THE ACHIEVEMENT GAP BETWEEN GOVERNMENT AND PRIVATE …

44
THE ACHIEVEMENT GAP BETWEEN GOVERNMENT AND PRIVATE SCHOOLS IN PAKISTAN A Thesis submitted to the Faculty of the Graduate School of Arts and Sciences of Georgetown University in partial fulfillment of the requirements for the degree of Master of Public Policy in Public Policy By Maryam Akmal, B.A. Washington, DC April 12, 2016

Transcript of THE ACHIEVEMENT GAP BETWEEN GOVERNMENT AND PRIVATE …

THEACHIEVEMENTGAPBETWEENGOVERNMENTANDPRIVATESCHOOLSINPAKISTAN

AThesis

submittedtotheFacultyofthe

GraduateSchoolofArtsandSciences

ofGeorgetownUniversity

inpartialfulfillmentoftherequirementsforthe

degreeof

MasterofPublicPolicy

inPublicPolicy

By

MaryamAkmal,B.A.

Washington,DC

April12,2016

ii

Copyright2016byMaryamAkmal

AllRightsReserved

iii

THEACHIEVEMENTGAPBETWEENGOVERNMENTANDPRIVATESCHOOLSINPAKISTAN

MaryamAkmal,B.A.

ThesisAdvisor:AdamThomas,Ph.D.

ABSTRACT

LearningoutcomesinPakistanhavetraditionallybeenpoor.However,overthelasttwo

decades,theeducationalmarketplacehaschangedsubstantially.Inparticular,enrollmentin

privateschoolshasincreaseddramaticallyacrossabroadrangeofincomegroupsinbothurban

andruralareas.GiventheimportantroleofprivateschoolsinPakistan'seducationallandscape,

thereisanincreasingfocusonthelearninggapbetweengovernmentandprivateschools.Using

household-leveldatafromruralandurbanareasofPakistan,thisstudyestimatestheextentto

whichprivateschoolstudentsperformbetterthangovernmentschoolstudents.

iv

IamgratefultoProfessorAdamThomasforhisguidance,supportandencouragementthroughoutthisproject.

Manythanks,MaryamAkmal

v

TABLEOFCONTENTS

Introduction..................................................................................................................................1

Background...................................................................................................................................3

LiteratureReview..........................................................................................................................4

ConceptualFrameworkandHypothesis.......................................................................................8

DataandMethods......................................................................................................................11

DescriptiveStatistics...................................................................................................................15

Results.........................................................................................................................................20

Discussion...................................................................................................................................30

References..................................................................................................................................34

Appendix.....................................................................................................................................38

vi

LISTOFTABLES

Table1:DefinitionsofVariables.................................................................................................14

Table2:DescriptiveStatisticsforDependent,KeyIndependentandControlVariables............17

Table3:KeyCharacteristicsDisaggregatedbySchoolType.......................................................19

Table4:RegressionResultsforMathTestScores......................................................................21

Table5:RegressionResultsforEnglishReadingScores..............................................................24

Table6:RegressionResultsforLocalReadingScores.................................................................27

1

INTRODUCTION

Pakistanhasthesecondhighestnumberofout-of-schoolchildrenintheworld(UNESCO,

2015).1Evenamongchildrenwhoareenrolledinschool,learninglevelsarelow.Estimates

suggestthatonly43percentofgrade-fivestudentscanperformgrade-twoleveldivision,while

only50percentofstudentsingradefivecanreadgrade-twolevelsentencesinPakistan's

nationallanguage,Urdu(ASERPakistan,2013).Basedonthecurrentstateofeducation,evenif

allPakistanichildrenwereenrolledinschool,manyofthestudentswouldstillbefunctionally

illiterateandinnumerate(Dasetal.,2006).

InordertoinstituteeffectivepolicyreforminPakistan'seducationsector,itisimportant

toidentifykeyfactorscontributingtothepoorlearninginsidetheclassrooms.Thecasefor

focusingonstudentlearningmaybestrengthenedevenfurtherbythepossiblelinkbetween

qualityofeducationandenrollment;improvedlearninginschoolsmayboostenrollmentand

retention,asparentsandchildrengainahigherreturnontheirinvestmentoftimeand

resources.

Educationisakeydriverofindividualearningsandnationaleconomicgrowth(Hanushek

andWößmann,2007).AccordingtothePakistanBureauofStatistics(PBS),morethanone-third

ofPakistan'spopulationisbelowtheageoften.Manypolicymakersbelievethatsustained

nationalgrowthwillrequirethatthissignificantsegmentofthepopulationreceivesahigh-

qualityeducation.ThegovernmentofPakistanhasdemonstrateditscommitmenttoimproving

educationaccessandqualitybymakingeducationakeypriorityofPakistan'sNationalPlanof

Action(2013).Pakistan’seducationbudget,whichhasaveragedataround2percentofGDPin

1AccordingtoUNESCO’smostrecentestimates,6.7millionPakistanichildrenareoutofschool,ofwhichmorethanhalfarefemale.

2

recentyears,isprimarilyusedforteachersalaries.AccordingtotheWorldBank(2008),90

percentofPakistan'seducationbudgetisspentonthesalariesofapproximately1.5million

teachers.However,notwithstandingthisinvestmentinimprovingthequalityofclassroom

instruction,studentlearningoutcomescontinuetobedisappointing.

Numerousstudieshaveexploredtheimpactofvariouseducationalinputs,suchas

teachers,facilitiesandcurricula,onstudentlearning.Hanushek(2003)criticizestheemphasis

on"input-based"ratherthan"incentive-based"educationpolicies,arguingthatincreasing

resourcesdoesnotsignificantlyimprovestudentlearning.InPakistan,governmentandprivate

schoolshavedifferentincentivestructures.Forexample,teachersingovernmentschoolstend

tobepaidbetterandhavemoretrainingandexperiencethanprivateschoolteachers.

However,salariesofteachersingovernmentschoolsareusuallytiedtoeducationandseniority

ratherthantostudentlearningoutcomes(Andrabietal.,2010).Keepinginmindthedifferent

incentivestructures,thisstudyanalyzeseducationoutcomesingovernmentandprivateschools

acrossPakistanusingsurveydatafromtheAnnualStatusofEducationReport(ASER)2014.The

surveycovers144districtsofPakistanandprovidesinformationaboutstudentlearning,family

characteristicsandhouseholdinformation.2

Theremainderofthispaperisorganizedasfollows.Inthenextsection,Iprovide

backgroundontheevolutionofprivateschoolinginPakistan.Ithenreviewtherelevant

literatureanddescribemyconceptualframework,data,econometricmethodsanddescriptive

statistics.Lastly,Idiscussmyfindingsandresults.

2ASERselectsdistrictsbasedonthepresenceoflocalcollaboratingpartners(ASER,2010).Whileatpresentonly144districtsaresurveyed,theultimategoalofASERistocoverall157districtsofPakistan.

3

BACKGROUND

Approximatelyone-thirdofallstudentsinPakistanattendprivateschools(Nguyenand

Raju,2014).Theseschoolsservearangeofincomegroupsandareprevalentinbothruraland

urbanareas(Andrabietal.,2002).Whilethereareeliteprivateschoolscateringtohigh-income

groupsinPakistan,thispaperfocusesonthemajorityofprivateschoolsthatarelow-cost

enterprisesservinglow-andmiddle-incomecommunities.Thesebusinessesarelargely

unregulatedandreceivealmostnogovernmentsupport(Andrabietal.,2010).Theytendto

havelowoperationalcostsandareoftenrunoutoftheowners'homes(Andrabietal.,2010).

Privateschoolsalsoemploymoreuntrainedstaffthangovernmentschools,whereteachersare

paidsignificantlymore(Andrabietal.2010;FrenchandKingdon2010).

AccordingtoAndrabietal.(2006),inmanydevelopingcountries,theper-childcostsin

privateschoolsaresignificantlylowerthantheper-childcostsingovernmentschoolsduetothe

payscalesforgovernmentteachersalaries.Privateschoolstendtoemployteacherswithlower

academicqualificationsandtraining.Asaresult,basedonqualificationsalone,teacherquality

seemslowerintheprivatesector.However,teachereffectivenessisafunctionofboth

qualificationsandmotivation,thelatterofwhichishardtomeasure.Itispossiblethatthe

differentincentivestructuresingovernmentandprivateschoolsmayinfluenceteachers'effort

andmotivation,andasaresult,thequalityoflearning(FrenchandKingdon,2010).Despite

theirlowercostsandinferiorhiringpractices,privateschoolsarelargelyperceivedasproviding

betterqualityeducationthantheirgovernmentcounterparts,asevidencedbytherising

demandforprivateeducationinPakistan.

4

ThenumberofprivateschoolsinPakistanmorethandoubledfrom30,000inthe1990s

to70,000in2008(NguyenandRaju,2014).3Inthelasttwodecades,enrollmentinprivate

schoolshasincreasedacrossabroadrangeofincomegroups,includinghigh-income,urban

householdsandlow-income,ruralones(Ibid.).UsingLEAPS(LearningandEducational

AchievementsinPunjabSchools)datacollectedbetween2004and2007inthreedistrictsof

Punjab,Andrabietal.(2010)concludethataveragestudentachievementissignificantlyhigher

inprivateschoolsthaningovernmentschools.Suchfindingscorroboratethecurrent

perceptionofsuperioreducationalqualityinprivateschools.

LITERATUREREVIEW

EvidenceforpoorlearningoutcomesacrossschoolsinPakistaniswell-documented.

Poorstudentperformanceisobservedacrossalltypesofschools,includinggovernmentand

privateschools(Andrabietal.2007;ASER2013).However,fewerstudiescontributedirectlyto

thegrowingdebateaboutthedifferenceineducationoutcomesbetweenpublicandprivate

schoolsinPakistan.Whiletherehasbeenincreasingrecognitionofthepotentialforlow-cost

privateschoolstoimproveeducationaccessandqualityinPakistan,mostexistingstudiesareof

restrictedgeographicalscopeduetolimiteddata.

DemandforPrivateSchoolsinPakistan

Thereissubstantialevidencedocumentingtheriseindemandforprivateeducationin

Pakistansincethe1980s.Andrabietal.(2005)findevidenceofsignificantgrowthinthenumber

ofprivateschoolsinruralareas,thusdispellingearlierassertionsbyJimenezandTan(1987)

3Enrollmentinprivateschoolsincreasedfromlessthan5percentin1990to35percentin2005(Andrabietal.,2010).In1999alone,8000newprivateschoolswerecreated.Roughlyhalfoftheseprivateschoolswereinruralareas(Andrabietal.,2006).

5

thatprivateschoolsareanurbanphenomenonandcatertoarelativelywealthyclientele(Arif

andSaqib2003;NguyenandRaju2014).Onthecontrary,Andrabietal.(2002)findthatmost

Pakistaniprivateschoolsarecateringtolow-andmiddle-incomepopulationsinruralareas,

ratherthantotheelite.MuralidharanandKremer(2006)observeasimilartrendinIndia,

whereprivateschoolsarewidespreadinruralareas.Furthermore,Andrabietal.(2010)find

thatthecostofeducatingaPakistanichildis40percentlowerinprivateschoolsthanin

governmentschools.Aldermanetal.(2001)alsofindevidenceofhighdemandforprivate

educationamonglow-incomehouseholdsinPakistan.Theirstudyshowsthatmanypoor

householdsuseprivateschoolsevenwhenfacedwithhighertuitioncosts,duetoparents'

preferencefortheperceivedhigherqualityofinstructioninprivateschools.

LearningOutcomesinGovernmentandPrivateSchoolsinPakistan

Severalstudiesfindthatprivateschoolspositivelyaffectlearningoutcomes.The

strongeststudyonlearninggapsbetweengovernmentandprivateschoolsisbyAndrabietal.

(2010)whoinstrumentforprivateschoolenrollmentusingthedistancetoaprivateschool

relativetothedistancetoagovernmentschool.Theyfindasignificantrelationshipbetween

privateschoolattendanceandlearningoutcomesinthreedistrictsofruralPunjabinPakistan.

Theirstudyrevealsthatforchildrenwithsimilarcharacteristicstestscoresare0.8to1standard

deviationshigheringovernmentschoolsthaninprivateschools.Inaddition,privateschool

studentshavehigherscoresontestsofcivicvaluesthatmeasureunderstandingofconcepts

suchasdemocracyandgenderequality.Whilethestudyislimitedingeographicalscope(asitis

confinedtothreedistrictsinruralPunjab),itpresentsthemostcredibleevidencetodatein

supportoftheclaimthatprivateschoolsproducebetterlearningoutcomes.

6

Dasetal.(2006),usingdatafromthePunjabprovinceofPakistan,findevidenceofpoor

academicoutcomesinbothgovernmentandprivateschools.Theauthorsfindthatatleast50

percentofthevariationinlearningoutcomescanbeaccountedforbydifferencesinschool

type,suchasgovernmentversusprivateschools,or"bad"governmentschoolsversus"good"

governmentschools.Forexample,thegapinEnglishtestscoresbetweenstudentsfrom

governmentandprivateschoolsistwelvetimeslargerthanthegapbetweenchildrenfromrich

andpoorfamilies.Inanotheranalysisthatuseshouseholdsurveydatafromlow-incomeareas

ofLahoreCity,Aldermanetal.(2001)findthatchildreninprivateschoolsoutperform

governmentschoolstudentsaftercontrollingforfamilycharacteristicsandschoolinputs.Using

thenationwideASERPakistan2011data,AmjadandMacLeod(2011)findthatthequalityof

educationispooracrossgovernmentschools,privateschoolsandschoolswithpublic-private

partnerships.Inaddition,theyfindthatstudentsfromprivateschoolsoutperformstudents

fromgovernmentschools.

Aslam(2009)usesaHeckmantwo-stepproceduretoovercomethepossibilityofsample

selectionbiasamongchildrenwhogotopublicandprivateschoolsinLahoreCity.Sample

selectionbiascouldariseifparentssendingchildrentoprivateschoolshaveagreaterinterest

intheirchild'sacademicsuccess,whichcouldaffectthestudent'sachievementregardlessof

thetypeofschoolattended.Theauthor'sHeckman-correctedresultsarenotsignificantly

differentfromherOrdinaryLeastSquares(OLS)estimates.Theseresultscorroborateearlier

findingsbyAndrabietal.(2010),whichshowthatprivateschoolsaremoreeffectiveatteaching

mathandliteracyskillsthangovernmentschools.

7

Notallevidencesupportsthenotionthatdifferencesinstudentperformancecanbe

attributedtoschooltype.ArifandSaqib(2003)sample50schoolsacrossthecountryandfind

evidenceofbetterlearningoutcomesamongprivateschoolstudents,buttheyalsofindthat

thedifferencecanlargelybeexplainedbyfamilybackgroundandschoolcharacteristics,suchas

teacherqualificationandstudent-teacherratio.Performancealsovariesbydistrict,with

governmentschoolsperformingbetterthanprivateschoolsinsomedistricts.Inaddition,the

authorsarguethatlow-incomehouseholdstendtosendtheirchildrentogovernmentschools,

asprivateeducationislargelyunaffordableforthem.

LearningOutcomesinGovernmentandPrivateSchoolsinOtherDevelopingCountries

Studiescomparingoutcomesacrossgovernmentandprivateschoolsinotherdeveloping

countrieslargelyprovidesupportfortheclaimthatstudentshavebetterlearningoutcomesin

privateschools.FrenchandKingdon(2010)examineASERdatafromIndiatoestimatethe

effectofprivateschoolenrollmentonlearningoutcomes.Theyuseavarietyoftechniquesto

estimatetheeffectofprivateschools,suchasOLSestimation,cross-sectionfixedeffect

techniquesatthelevelofstate,district,villageandhouseholds,andpaneldataanalysisusing

villageandtimefixedeffects.Theauthorsfindthatthereisaprivateschooleffectonchild

achievementof0.17standarddeviations.AnotherstudybyMuralidharanandKremer(2006)in

Indiafindsevidenceofasizeableandsignificantassociationbetweenprivateschoolattendance

andlearningoutcomesaftercontrollingforfamilyandhouseholdcharacteristics.

AdescriptivestudybyTooleyandDixon(2005)findsthatprivateschoolsinIndia,

Ghana,NigeriaandKenyaarepopularamonglow-incomesegments,havebetterachievement

scoresandhavelowerteachercosts.Jimenezetal.(1991)usedatafromColombia,the

8

DominicanRepublic,thePhilippines,TanzaniaandThailand,toshowthatprivateschool

studentsperformbetteronstandardizedmathandEnglishteststhangovernmentschool

students,evenaftercontrollingforincome.Inaddition,theyfindevidencethatprivateschools

havelowerper-unitcoststhangovernmentschools,substantiatingclaimsoflowerprivate

schoolcostsinPakistanbyAndrabietal.(2010).

ThePresentStudy

Manystudieshaveexaminedthelearninggapbetweengovernmentandprivateschools

invariousdevelopingcountries.However,onlyafewhavefocusedonlearningoutcomesin

governmentandprivateschoolsinPakistan.Todate,themostinformativestudiesfrom

Pakistanhaveutilizeddatafromarestrictedgeographicalarea(Andrabietal.2006;Dasetal.

2006).ThisstudyusesASERsurveydatafrom2014,coveringmostregionsofPakistan,to

presentacomprehensivenation-wideassessmentoflearningingovernmentandprivate

schools.

CONCEPTUALFRAMEWORKANDHYPOTHESIS

Basedonthefindingsintheexistingliterature,Ihypothesizethatattendingprivate

schoolsispositivelycorrelatedwithlearningoutcomes.Inotherwords,Ipredictthatthereisa

significantlearninggapbetweenstudentsattendingprivateschoolsandstudentsattending

governmentschools.Inaddition,Iexpectthesizeoftheprivateschooladvantagetovaryby

district(Dasetal.2006).Whileprivateschoolshavegrowninbothruralandurbandistricts,

theireffectsmayvaryduetothedifferentsocialandeconomiccharacteristicsofthedistricts.

9

Whilethepurposeofthisstudyistoinvestigatetheeffectofschooltype(government

versusprivate)onlearninglevels,manyotherfactorsalsocontributetothedifferencein

achievementamongstudents.Thesefactors,whicharerelatedtobothlearninglevelsand

schooltype,canbebroadlycategorizedintoindividualchildcharacteristics,parent

characteristicsandhouseholdcharacteristics.Inordertoisolatetherelationshipbetween

schooltypeonlearningoutcomes,itisimportanttocontrolforthesefactors.Thisstudyuses

controlvariablesfromthenationallyrepresentativeASERPakistansurvey,whichprovideschild-

,parent-andhousehold-leveldataacrossruralandurbandistrictsofPakistan.

ChildCharacteristics

Apartfromthewell-documentedlinkbetweenachild'snaturalcognitiveabilityand

learningoutcomes,otherchildcharacteristics,suchasnumberofsiblings,genderandaccessto

tutoringoutsideofschoolarealsorelatedtoeducationalachievement.Numerousstudiesshow

aninverserelationshipbetweenfamilysizeandlearningoutcomes(Alwin1991;Shavitand

Pierce1991;Downey1995).Onepossibletheoryexplainingthisrelationshipisthe"resource

dilutioneffect":asparent'sfiniteresources,suchasincomeandtime,arespreadoutamong

moresiblings,thelowerresourcesperchildarerelatedtolowereducationoutcomes(Downey,

2001).GiventhestronggenderbiasinenrollmentandlearninginPakistan,parentsmaychoose

tospendmoreonamalechild'seducation.4Forexample,Aslam(2009)findsthatboysin

Pakistanaremorelikelytobeenrolledincomparativelyhigh-qualityprivateschoolsthangirls.

4Accordingtoa2014reportbytheWorldEconomicForum,Pakistanranks132outof142countriesintermsofthegendergapineducationalattainment.AnotherstudybytheBrookingsInstitution(KingandWinthrop,2015)findsthatthereare74girlsforevery100boysenrolledinsecondaryschoolinPakistan.

10

Privatetutoringoutsideofschoolisprevalentamongstudentsfrombothruraland

urbanareas(Dundaretal.,2014).Aslam(2009)positsthatpartofthereasonfortheuptakeof

privatetutoringcouldbethepoorqualityofeducationprovidedinschools,withqualitybeing

worseingovernmentschools.Theprobabilityofreceivingextratutoringisalsorelatedto

income,genderandmaternaleducation(Macphersonetal.,2014).

HouseholdCharacteristics

Manystudiesfromhigh-incomecountrieshavedocumentedthepositiverelationship

betweenincomeandtestscores(DahlandLochner2012;DuncanandMagnuson2005;Davis-

Kean2005).Dasetal.(2006)findasimilarrelationshipbetweenparentalincomeandchild's

educationalachievementinPakistan.However,theauthorsfindthattheachievementgap

attributabletoparents'incomeandeducationissignificantlyreducedonceoneaccountsfor

differencesinschooltype.Incomealsoaffectsavailabilityofelectricity,waterandotheritems

thatfacilitateaproductivelearningenvironmentathome.Furthermore,incomeand

transportationavailabilitydeterminewhetherschoolsarephysicallyaccessible.InPakistan,

wherefemalemobilityisrestrictedduetoculturalexpectations,distancefromschoolaffects

girlsmorethanboys(Aldermanetal.,2001).Andrabietal.(2006)findthatprivateschoolshelp

tomitigatethenegativeeffectofdistanceonschoolattendanceandboostoverallenrollment

forbothboysandgirls.

ParentCharacteristics

Thereisampleevidencedocumentingapositiverelationshipbetweenparents'levelof

educationandchildren'slearningoutcomes(Davis-Kean2005;Dubow2010).InPakistan,

Andrabietal.(2009)findthatthetestscoresofchildrenwhosemothershavesomeeducation

11

are0.24to0.35standarddeviationshigherthanchildrenwithuneducatedmothers.

Furthermore,motherswithsomeeducationaremorelikelytospendtimehelpingchildrenwith

homework(Ibid.).Itispossiblethateducatedparentsarealsomorelikelytochoosebetween

publicandprivateschoolsbasedonperceivedquality.

DATAANDMETHODS

MyempiricalanalysesusedatafromtheAnnualStatusofEducationReport(ASER)

Pakistan2014surveycoveringchildrenaged3to16yearsin123ruraldistrictsand21urban

centersofPakistan.ASERassessmenttoolstestchildren'slearningthroughagradeonetotwo

leveltestonarithmetic,Englishandreading(Urdu,SindhiorPashto).Eachstudentisassigneda

learninglevelbetweenone(lowest)tofive(highest).Thegoaloftheassessmentistoobtain

informationonthebasicreadingandarithmeticabilitiesofchildren.Forthepurposesofthis

study,Irecodethetestscorevariablesintodichotomousvariablestakingthevalueofoneif

studentsscoreabovetheaverageforthesubjectandzeroifstudentsscorebelowtheaverage.

Thisapproachallowsmetoproducemoreeasilyinterpretableregressionresults.

SincetheASERtestisonlyadministeredtochildrenabovefiveyearsofage,Iexcludeall

childrenwhoarebelowfiveyearsofagefrommyanalysis.Furthermore,Iexcludeanychildren

whowenttoreligiousorothertypesofschoolsfromthesample,asthisstudyisfocusedonly

ondifferencesinlearningoutcomesbetweenstudentsenrolledingovernmentandprivate

schools.5

5Religiousschoolsormadrassahs,unlikepublicandprivateschools,primarilyteachreligioussubjects.However,somemadrassahsalsoteachnon-religioussubjectssuchasmathincombinationwithreligiousstudies.

12

Icontrolforvariablesthatareplausiblyrelatedtobothlearningoutcomes(the

dependentvariable)andtheprobabilityofgoingtoprivateschoolsovergovernmentschools

(thekeyindependentvariable).Asisimpliedinthediscussionintheprevioussectionthese

controlvariablescanbecategorizedintothreebroadcategories:childcharacteristics,

householdcharacteristicsandparentcharacteristics(SeeTable1).

Toanalyzetherelationshipbetweenschooltype(adummyvariablethattakesona

valueofoneifthechildgoestoaprivateschoolandzerootherwise)andlearningoutcomes(a

dummyvariablethattakesonavalueofoneifthechildscoresabovetheaverageandzero

otherwise),IestimateaLinearProbabilityModel(LPM)withdistrict-levelfixedeffects.TheLPM

allowsmetopredictthechangeinprobabilityofscoringabovetheaveragegivenachangein

schooltypefromgovernmenttoprivate.Myfixedeffectsspecificationcontrolsfortime-

invariantcharacteristicsoftheindividualdistricts–forexample,unobservedadministrative

differencesbetweendistrictsthatcouldaffectresourcesallocatedtoschools.Thisapproachis

consistentwiththatofAslam(2009),whoestimatedavillage-levelfixed-effectsmodeltostudy

therelativeeffectivenessofschooltypesinLahore,Pakistan.Isubsequentlyapplymore

stringentfixedeffectscontrolsatthevillageandsiblinglevel.

IcannotuseasimpleOLSmodeltoestimatetheeffectofprivateschoolingonlearning

outcomesduetosampleselectionbias.Itispossiblethatthechoiceofprivateschoolsisrelated

tounobservedcharacteristicsofthechildandthefamilythatarealsorelatedtoachievement

scores.AssuggestedbyFrenchandKingdon(2010),OLSestimatesprovideanupperboundof

theprivateschooleffect,astheyincludetheeffectofunobservablefactorsinadditiontothe

privateschooleffect.Toapplystrictercontrols,Iusedistrict-levelfixedeffects.

13

Asimpleversionofmyregressionequationisasfollow:

LearningLevelid=β0+β1Privateid+βxxXXid+αd+uid

whereXXrepresentsallthecontrolvariableslistedinTable1.Thesubscripts"i"and"d"

representeachindividualchildanddistrictrespectively.

Iestimatethreeseparateregressionequations,witheachequationusingoneofthree

dependentvariablesmeasuringlearningoutcomes:math,Englishandreadinginthelocal

language.

14

Table1:DefinitionsofVariables

DependentVariableLearningLevel

MathHighestLevel

Adichotomousvariableindicatingwhetherthechildscoredabovetheaverageinmath(0=no,1=yes).6

EnglishReading

AdichotomousvariableindicatingwhetherthechildscoredabovetheaverageinEnglish(0=no,1=yes).7

LocalReading

Adichotomousvariableindicatingwhetherthechildscoredabovetheaverageinlocallanguage:Urdu,Sindhi,Pashto(0=no,1=yes).8

IndependentVariable

PrivateAdichotomousvariableindicatingthetypeofschoolattendedbythechild(0=government,1=private).

ChildCharacteristics

FemaleAdichotomousvariableindicatingthechild’sgender(0=male,1=female).

AgeAcontinuousvariableindicatingthechild’sage(between5-16years).

Grade Acontinuousvariableindicatingthegradeofthechild:0-12.

ChildTakingPaidTutoringAdichotomousvariableindicatingwhetherthechildtakespaidtutoringoutsideofschool(0=no,1=yes).

TuitionFee9Acontinuousvariableindicatingtheschool'stuitionfeeinPakistaniRupees.

HouseholdCharacteristics

HouseholdSizeAcontinuousvariableindicatingthetotalnumberofmembersinthesurveyedhousehold.

HouseTypeMixedAdichotomousvariableindicatingthattheconstructiontypeofhouseismixed(0=notmixed,1=mixed).

HouseTypeSolid Adichotomousvariableindicatingthattheconstructiontypeof

6Theaveragescoreformathis3.450onascaleof1to5(1=beginner/nothing,2=recognitionof1-9,3=recognitionof10-99,4=subtraction,5=division).7TheaveragescoreforEnglishreadingis3.468onascaleof1to5(1=beginner/nothing,2=capitalletters,3=smallletters,4=words,5=sentences).8Theaveragescoreforlocallanguagereadingis3.476onascaleof1to5(1=beginner/nothing,2=letters,3=words,4=sentences,5=story).9GovernmentschoolsinPakistantendtochargeasmalltuitionfee(Kattan&Burnett,2004).ForthegovernmentschoolssampledinASER2014data,tuitionfeesrangefromRs.20($0.19)toRs.8000($75.88)permonth.

15

houseissolid(0=notsolid,1=solid).

HouseOwnedAdichotomousvariableindicatingwhetherthehouseisownedbytheresidents(0=no,1=yes).

ElectricityConnectionAvailable

Adichotomousvariableindicatingwhetherthehousehasanelectricalconnection(0=no,1=yes).

MobileAvailableAdichotomousvariableindicatingwhetherthereisamobilephoneavailableforusebyhouseholdmembers(0=no,1=yes).

TVAvailableAdichotomousvariableindicatingwhetherthereisatelevisioninthehousehold(0=no,1=yes).

NearestSchoolDistanceAcontinuousvariableindicatingtheone-waydistancefromthehousetothenearestschoolinkilometers.

ParentCharacteristicsMother'sAge Acontinuousvariableindicatingtheageofthechild'smother.Mother'sEducationSecondary

Adichotomousvariableindicatingwhetherthechild'smotherhascompletedsecondaryschool(0=no,1=yes).

Mother'sEducationHighSchool

Adichotomousvariableindicatingwhetherthechild'smotherhascompletedhighschool(0=no,1=yes).

Mother'sEducationSomeCollege

Adichotomousvariableindicatingwhetherthechild'smotherhascompletedsomecollege(0=no,1=yes).

Father'sAge Acontinuousvariableindicatingtheageofthechild'sfather.Father'sEducationSecondary

Adichotomousvariableindicatingwhetherthechild'sfatherhascompletedsecondaryschool(0=no,1=yes).

Father'sEducationHighSchool

Adichotomousvariableindicatingwhetherthechild'sfatherhascompletedhighschool(0=no,1=yes).

Father'sEducationSomeCollege

Adichotomousvariableindicatingwhetherthechild'sfatherhascompletedsomecollege(0=no,1=yes).

DESCRIPTIVESTATISTICS

Table2providesdescriptivestatisticsforthedependentandkeyindependentvariables,

andforchild,householdandparentcharacteristics.AppendixAsupplementsTable2by

providingdescriptivestatisticsbyregiontype(urbanversusrural).Table3presentsdifferences

inkeycharacteristicsbyschooltype(governmentversusprivate).1011

10Alldescriptivestatisticsareunweighted.AccordingtoASERsurveydocumentation,weightedestimatesarethesameasunweightedestimatesatthedistrictlevelduetotheuseofprobability-basedsampling(PBS).

16

AsshowninTable2,averagelearninglevelsamongstudentsinbothgovernmentand

privateschoolsarelow.Themeanvalueforallthreelearningoutcomevariablesis

approximately0.52,implyingthatmoststudents'ASERtestscoresarebelowthebasic

proficiencylevelsexpectedingradeonetotwo.Furthermore,themeanandstandard

deviationsforstudentoutcomesaresimilaracrossmath,readinginEnglishandreadinginlocal

language,whichshowsthatstudentstendtodopoorlyacrossalltests.Inlinewiththenational

gapinmale-femaleliteracyrates,fatherstendtobemoreeducatedthanmothers.12The

averagetuitionfeeforgovernmentandprivateschoolsisRs.359($3.42)permonth.However,

thereissubstantialvariationintuitionfees,rangingfromRs.20($0.19)permonthtoRs.8,000

($75.88)permonth.13Themaximumhouseholdsizeis74,implyingthatsomefamiliesinthe

sampleresideinsharedhouseholds.14

11Theoriginalsamplehad238,843observations.Missingvaluesfortuitionfeewerederivedusingregressionmeanimputation.Intheoriginalsample,213,018valuesweremissingandonly25,825observationswereavailable.Amongthe213,018imputedvalues(89.19%oftheobservations),anyimputedvaluesbelowthevalueof20(minimumvalueforthetuitionfeevariableacrossprivateandgovernmentschools)ornegativevaluesweredropped(275observations).Lastly,allmissingobservationsforthedependent,keyindependentandcontrolvariables(exceptforthetuitionfeevariable)weredroppedfromthesample.Fordependentvariables,Idroppedthefollowingmissingvalues:1,134for"MathHighestLevel,"1,027for"EnglishReading"and46,214for"LocalReading."Idropped27,234observationsfor"Private"(keyindependentvariable).Idropped44,993observationsforthemissingcontrolvariables(excluding"TuitionFee").Asaresult,thefinalsamplesizeis117,966.12AccordingtoUNICEF,theyouthliteracyrateformalesin79.1percent.Thecomparablefigureforfemalesis61.5percent.(UNICEF,2013)13AccordingtothePakistanBureauofStatistics,Pakistan'sannualpercapitaincomeis$1,513,implyingthattheaveragePakistanimakesRs.13,292($126)permonth.(PakistanBureauofStatistics,2015)14Twoextremelyhighhouseholdsizevaluesof7773and307werecodedasmissinganddropped,astheyareextremeoutliers.

17

Table2:DescriptiveStatisticsforDependent,KeyIndependentandControlVariables

Mean SD Min MaxDependentVariables

MathHighestLevel 0.50 0.50 0 1EnglishReading 0.55 0.50 0 1LocalReading 0.51 0.50 0 1

KeyIndependentVariable Private 0.34 0.48 0 1

ChildCharacteristics Female 0.36 0.48 0 1Age 9.75 3.19 5 16Grade 3.97 2.90 0 12ChildTakingPaidTutoring 0.16 0.37 0 1TuitionFee 358.55 227.37 20 8000

HouseholdCharacteristics HouseholdSize 7.27 3.54 1 74HouseTypeMixed 0.30 0.46 0 1HouseTypeSolid 0.34 0.47 0 1HouseOwned 0.91 0.29 0 1ElectricityConnectionAvailable 0.91 0.29 0 1MobileAvailable 0.84 0.36 0 1TVAvailable 0.65 0.48 0 1NearestSchoolDistance(km) 1.53 14.02 0 2200

ParentCharacteristics Mother'sAge 36.47 7.20 16 97Mother'sEducationSecondary 0.17 0.38 0 1Mother'sEducationHighSchool 0.13 0.34 0 1Mother'sEducationSomeCollege 0.03 0.17 0 1Father'sAge 41.25 7.98 18 92Father'sEducationSecondary 0.21 0.40 0 1Father'sEducationHighSchool 0.26 0.44 0 1Father'sEducationSomeCollege 0.12 0.32 0 1Thesamplesizeis117,966,with15,681observationsforurbanareasand102,285observationsforruralareas.Fordescriptivestatisticsdisaggregatedbyregiontype,refertoAppendixA.

18

Table3showsthedifferenceincharacteristicsbetweenstudentswhoattend

governmentandprivateschools.Mostdifferencesarestatisticallysignificantattheonepercent

level.Thetableshowsthat,onaverage,privateschoolstudentspayhighermonthlyfees(Rs.

452.20versusRs.309.33)forschoolsandarealsomorelikelytotakepaidtutoringlessons

outsideofschool.Privateschoolstudentsaremorelikelytohaveasolidhouseandhaveaccess

toelectricity,mobilephoneandtelevisionathome.Onaverage,parentsofchildrenwhogoto

privateschoolshavecompletedagreaternumberofyearsofschoolingthanparentsofchildren

whogotogovernmentschools.Itisnoteworthythatmanyofthedifferencesinkey

characteristicsbetweenprivateandgovernmentschoolsaresimilartothedifferencesinkey

characteristicsbetweenurbanandruralareas(AppendixA).

19

Table3:KeyCharacteristicsDisaggregatedbySchoolType

Private Government Difference SEDependentVariables

MathHighestLevel 0.51 0.49 0.02*** 0.00EnglishReading 0.58 0.53 0.04*** 0.00LocalReading 0.51 0.51 0.00ns 0.01

ChildCharacteristics Female 0.39 0.35 0.04*** 0.00Age 9.22 10.02 -0.81*** 0.02Grade 3.52 4.20 -0.68*** 0.02ChildTakingPaidTutoring 0.33 0.08 0.25*** 0.00TuitionFee 452.20 309.33 142.27*** 1.33

HouseholdCharacteristics HouseholdSize 6.83 7.50 -0.66*** 0.02HouseTypeMixed 0.27 0.31 -0.03*** 0.00HouseTypeSolid 0.52 0.25 0.26*** 0.00HouseOwned 0.88 0.92 -0.04*** 0.00ElectricityConnectionAvailable 0.96 0.88 0.08*** 0.00MobileAvailable 0.92 0.80 0.12*** 0.00TVAvailable 0.78 0.59 0.19*** 0.00NearestSchoolDistance(km) 1.72 1.44 0.28** 0.09

ParentCharacteristics Mother'sAge 35.52 36.93 -1.41*** 0.04Mother'sEducationSecondary 0.20 0.15 0.05*** 0.00Mother'sEducationHighSchool 0.23 0.08 0.15*** 0.00Mother'sEducationSomeCollege 0.06 0.01 0.05*** 0.00Father'sAge 40.50 41.65 -1.15*** 0.00Father'sEducationSecondary 0.19 0.22 -0.03*** 0.03Father'sEducationHighSchool 0.34 0.22 0.12*** 0.00Father'sEducationSomeCollege 0.19 0.08 0.11*** 0.00Totalsamplesizeis117,966,with77,345observationsforgovernmentschoolsand40,621observationsforprivateschools.

20

RESULTS

MyregressionresultsaresummarizedinTables5,6and7,whichusemath,Englishand

localreadingtestscoresasthedependentvariable,respectively.Eachregressionalsocontrols

forchild,parentandhouseholdcharacteristics.Myanalysisusescross-sectionfixedeffects

techniques,andIapplyprogressivelystringentfixedeffectscontrolsatthedistrict,villageand

siblinglevels.Ineachtable,Model1reportstheresultsofmyLPMregressionwithoutfixed

effects,providinganupperlimitfortheprivateschoolrelationshipwithstudentachievement.

Model2displaystheresultsofregressionsthatincludedistrict-levelfixedeffects,whichcontrol

forthepermanentcharacteristicsofeachdistrict,bothobservableandunobservable,suchas

thelevelofurbanizationortheextentofcorruptionwithinlocaleducationbodiesandother

fixedsocial,politicalandgeographicdifferences.Model3includestheresultsofvillage-level

fixedeffectsregressions,whichcontrolforpermanentcharacteristicsofthevillage,suchas

schoolquality(FrenchandKingdon,2010).Model4testswhethertheeffectofprivate

schoolingdiffersbygenderusinginteractions.Finally,Model5showstheresultsofasibling-

levelfixedeffectsmodel.Bycontrastingchildrenwhoareenrolledinprivateschoolswiththeir

siblingswhoarenot,Icontrolfortheeffectoffamilybackgroundontestoutcomes.Robust

standarderrorsarereportedbeneathallcoefficients.

Overall,theresultsreportedinTables5,6and7fortheLPM,district-levelandvillage-

levelfixedeffectsspecificationsconfirmmyhypothesisthatprivateschoolinghasasmall,

significantandpositiveassociationwithtestscores.However,therelationshipbetweenprivate

schoolingandtestscoresdisappearswhenincludingsiblingfixedeffectsintheregression.

Nevertheless,itisadvisabletoexercisecautionwheninterpretingsiblingfixedeffects,asthe

informationusedtoestimatetheprivateschoolcoefficientisderivedfromlessthanfive

21

percentofthesamplememberswholiveinhouseholdswithatleasttwosiblings,andinwhich

atleastonesiblingattendsprivateschoolandatleastonesiblingattendspublicschool.

Table4:RegressionResultsforMathTestScores

15Thedummyvariable"MathHighestLevel"takesonavalueof1ifthestudentscoresabovetheaveragemathscoreof3.450andtakesonavalueof0ifthestudentscoresbelowtheaverage.

DependentVariable MathHighestLevel(Score0-1)15 (1) (2) (3) (4) (5)

LPM DistrictFE VillageFE

VillageFEw/

Interaction SiblingFEKeyIndependent

Variable Private 0.0554*** 0.0180*** 0.0149*** 0.0120*** 0.00355 (0.00264) (0.00277) (0.00290) (0.00336) (0.00540)

ChildVariables Female -0.00726*** -0.0101*** -0.00984*** -0.0125*** -0.00593* (0.00229) (0.00223) (0.00217) (0.00272) (0.00315)Female*Private

0.00742*

(0.00440)Age 0.0171*** 0.0200*** 0.0282*** 0.0282*** 0.0445***

(0.000823) (0.000816) (0.000844) (0.000844) (0.00141)

Grade 0.0938*** 0.0883*** 0.0806*** 0.0805*** 0.0667***

(0.000946) (0.000959) (0.00101) (0.00101) (0.00174)

ChildTakingPaidTutoring 0.0388*** 0.0360*** 0.0382*** 0.0383*** 0.0553***

(0.00320) (0.00325) (0.00362) (0.00363) (0.00881)

TuitionFee/100 0.000244 0.000391 -0.00108 -0.00106 -0.00396***

(0.000711) (0.000696) (0.000705) (0.000704) (0.00149)

HouseholdVariables HouseholdSize 0.000303 -0.000392 -0.0416 -0.0387

(0.000320) (0.000353) (0.000394) (0.000394)HouseTypeMixed 0.0378*** 0.0164*** 0.00749** 0.00748**

(0.00291) (0.00311) (0.00354) (0.00354)

HouseTypeSolid 0.0585*** 0.0294*** 0.0157*** 0.0158***

(0.00321) (0.00352) (0.00406) (0.00406)

HouseOwned -0.0188*** 0.00178 0.0113** 0.0114**

(0.00382) (0.00395) (0.00443) (0.00443)

ElectricityConnectionAvailable 0.0321*** -0.00441 -0.00951 -0.00949

(0.00419) (0.00465) (0.00615) (0.00615)MobileAvailable 0.0190*** 0.0189*** 0.00676 0.00676

22

Robuststandarderrorsareinparentheses.P-valuesaregiveninparenthesesunderF-statisticsforjointhypothesistests.***p<0.01,**p<0.05,*p<0.1

(0.00353) (0.00392) (0.00460) (0.00460)

TVAvailable -0.0182*** 0.00802*** 0.00366 0.00369

(0.00282) (0.00298) (0.00324) (0.00324)

NearestSchoolDistance(km) 0.000186** 0.0357 0.0637 0.0657

(0.0534) (0.0476) (0.000119) (0.000119)ParentVariables

Mother'sAge 0.00172*** 0.00174*** 0.000136 0.000137 -0.00193

(0.000301) (0.000302) (0.000317) (0.000317) (0.00291)

Mother'sEducationSecondary 0.0112*** 0.00483 0.00810** 0.00810** 0.00847

(0.00328) (0.00331) (0.00335) (0.00335) (0.0376)

Mother'sEducationHighSchool 0.0333*** 0.0206*** 0.0146*** 0.0145*** 0.0769

(0.00397) (0.00408) (0.00420) (0.00420) (0.0561)

Mother'sEducationSomeCollege 0.0350*** 0.0295*** 0.0271*** 0.0269*** -0.0863

(0.00722) (0.00724) (0.00735) (0.00735) (0.0788)

Father'sAge -0.00142*** -0.00152*** -0.0332 -0.0332 0.00340

(0.000271) (0.000271) (0.000283) (0.000283) (0.00267)

Father'sEducationSecondary -0.00128 -0.0157*** -0.00795** -0.00793** -0.0170

(0.00308) (0.00309) (0.00321) (0.00321) (0.0343)

Father'sEducationHighSchool 0.0158*** 0.00579* 0.0109*** 0.0109*** -0.0446

(0.00319) (0.00319) (0.00333) (0.00333) (0.0369)

Father'sEducationSomeCollege 0.0287*** 0.0278*** 0.0222*** 0.0221*** -0.00784 (0.00441) (0.00449) (0.00475) (0.00475) (0.0490)Constant -0.132*** -0.0940*** -0.127*** -0.126*** -0.256***

(0.00902) (0.00939) (0.0107) (0.0107) (0.0605)

Observations 117,966 117,966 117,966 117,966 117,966R-squared 0.437 0.476 0.561 0.561 0.768F-statisticforjointhypothesis:private=female*private=0

24.16***(0.0000)

23

MathTestScores

ThesimpleLPMmodelinTable5showsthatgoingtoaprivateschoolinsteadofa

governmentschoolisassociatedwitha5.54percentagepointincreaseintheprobabilityof

scoringabovetheaverageformathtestscores.Applyingdistrict-level(orvillage-level)fixed

effectsreducesthemagnitudeofthisassociationconsiderably.ThisimpliesthatourinitialLPM

estimatewaspositivelybiasedduetoomissionofcontrolsforthecharacteristicsofvillages(for

example,whetheravillageisclosertoanurbancenter,hasbetterqualityschoolsorlower

levelsofcorruptionwithinlocaladministrativebodies)thatarelinkedtomathtestscoresand

probabilityofprivateschoolattendance.Forthedirectionofthebiastobepositive,village

characteristicsmusteitherbepositivelyassociatedwithbothmathtestscoresandthe

probabilityofattendingprivateschools,ortherelationshipofvillagecharacteristicswithboth

variablesmustbenegative.Forexample,avillagewithproximitytoanurbancenterislikelyto

havebetteraccesstoelectricity,whichcouldthereforebelinkedtobettertestscores.Sucha

villagecouldalsohaveabetterroadinfrastructure,whichcouldbepositivelylinkedtoprivate

schoolattendance.Ontheotherhand,itispossiblethatavillagewithhighlevelsofcorruption

doesnotspendenoughfundsonteachertrainingorschoolfacilities,whichcouldcontributeto

poortestscores.Similarly,highcorruptioncouldbeassociatedwithdifficultyinestablishing

privateschoolsduetolicensinghurdles,therebynegativelyaffectingprivateschoolattendance.

Bothscenarioswouldresultinapositivebiasintheprivateschoolcoefficient.Lastly,theresults

showthattheassociationbetweenprivateschoolingandmathscoresoffemalestudentsis

positiveandhighlysignificant.Inaddition,theassociationbetweenprivateschoolattendance

andmathtestscoresislargerforfemalesthanformales.

24

Table5:RegressionResultsforEnglishReadingScores

16Thedummyvariable"EnglishReading"takesonavalueof1ifthestudentscoresabovetheaverageEnglishreadingscoreof3.468andtakesonavalueof0ifthestudentscoresbelowtheaverage.

DependentVariable EnglishReading(Score0-1)16 (1) (2) (3) (4) (5)

LPM DistrictFE VillageFE

VillageFEw/

Interaction SiblingFEKeyIndependent

Variable Private 0.0686*** 0.0203*** 0.0111*** 0.00738** -0.0186*** (0.00271) (0.00283) (0.00299) (0.00346) (0.00552)

ChildVariables Female -0.00455* -0.0107*** -0.0131*** -0.0165*** -0.0127*** (0.00235) (0.00229) (0.00223) (0.00278) (0.00321)Female*Private

0.00953**

(0.00456)

Age 0.0204*** 0.0233*** 0.0291*** 0.0291*** 0.0457***

(0.000837) (0.000823) (0.000855) (0.000855) (0.00142)

Grade 0.0837*** 0.0780*** 0.0726*** 0.0726*** 0.0578***

(0.000963) (0.000966) (0.00102) (0.00102) (0.00175)

ChildTakingPaidTutoring 0.0534*** 0.0462*** 0.0394*** 0.0396*** 0.0541***

(0.00324) (0.00332) (0.00370) (0.00370) (0.00891)

TuitionFee/100 0.00138* -0.000224 -0.00116 -0.00113 -0.00469*** (0.000732) (0.000698) (0.000715) (0.000715) (0.00146)HouseholdVariables HouseholdSize 0.00245*** -0.0414 0.0393 0.0430

(0.000336) (0.000371) (0.000408) (0.000408)HouseTypeMixed 0.0351*** 0.0188*** 0.0108*** 0.0108***

(0.00300) (0.00318) (0.00365) (0.00365)

HouseTypeSolid 0.0608*** 0.0353*** 0.0165*** 0.0165***

(0.00330) (0.00359) (0.00417) (0.00417)

HouseOwned -0.0229*** -0.00155 0.00485 0.00492

(0.00390) (0.00407) (0.00458) (0.00458)

ElectricityConnectionAvailable 0.0444*** 0.00507 0.00137 0.00139

(0.00433) (0.00475) (0.00632) (0.00632)MobileAvailable 0.0225*** 0.0227*** 0.0112** 0.0112**

(0.00364) (0.00398) (0.00472) (0.00472)

TVAvailable -0.0125*** 0.00998*** 0.00607* 0.00611*

(0.00292) (0.00305) (0.00333) (0.00333)

25

Robuststandarderrorsareinparentheses.P-valuesaregiveninparenthesesunderF-statisticsforjointhypothesistests.***p<0.01,**p<0.05,*p<0.1

NearestSchoolDistance(km) 0.000172** 0.0206 0.000102 0.000103

(0.0567) (0.0455) (0.000123) (0.000123)ParentVariables

Mother'sAge 0.000495 0.00123*** 0.000117 0.000119 -0.00265

(0.000306) (0.000308) (0.000330) (0.000330) (0.00314)

Mother'sEducationSecondary 0.0219*** 0.00687** 0.00497 0.00497 -0.0840**

(0.00336) (0.00337) (0.00343) (0.00343) (0.0380)

Mother'sEducationHighSchool 0.0479*** 0.0237*** 0.0139*** 0.0138*** -0.00527

(0.00405) (0.00414) (0.00430) (0.00430) (0.0560)

Mother'sEducationSomeCollege 0.0513*** 0.0218*** 0.0176** 0.0174** -0.149*

(0.00745) (0.00746) (0.00754) (0.00754) (0.0779)

Father'sAge -0.000533* -0.00133*** -0.000245 -0.000245 0.00421

(0.000276) (0.000276) (0.000293) (0.000293) (0.00286)

Father'sEducationSecondary 0.0149*** -0.00353 0.000958 0.000990 0.000160

(0.00316) (0.00315) (0.00330) (0.00330) (0.0362)

Father'sEducationHighSchool 0.0141*** 0.00816** 0.0142*** 0.0142*** -0.0192

(0.00327) (0.00326) (0.00343) (0.00343) (0.0380)

Father'sEducationSomeCollege 0.0127*** 0.0248*** 0.0242*** 0.0241*** 0.0271 (0.00447) (0.00454) (0.00485) (0.00484) (0.0516)Constant -0.114*** -0.0433*** -0.0576*** -0.0566*** -0.163**

(0.00924) (0.00958) (0.0109) (0.0109) (0.0650)

Observations 117,966 117,966 117,966 117,966 117,966R-squared 0.402 0.448 0.536 0.536 0.758F-statisticforjointhypothesis:private=female*private=0

16.98***(0.0000)

26

EnglishReadingTestScores

ThemagnitudeoftherelationshipbetweenprivateschoolingandEnglishtestscoresis

higheracrossthedifferentregressionspecificationsascomparedtothemagnitudeofthe

relationshipbetweenprivateschoolattendanceandmathtestscores.ThesimpleLPMmodelin

Table6showsthatgoingtoaprivateschoolinsteadofagovernmentschoolisassociatedwitha

6.86percentagepointincreaseintheprobabilityofscoringaboveaverageforEnglishtest

scores.Applyingdistrict-levelfixedeffectsreducesthemagnitudeoftherelationshipto2.03

percentagepoints.Village-levelfixedeffectsfurtherreducethemagnitudeoftherelationship

to1.11percentagepoints.SimilartotheresultsforMathscores,capturingpermanentdistrict

andvillagecharacteristicslowerstheestimatedcoefficientonprivateschooling,implyingthat

thesimpleLPMmodelfailedtoaccountforpermanentdistrictandvillagecharacteristicsthat

haveasimilarlysignedrelationshipwithEnglishreadingtestscoresandtheprobabilityof

attendingprivateschools.However,afterIapplysibling-levelfixedeffects,Ifindthatthereisa

negativeassociationbetweenprivateschoolattendanceandtestscores.Theimplicationsof

thisfindingarediscussedingreaterdetailinthenextsection.Lastly,consistentwiththe

findingsformathscores,theresultsshowapositiveandsignificantassociationbetweenprivate

schoolattendanceandEnglishtestscoresoffemalestudents.Inaddition,theassociation

betweenprivateschoolattendanceandEnglishtestscoresislargerforfemalesthanformales.

27

Table6:RegressionResultsforLocalReadingScores

DependentVariable LocalReading(Score0-1)17 (1) (2) (3) (4) (5)

LPM DistrictFE VillageFEVillageFEw/Interaction SiblingFE

KeyIndependentVariable

Private 0.0477*** 0.0142*** 0.00979*** 0.00727** -0.00758 (0.00263) (0.00277) (0.00292) (0.00339) (0.00541)

ChildVariables

Female 0.000719 -0.00547** -0.00762*** -0.00999*** -0.00778**

(0.00228) (0.00224) (0.00218) (0.00274) (0.00315)

Female*Private

0.00654

(0.00441)

Age 0.0220*** 0.0239*** 0.0290*** 0.0290*** 0.0448***

(0.000822) (0.000820) (0.000849) (0.000849) (0.00142)

Grade 0.0897*** 0.0865*** 0.0820*** 0.0819*** 0.0682***

(0.000945) (0.000960) (0.00102) (0.00102) (0.00174)

ChildTakingPaidTutoring 0.0511*** 0.0449*** 0.0441*** 0.0442*** 0.0538***

(0.00319) (0.00327) (0.00361) (0.00361) (0.00882)

TuitionFee/100 -0.00169** -0.00139** -0.00114 -0.00113 -0.00406*** (0.000720) (0.000701) (0.000736) (0.000736) (0.00145)

HouseholdVariables

HouseholdSize 0.000260 -0.000460 0.000106 0.000109

(0.000327) (0.000364) (0.000402) (0.000402)

HouseTypeMixed 0.0412*** 0.0272*** 0.0106*** 0.0106***

(0.00290) (0.00313) (0.00358) (0.00358)

HouseTypeSolid 0.0593*** 0.0322*** 0.0126*** 0.0127***

(0.00318) (0.00352) (0.00409) (0.00409)

HouseOwned -0.0222*** 0.00330 0.00743* 0.00749*

(0.00383) (0.00398) (0.00447) (0.00447)

ElectricityConnectionAvailable 0.0248*** 0.00521 0.00216 0.00218

(0.00422) (0.00468) (0.00620) (0.00620)MobileAvailable 0.00911*** 0.0118*** 0.0129*** 0.0129***

(0.00352) (0.00390) (0.00467) (0.00467)

TVAvailable -0.0146*** 0.000367 0.000912 0.000939

17Thedummyvariable"LocalReading"takesonavalueof1ifthestudentscoresabovetheaveragelocalreadingscoreof3.476andtakesonavalueof0ifthestudentscoresbelowtheaverage.

28

(0.00280) (0.00296) (0.00326) (0.00326)

NearestSchoolDistance(km) 0.000214*** 0.0385 0.000101 0.000103

(0.0548) (0.0470) (0.000121) (0.000121)ParentVariables

Mother'sAge 0.00230*** 0.00160*** 0.000468 0.000469 -0.00432

(0.000300) (0.000305) (0.000326) (0.000326) (0.00301)

Mother'sEducationSecondary 0.0192*** 0.00639* 0.00869*** 0.00869*** -0.0256

(0.00326) (0.00329) (0.00336) (0.00336) (0.0385)

Mother'sEducationHighSchool 0.0406*** 0.0251*** 0.0213*** 0.0212*** 0.0430

(0.00395) (0.00405) (0.00420) (0.00420) (0.0512)

Mother'sEducationSomeCollege 0.0435*** 0.0323*** 0.0355*** 0.0353*** -0.0466

(0.00732) (0.00735) (0.00749) (0.00749) (0.0758)

Father'sAge -0.00103*** -0.00109*** -0.000279 -0.000279 0.00683**

(0.000270) (0.000274) (0.000291) (0.000291) (0.00276)

Father'sEducationSecondary 0.00629** -0.00106 0.000959 0.000981 0.00203

(0.00306) (0.00309) (0.00323) (0.00323) (0.0352)

Father'sEducationHighSchool 0.00872*** 0.0107*** 0.0117*** 0.0117*** -0.0374

(0.00316) (0.00318) (0.00334) (0.00334) (0.0356)

Father'sEducationSomeCollege 0.0134*** 0.0311*** 0.0219*** 0.0219*** -0.00640 (0.00438) (0.00449) (0.00475) (0.00475) (0.0491)Constant -0.170*** -0.129*** -0.145*** -0.144*** -0.299***

(0.00897) (0.00936) (0.0107) (0.0107) (0.0606)

Observations 117,966 117,966 117,966 117,966 117,966R-squared 0.440 0.473 0.558 0.558 0.767F-statisticforjointhypothesis:private=female*private=0

12.14***(0.0005)

Robuststandarderrorsareinparentheses.P-valuesaregiveninparenthesesunderF-statisticsforjointhypothesistests.***p<0.01,**p<0.05,*p<0.1

29

LocalReadingTestScores

Themagnitudeoftherelationshipbetweenprivateschoolingandlocalreadingtest

scoresisloweracrossthedifferentregressionspecificationsascomparedtothemagnitudeof

therelationshipbetweenprivateschoolingandmathandEnglishtestscores.ThesimpleLPM

modelinTable7showsthatgoingtoaprivateschoolinsteadofagovernmentschoolis

associatedwitha4.77percentagepointincreaseintheprobabilityofscoringabovetheaverage

forlocalreadingtestscores.Applyingdistrict-levelfixedeffectsreducesthemagnitudeofthe

relationshipto1.42percentagepoints.Controllingforvillage-levelfixedcharacteristicsreduces

themagnitudeoftherelationshipevenfurtherto0.98percentagepoints.Therelativelysmall

relationshipbetweenprivateschoolattendanceandlocalreadingtestscorescouldbe

explainedbythefactthatprivateschoolstendtofocusonlearninginEnglishmorethan

governmentschools.Infact,parentstendtoassociatethehigherqualityofprivateschool

educationwithinstructioninEnglish(SocietyforAdvancementofEducation,2015).Therefore,

itispossiblethatlocallanguageinstructioninprivateschoolsisnotashigh-qualityasthe

instructioningovernmentschools.Lastly,consistentwiththefindingsformathandEnglishtest

scores,theresultsshowapositiveandsignificantassociationbetweenprivateschool

attendanceandlocalreadingtestscoresoffemalestudents.Inaddition,theassociation

betweenprivateschoolattendanceandlocalreadingtestscoresislargerforfemalesthanfor

males.

ControlVariables

Myestimatesoftherelationshipbetweenprivateschoolingandgender,andbetween

privateschoolingandindicatorsoffamily'swealth,areunsurprising.Boystendtoscorehigher

inmathandEnglishthangirlsacrossthevariousspecifications.Takingpaidtuitionoutsideof

30

regularschoolhasastatisticallysignificantandpositiveassociationwithmath,Englishandlocal

readingtestscores.Havingabrickhouseinsteadofahutisalsosignificantlypositively

associatedwithtestscoresacrossthethreesubjects,underscoringtheimportantroleoffamily

income.Inaddition,itisnotablethatparentaleducationlevelsplayanimportantrolein

studentperformance.InlinewiththefindingsbyAndrabietal.(2009),childrenwithmothers

whohavesecondary-leveleducationperformsignificantlybetterthanchildrenwithmothers

whohavenoeducation.Similarly,childrenwhosefathershavesomecollegeeducationtendto

havesignificantlyhighertestscores.

DISCUSSION

OverviewofFindings PrivateschoolsinPakistanhaveexpandeddramaticallyoverthepastfewdecades.In

lightofthisfact,policymakershavequestionedtheroleofprivateschoolinginimprovingaccess

toqualityeducationinPakistan.Whileasubstantialbodyofliteraturefindsthatprivate

schoolinghasapositiverelationshipwithtestscores,themajorityofstudiesusedatathathas

limitedgeographicalscope.MyresearchusesdatafromthenationallyrepresentativeASER

surveytoanalyzetherelationshipbetweenprivateschoolsandtestscores.

Theresultsofmyprimaryspecificationssuggestthat,consistentwithpreviousresearch

(Aldermanetal.2001;Andrabietal.2010;Aslam2009;Dasetal.2006;FrenchandKingdon

2010),privateschoolshaveamodestandpositiverelationshipwithtestscores.Theregression

resultsforallthreetests(math,readinginEnglishandreadinginlocallanguage)reveala

significantandpositiveassociationbetweenprivateschoolingandtestscores.However,this

relationshipdisappearswhensiblingfixedeffectsareincludedinthemodel.Whiletheresults

formathandlocalreadingtestscoresarepositiveandinsignificant,theresultsforEnglishtest

31

scoresrevealthatgoingtoprivateschoolsisassociatedwitha1.86percentagepointdecrease

inprobabilityofscoringabovetheaverage,implyingthatthereisaprivateschool

disadvantage.Thisresultissurprisingconsideringprivateschoolsareperceivedasfocusing

moreoninstructioninEnglishthantheirgovernmentcounterparts.However,asexplained

previously,mysiblingfixedeffectsspecificationsrelyonmuchlessvariationthanmyother

specificationstoestimatetherelationshipbetweenprivateschoolattendanceandtestscores.

Themixedresultsofthesiblingfixedeffectsmodelhighlighttheneedformoresophisticated

identificationstrategiesusingnationallyrepresentativedatatogainabetterunderstandingof

therelationshipbetweenprivateschoolattendanceandtestscores.

MyresultsareparticularlyintriguingwhenIexaminetherelationshipbetweenprivate

schoolingandtestscoresseparatelybygender.Forfemales,goingtoaprivateschoolinsteadof

agovernmentschoolisassociatedwithalargerchangeinprobabilityofscoringabovethe

average(incomparisontomales)acrossallthreetesttypes.Theassociationbetweenprivate

schoolingandtestscoresoffemalestudentsisconsistentlysignificantattheonepercentlevel.

AccordingtoAndrabietal.(2006),low-costprivateschoolstendtohirefemaleteachers

residinginthelocalareastokeepsalarycostslow.Itispossiblethatthegreaterpresenceof

femaleteachersinprivateschoolsfacilitatesproductivein-classinteractionforfemalestudents

thatmayleadtobettertestperformance.Inaddition,previousevidencebyAslam(2009)has

suggestedthatgirlstendtohavepooreraccesstoprivateschoolingthanboys.However,no

priorstudiesfromPakistanhavepresentedevidenceforthedifferentialrelationshipbetween

privateschoolingandtestscoresforboysandgirls.Thisfindingprovidesanopportunityfor

furtherresearch.

32

Althoughmyanalysiscontrolsforvariouschild-,parent-andhousehold-level

characteristics,myresultsmaybesubjecttoomittedvariablebias.Factorssuchasparental

motivationarehardtomeasureandarelikelylinkedtobothprivateschoolattendanceandtest

scoreoutcomes.Moremotivatedparentsarelikelytosendtheirchildrentoschoolsthatare

perceivedashavinghigherquality,thatis,privateschools.Similarly,parentswithhigherlevels

ofmotivationarelikelytoinvestmoretimeandresourcesintheirchild'seducation,for

example,byhelpingtheirchildwithhomework.Theabsenceofthisvariablefromtheanalysis

wouldupwardlybiasourcoefficient.

Furthermore,parentalattitudesandexpectationscansignificantlydifferdependingon

thechild'sgender.Forexample,someresearchhassuggestedthatparentshavemore

confidenceinamalechild'smathabilitiesthanintheabilitiesofafemalechild(Stockdale,

1995).Itispossiblethatparentswhosendafemalechildtoaprivateschoolarelikelytohave

lessdiscriminatoryattitudestowardsgirls'education,whichcouldexplainthestrongerprivate

schooleffectforgirlsthanforboys.Becausemymodeldoesnotcaptureunobservableparental

attitudesandexpectationstowardsfemaleeducation,itispossiblethatthecoefficienton

privateschoolingisupwardlybiased:parentalsupportandencouragementislikelytobe

positivelylinkedtoprivateschoolattendanceforfemalesaswellasbettertestscores.

PolicyImplications

Thepolicyimplicationsofmystudyaremixed.Inlightoftheriseofprivateschoolsand

poorqualityoflearningingovernmentschools,thedebateregardingtheroleofthepublicand

privatesectorsineducationremainsunresolved.However,itisclearthatalargeanddiverse

privatesector,comprisingbotheliteandlow-feeprivateschools,isanintegralpartofPakistan's

33

educationallandscape.Inlinewiththestanceofsupportersofprivatizededucation,thisstudy

presentssomeevidencethatprivateschoolingcanhelpimproveaccesstoqualityeducation.

Myfindingssuggestthattheimpactofprivateeducationonfemalelearningoutcomes

maybemeaningfulandsignificant.Therefore,privateschoolsmayplayanimportantrolein

narrowingthegenderdisparitiesinachievement.Thereisaneedforbettercollectionand

utilizationofgender-disaggregateddatatofurtherexaminetherelationshipbetweenprivate

schoolingandfemaleachievement.

IrrespectiveofthemarginalsuperiorityofprivateschoolsinPakistan,childreninboth

governmentandprivateschoolshavepoorlearningoutcomesonaverage.Thepolicysolution

toPakistan'seducationcrisisistounderstandwhybothgovernmentandprivateschoolsare

failingtodeliverqualityeducation.Thereisaneedtobetterutilizeexistingdatatounderstand

thereasonsforpoorlearningtakingplaceinPakistanischools.Oneimportantsteptowardsthat

goalistogathermoredetailedandin-depthsurveyinformationtounderstandhowprivate

schoolscanbeusedtoprovideaccessibleandqualityeducation.

34

REFERENCES

Alderman,H.,Orazem,P.,&Paterno,E.(2001).SchoolQuality,SchoolCost,andthePublic/PrivateSchoolChoicesofLow-IncomeHouseholdsinPakistan.JournalofHumanResources,36,pp.304-326.Retrievedfromhttp://econ2.econ.iastate.edu/faculty/orazem/lahore.pdf

Alwin,D.(1991).FamilyofOriginandCohortDifferencesinVerbalAbility.AmericanSociologicalReview,Vol.56,No.5,pp.625-638. Retrievedfromhttp://www.jstor.org/stable/2096084

Amjad,R.,&MacLeod,G.(2011).EffectsofPrivate,Public,andPrivate-PublicPartnershipSchoolsinPakistan.Retrievedfromhttp://www.aserpakistan.org/document/learning_resources/2014/Private_Tuitions/Effectiveness%20of%20PPP%20by%20Ravish%20&%20Gordon.pdf

Andrabi,T.,Das,J.,&Khwaja,A.(2002).TheRiseofPrivateSchoolinginPakistan:CateringtotheUrbanEliteorEducatingtheRuralPoor?Retrievedfromhttp://www.innovations.harvard.edu/sites/default/files/2608043.pdf

Andrabi,T.,Das,J.,&Khwaja,A.(2005).PrivateSchooling:LimitsandPossibilities.Retrievedfromhttp://www.hks.harvard.edu/fs/akhwaja/papers/PrivateSchoold_Final_Nov5.pdf

Andrabi,T.,Das,J.,&Khwaja,A.(2006).StudentsToday,TeachersTomorrow?TheRiseofAffordablePrivateSchools.Retrievedfromhttp://www.cgdev.org/doc/event%20docs/MADS/ADK_PriSchools_Feb06.pdf

Andrabi,T.,Das,J.,Khwaja,A.,&Vishwanath,T.(2007).LearningandEducationalAchievementsinPunjabSchools(LEAPS):InsightstoInformtheEducationPolicyDebate.Retrievedfromhttp://leapsproject.org/assets/publications/LEAPS_report.pdf

Andrabi,T.,Das,J.,Khwaja,A.(2009).WhatDidYouDoAllDay?MaternalEducationandChildOutcomes.Retrievedfromhttp://www.hks.harvard.edu/fs/akhwaja/papers/WhatDidYouDoAllDay.pdf

Andrabi,T.,Bau,N.,Das,J.,&Khwaja,A.(2010).AreBadPublicSchoolsPublic“Bads”?TestScoresandCivicValuesinPublicandPrivateSchools.Retrievedfromhttp://www.hks.harvard.edu/fs/akhwaja/papers/badpublicschools.pdf

Arif,G.,&Saqib,N.(2003).ProductionofCognitiveandLifeSkillsinPublic,Private,andNGOSchoolsinPakistan.ThePakistanDevelopmentReview,Vol.42,No.1,pp.1-28.Retrievedfromhttp://www.jstor.org.proxy.library.georgetown.edu/stable/pdf/41260519.pdf?acceptTC=true

ASERPakistan(2010).Retrievedfromhttp://pc.gov.pk/usefull%20links/ASER-2010/ASER-2010%20FAQs.pdf

ASERPakistan.(2013).Retrievedfromhttp://aserpakistan.org/document/aser/2013/reports/national/ASER_National_Report_2013.pdf

35

Aslam,M.(2009).TheRelativeEffectivenessofGovernmentandPrivateSchoolsinPakistan:AreGirlsWorseOff?Retrievedfromhttp://web.a.ebscohost.com/ehost/pdfviewer/pdfviewer?sid=f15691d7-2ecc-47cb-84db-67258d057f06%40sessionmgr4004&vid=4&hid=4101

Dahl,H.,&Lochner,L.(2012).TheImpactofFamilyIncomeonChildAchievement:EvidencefromtheEarnedIncomeTaxCredit.AmericanEconomicReview,102(5),pp.1927–1956.Retrievedfromhttp://econweb.ucsd.edu/~gdahl/papers/children-and-EITC.pdf

Das,J.,Pandey,P.,&Zajonc,T.(2006).LearningLevelsandGapsinPakistan.TheWorldBank.Retrievedfromhttps://www.openknowledge.worldbank.org/bitstream/handle/10986/8866/wps4067.txt?sequence=2

Davis-Kean,P.(2005).TheInfluenceofParentEducationandFamilyIncomeonChildAchievement:TheIndirectRoleofParentalExpectationsandtheHomeEnvironment.JournalofFamilyPsychology,Vol.19,No.2,pp.294-304.Retrievedfromhttp://www.rcgd.isr.umich.edu/garp/articles/davis-kean05.pdf

Downey,D.(1995).WhenBiggerIsNotBetter:FamilySize,ParentalResources,andChildren'sEducationalPerformance.AmericanSociologicalReview, 60.5,pp.746.Retrievedfromhttp://search.proquest.com/openview/ce5065d0a332bef44f0dae95fb09784c/1?pq-origsite=gscholar&cbl=41945

Downey,D.(2001).NumberofSiblingsandIntellectualDevelopment:TheResourceDilutionExplanation.AmericanPsychologist,56(6-7),pp.497-504.Retrievedfromhttp://www.ncbi.nlm.nih.gov/pubmed/11413873

Dubow,E.,Boxer,P.,&Huesmann,L.(2010).Long-termEffectsofParents’EducationonChildren’sEducationalandOccupationalSuccess:MediationbyFamilyInteractions,ChildAggression,andTeenageAspirations.Retrievedfromhttp://www.ncbi.nlm.nih.gov/pmc/articles/PMC2853053/

Duncan,G.&Magnuson,K.(2005).CanFamilySocioeconomicResourcesAccountforRacialandEthnicTestScoreGaps?TheFutureofChildren,Vol.15,No.1,pp.35-54.Retrievedfromhttps://muse.jhu.edu/login?auth=0&type=summary&url=/journals/future_of_children/v015/15.1duncan.html

Dundaretal.(2014).StudentLearninginSouthAsia:Challenges,Opportunities,andPolicyPriorities.Retrievedfromhttp://www-wds.worldbank.org/external/default/WDSContentServer/WDSP/IB/2014/05/23/000442464_20140523133319/Rendered/PDF/882670PUB0978100Box385205B00PUBLIC0.pdf

EducationforAllNationalReviewReport:Pakistan.(2015).UNESCO.Retrievedfromhttp://unesdoc.unesco.org/images/0022/002297/229718E.pdf

36

French,R.,&Kingdon,G.(2010).TherelativeeffectivenessofprivateandgovernmentschoolsinRuralIndia:EvidencefromASERdata.Retrievedfromhttp://www.researchgate.net/publication/46463037

Hausmanetal.(2014).TheGlobalGenderGapReport.TheWorldEconomicForum.Retrievedfromhttp://www3.weforum.org/docs/GGGR14/GGGR_CompleteReport_2014.pdf

Hanushek,E.(2003).TheFailureofInput-basedSchoolingPolicies.TheEconomicJournal,113,pp.64-98.Retrievedfromhttp://hanushek.stanford.edu/publications/failure-input-based-schooling-policies

Hanushek,E.,&Wößmann,L.(2007).TheRoleofEducationQualityinEconomicGrowth.TheWorldBank.Retrievedfromhttps://openknowledge.worldbank.org/bitstream/handle/10986/7154/wps4122.pdf?sequence=1

Jimenez,E.,&Tan,J.(1987).DecentralizedandPrivateEducation:TheCaseofPakistan.TheWorldBank.Retrievedfromhttp://www-wds.worldbank.org/external/default/WDSContentServer/WDSP/IB/1987/03/01/000009265_3980623151848/Rendered/PDF/multi_page.pdf

Jimenez,E.,Lockheed,M.,&Paqueo,V.(1991).TheRelativeEfficiencyofPrivateandPublicSchoolsinDevelopingCountries.TheWorldBankResearchObserver,Vol.6,No.2,pp.205-218.Retrievedfromhttp://www.jstor.org/stable/3986318?seq=1#page_scan_tab_contents

Kattan,R.,&Burnett,N.(2004).UserFeesinPrimaryEducation.TheWorldBank.Retrievedfromhttp://siteresources.worldbank.org/EDUCATION/Resources/278200-1099079877269/547664-1099079993288/EFAcase_userfees.pdf

King,E.,&Winthrop,R.(2015).Today'sChallengesforGirls'Education.TheBrookingsInstitution.Retrievedfromhttps://books.google.com/books?isbn=0815728611

Macphersonetal.(2014).Microeconomics:PrivateandPublicChoice.

Muralidharan,K.,&Kremer,M.(2006).PublicandPrivateSchoolsinIndia.Retrievedfromhttp://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.168.3057&rep=rep1&type=pdf

Nguyen,Q.&Raju,D.(2014).PrivateSchoolParticipationinPakistan.TheWorldBank.Retrievedfromhttp://www-wds.worldbank.org/external/default/WDSContentServer/WDSP/IB/2015/06/17/090224b0828c82e0/1_0/Rendered/PDF/Private0school0participation0in0Pakistan.pdf

PakistanBureauofStatistics.(2015).PakistanSocialandLivingStandardsMeasurements.Retrievedfromhttp://www.pbs.gov.pk/content/pakistan-social-and-living-standards-measurement

SocietyforAdvancementofEducation.(2015).EducationMonitor.Retrievedfromhttp://www.cqe.net.pk/pdf/EM2EducationMonitor.pdf

37

Shavit,Y.&Pierce,J.(1991).SibshipSizeandEducationalAttainmentinNuclearandExtendedFamilies:ArabsandJewsinIsrael.AmericanSociologicalReview,Vol.56,No.3,pp.321-330.Retrievedfromhttp://www.jstor.org/stable/2096107

Stockdale,S.(1995).GenderDifferencesinHighSchoolMathematicsAchievement:AnEmpiricalApplicationofthePropensityScoreAdjustment.

Tooley,J.,&Dixon,P.(2005).PrivateEducationisGoodforthePoor.CatoInstitute.Retrievedfromhttp://object.cato.org/sites/cato.org/files/pubs/pdf/tooley.pdf

UNICEF.(2013).Retrievedfromhttp://www.unicef.org/infobycountry/pakistan_pakistan_statistics.htmlTheWorldBank.(2008).Retrievedfromhttp://www.worldbank.org/en/news/feature/2014/10/29/teachers-hold-key-student-learning-education-review

38

APPENDIX

AppendixA:DescriptiveStatisticsDisaggregatedbyRegionType

Urban Rural Difference SEDependentVariables

MathHighestLevel 0.63 0.48 0.15*** 0.00EnglishReading 0.70 0.53 0.17*** 0.00LocalReading 0.65 0.49 0.16*** 0.00

ChildCharacteristics Age 10.01 9.71 0.30*** 0.03Female 0.44 0.35 0.08*** 0.00Grade 4.57 3.88 0.70*** 0.02ChildTakingPaidTutoring 0.24 0.13 0.11*** 0.00TuitionFee 518.79 333.95 184.83*** 1.86

HouseholdCharacteristics HouseholdSize 5.57 7.53 -1.96*** 0.03HouseTypeMixed 0.19 0.31 -0.13*** 0.00HouseTypeSolid 0.77 0.28 0.50*** 0.00HouseOwned 0.75 0.93 -0.18*** 0.00ElectricityConnectionAvailable 0.99 0.89 0.10*** 0.00MobileAvailable 0.99 0.82 0.17*** 0.00TVAvailable 0.95 0.61 0.34*** 0.00NearestSchoolDistance(km) 1.41 1.55 -0.14ns 0.12

ParentCharacteristics Mother'sAge 36.16 36.51 -0.35*** 0.06Mother'sEducationSecondary 0.20 0.17 0.03*** 0.00Mother'sEducationHighSchool 0.38 0.10 0.28*** 0.00Mother'sEducationSomeCollege 0.10 0.02 0.08*** 0.00Father'sAge 40.70 41.34 -0.64*** 0.07Father'sEducationSecondary 0.13 0.22 -0.09*** 0.00Father'sEducationHighSchool 0.39 0.24 0.16*** 0.00Father'sEducationSomeCollege 0.29 0.09 0.19*** 0.00Totalsamplesizeis117,966,with15,681observationsforurbanareasand102,285observationsforruralareas.