 3day
Post Treatment
ispinfo_fixed_1half pc_fixed_1halfsmsonly_fixed_1half
Notes: This figure presents estimated e↵ects for each treatment group in day around the reception of an SMSmessage. Day 0 marks the day on which the SMSs are received each week. See Appendix Table 11 for more details.
Table A1: Sample Comparisons
Sample: Full Experiment Survey
Panel A: Student CharacteristicsDaily mean download, pre 157.49 153.55 155.18
( 183.55) ( 179.04) ( 185.56)Average daily use in minutes, predicted, pre 189.51 186.52 185.67
( 164.21) ( 162.85) ( 162.49)Female 0.41 0.43 0.43
( 0.49) ( 0.49) ( 0.49)ADHD 14.21 14.48 14.52
( 3.94) ( 4.14) ( 4.20)Number of siblings 1.77 1.72 1.69
( 1.32) ( 1.28) ( 1.26)Rural School 0.15 0.15 0.14
( 0.36) ( 0.35) ( 0.35)Computer Skills 8.69 8.72 8.75
( 2.06) ( 2.02) ( 2.00)
Panel B: Mother CharacteristicsMother’s Education Level
Elementary incomplete 0.12 0.10 0.10( 0.32) ( 0.30) ( 0.29)
Elementary complete 0.14 0.14 0.14( 0.35) ( 0.34) ( 0.34)
Secondary incomplete 0.16 0.15 0.15( 0.36) ( 0.36) ( 0.36)
Secondary complete 0.45 0.47 0.48( 0.50) ( 0.50) ( 0.50)
High incomplete 0.03 0.04 0.04( 0.18) ( 0.19) ( 0.19)
High complete 0.06 0.06 0.06( 0.23) ( 0.23) ( 0.24)
Mother’s Employment StatusEmployed 0.34 0.32 0.30
( 0.47) ( 0.47) ( 0.46)Unemployed 0.04 0.04 0.04
( 0.21) ( 0.20) ( 0.20)Home-maker 0.58 0.61 0.63
( 0.49) ( 0.49) ( 0.48)Retired 0.00 0.00 0.00
( 0.07) ( 0.06) ( 0.07)
Panel C: Household CharacteristicsRules Index 4.02 4.06 4.08
( 1.11) ( 1.08) ( 1.07)Student has Computer at Home 0.40 0.41 0.42
( 0.49) ( 0.49) ( 0.49)Student has Internet at Home 0.24 0.25 0.25
( 0.43) ( 0.43) ( 0.43)
Sample Size 29833 7707 5001
Note: This table presents estimated means for students in the YEMPC program who applied for a computer withInternet connection (Column 1), students in the experimental sample (Column 2), and students in the sample ofthe phone survey (Column 3). Estimated standard deviations are reported in parentheses.
Table A2: Balance by Treatment
(1) (2) (3)Random Fixed P-Value
Panel A: Student CharacteristicsDaily mean download, pre 153.47 153.54 0.970
( 177.62) ( 181.21)Average daily use in minutes, predicted, pre 186.12 186.54 0.832
( 162.75) ( 162.87)Live with mother 0.96 0.96 0.497
( 0.20) ( 0.19)Live with father 0.61 0.62 0.113
( 0.49) ( 0.48)Live with Brother/Sister 0.76 0.77 0.296
( 0.43) ( 0.42)Live with Grandfather/Grandmother 0.15 0.15 0.444
( 0.36) ( 0.36)Female 0.43 0.42 0.757
( 0.49) ( 0.49)Number of siblings 1.71 1.72 0.519
( 1.28) ( 1.27)
Panel B: Parents CharacteristicsGuardian Age 40.58 40.30 0.262
( 8.15) ( 7.52)What is your education level?
Elementary incomplete 0.10 0.10 0.443( 0.30) ( 0.30)
Elementary complete 0.14 0.14 0.866( 0.35) ( 0.35)
Secondary incomplete 0.15 0.16 0.525( 0.36) ( 0.36)
Secondary complete 0.47 0.47 0.990( 0.50) ( 0.50)
High incomplete 0.04 0.05 0.072( 0.19) ( 0.21)
High complete 0.10 0.09 0.162( 0.30) ( 0.29)
What is your current employment status?Working full time 0.33 0.32 0.374
( 0.47) ( 0.47)Working part-time 0.13 0.13 0.342
( 0.33) ( 0.34)Not working looking for a job 0.06 0.06 0.515
( 0.23) ( 0.23)Not working not looking for a job 0.46 0.48 0.289
( 0.50) ( 0.50)Other 0.02 0.02 0.049
( 0.15) ( 0.13)
Note: This table presents estimated di↵erences between students in the di↵erent experimentalgroups. Columns 1 and 2 present means and stadard deviations in parentheses. Column 3 presentsthe p-value of a t-test for di↵erences between the Random day and Fixed day groups.
Table A3: Balance by Treatment for the FU Survey Sample
(1) (2) (3) (4) (5)T1 T2 T3 SMS-Only P-Value(F-Test)
Panel A: Student CharacteristicsDaily mean download, pre 149.38 158.72 151.25 163.15 0.305
( 177.19) ( 184.23) ( 171.92) ( 205.38)Average daily use in minutes, predicted, pre 180.47 191.39 185.34 187.61 0.505
( 159.27) ( 166.27) ( 162.66) ( 162.93)Live with mother 0.96 0.97 0.97 0.95 0.041
( 0.19) ( 0.17) ( 0.17) ( 0.22)Live with father 0.65 0.65 0.64 0.62 0.345
( 0.48) ( 0.48) ( 0.48) ( 0.49)Live with Brother/Sister 0.77 0.79 0.80 0.74 0.030
( 0.42) ( 0.41) ( 0.40) ( 0.44)Live with Grandfather/Grandmother 0.16 0.14 0.14 0.15 0.637
( 0.37) ( 0.35) ( 0.35) ( 0.36)Female 0.42 0.42 0.42 0.46 0.205
( 0.49) ( 0.49) ( 0.49) ( 0.50)Number of siblings 1.62 1.75 1.74 1.71 0.109
( 1.20) ( 1.30) ( 1.33) ( 1.28)
Panel B: Guardian CharacteristicsGuardian Age 40.54 40.47 40.54 40.85 0.693
( 7.62) ( 7.26) ( 7.79) ( 7.77)What is your education level?
Elementary incomplete 0.09 0.09 0.10 0.09 0.856( 0.28) ( 0.29) ( 0.30) ( 0.29)
Elementary complete 0.12 0.14 0.15 0.16 0.139( 0.33) ( 0.35) ( 0.36) ( 0.36)
Secondary incomplete 0.17 0.16 0.14 0.16 0.315( 0.37) ( 0.37) ( 0.34) ( 0.37)
Secondary complete 0.49 0.48 0.48 0.46 0.724( 0.50) ( 0.50) ( 0.50) ( 0.50)
High incomplete 0.03 0.04 0.04 0.03 0.290( 0.17) ( 0.20) ( 0.21) ( 0.18)
High complete 0.11 0.09 0.09 0.09 0.543( 0.31) ( 0.29) ( 0.29) ( 0.29)
What is your current employment status?Working full time 0.31 0.30 0.31 0.29 0.632
( 0.46) ( 0.46) ( 0.46) ( 0.45)Working part-time 0.12 0.14 0.13 0.14 0.510
( 0.33) ( 0.34) ( 0.33) ( 0.35)Not working looking for a job 0.05 0.05 0.06 0.05 0.916
( 0.22) ( 0.23) ( 0.23) ( 0.22)Not working not looking for a job 0.50 0.49 0.48 0.50 0.809
( 0.50) ( 0.50) ( 0.50) ( 0.50)Other 0.02 0.02 0.02 0.01 0.572
( 0.14) ( 0.14) ( 0.15) ( 0.12)
Note: This table presents estimated di↵erences between students in the di↵erent experimental groups who participated in the follow-uptelephone survey. Columns 1 to 4 present means and stadard deviations in parentheses. Column 5 presents the p-value of a of joint test fordi↵erences between the T1, T2 and T3 and SMS-only groups.
Table A4: Take-up: Extensive Margin using Administrative Data
(1) (2) (3) (4)SMS ISP SMS ISP SMS W8 SMS W8
T1 0.976*** 0.975*** 0.000 -0.000(0.004) (0.004) (0.004) (0.004)
T2 0.000 -0.000 0.978*** 0.978***(0.004) (0.004) (0.004) (0.004)
T3 0.967*** 0.967*** 0.967*** 0.967***(0.004) (0.004) (0.004) (0.004)
Observations 7,707 7,707 7,707 7,707Control Mean 0 0 0 0BL Controls X X
Note: This table presents estimated e↵ects for di↵erent treatmentgroups with respect to the control group. Columns 1 and 2 presentestimates on the reception of at least one SMS including ISP informa-tion. Columns 3 and 4 present estimates on the reception of at leastone SMS including an o↵er of help to install parental control settings.Odd-numbered columns present estimates without controls and even-numbered columns present estimates including baseline control vari-ables. Control variables include the baseline values of mean of MBs ofInternet use; guardian gender, age, education level and employment sta-tus; number of siblings; and dummies for family composition (indicatingwhether the child lives with mother, father, step-mother or father’s part-ner, step-father or mother’s partner, uncle/aunt, brother/sister, grand-father/grandmother, other relatives, and other non-relatives). Robustestimated standard errors are reported in parentheses. *** Significantat the 1 percent level. ** Significant at the 5 percent level. * Significantat the 10 percent level.
Table A5: Impact of Treatments on Time of Internet Use
(1) (2) (3) (4) (5) (6)Period: All All Weekdays Weekdays Weekend Weekend
Panel A: T1, T2, T3T1 -7.290* -7.641** -7.039* -7.360** -7.929** -8.356**
(3.770) (3.388) (3.750) (3.373) (3.976) (3.599)T2 3.227 0.834 3.240 0.879 3.194 0.720
(3.831) (3.459) (3.806) (3.437) (4.058) (3.697)T3 -2.563 -5.507 -2.780 -5.688* -2.011 -5.045
(3.696) (3.386) (3.663) (3.351) (3.944) (3.651)
Observations 7,707 7,707 7,707 7,707 7,707 7,707Control Mean 145.9 145.9 145.7 145.7 146.6 146.6BL Controls X X X
Panel B: ISP Information and Parental ControlISP Information -6.540** -6.991*** -6.530** -6.963*** -6.568** -7.060***
(2.634) (2.389) (2.617) (2.370) (2.792) (2.560)Parental Controls 3.976 1.484 3.749 1.275 4.555 2.016
(2.637) (2.393) (2.619) (2.375) (2.796) (2.562)
Observations 7,707 7,707 7,707 7,707 7,707 7,707Control Mean 145.9 145.9 145.7 145.7 146.6 146.6BL Controls X X X
Note: This table presents estimated e↵ects on Internet use measured as predicted time connected to Internet. Odd-numbered columns present estimates without controls and even-numbered columns present estimates includingbaseline control variables. Control variables include the baseline values of mean of MBs of Internet use; guardiangender, age, education level and employment status; number of siblings; and dummies for family composition(indicating whether the child lives with mother, father, step-mother or father’s partner, step-father or mother’spartner, uncle/aunt, brother/sister, grandfather/grandmother, other relatives, and other non-relatives). Robustestimated standard errors are reported in parentheses. *** Significant at the 1 percent level. ** Significant at the5 percent level. * Significant at the 10 percent level.
Table A6: Event Study of SMS messages by Treatment
(1) (2) (3) (4) (5) (6) (7)Day -3 Day -2 Day -1 Day 0 Day 1 Day 2 Day 3
Panel A: Treatment e↵ectISP Info 1.347 -0.542 0 -6.613 -10.660** -4.793 -4.885
( 4.753) ( 4.660) (0) ( 4.526) ( 4.529) ( 4.542) ( 4.566)PC -1.603 -0.256 0 -1.550 -4.419 -3.175 -3.960
( 4.759) ( 4.661) (0) ( 4.522) ( 4.526) ( 4.539) ( 4.564)SMS-Only 2.243 1.295 0 0.834 -5.519 1.101 -1.135
( 4.691) ( 4.596) (0) ( 4.463) ( 4.467) ( 4.479) ( 4.504)
Panel B: Above EventsISP Info 9.420 4.375 0 -17.484** -27.921*** -18.247** -21.621***
( 8.134) ( 7.979) (0) ( 7.757) ( 7.764) ( 7.783) ( 7.823)PC -2.243 -0.889 0 -3.002 -4.930 -4.679 -4.502
( 5.847) ( 5.727) (0) ( 5.556) ( 5.561) ( 5.577) ( 5.608)SMS-Only 1.457 0.178 0 -0.148 -7.422 0.215 -2.856
( 5.778) ( 5.664) (0) ( 5.503) ( 5.507) ( 5.522) ( 5.552)
Panel C: Not-Above EventsISP Info -5.504 -5.422 0 0.814 1.036 2.702 5.447
( 4.929) ( 4.828) (0) ( 4.687) ( 4.691) ( 4.705) ( 4.731)PC -0.097 1.911 0 -0.899 -4.638 -1.119 -3.600
( 4.186) ( 4.099) (0) ( 3.976) ( 3.980) ( 3.991) ( 4.014)SMS-Only 0.596 -0.852 0 0.543 -4.954 -0.583 -1.208
( 4.117) ( 4.034) (0) ( 3.917) ( 3.920) ( 3.931) ( 3.953)
Note: This table presents estimated e↵ects on Internet use measured as MBs downloaded and uploaded for di↵erentdays around the reception of a SMS. Day 0 marks the day on which the SMSs are received each week. The coe�cientfor Day -1 is imposed to be 0 for each treatment group. Regressions also include dummies for each event. Robustestimated standard errors are reported in parentheses. *** Significant at the 1 percent level. ** Significant at the5 percent level. * Significant at the 10 percent level.
Table A7: W8 Parental Control Settings Installation, Event Study
(1) (2) (3) (4) (5) (6) (7)Day -3 Day -2 Day -1 Day 1 Day 2 Day 3
Panel A: Parental ControlsMB Use -12.852 -10.555 -21.823 -21.960 -15.711 -1.548
(28.384) (28.375) (28.367) (28.362) (28.372) (28.382)Panel B: T2MB Use -43.390 -22.280 -22.451 -14.188 -0.901 25.217
(49.181) (49.166) (49.153) (49.143) (49.153) (49.166)Panel C: T3MB Use 13.607 -0.367 -21.302 -28.914 -29.069 -25.277
(31.493) (31.483) (31.474) (31.470) (31.486) (31.500)
Notes: This table presents estimated e↵ects on Internet use measured as MBs downloaded and uploaded fordi↵erent days around the installation of W8 parental control settings.. Day 0 marks the day on which theprogram was installed. The coe�cient for Day 0 is imposed to be 0 for each treatment group. Regressionsalso include dummies for weeks in which the installation took place. Robust estimated standard errors arereported in parentheses. *** Significant at the 1 percent level. ** Significant at the 5 percent level. *Significant at the 10 percent level.
Table A8: Event Study for 1st and 2nd Half of Treatment
(1) (2) (3) (4) (5) (6) (7)Day -3 Day -2 Day -1 Day 0 Day 1 Day 2 Day 3
Panel A: 1st HalfISP Info -3.677 -4.411 0 -13.005** -19.689*** -4.398 -10.291*
( 6.254) ( 6.135) (0) ( 5.808) ( 5.849) ( 5.902) ( 5.939)PC -7.752 -0.765 0 -4.521 -7.171 -4.359 -5.746
( 6.262) ( 6.055) (0) ( 5.822) ( 5.863) ( 5.919) ( 5.953)Placebo -1.173 -1.818 0 -0.415 -10.337* 2.670 -2.645
( 6.186) ( 6.059) (0) ( 5.729) ( 5.767) ( 5.822) ( 5.860)
Panel B: 2nd HalfISP Info 4.832 2.515 0 -0.461 -3.282 -8.304 -2.453
( 7.598) ( 7.354) (0) ( 7.020) ( 6.983) ( 6.967) ( 7.001)PC 7.169 1.322 0 0.771 -3.606 -5.047 -5.145
( 7.569) ( 7.330) (0) ( 6.990) ( 6.956) ( 6.939) ( 6.975)Placebo 6.551 6.456 0 1.133 -2.687 -3.744 -2.818
( 7.489) ( 7.253) (0) ( 6.929) ( 6.896) ( 6.878) ( 6.911)
Note: This table presents estimated e↵ects on Internet use measured as MBs downloaded and uploaded for di↵erentdays around the reception of a SMS. Day 0 marks the day on which the SMSs are received each week. The coe�cientfor Day -1 is imposed to be 0 for each treatment group. Regressions also include dummies for each event. Robustestimated standard errors are reported in parentheses. *** Significant at the 1 percent level. ** Significant at the5 percent level. * Significant at the 10 percent level.
Table A9: Take-up by Subtreatment: using Administrative Data
(1) (2) (3) (4) (5) (6)SMS ISP SMS ISP W8 ISP W8 ISP W8 Installed W8 Installed
Panel A: Within Separate TreatmentsWithin T1
Random -0.007 -0.006 0.000 0.000 0.000 0.000( 0.011) ( 0.011) ( 0.000) ( 0.000) ( 0.000) ( 0.000)
Observations 1,927 1,927 1,927 1,927 1,927 1,927BL Controls X X X
Within T2
Random 0.000 0.000 0.002 0.003 0.034** 0.033**( 0.000) ( 0.000) ( 0.011) ( 0.011) ( 0.016) ( 0.016)
Observations 1,928 1,928 1,928 1,928 1,928 1,928BL Controls X X X
Within T3
Random -0.025** -0.027** -0.023** -0.024** 0.000 0.002( 0.011) ( 0.011) ( 0.011) ( 0.011) ( 0.017) ( 0.017)
Observations 1,924 1,924 1,924 1,924 1,924 1,924BL Controls X X X
Panel B: Within Joint TreatmentsWithin ISP Information
Random -0.016** -0.016** -0.011 -0.012 0.000 0.000( 0.008) ( 0.008) ( 0.014) ( 0.014) ( 0.009) ( 0.009)
Observations 3,851 3,851 3,851 3,851 3,851 3,851BL Controls X X X
Within Parental Controls
Random -0.013 -0.014 -0.010 -0.009 0.017 0.017( 0.014) ( 0.014) ( 0.008) ( 0.008) ( 0.011) ( 0.011)
Observations 3,852 3,852 3,852 3,852 3,852 3,852BL Controls X X X
Note: This table presents estimated e↵ects on take-up for the random-day sub-treatment group with respect tothe fixed-day sub-treatment group. Columns 1 and 2 present estimates on the reception of SMSs including ISPinformation. Columns 3 and 4 present estimates on the reception of SMSs including an o↵er of help to installparental control settings. Columnd 5 and 6 present estimates on the installation of parental control settingsthrough the call center of the experiment. Odd-numbered columns present estimates without controls and even-numbered columns present estimates including baseline control variables. Control variables include the baselinevalues of mean of MBs of Internet use; guardian gender, age, education level and employment status; number ofsiblings; and dummies for family composition (indicating whether the child lives with mother, father, step-motheror father’s partner, step-father or mother’s partner, uncle/aunt, brother/sister, grandfather/grandmother, otherrelatives, and other non-relatives). Robust estimated standard errors are reported in parentheses. *** Significantat the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
Table A10: Interactions of Treatments with Random
(1) (2) (3) (4) (5) (6)Period: All All Week Week Weekend Weekend
T1 -10.943 -17.147 -9.138 -15.498 -15.549 -21.357*(12.714) (11.061) (13.154) (11.433) (12.553) (11.187)
T2 -5.835 -6.839 -5.250 -6.324 -7.326 -8.152(12.244) (11.523) (12.457) (11.782) (12.631) (11.854)
T3 -19.407 -21.653* -19.057 -21.408* -20.300 -22.279*(12.603) (11.340) (12.828) (11.559) (12.917) (11.771)
Random -21.642* -22.497** -22.146* -22.986** -20.357 -21.250*(12.044) (11.212) (12.037) (11.290) (13.098) (12.130)
T1 ⇥ Random -2.972 7.921 -4.047 7.014 -0.228 10.233(15.947) (14.141) (16.218) (14.358) (16.555) (14.999)
T2 ⇥ Random 11.796 12.572 13.072 13.836 8.541 9.348(16.419) (15.057) (16.455) (15.154) (17.568) (16.165)
T3 ⇥ Random 19.627 19.867 18.384 18.577 22.797 23.159(15.569) (14.205) (15.573) (14.249) (16.907) (15.565)
Observations 7,707 7,707 7,707 7,707 7,707 7,707BL Controls X X X
Note: This table presents estimated e↵ects on Internet use measured as daily average MBs uploaded and down-loaded. Odd-numbered columns present estimates without controls and even-numbered columns present estimatesincluding baseline control variables. Control variables include the baseline values of mean of MBs of Internetuse; guardian gender, age, education level and employment status; number of siblings; and dummies for familycomposition (indicating whether the child lives with mother, father, step-mother or father’s partner, step-father ormother’s partner, uncle/aunt, brother/sister, grandfather/grandmother, other relatives, and other non-relatives).Robust estimated standard errors are reported in parentheses. *** Significant at the 1 percent level. ** Significantat the 5 percent level. * Significant at the 10 percent level.
Table A11: Impact for Fixed Subgroup for 1st Half/2nd Half/After Treatment
(1) (2) (3) (4) (5) (6) (7)Day -3 Day -2 Day -1 Day 0 Day 1 Day 2 Day 3
Panel A: 1st Half of TreatmentISP Info -10.773 -9.457 0 -7.499 -23.529*** -3.387 -8.632
( 8.307) ( 8.253) (0) ( 8.052) ( 8.109) ( 8.163) ( 8.169)PC -8.570 1.832 0 -7.039 0.391 -9.177 -9.394
( 8.345) ( 8.137) (0) ( 8.141) ( 8.200) ( 8.257) ( 8.262)SMS-Only -6.137 -1.255 0 1.381 -5.785 3.008 -7.202
( 8.202) ( 8.159) (0) ( 7.962) ( 8.017) ( 8.070) ( 8.074)
Panel B: 2nd Half of TreatmentISP Info 7.684 2.810 0 -8.582 -3.671 -8.477 1.099
( 10.650) ( 10.495) (0) ( 10.348) ( 10.289) ( 10.233) ( 10.232)PC 8.267 2.850 0 5.434 0.924 7.317 -1.573
( 10.620) ( 10.465) (0) ( 10.315) ( 10.264) ( 10.208) ( 10.206)SMS-Only 6.203 13.041 0 -0.109 -0.379 5.897 1.154
( 10.479) ( 10.320) (0) ( 10.180) ( 10.128) ( 10.075) ( 10.071)
Panel C: After TreatmentISP Info 7.589 -0.275 0 3.744 7.274 -5.752 1.996
( 6.122) ( 6.073) (0) ( 6.052) ( 6.055) ( 6.064) ( 6.018)PC 0.542 -6.553 0 2.427 4.968 4.880 0.848
( 6.124) ( 6.074) (0) ( 6.053) ( 6.055) ( 6.065) ( 6.018)SMS-Only 1.187 -2.627 0 6.528 10.034* -1.346 0.384
( 6.009) ( 5.971) (0) ( 5.949) ( 5.953) ( 5.962) ( 5.911)
Note: This table presents estimated e↵ects on Internet use measured as MBs downloaded and uploaded for di↵erentdays around the reception of a SMS. Day 0 marks the day on which the SMSs are received each week. The coe�cientfor Day -1 is imposed to be 0 for each treatment group. Regressions also include dummies for each event. Robustestimated standard errors are reported in parentheses. *** Significant at the 1 percent level. ** Significant at the5 percent level. * Significant at the 10 percent level.
Table A12: Impact of Treatments on Intensity of Internet Use, Pre/Post School
(1) (2)Pre-School Post-School
Panel A: T1, T2, T3T1 -12.992 -22.580**
(12.056) (9.927)T2 1.017 -8.536
(12.889) (10.766)T3 -12.551 -10.991
(12.220) (10.532)
Observations 7,707 7,707BL Controls X XControl Mean 204.9 153.7
Panel B: ISP Information and Parental ControlISP Information -13.280 -12.514*
(8.429) (6.755)Parental Controls 0.729 1.527
(8.489) (6.787)
Observations 7,707 7,707BL Controls X XControl Mean 204.9 153.7
Note: :This table presents estimated e↵ects on Internet use measuredas daily average MBs uploaded and downloaded. Column 1 presents es-timates for the last two weeks of the vacation period and Column 2 forthe first two weeks of the school period. Control variables include thebaseline values of mean of MBs of Internet use; guardian gender, age, ed-ucation level and employment status; number of siblings; and dummiesfor family composition (indicating whether the child lives with mother,father, step-mother or father’s partner, step-father or mother’s partner,uncle/aunt, brother/sister, grandfather/grandmother, other relatives,and other non-relatives). Robust estimated standard errors are reportedin parentheses. *** Significant at the 1 percent level. ** Significant atthe 5 percent level. * Significant at the 10 percent level.