risna

Upload: cut-nyak-rae

Post on 07-Apr-2018

225 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/6/2019 risna

    1/37

    The Board of Regents of the University of Wisconsin System

    Why Does Mother's Schooling Raise Child Health in Developing Countries? Evidence fromMoroccoAuthor(s): Paul GlewweSource: The Journal of Human Resources, Vol. 34, No. 1 (Winter, 1999), pp. 124-159Published by: University of Wisconsin PressStable URL: http://www.jstor.org/stable/146305 .Accessed: 20/05/2011 02:27

    Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless

    you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you

    may use content in the JSTOR archive only for your personal, non-commercial use.

    Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .http://www.jstor.org/action/showPublisher?publisherCode=uwisc. .

    Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

    JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

    content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

    of scholarship. For more information about JSTOR, please contact [email protected].

    University of Wisconsin Press and The Board of Regents of the University of Wisconsin System are

    collaborating with JSTOR to digitize, preserve and extend access to The Journal of Human Resources.

    http://www.jstor.org

    http://www.jstor.org/action/showPublisher?publisherCode=uwischttp://www.jstor.org/stable/146305?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/action/showPublisher?publisherCode=uwischttp://www.jstor.org/action/showPublisher?publisherCode=uwischttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/146305?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=uwisc
  • 8/6/2019 risna

    2/37

    Why Does Mother's SchoolingRaise Child Health in DevelopingCountries?Evidence from Morocco

    Paul Glewwe

    ABSTRACTMother's education is oftenfound to be positively correlated with childhealth and nutrition in developing countries, yet the causal mechanismsare poorly understood. Threepossible mechanisms are: (1) Formal educa-tion directly teaches health knowledge to future mothers; (2) Literacy andnumeracy skills acquired in school assist future mothers in diagnosingand treating child health problems; and (3) Exposure to modern societyfrom formal schooling makes women more receptive to modern medicaltreatments. This paper uses data from Morocco to assess the role playedby these different mechanisms. Mother's health knowledge alone appearsto be the crucial skill for raising child health. In Morocco, such knowl-edge is primarily obtained outside the classroom, although it is obtainedusing literacy and numeracy skills learned in school; there is no evidencethat health knowledge is directly taught in schools. This suggests thatteaching of health knowledge skills in Moroccan schools could substan-tially raise child health and nutrition in Morocco.

    I. IntroductionChild health s a key indicatorof the qualityof life in developingcountries.Mother'syearsof educations oftenpositivelyassociatedwithimprovedchildhealthandnutritionaltatus seeBehrman, 990).Thereareavarietyof mecha-

    Paul Glewwe is a Senior Economist in the Development Research Group at the The World Bank. Hewould like to thank Hanan Jacoby, Martin Ravallion, and two anonymous reviewers for helpful com-ments on previous drafts, and Nauman Ilias for excellent computational assistance. This research wassupported by a grant from the World Bank Research Committee (RPO 679-84). Thefindings, interpreta-tions and conclusions expressed in this paper are entirely those of the author. They do not necessarilyrepresent the views of the World Bank, its Executive Directors, or the countries they represent. Thedata used in this article can be obtained beginning May, 1999, through April, 2002, from PaulGlewwe, The World Bank, 1818 H Street NW, Washington, DC 20433.[SubmittedOctober1996;acceptedFebruary 998]THE JOURNAL OF HUMAN RESOURCES * XXXIV * 1

  • 8/6/2019 risna

    3/37

    Glewwe 125

    nismsthroughwhichmother'seducation ould raise child health: 1) Directacqui-sition of basic healthknowledge n schoolmay providefuturemotherswith infor-mationuseful for diagnosingandtreatingchild healthproblems; 2) Literacyandnumeracy kills learned n school may enhancemothers'abilitiesto treatchild ill-nesses,conditional nhealthknowledge,andalsoshouldhelpmothers ncrease heirstock of healthknowledgeafter eavingschool;and(3) Exposure o modemsocietyin generalvia schoolingmay changewomen'sattitudes oward raditionalmethodsof raisingchildrenandtreating heirhealthproblems.Thispaperattemptso assess therelative mportance f these threemechanisms,usingthe 1990-91 MoroccanEnqueteNationaledes Niveauxde Vie des Menages(ENNVM).Knowledgeof the relativeimportance f these mechanismscan haveimportant olicy implications.Forexample, f themainimpactof education omesfromdirectlyraisingmothers'basic healthknowledge,suchknowledgeshouldbetaught n schools as early as possible (thatis, beforegirls dropout) andperhapsshould also be taught n specialeducationcoursesfor women of child-bearing gewho have already eft school.Thepaper s organizedas follows. Section II reviews,in broad erms, he impactof mother'seducationon childhealthandbrieflyreviews the recent iterature. ec-tionIII discusses he dataand heestimation trategy.SectionIVpresents heempiri-cal results. SectionV decomposes he total impactof mother'sschoolingon childhealth.Section VI summarizeshe results.

    II. Mother's Education and Child HealthA. GeneralDiscussionFigure 1 providesa schematic ramework or thinkingabout the determinants fchildhealthandnutritionaltatus.As seen at the bottomof thatfigure,child healthis ultimatelydetermined y threedistinctsets of factors:1. Health andnutritionalinputsprovidedby thehousehold arrow ); 2. Thelocal healthenvironmentarrowf); and 3. The child's healthendowment arrowh). Health andnutritionalnputsprovidedby the household ncludeprenatal are,breastmilk, reastmilk ubstitutessuchas infantformula,caloriesfrom adult oods (forweanedchildren),medicines,andmedicalcare.Thequalityof householddrinkingwatersources, oiletfacilities,andotherhygienicconditions analsobe considered shealthandnutritionalnputsprovidedby thehousehold.Thelocalhealthenvironmentonsists of all communitycharacteristicshatdirectlyaffectchild healthand aregenerallybeyondthe controlof theparents, uchasprevalence f parasites nd heincidenceof contagiousdiseaseamongthe generalpopulation.Finally,the child healthendowmentconsists of allcomponentsof the child's geneticinheritance hathave implications or his or herhealth.Householdhealthandnutritionalnputsare determinedby householddecisionsthatreflectthecharacteristics f thehousehold, he local community,andthechild,such as (initial) householdassets, parentalschooling, communityeconomic andhealth-relatedharacteristicssuchas theavailability ndpricesof medicalservices),andeach child'shealthendowment.Thispaper ocuses onparentalchooling,partic-

  • 8/6/2019 risna

    4/37

    ParentalSchooling

    HouseholdAssets

    d

    a' \a a c'a

    ParentalHealth HouseholdKnowledge Income

    abc \ v I/ acdHousehold HealthandNutritional nputs

    Child Health

    Figure 1The Determinants of Child Health

    e

    ExogenousVariables

    EducationOutcomes

    andEndogenous

    Variables

    HealthOutcome

    4--c

  • 8/6/2019 risna

    5/37

    Glewwe 127

    ularlymother'sschooling;father'sschooling,apart rom its income effect, is lesslikely to be importantor maintaining hildren'shealth.Schultz(1984) argues hat mother'seducationcan influencechild health n fiveways:1 1) Educationmaylead to a more efficientmix of healthgoods used to pro-ducechildhealth; 2) Bettereducatedmothersmay be more effective at producingchild healthfor a given amountandmix of healthgoods; (3) Schoolingcan affectparents'preferencesn systematicways-for example,educatedmothers endto optfor fewer but healthierchildren; 4) Moreschoolingshould raise family incomes,eitherthroughhigherwages or increasedproductivityn self-employment,whichshould mprovechild healthstatus;and(5) Education aisesthe opportunity ostsof time, which tends to increase he timemothers pendworkingoutside the homeand thus reduce time for child care-this effect of schoolingcould reduce childhealth by reducingboth maternaltime devoted to child care and durationofbreastfeeding. n Figure 1, the third and fourthpathwaysare representedby thearrowsa-a' (andalso by a-a")andacd (via a-a"'andc-c"'),respectively.2The firsttwo pathways,whichreflect he directeffect of the healthknowledgeandcognitiveskills thateducation mparts,havereceived ittle attentionn the literature.What sit aboutschoolingthat makes mothersmore efficient n producing hild health?Figure1 presents wo mechanisms hroughwhich schoolingcould influence hechoice of healthand nutritionalnputsvia the knowledgeand skills it provides.3First, schools may directlyteach effective health care practicesto students.Thispathway s denotedby b-abc. Forexample,the impactof diarrhea n child healthcan be reducedby oral rehydrationherapy ORT),which can be taughteven in

    primary chools(seeCash1983).Second,schoolingcan influence hild health nputsthrough hecognitiveskills imparted,uch as literacyandnumeracy.Literatemoth-ersarebetterable to readwritten nstructionsortreatingof childhooddiseases,andnumeracyenables mothersto bettermonitor llnesses and applytreatments.Thisdirecteffect is shownby c-c" inFigure1.Literacy ndnumeracy lsoenablemothersto increasetheir healthknowledgeby enablingthem to gather nformation romwrittensources.This indirecteffect is pathc-c'-abcin Figure1.Figure 1 also depictshow factors other than schooling influence child health.Householdphysicalassets raisehousehold ncomes(arrowd), which should havea positiveeffecton bothnutritionalnputs suchas calories)andenvironmentalon-ditions around he home. The choice of health andnutritionalnputswill also beaffectedby factorsassociatedwith the supply of these inputs in the community(arrow ). Forexample, heavailability ndqualityof healthandnon-health ommu-nityfacilitiesaffectsthedecisionshouseholdsmakeregarding ealthandnutritional1. Schultz'sframeworks primarily oncernedwith childmortality.Yet broadeningt to includeother,less severe,aspectsof child healthdoes not require ignificantmodification.2. The fifthpathway,via mother'stime, could be addedto Figure1 but is omittedto reduce clutter.Similarly, he impactof the thirdpathwayvia reduced amilysize could also be made moreexplicit(asa box labeled"familysize," another ndogenousvariable),but this is also omitted o reduceclutter.3. The distinctionbetweenpathways 1) and(2) in the previousparagraphs that the firstconcernsanefficient mix of physicalhealth nputs(for example,medicines)while the second addsefficient use ofnon-physicalnputs suchas caregivento the sick child).In thispaper, he distinctionbetweenphysicalandnonphysicalnputss not of primarynterest, atherheemphasiss onthe differentypesof knowledgeandskills learnedn school,and how theyaffectefficientuse of bothtypesof inputs.

  • 8/6/2019 risna

    6/37

    128 The Journalof HumanResources

    inputs.Finally,the child'shealthendowmentwill also affecthouseholdhealthandnutritionalnputs(via arrowg), since more sickly childrenusuallyreceive largeramountsof health andnutritionalnputs.B. RecentEmpiricalEvidenceManyrecentstudieshave examined heimpactof mother's andfather's)educationon child health.Forcomprehensiveeviews of the literatureee Behrman ndDeo-lalikar(1988), Behrman 1990), and Straussand Thomas(1995). The discussionherewill be limitedto an overview of a few recentstudies, ocusingon the impactof mother'seducationon height-for-age ndweight-for-height.Studiesof the determinants f child height andweight in manycountrieshavefoundpositiveeffectsof mother's ducation.Mostof these studiespresented educedformestimates,but a few wentfurther, xamining hepathwaysby which mother'seducationmproveschild health.In thePhilippines,Barrera1990)foundthatbettereducatedmothersended o weantheirchildren ooner,butthey compensatedor thisshortened reastfeedingime withbetter are;overall, heirchildrenwerehealthier smeasuredby higherheight-for-age -scores.4Theonly published tudythat focuseson the "information rocessing"attributes f schooling s by Thomas,Strauss,andHenriques 1991), whichused Braziliandata that includedvariables or whetherawoman reads a newspaper, istens to the radio, or watches television. Mother'sschoolingwasnotsignificantwhendummyvariableswere ncluded orthese"infor-mationprocessing"activities; henewspaper ndradiovariablesweresignificantnruralareasbutonly the television variablewas significantn urbanareas.Amongthe most interesting tudiesare those basedon the Cebu LongitudinalHealthandNutritionSurvey.Severalstudieshave used thesedata o modelthepath-waysby whichexogenousvariablesnfluence hildnutritional tatusandmorbidity.TheCebuStudyTeam(1991, 1992)found hatmother's ducation eadsto improvedwaste disposalandhighernon-breastmilkalorieintake,both of whichreducetheincidenceof diarrhea.Maternal ducationalso leadsto earlierweaning,whichcanincreaseepisodesof diarrhea, ut the net effect of maternal ducations to reducethe incidenceof diarrhea.An important ritiqueof findings hat mother'seducation mproveschildhealthis thehypothesis hateducation implyreflectsunobservedmaternal haracteristics.Wolfe and Behrman 1987) used Nicaraguandataon mothers'siblingsto controlfor unobservedamilyfixedeffects.Theyfoundthatapplying hesecontrols eavesno significant ffectof mother'seducation n childanthropometrictatus.However,Strauss 1990) foundin Cote d'Ivoirethat mother'seducationraiseschild height-for-ageandweight-for-height,ven afterusingfamily fixed effects estimators.Insummary,here s considerablevidence hatmother's ducationmproves hildhealth,andsomeevidenceon how this occurs.Still, thereareno studies hatdistin-4. Height oragez-scores,whichwill be usedin theempiricalworkbelow,arebasedonfittinga standardnormaldistributiono the growthcurvesof a healthypopulation f children.A childwith a z-scoreofzero is exactlyat the median n termsof heightforage, whilechildrenwithpositive(negative)z-scoresare taller(shorter)hanaverage.Low heightfor age z-scores ndicatestuntingdueto repeated pisodesof malnutritionver the life of the child,while low weightforheightz-scores ndicatewasting (weightloss) due to a current pisodeof malnutritionsee Gibson1990).

  • 8/6/2019 risna

    7/37

    Glewwe 129

    guish between the literacy and numeracy impacts of schooling and other, more gen-eral, impacts. Also, there are no studies that attempt to assess directly the impactof mother's health knowledge on child health.

    III. Analytical Framework, Data and EstimationStrategyA. Analytical FrameworkEstimation of the pathways by which mothers' schooling affects child health is notnecessarily straightforward.This subsection provides a framework for thinking abouthow to estimate these relationships. Recall Figure 1. The bottom of that figure showshow health and nutritional inputs, the environment and a child's health endowmentjointly determine child health. This can be expressed in terms of a production func-tion for child health:(1) Hi = f(HIi, Ei, Ei)where Hi is the health of child i, HIi is a vector of health inputs chosen by child i'shousehold, Ei is a vector summarizing the environmental conditions surroundingchild i, andEi s the child's genetic health endowment. Parents take this technologicalrelationship into account as best they can when making decisions that affect theirchildren's health. Although Ei and Eiareoutside the household's control,5health andnutritional input choices are chosen by the household.Estimation of Equation 1 would require detailed information on a large numberof health inputs, which is not feasible with the 1990-91 ENNVM data. However,as seen in Figure 1, one can substitute out these health inputs and obtain a reducedform relationship that shows how exogenous variables (those shown at the top ofFigure 1) determine child health:6(2) Hi = g(FSi, MSi, HAi, Ei, Ei)where FSi and MSi are father's and mother's schooling, respectively, and HAi is theinitial assets of child i's household.

    Although Equation 2 is much easier to estimate, and often has been estimated, itdoes not indicate what aspects of mother's schooling lead to improved child health.Referring again to Figure 1, one can obtain a better understandingof the impact ofmother's schooling by replacing it in Equation 2 with the educational outcomes itdirectly affects, namely cognitive skills, parental values and health knowledge:5. Thelocal healthenvironments not chosenby parentsf: a) migrationorpurposesof findinga betterhealthenvironments rare;andb) households annotpressureocal authoritieso improve helocal healthenvironment.The formerassumptions supported y migrationdata from the 1990-91 ENNVM;only0.5 percentof respondents eport hat"healthreasons"were the mainreason or theirmost recentmove.The latterassumption,whileharder o check,is plausible or Moroccobecause healthcareprovision shighlycentralized,withfew fundsunder he controlof local governmentssee WorldBank1994).6. The assumptionhatparental ducation s exogenousseemsreasonable or Morocco,whereaverageschooling ormenandwomanbetween heagesof 18and65 is only4.7 and2.3 years,respectively.Evenso, thisassumptionwill be checked n Section IV.

  • 8/6/2019 risna

    8/37

    130 The Journalof HumanResources

    (3) Hi = h(FSi, Li, Ni, Vi, HKi, HAi, Ei, ?i)whereLi, Ni, ViandHKistandfor mother's iteracy,numeracy,values andhealthknowledge, espectively.Thisequations aconditional emand elationship ecause,as explainedbelow, healthknowledge,andperhapsiteracyandnumeracy,maybeendogenous.Estimation f Equation wouldclarify hepathwaysby whichmother's choolingaffects childhealth,but such estimation s complicatedby severalproblems.First,it is difficult o observemother'svalues(Vi),and ndeed hereareno suchdata n the1990-91 ENNVMsurvey.Second,a child's healthendowment?i) is also virtuallyimpossibleto observe.Third,becauseparents' reatment f their children'shealthproblemsoften causes themto acquireadditionalhealthknowledge,healthknowl-edgeis likelyto be anendogenous ariable. nparticular, ealthknowledge s likelyto be negativelycorrelatedwith a child's (unobserved) enetichealthendowmentbecauseparentswith "healthy"childrenneednotacquire s muchhealthknowledgeas parentswith "sickly" children,ceterisparibus.7 ourth, t is also possiblethatliteracyandnumeracy reendogenousbecause actionsto acquireadditionalhealthknowledgemay lead to greateruse of those skills, thoughthe impactof a child'sgenetichealthendowmenton these variables s likely to be considerably mallerthan ts impacton healthknowledge.Theapproachesaken o dealwiththeseprob-lems will be discussed n detail in Subsection IICbelow.Anotherrelationship f interest s a variationof Equation3; whenattemptingoassess thepathwaysby which mother'seducationaffectschildhealthone maywishto isolate tsimpacton householdncome.8 nthiscase one canaddhouseholdncome(Yi)to Equation3 and removehouseholdassets(since their mpacton childhealthwouldoperateonlythrough heir mpacton household ncome).Thisyieldsthefol-lowing conditionaldemandrelationshipor estimation:(4) Hi = h'(FSi, Li,Ni, Vi,HKi,Yi,Ei, ?i).The relationships f primarynterest n this paperareEquations3 and 4.A finalissueto consider s thepossibility hatmother'seducationmproveschildhealthby reducing he numberof childrenwomen bear-with fewerchildren, hemothershouldbe able to allocatemore time and health nputsperchild.As men-tionedin SectionII, the desirefor fewer childrencanbe depictedas the impactofschoolingonparental alues.Thus,one couldmodifyEquations and4 byreplacingViwith the number f childrenbornor,moregenerally,by adding he lattervariablewhileretainingVi(since valuesmay affectchild health n otherways). Of course,since the numberof childrenborn s clearlyan endogenousvariable, stimation e-7. Thiscouldbe shownin Figure1 by an arrow eadingfrom the child'shealthendowment o parentalhealthknowledge.8. One couldgo even further. ncreased ducation an raise household ncomenot only by increasingwage ratesbut also by increasing he amountof time the motherworksoutside the home. Moreover,increasedime of the motherawayfromhomemayhavea direct,negative mpacton childhealth.Thusone could addbothhouseholdncome and mother's imespentworking o Equation . Thiswas tried nan earlierversionof thispaper see Glewwe1997),butit proved mpossible o find nstrumentalariablesthatcouldplausiblybe excluded romEquation andwere also good predictors f mother's imespentworking. n thispapermother's imespent workinghas beensubstitutedutof Equation .

  • 8/6/2019 risna

    9/37

    Glewwe 131

    quiresplausible nstrumentalariables; his will be discussedfurthern SubsectionIIIC.

    B. TheDataThis paperuses data from the 1990-91 EnqueteNationale sur le Niveau de Viedes Menages(ENNVM),which was implementedby Morocco's Directionde laStatistique.Thesurvey,which s basedon the WorldBank'sLSMSsurveys,covered3,323households romall areasof Morocco.Thesurveycollecteda varietyof infor-mation romeachhousehold,ncludinghousehold xpendituresnd ncome,employ-ment,education, ssets,agriculturalctivities,andmuchmore.A key aspectof thesedatafor thispaper s thattheycontain heheightandweightof all householdmem-bers.Even more mportants thata batteryof testswas givento householdmembersin two thirds 2,171)of thesampledhouseholds.The tests ncluded:1. Fivequestionson healthknowledge;2. Twelve questionson general knowledge(how to mail aletter,how to readan electricitybill, andso on); 3. An oralmathematics est of tenquestions;4. A set of writtenmathematics ests of varying degreesof difficulty;5. A set of Arabicreadingand writingtests; and 6. A set of Frenchreadingandwriting ests. The tests aredescribedn detail n Glewwe(1997).The healthknowl-edge test is of particularnterest,since it is rarelya partof any householdsurvey.It consists of five questionson vaccinations, reatingnfections,polio, diarrhea ndsafe drinkingwater.The test is fully described n AppendixI.All persons n the 2,171 selectedhouseholdsbetweenthe ages of 9 and 69 wereto be testedexcept: 1. Individualswith a baccalaureate egree9or higherlevel ofeducation ookonly the healthknowledge est since it was assumed hattheycouldobtainnearlyperfectscores on all other ests;and 2. Thehealthknowledge est wastakenonly by individuals etween heagesof 20 and50. The2,171 householdswhoparticipatedontained1,612 childrenage 5 or younger,of which 81 had motherswho did not participaten the tests for one reasonor another,eavinga sampleof1,531children. t is assumed hatthe 39 motherswitha baccalaureateegreewouldhave receivedperfectscoreson all the tests (exceptthe healthtest, whichthey didtake),which boosts the samplesize to 1,570. Droppingobservationswith missingvalues leaves a sampleof 1,495 children.Table 1 providesdescriptivestatisticson all variablesused in the analysis.Ofparticularnterestare the test scorevariables,which aredefined as the numberofquestionscorrectlyansweredby the respondent.They show substantial ariation,which is necessary o assess theunderlyingpathwaysby whichmothers'schoolingraises child health.In addition, hese scores shouldnot be highly correlatedwithyearsof schooling,or witheachother; f they are,regressionanalysis s less likelyto identify heunderlyingmechanisms.Table2 showscorrelation f years n schoolwith the test scores(thetablealso includesa test on readinga medicinebox-thiswill be discussed in Section IV). Mathematics,Frenchand Arabic scores are allhighlycorrelatedwith each otherandwithyearsin school (correlation oefficients9. Roughly speaking,a baccalaureateegree ies somewherebetween a U.S. high schooldegreeandacollege degree.It is only awarded fterpassinga rigorous et of examinations.

  • 8/6/2019 risna

    10/37

    132 The Journalof HumanResources

    Table 1DescriptiveStatisticsof VariablesUsed

    StandardVariable Mean DeviationHeightfor age Z-score -0.94 1.86Percapitaexpenditure 5,398.82 5,081.68Sex of child (female) 0.53 0.50Age of child (in months) 35.86 20.00Mother'sheight 157.02 6.20Father'sheight 168.49 5.95Father'sheight missing 0.31 0.46Mother'syearsschooling 1.98 2.97Father'syears schooling 3.14 4.10Healthknowledge 2.89 1.57Generalknowledge 1.53 3.18Oralmathematics 1.71 1.90Readingandwritingmathematics 1.18 2.62Arabicreading 2.33 4.98Arabicwriting 0.57 1.54Frenchreading 1.48 4.44Frenchwriting 0.26 1.06Rental ncome 1,460.79 13,920.17Childrenoverseas 0.01 0.09Irrigated ropland(hectares) 2.36 12.62Unirrigated ropland(hectares) 30.88 94.59Treecropland(hectares) 0.32 1.42Mother'smarried isters 1.88 1.56Father'smarried isters 1.51 1.56Father'smarried istersmissing 0.22 0.41Fatherbornhere 0.64 0.48Numberof televisions 0.55 0.58Numberof radios 0.89 0.55Availabilityof newspapers 0.21 0.41Mother's ather'sschooling 0.03 0.16Mother'smother'sschooling 0.01 0.10SampleSize: 1,495.

  • 8/6/2019 risna

    11/37

    Glewwe 133

    Table 2CorrelationAmongSchoolingand Test Score Variablesof MothersReadingYears Arabic French Health Box of

    Schooling Literacy Literacy Numeracy Knowledge MedicineYearsschooling 1.0000Arabic iteracy 0.8938 1.0000French iteracy 0.8869 0.8681 1.0000Numeracy 0.8665 0.8695 0.8383 1.0000Healthknowledge 0.3343 0.3867 0.3138 0.4356 1.0000Readingbox of medicine 0.8152 0.8553 0.7775 0.8198 0.4387 1.0000Note: All variablesare in logarithms.

    from 0.84 to 0.89).10Healthknowledgeis less highly correlatedwith these othervariables correlation oefficientsfrom0.31 to 0.44). Whetherregressionanalysiscandistinguishbetween heimpactsof the mosthighlycorrelatedkills is uncertainandwill become clearonly by examiningestimation esults.C. EstimationStrategy:GeneralConsiderestimationof equation(4) using the 1990-91 ENNVM.1 Child healthcan be measuredby the Z-scoreof childheightfor age (see endnote4), which indi-cates chronic malnutritionover a child's lifetime (stunting).In principle,onecould also use weight for height, but the 1990-91 ENNVM weight data sufferfrom seriousmeasurement rrorbecauseweight was recordedonly to the nearestkilogram.Severalestimationproblemsariseconcerning he explanatory ariables n Equa-tion 4. Begin with the child's local healthenvironment,Ei. Althoughthe 1990-91 ENNVM datacontain nformation bout ocal healthclinics and otherpertinentvariables, ome of the dataaremissingandcomparabilitys a problem. n addition,thesamplingprocedure sed forchoosingthehealth acilities coveredby thesurveyis unclear.Because the main interestof thispaper ocuses on household evel vari-ables(theskills andeducation evels of mothers),a simple community ixedeffectsprocedures used to avoidbias causedby omittedcommunityevel healthenviron-ment variables.This is possiblebecause the sampledhouseholdswere drawn rom140 primary amplingareas.Of theremaining xplanatory ariablesnEquation , two areunobserved:mater-nal valuesacquiredn school(Vi)andthe child's healthendowmente6). f the effect10. For simplicity, n the remainder f this paper(the logarithmsof) the two mathematics coresaresummed o createa singlemathematicsariable, nd hereading ndwriting coresaresummed orFrenchandArabic.For an analysisof themoredisaggregatedcores,see Glewwe(1997),thefindingsof whicharebasically he same as thosein thispaper.11. Thefollowingparagraphslso applyto Equation3, exceptthe discussionon choosing nstrumentalvariables orhousehold ncomeis irrelevanthouseholdncome hasbeen substituted utof Equation3).

  • 8/6/2019 risna

    12/37

    134 The Journal f HumanResources

    of schoolingon maternal aluesis important, ne could detectthisby addingyearsof schooling o Equation . A positiveeffect of yearsof schoolingon heightforagewould indicatethatvalues (or perhapssome otheraspectof schoolingother thanliteracy, numeracy,and healthknowledge)is an importantdeterminant f childhealth.If the years of schoolingvariablehas no perceptibleeffect, it is unlikelythatvaluesacquiredby mothers romschooling s an important athwayby whichschoolingaffectschild health.The inabilityto directlyobservea child's healthendowment E?)could lead tobiasedparameterstimatesdueto its correlationwith observedvariables.Onewayto reducesuch bias is to enterthe heightsof bothparentsas explanatory ariables,since tallerparentsarelikely to have betterhealthendowments,which in turnareinheritedby theirchildren. naddition,parents and heirchildren)displayvariationin height that is not relatedto healthstatus-healthy people can vary in height.Enteringparentalheight in the regressionscontrols for this as well, purgingthedependentvariableof variation n height that is not indicative of health status.Note that father'sheightis missingfor aboutone thirdof the children,eitherbe-causethe fatherdid not live withthe childrenor was unavailable or measurementat the time of the interview.To avoid losing this portionof the sample,whichcould lead to sample selectivity biases, a dummyvariable s createdindicatingthatfather'sheightwas missing; n such cases father'sheightvariable s set equalto the mean.Evenafteraddingparental eight o reducebias causedbyunobservedhildhealthendowments,t is stillpossiblethat heinability o observechildren'shealthendow-mentscouldbias estimatedmpactsof observedvariables.Healthknowledge,house-holdincome,andperhaps iteracyandnumeracy,maywell be endogenous.Particu-larlyworrisomes the fact thathealthknowledgemaybe negativelycorrelatedwiththe child's unobservedhealthendowment. n principle,usinginstrumental ariablemethodscan removebias,butthisrequiresplausible nstruments.Householdassetscan be used to instrument urrent ncome.12 he 1990-91 ENNVM containsdatathat can be used to construct everalhouseholdasset variables.The following areused in this paper:1. Three variableson agriculturaland (in hectares)ownedbythehousehold; . Household ental ncome(from and,buildingsanddurable oods);and 3. The numberof adultchildrenof householdmembers iving overseas(whomay send sizableremittances).Finding nstrumental ariables or mother'shealthknowledge s more difficult.Threedifferent ypes seem plausible: ndicatorsof the existenceof close relativeswho could be sourcesof healthknowledge;exposure o massmedia;andmother'seducation whichcan be excluded romEquation if one findsthat t is not neededas aproxyof theimpactof valueson childhealth).Onewaythatmothers anacquirehealthknowledge s from close relatives,especiallythose who have had children.The ideahere s thatmothers onsultwithotherrelativesconcerningheirchildren'shealth,andby doingso they add to their stockof healthknowledge.Mostof theserelatives do not directlycarefor the mother'syoung children,except perhaps hepaternalgrandmother,o theirimpacton the child's health comes aboutonly by12. In the estimatesgivenin latersections,householdexpenditures usedinsteadof household ncomebecause t is likely to be more accurate nd morecloselyrelated o households'permanentncomes.

  • 8/6/2019 risna

    13/37

    Glewwe 135

    raisingthe mother'shealthknowledge(thusthey can be excludedfromEquations3 and4).13The 1990-91 ENNVM containsdataon the numberof married istersof the motherand of her husband. t can also be used to add a dummyvariableindicatingwhether hehusbandwasborn n thecurrent laceof residence;f hewas,thereshouldbe severalmembersof his familyin the areafromwhom his wife canobtainhealthknowledge.Finally,the educationof the mother'sown parentscouldalso affect her healthknowledge(andher cognitiveskills as well), but can be ex-cluded romEquations and4 because he motherno longer ives with herparents.Mass media s also a useful sourceof healthknowledge nformationsee Thomas,StraussandHenriques,1991).It is unlikelythat thesevariableshave any effect onchild healthapart rom their mpacton mother'shealthknowledge; n particular,twould be rarefor parents o purchase elevisions or radios n response o havingasick child. The ENNVM data collect data on the numberof radios andtelevisionsin the household and on the availabilityof local newspapers.These mass mediavariablescould also be used as instrumentsor numeracyandliteracy.Finally,this paperwill also investigatewhetherpartof the impactof mother'seducation n childhealthworksby reducing amilysize,andsincefamilysize is alsoendogenous neneeds instrumentalariables or children verborn.Itis particularlydifficult to findplausible nstruments or this variable.Some possibilitiesare thenumberof married istersof both the womanandherhusband,which could reflectpreferences or children on both sides of the family, the education evels of thewoman'sparents,whichagainmayreflect amilypreferencesorchildren, nd inallytheage of thewoman,sinceolderwomen will havehadmoretime to bearchildren.Of course,one can imagineplausiblereasons or why these variablesmay directlyaffect childhealth,but thereareno better nstrumental ariablesavailable romthisdataset.Theinstrumentalariablesdescribedn thepreviousparagraphsenerallyappearreasonable, ut onecannotprove hat heydonotbelong nEquation . Thisproblemplagues most, if not nearlyall, applicationsof instrumental ariables.Thus oneshouldapplya specificationest to check theplausibility f theunderlying xclusionrestrictions.This is doneusing standard veridentificationests (see Davidson andMacKinnon1993).D. EstimationStrategy:Decomposingthe SeparateImpacts ofMother'sEducationFinally, considerhow estimates of Equations3 and 4 can be used to assess themechanismsby which mother'sschooling eads to improvedchild health.As seenin Equation , andFigure1, theoverall mpactof mother's choolingon childhealthcan be decomposedas follows:

    ) aHi aHi aLi a- - aHi avi(5) HH +aMSi aLi aMSi aNi aMSi av,i MSi13. InMoroccan ulture,whenwomenmarryhey ointheirhusband's ousehold.Thus hemothermovesawayfromherparents ndsiblings,andherhusband'smarried istershave movedawayfromhis family.Only the husband'smother,his unmarriedisters,and the wives of any marriedbrothershe may havebelongto his (andthus to his wife's) household.

  • 8/6/2019 risna

    14/37

    136 The Journalof HumanResources

    + Hji (3HKi aHKi a3L aHKiaNi aHi 3YiaHK, aMSi aLi aMSi aN Si aYi MSi

    The first,second, third,fourth,and fifth terms to the rightof the equal sign showtheimpactof mother's choolingvia its impacton literacy,numeracy, alues,healthknowledge,and ncome,respectively.Note thattheimpactsvia literacyandnumer-acy (the first and secondterms)are direct effectsonly (arrowc" in Figure1); theindirect ffects via theimpacton healthknowledge arrow ' in Figure1) arerepre-sentedby the second and thirdterms nside the parentheses f the fourth erm.Asimilardecomposition asedon Equation3 is identicalexceptthatthe incometermis dropped; he remainingpartialderivativesmay differ fromthose in Equation5because heincomeeffectis "dividedup" among heremainingerms forexample,the impactof literacynow incorporates oth a direct effect andthe indirecteffectof literacy hroughts effect on household ncome).Assumesimplelinearfunctional ormsfor Equations2, 3, and 4:(2') Hi = Po + FFSi + P2MSi + P3HAi + Ul(3') Hi = Yo+ ylFSi + y2Li + y3Ni + Y4Vi+ y5HKi + y6HAi + u2(4') Hi = 80 + 8lFSi + 82Li + 63Ni + 64Vi + 65HKi + 66Yi + U3.In each specification,an error ermis addedto account or randommeasurementerror n childheightandthe child's unobserved ealthendowment recall hatcom-munity ixed effects estimationallows one to dropEifrom eachequation). nEqua-tion 2', P2 estimates the overall impact of mother'sschooling on child health,aHil/MSi, which is the left hand side of Equation5. Estimatesof all the partialderivatives n therighthandsideof Equation comefrom:1. Equation , orEqua-tion 3 if the income termis dropped romEquation5; 2. Reduced orm estimatesof the impact of mother's schooling on literacy (Li), numeracy (Ni), values (Vi),healthknowledge(HKi),and household ncome(Yi),whichcan be denotedas oL,aN, a, (HK, and oty, respectively; and 3. Estimates of the impact of literacyand numeracy on health knowledge, which can be denoted as rL and TH, respec-tively.14

    ThusEquation5 becomes:(5') 12 = 620L + 63(aN + 64aV + S5(OHK + LaL + rTNO(N) + 66aY.This decomposition s based on Equation4' and thus explicitly accountsfor theimpactof mother'seducation n income.If income s substituted ut,as in Equation3', the last term s dropped romEquation5 and the decomposition ecomes:(5") 32 = Y2a2L + Y3aN + Y40v + Y_5(OHK + I1LaL + T_NAN).

    14. These mpacts rL and1nH)are shownas arrow ' in Figure1.If literacyandnumeracy anbe consid-eredexogenouswithrespect o healthknowledge,namely, iteracyandnumeracy hangeverylittle afteroneleavesschoolbut healthknowledge anchange, henthisrelationships a reduced orm.On theotherhand, f literacyandnumeracy reendogenoushenthis is a conditionaldemand unction.Thiswill befurtherdiscussedwhen thisrelationships estimatedn SectionIV.

  • 8/6/2019 risna

    15/37

    Glewwe 137

    Estimation f thedifferentparametersf Equations ' and5"allows one to examinenotonlywhichpathwaysare mportant, ut alsoto assess theirrelativecontributions.

    IV. Estimation ResultsA. Reduced Form Estimatesof the Determinantsof ChildHeightTable 3 presentsordinary east squares(OLS) and community ixed effects (FE)estimatesof Equation2, that s reduced orm estimatesof the determinantsf childheightfor age.15TheOLSand FE estimates n Table 3 aresimilar,butspecificationtests favor the fixed effects specification.16ecause of this, all remaining stimatesin this paperwill incorporate ommunity ixedeffects.

    Table 3 shows thatmother'syearsof schoolinghas a significantlypositiveeffecton child height,which is consistentwith evidence fromother countriesdiscussedin SectionII.17 his s themain indingof interestnTable3;the restof thisparagraphbrieflydiscusses he otherexplanatory ariables.A dummyvariable orfemalechil-dren s negativebutinsignificant, ieldingno strongevidenceof genderdiscrimina-tion in child health.Child height for age varies substantiallywith child age (inmonths); his reflectscommonpatternswherebymalnutritionises with age in thefirst woyearsof life (untilweaningends)but then evels off (andmayevendecline).Bothmother'sandfather'sheightarepositivelycorrelatedwithheight orage,whichpicksupbothvariationn the child'sunobserved ealthendowment ?i)andnaturalvariation n height amonghealthychildren.Father'syearsof schooling s positivebutinsignificant,whichsuggests ittle directeffect on childheight.Perhapsa weakindirecteffect exists via household ncome; his will be checkedbelow.The house-hold asset variablesare all insignificant,hough hreeof the five have the expectedsigns.Before moving to conditionaldemandestimates, t is useful to check whethermother'syearsof schoolingcanreallybe specifiedas exogenous,andwhetherpartof the effect of mother'sschoolingon child healthworksthroughreduced amilysize. To investigate heformer,mother'syearsof educationwas instrumentedsingthe educational evels of bothherparents,as well as the numberof married istersshe has. These instrumentswere good predictorsof mother'sschooling (jointF-statisticsover20 for bothOLS andfixedeffectsestimates,withp-valuesof 0.000).15. All yearsof schoolingandtest scorevariablesarespecified n logarithmicormbecause:1. It seemsreasonable o assume hatattainmentf the mostbasic skillswouldhavelarger mpact hanwould attain-ment of additional killsamongpersonswho alreadyhave basicskills;and2. In general, aking he logsof these variables lmostalwaysfit the databetter asmeasured y R2statistics). f yearsof schoolingoranytest scoreis zero,the log of it is set to zero,and the sameapplies o rental ncome.16. A Hausmanestof fixedeffects versus he nullhypothesis f random ffectsyieldsaX2(d.f. 3)statisticof 20.29, which is statistically ignificantat the 10 percent evel. Since Hausman ests often have lowpowerto reject he null,it is prudento reject he random ffects specification.n turn, hatspecification(notshown n Table3, but similar o theOLSresults) s favoredoverOLS; heBreusch-Paganagrangemultiplieresthas a X2(d.f.1) statisticof 46.70, clearlyrejectinghenullhypothesisof homoscedasticity.17. Inregressionsnot shownhere,a squaredermof (thelogarithm f) mother'syearsof schoolingwasadded to both specificationsn Table 3 to allow for a more flexiblefunctional orm. The linear termremained ignificantat the 10 percent evel), butthe squared erm was completely nsignificant.n therest of thispaperonly the linear erm s used.

  • 8/6/2019 risna

    16/37

    138 TheJournalof HumanResources

    Table 3ReducedFormEstimatesof Determinants f Height or Age Z-ScoresCommunityOLS Fixed Effects

    Variable Coefficient t-Statistic Coefficient t-StatisticConstant -16.931 -10.83Sex (female) -0.149 -1.69 -0.116 -1.31Age (months) -0.081 -9.14 -0.076 -8.55Age2(months) 0.001 7.70 0.001 7.11Mother'sheight 0.066 8.95 0.052 6.39Father'sheight 0.041 5.39 0.041 4.96Father'sheight missing 0.155 1.61 0.167 1.60Log mother'sschooling 0.226 3.17 0.165 1.97Log father'sschooling 0.076 1.43 0.050 0.82Log rental ncome 0.020 1.03 -0.016 -0.74Childrenoverseas 0.517 0.99 0.659 1.21Irrigated ropland -0.001 -0.28 0.001 0.14Unirrigated ropland 0.001 1.21 0.000 0.61Treecropland -0.038 -1.23 -0.014 -0.39R2 0.170 0.299Lagrangemultiplierest(OLSvs. random ffects),X2(l) = 46.70,p-value= 0.000Hausman est (random ffects vs. fixedeffects),X2(13) = 20.29,p-value = 0.088Samplesize: 1,495.

    The impactof mother'sschoolingon childhealthwas substantiallyower,andnotsignificantly ifferent romzero for bothOLS andfixed effects estimates.However,the standard rrorswere muchlarger(0.256 for OLS and 0.374 for fixed effects),so thatstandardHausman ests could notreject hehypothesis hatthe instrumentalvariableestimateswereequalto those given in Table 3.Regardingheimpactof education n familysize, simplereduced ormOLSesti-mates(not shownhere)confirm hatmother'syearsof schoolinghas a strongandsignificantlynegativeimpacton family size. Unfortunately,he only instrumentalvariable hathada significant mpacton fertilitywas mother'sage. This was truefor two distinctfertilityvariables ried,numberof childrenand age of motheratfirst birth.The fixed effects specification n Table3 was reestimated ddingeachvariable eparately, ndboth instrumentednd uninstrumentedersionswere tried.In eachcase, thefertilityvariablewas not statistically ignificant ndtheparameterestimatefor mother'syears of school showed little change.Thus it appears hatalthoughmother'seducationdoes reducefamily size, there s no strong mpactoffamily size on childhealthaftercontrolling or mother'seducation.

  • 8/6/2019 risna

    17/37

    Glewwe 139

    B. ConditionalDemand Estimatesof ChildHeight (not Conditioningon Income)Table4 presentsestimatesof Equation3, the conditionaldemanddeterminants fchildheightwhere income is not one of the conditioning ariables.Threekindsofestimatesare shown:community ixed effects (FE);two stage least squareswithcommunity ixed effects(2SLSFE)withonlyhealthknowledgespecifiedas endoge-nous;and 2SLSFEwith all four skills test (mathematics,Arabic,French,andhealthknowledge) pecifiedas endogenous.The lasttwo kindswereestimated wice,oncewith mother'seducationandonce without t, for reasonsexplainedbelow.The FE estimatesshown in Table4 areidentical o the FE estimates n Table3,exceptthatthefourtest scorevariableshave beenaddedas (exogenous)explanatoryvariables.Mother's chooling s nowcompletely nsignificant, ndevenslightlyneg-ative,which suggeststhat the fourtest score variables ully capture he impactofmother'seducationon childhealth.However,althoughall four test scorevariablesshowpositiveeffects noneis evenclose to beingstatistically ignificant, ndneitheraretheyjointly significant.Theseresultsarequitepuzzling.Oneproblemwith thisspecifications thathealthknowledge,andperhapsheother estscorevariables,maybeendogenous.Theremaining pecificationsnTable4 use2SLSFE pecificationsoaddress hisproblem.Thesecondand hirdcolumnsof Table4 are dentical othe firstexceptthathealthknowledge s specifiedas endogenous,usingtheinstrumentalariablesdescribed nsubsectionIIIB.The one differencebetweenthe secondand thirdcolumns is thatthe former ncludesmother'sschooling(to account or the impactof values)whilethelatter xcludes t (toprovidean additionalnstrumentalariable orhealthknowl-edge). The mainfindingwhenhealthknowledge s specifiedas endogenous s thatsuch knowledgehas a strongand statisticallysignificant mpacton child health.Indeed, n the specification hatincludesmother'seducation he pointestimate n-creases50-foldandis significantat the 1 percent evel."8Whenmother'seducationis excluded from the specification he impactof healthknowledgeis somewhatsmaller,but still highly significant t-valueof 2.48). None of the othertest scorevariables mathematics,Arabic,andFrench)have significantlypositiveeffects, andin factbothArabicandmathematics ave negativeeffects.Oneissue thatarisesat thispoint s what s meantby healthknowledge.The fivequestions hatcomprise hehealthknowledge est areshown in AppendixI. Couldit bethatwhatreallymatterss onlya subsetof thesequestions?Ofthefivequestions,theoneconcerning accinationsprovidedverylittleinformation ecause95 percentof the womenansweredt correctly.Thepercentage f womenanswering he otherfourquestions orrectlyvariedbetween44 percentand54 percent.Thesefourques-tionswere alsofairlyhighlycorrelatedwith eachother,withcorrelation oefficientsbetween0.40 and0.50. To checkwhether omequestionsmatteredmore hanothers,the regression n the second columnof Table 4 was reestimated ive times, eachtime withtheresponse o a singlequestionreplacing hehealthknowledgevariable.18. At thesuggestion f onereviewer, eparateegressionswererunby sex andage(thetwoage categoriesbeing0-35 monthsand36-71 months) o see whether he impactof healthknowledgevariedby sex orage. Although hereweresome differences theimpactwas larger or girlsthanforboys, andlarger oryoungerchildren),heywere not statistically ignificant.

  • 8/6/2019 risna

    18/37

    Table 4Conditional Demand Estimates of Determinants of Height for Age Z-Scores

    Two Stage Least(Including Community

    CommunityFixed Only Health KnowledgeEffects EndogenousFather's schoolingArabicFrenchMathMother's schoolingMother's health knowledgeSex (female)Age (months)

    Age2 (months)Mother's heightFather's height

    0.012(0.20)0.048(0.63)0.046(0.45)0.024(0.27)-0.012

    (-0.08)0.037(0.41)-0.109

    (-1.25)-0.075(-8.06)

    0.001(6.92)0.052(6.69)0.041(5.33)

    -0.114(-1.32)-0.129(-1.04)0.199

    (1.37)-0.370(-1.97)0.357(1.40)2.020(2.71)-0.076

    (-0.72)-0.072(-6.84)

    0.001(5.73)0.040(3.70)0.040(4.01)

    -0.073(-0.96)-0.033(-0.34)0.216

    (1.60)-0.221(-1.52)

    1.452(2.48)-0.084

    (-0.85)-0.073(-7.43)0.001

    (6.19)0.043(4.46)0.040(4.33)

  • 8/6/2019 risna

    19/37

    Father'sheight missingRental ncome

    ChildrenoverseasIrrigated ropland(hectares)Unirrigatedropland(hectares)Treecropland(hectares)Overidentificationest (d.f.)F-tests of identifying nstrumentsMother'shealthknowledge

    ArabicFrenchMath

    Hausman ests (d.f.)(Endogenousparameters nly)

    0.136(1.34)-0.017

    (-0.66)0.224(0.39)-0.001

    (-0.13)0.000(0.70)-0.019

    (-0.51)

    0.262(1.96)-0.030

    (-1.13)0.354(0.53)-0.002

    (-0.38)0.001(1.29)0.048(0.96)10.16(8)[0.254]4.42[0.000]

    7.16(1)[0.007]

    0.220(1.81)-0.026

    (-1.06)0.290(0.46)-0.002

    (-0.34)0.001(1.16)0.029(0.65)14.07(9)[0.120]5.76[0.000]

    5.95(1)[0.015]Notes:1. Samplesize is 1,451.Thisis lowerthan n Table3 because scoreson the healthknowledge est weremissingfor 22 observmissingat leastone test score for themathematics,ArabicandFrench ests.2. Asymptotic -statistics hown n parentheses.3. P-valuesof specificationestsshown n brackets.

  • 8/6/2019 risna

    20/37

    142 TheJournalof HumanResourcesAs one wouldexpectgivenits lack of variation,heanswer o the vaccinationques-tion had no significanteffect on child health.In contrast,each of the four otherquestionshada significantly ositiveeffecton childhealth,with t-statistics angingfrom 1.71 to 1.99 andparameterstimatesranging rom 1.24to 1.79.Thissuggeststhat the healthknowledgeembedded n each of these questions s important, ndbecause the different ypes of healthknowledgeare fairly highly correlatedwitheach other hey maywell reflect heimpactof other ypesof basic healthknowledgenot measured n this test. In addition, hese resultsalso suggestthatspecifyingtheoverall mpactof healthknowledgeas the sum of the scoreson this testis a reason-able way to aggregatehealthknowledge nto a single variable.Given how the resultschangewhenhealthknowledge s specifiedas endogenousin the secondandthirdcolumnsof Table4, it is worthwhile o applysomespecifica-tion tests to theseregressions. n general, nstrumental ariablesmust notbe corre-lated with the error erm of the equationof interest u2 in equation3'), and theymust provide strongexplanatorypower for the endogenousvariable(s).The firstrequirementanbe checkedby anoveridentificationest. In bothcolumns he exclu-sionrestrictions re notrejected thep-valuesbeing0.254 and 0.120 in the secondand thirdcolumns,respectively).The secondrequirements verifiedby F-tests onthe explanatory owerof the identifying nstruments,he nullhypothesisof no ex-planatory ower s resoundingly ejected see Appendix I for the firststage regres-sions andpartialR2statistics, he latterof whicharerecommendedyBound,Jaeger,andBaker1993).Finally,a Hausman est is usedto examinewhether he 2SLSFEspecifications preferredo the FE specification.Thistest rejectsthe FE specifica-tion.19Overall, hespecificationests ndicate hathealthknowledge houldbetreatedas endogenousand that the instrumental ariablesused satisfybothrequirements.There s one moreset of regressionso checkbeforeconcludinghathealthknowl-edge is the mostimportant athwayby which mother's ducation eads to improvedchild health.Perhaps he mathematics,Arabic and Frenchtest score variables nEquation3 are also endogenous,so thatwhen they are specifiedas such duringestimation heywill also yield significant mpactson childhealth.This is examinedin the last two columnsof Table4, usingthe same instrumental ariablesusedforhealthknowledge.The basic resultsareunchanged-healthknowledgehas a large,positive and statistically ignificant mpacton child health,and none of the othervariablesdoes. Thisis trueregardless f whethermother'seducations includedasan explanatory ariable.Morespecifically,when mother'seducation s includedas a regressor Column4) Arabichas a small (relative o the impactof healthknowledge)positiveeffect,but it is completely nsignificant t-statisticof 0.16); when mother'seducation sexcluded (Column5) the impact is slightly negative and even less significant(t-statisticof -0.00). French anguageskills have implausiblenegativeeffects inbothColumns and5, andarecompletelynsignificant. inally, he mpactof mathe-matics s stronger in termsof the size of thecoefficient) hanArabicorFrench,but19. ThisHausman estexaminedonlythecoefficienton healthknowledgen order o increase hepowerof the test to reject henullhypothesis.All Hausman estsin theremainderf thispaperareappliedonlyto the parametersssociatedwithpotentially ndogenousvariables, or the samereason.Hausmanestswere alsorunon theentire et of explanatoryariables;neverycasetheyfailedtoreject he nullhypothe-sis, whichprobably eflects heir ow powerwhenjointly testinga largenumber f parameters.

  • 8/6/2019 risna

    21/37

    Glewwe 143

    it is still statisticallynsignificant t-statisticsof 0.92 and0.57 in Columns4 and5,respectively).To see whether he statistical ignificance f mathematics killsmightimprove f onlyit andhealthknowledgewere nstrumented,heregressionn Column4 was reestimatednotshownhere)withonlythesetwo test scorevariablesasendog-enous. The coefficienton mathematics killsdrops o half its value(0.402) and thet-statistics correspondinglymuch ower(0.55).Thusspecifyingmathematics,Ara-bic, andFrenchas endogenousdoes not changethe basic resultsfoundwhen onlyhealthknowledge s treatedas endogenous.Thespecificationests donefor Columns2 and3 werealsodoneforColumns4 and5. Bothspecifications assed heoveridentificationest,indicating hat he exclusionrestrictions re reasonable nd theexplanatory owerof theidentifying nstrumentsis quitegood (especiallywhen mother'seducations not used as aregressor).Unlikethespecificationswhenonlyhealthknowledge s endogenous,Hausman estsof thejointendogeneity f the fourtest scorevariablesonlyweaklyreject he FEspecifica-tion (p-values of 0.065 and 0.073 for Columns4 and 5, respectively).In fact, aHausman est of thejoint endogeneityof the mathematics,Arabic,and French estscoresdoes notreject he nullhypothesis f exogeneity thex2statisticwith 3 degreesof freedom s 3.83,whichhas ap-valueof 0.280).Thus t is not clear hat hespecifi-cationwith all test scoresendogenous s preferableo that withonly healthknowl-edge endogenous.A final ssueto consider egardingTable4 is whetherpartof the effect of mother'seducationworksthroughreductionsn the numberof childrenborn.Using age ofthemother,and ts square,as additionaldentifyingvariables, heregressionsn thelast four columns of Table4 were reestimated wice, once addingthe numberofchildrenandagain ryingageof motheratfirstbirth.Neithervariablewassignificantat the5 percent evel, and the impactof mother'shealthknowledgedid notchangeappreciablythough n one of the eight cases it lost statistical ignificance).More-over,the sameresultshold whenthese two variablesare treatedas exogenous.Thusthere is no evidence thatmost, or even some, of the effect of mother'seducationon child healthoperates hrough eductionsn family size.To summarize heresultsof Table4, it appearshathealthknowledge s the mainpathwayby which mother's education eads to healthierchildren.This was seenonly when mother'shealthknowledgewas specifiedas endogenous; reating t asexogenousgreatlyunderestimatedts impacton childhealth.Numeracyand iteracyskills never showedany significantlypositive impact,whether hey were specifiedas endogenousor exogenous. In the three specifications hat includedmother'sschoolingas a regressor,ts coefficientwas neversignificant,which casts doubtonthehypothesis hatanimportant athwayby which mother's choolingaffects childhealth s by changing he values of the mother.C. ConditionalDemand Estimatesof ChildHeight (Conditioningon Income)Turnnow to the finalpathway o investigate, hatvia household ncome. Table5shows threeregressions hat estimateEquation4, the conditionaldemand or childhealth hat ncludes ncomeas a conditioning ariable.TheFE estimates n the firstcolumnare n manyways similar o thosein Table4; in particular, lthoughall fourtest score variables how positiveeffects noneis close to being statistically ignifi-

  • 8/6/2019 risna

    22/37

    Table 5Conditional Demand Estimates, Conditioning on Income, of Height for Age Z-Scores

    Two-StageLe(IncludingCommuCommunity Expenditures ndHealthFixed Effects KnowledgeEndogenous

    Father's choolingArabicFrenchMathMother'sschoolingMother'shealthknowledgeExpenditures ercapitaSex (female)Age (months)

    -0.009(-0.15)0.071(0.93)0.025

    (0.25)0.013(0.15)-0.051(-0.32)0.025

    (0.28)0.224(2.14)-0.105

    (-1.21)-0.074(-8.03)

    -0.134(-1.59)-0.027(-0.23)0.113

    (0.84)-0.295(-1.84)0.159(0.68)1.443(2.47)0.587(1.79)-0.070(-0.71)-0.071

    (-7.28)

  • 8/6/2019 risna

    23/37

    Age2(months)Mother'sheightFather'sheightFather'sheightmissingOveridentificationest (d.f.)F-testsof identifying nstrumentsMother'shealthknowledge

    IncomeArabicFrenchMath

    Hausman ests (d.f.)(Endogenous arametersnly)Notes:1. Sample size is 1,451.2. Asymptotic t-statistics shown in parentheses.3. P-values of specification tests shown in brackets.

    0.001(6.89)0.050(6.31)0.040(5.27)0.140(1.39)

    0.001(6.05)0.037(3.55)0.038(4.10)0.218(1.79)11.46(12)[0.490]3.74[0.000]14.89

    [0.000]

    7.32(2)[0.026]

  • 8/6/2019 risna

    24/37

    146 The Journalof HumanResources

    cant,andmother's choolinghasno explanatory ower.Household xpenditures ercapitahas a significantly ositiveeffect,as expected.Of course,household ncome,healthknowledge,andpossiblymathematics,ArabicandFrench,may be endoge-nous.The secondandthirdspecificationsn Table5 allow for this.20The secondcolumn of Table 5 treatsboth healthknowledgeandhouseholdpercapita expendituresas endogenous,using the instrumental ariablesdescribed nSubsection IIB. As in Table4, healthknowledgehas a significantly ositiveeffecton childhealth,butthe other est scoresare nsignificant nd/ornegative.Householdincome has a positiveeffectthat s significant t the 10percentevel, andmorethantwice as largeas the effect shown in the FE estimatesof Column 1.21The samespecificationests shown n Table4 are also shownhere. The overidentificationestdoes not rejectthe exclusionrestrictionsmpliedby the choice of instruments, ndthose instruments ave strongexplanatory ower. Finally,the Hausman est showsthat the 2SLSFEresults aresignificantlydifferent rom the FE results(p-valueof0.026). Thus,assuming hatmother'seducationaffects household ncome(which sverifiedbelow), there s a pathwayother hanhealthknowledgeby whichmother'seducationaffectschild health.The thirdcolumnof Table 5 treatsall test scorevariables,as well as per capitaexpenditures, s endogenous.As in Table4, this doesnotchange hegeneral indingthathealthknowledge s thekey skill thateducatedmotherspossessthatraisestheirchildren'shealth.Thespecificationestsshow thatthe instrumental ariablesappearreasonable, ndtheHausman est doesnotsupporthehypothesis hatall fiveendog-enous variablesare ndeedendogenous. ndeed,whenonlythemathematics,Arabic,andFrench est scoresarespecifiedas endogenous, he Hausmanest cannotrejectthenullhypothesisof exogeneity(thex2statisticwith3 degreesof freedom s 2.57,whichhasap-valueof 0.463).Notefinally hat he coefficientonpercapitaexpendi-tures s almostthe samein Column3 as in Column2, thoughno longersignificantat the 10 percent evel (the t-statistic s 1.45).Overall, he results n Table 5 showthat household ncome is anotherpathwayby which mother'seducation an affectchildhealth(andthe onlypathwayby which father'sschoolingaffectschildhealth,sincethefirststageregressionsn TableAl of Appendix Ishowa significantmpactof father'sschoolingon householdexpenditures).The relativemagnitudesof thedifferent mpactsof mother'sschoolingwill be examined n Section V.D. Health Knowledgeor the Abilityto Read a MedicineBox?:A Brief DigressionAs mentioned n Section III, another est given to respondentsn the 1990-91ENNVM was one on "generalknowledge."This test consistedof 12 questions20. In addition othethree pecificationshown,analogous egressionswererun nwhichmother's duca-tion was omittedas an explanatory ariable.Theywerevery similar o the resultsshown here andthusarenotpresented o reducecluttern this table.21. To check for nonlinearities,his regressionwas repeatedwith a squared xpendituresermadded(notshownhere).Thesquaredermwascompletely nsignificant. ecauseaddinga squaredermhasthedisadvantagef addinganother ndogenousvariable o theregression,he squaredermwas not usedintheestimation esultsreported ere.

  • 8/6/2019 risna

    25/37

    Glewwe 147

    concerning veryday ife in whicha commonobject s handed o therespondent ndtwo or threequestionsare asked about t. Theobjectswere a national dentitycard,a letter,a box of medicine,a newspaper ndanelectricitybill. Of particularnterestfor this studyare the questionsconcerninga box of medicine.The threequestionsaskedwere:1.Wheredoes it show how manypills are n thebox?;2. Wherearetheinstructionsorusingthemedicine?; nd3. Where s the dateof expirationndicated?Presumably,motherswho are more able to answerthese questions correctlycanprovidebetterhealthcare for their children.For purposesof this paper,two questionsarise.Do the medicine box questionsmeasure omething hat heother estsdo not?Andif theydo, whatare theimplica-tionsregardingherelationship etweenmothers'healthknowledgeandchildhealth?A mother'sabilityto read a medicinebox may simplyreflect her abilityto readArabic(all writingon the medicine box was in Arabic),the impactof which onchild healthhas alreadybeen examined.Alternatively,t may be thatthe abilitytoreada medicinebox is a formof healthknowledge. n thiscase,it would be interest-ing to examinewhetherreadinga medicinebox is the "most important"kind ofhealthknowledge inwhich casethatabilitywould "displace"most of theexplana-tory power of healthknowledge)or whether t plays only a smallrole (andthuswouldnot "displace"theexplanatory owerof healthknowledge).Thecorrelationsshownin Table 2 suggestthatthe abilityto reada medicinebox maybe little morethananotherversionof the test for Arabic iteracy,since the correlation oefficientbetween hesetwo variabless 0.85. However, t is alsofairlyhighlycorrelatedwithhealthknowledge correlation oefficientof 0.44). Furthernvestigation equires e-gressionanalysis.Therole of theabilityto reada medicinebox is examined n Table6. In order ofocusontherelationship etween hatabilityandhealthknowledge, hemathematics,Frenchand Arabictest score variableshave been omitted(recallthatthey hadnoexplanatorypowerin Tables4 and 5). The firsttwo columnsshow FE estimates,with (Column1) and without(Column 2) mother'sschooling.As usual with FEestimation,mother'shealthknowledge s completely nsignificant.However,moth-er'sability o reada medicineboxis significant tthe 10percentevel whenmother'sschooling s includedas a regressor ndnearlyat the 1 percent evel whenmother'sschooling s excluded.Thus thisvariableappears o be capturingmorethan ust theability o readArabic,whichwasneversignificantn anyof theprevious egressions.Becausethepreviousregressionspresentedairlyconvincingevidencethatmoth-er's healthknowledge s endogenous, hatvariable houldbe instrumented.n addi-tion, it is prudent, ndintuitivelyplausible parentswith sickly childrenhave moreexperiencereadingmedicineboxes), to specify the abilityto reada medicineboxas endogenous.Theseregressionsare shown in Columns3 and4 of Table 6. Whenmother'seducation s included n the regression,both mother'shealthknowledgeandthe abilityto read a medicinebox have muchlargereffects.22While neither ssignificantat the 5 percent evel, healthknowledge s significantat the 10 percent22. Note thatmother'sability o reada medicinebottlehasnot been transformedntologarithms ecausethe originalvariable angesonly from0 to 3. Moreover, ransformingt to logs would haveequated e-sponsesof 0 and1, whichtogetheraccountedor 85 percentof theresponses.

  • 8/6/2019 risna

    26/37

  • 8/6/2019 risna

    27/37

    Glewwe 149

    level, and the estimatedcoefficientis aboutthe same magnitudeas it was in theregressions n Tables4 and 5. Note, however,that mother'seducationhas a verylarge negativecoefficient,with a t-statisticof -1.51. This suggestspossible col-inearity problems.When mother's education s dropped rom the regressiontheimpactof the ability to read a medicinebox becomes completely insignificant,and even becomes slightly negative, while health knowledge remains stronglysignificant.Overall, he resultsshownin Table6 do not alterthe conclusionreachedabovethathealthknowledge s thekey aspectof mothers ducation hat eads to improvedchildhealth.Although heabilityof motherso reada medicineboxinitiallyappearedto contain nformation otpickedupin healthknowledge,2SLSFEestimation astsseriousdoubton thisproposition.ncontrast, slongas healthknowledge s specifiedasendogenoust yields statistically ignificant esults at eastatthe 10percentevel)and the parameterstimatesarefairlystable.To summarizeSectionIV, the fundamental esult s thatmother'shealthknowl-edge is the key mechanismby which mother'seducation eads to improvedchildhealth.A secondfinding s of a moremethodological ature:gnoring heendogene-ity of mother'shealthknowledgemay seriouslyunderestimatets role in promotingchildhealth.

    V. Decomposing the Impact of Mother's Schoolingon Child HealthThe finding n the previoussectionthatmother'shealthknowledgeplays by far the mostimportantole in determininghildhealthdoes notnecessarilyimplythat iteracyandnumeracy o notmatter.As explained n SectionII,mother'shealthknowledgemay developafter eavingschool, and in a way thatwill dependon mother's iteracy,and perhapson numeracyas well. This section attempts odecompose he overall mpactof mother's choolingon childhealthby decomposingthe parameter2 given in Equation2' in Section III.RecallfromSectionIIIthat he overalleffectof mother's ducation n childhealthin (2'), 32,canbe decomposed n two ways. If one explicitlyincludesthe pathway

    thatoperatesvia household ncome,the decompositions thatgiven in Equation5'of SectionIII. If income effects are substituted ut, the decomposition s that inEquation5". The resultsin Section IV indicate that the directeffect of literacy,numeracy nd valueson child health(that s thecoefficients82, 63 and84)on childhealth were not statistically ignificant rom zero. The 2SLSFE estimates n Table5 indicatethatthe impactof household ncome on child health(measuredby 66)was approximatelyqualto 0.59.23Finally,the best pointestimates or the impact

    23. Tables4, 5, andAl present everalspecifications.Based on the resultsof theprevioussection,thespecifications sed in this sectionarethosewhere:1. Heathknowledges endogenousbutArabic,French,andmathematicskills areexogenous;and2. The directeffect of yearsof schoolingon childhealth sconstrainedo equalzero.Analternatives tobe moreagnostic, akingaveragesacrossdifferent pecifica-tions,yet doingso produced esultsvery similar o thosegivenhere.

  • 8/6/2019 risna

    28/37

    150 The Journalof HumanResources

    of healthknowledgeon child health 65) are 1.44 when incomeeffects areexplicitand 1.45 whenincome effects are substituted ut.To complete hedecomposition f 32oneneeds estimatesof theimpactof school-ing on health knowledge (either directly via aHKor indirectly via TrLacLnd rlNtN)and the impactof mother'sschoolingon household ncome(cay).TableAl in theAppendixprovidesa pointestimateof 0.153 for ay. Theremainingparametersti-matesareshown n Table7. The first hreecolumnspresent educed ormestimatesof the determinants f healthknowledgeunder he assumptionhatArabic,French,andmathematics corescan be considered s exogenous.24 heregressionn column1 shows that Arabicand mathematics kills have significantlypositiveimpactsonhealthknowledge.French kills haveanunexpectednegativeeffect,butthis is sig-nificantlydifferent rom zeroonly at the 10 percent evel. Finally,aftercontrollingfor these effects years in school has a significantlynegativeeffect. Takingtheseestimates at face value implies that cHK = -0.185, rN = 0.184 and the two partsof TrL re 0.088 and -0.052 for Arabic andFrench,respectively.Estimatesof the

    impact of years of schooling on literacy (otL) and numeracy (aN) skills are shownin columns4-6 of Table7. Briefly,aN = 1.268 and the two partsof aL are 1.734and 1.266 for ArabicandFrench,respectively.Usingthe estimatesof aHK, TL andaN givenin thefirstcolumnof Table7 yieldsan estimateof 32 of 0.196, as shownin the firstrow of the last columnof Table 8.Given the simplefunctional orms used and the imprecisionof the estimates, hisis surprisingly imilar o the estimated igureof 0.165 from the FE reduced ormestimate n Table3. The other columns n Table 8 show how this is decomposedaccording o Equation5".Perhaps he most unusual inding s thatyearsin schooldoes not raise healthknowledge;ndeed, t has a strongly ignificant egativeeffect.Is this plausible?It may be. In Moroccanschools, basic healthknowledge s notpartof the standard urriculum,25o one should not be surprisedhatthe impactof schoolingis not positive. Even a negative impact may occur-because schoolattendance educesthe time girls spendat home with theirmothers, t mayreduceopportunitiesor them o acquirehealthknowledgeathome.That s, timegirls spendat home is an omittedvariable hatis negativelycorrelatedwith girls' schooling;schoolingitself does not reduce healthknowledge,but it implies an allocationoftime that results n lower healthon knowledge.

    Turning o the rest of the decomposition, he main avenueby which schoolingraiseshealthknowledge s by raisingArabicandmathematicskills,particularlyhelatter,whichcan in turnbe usedto acquirehealthknowledge.In contrast,Frenchskills have a small negativeimpact,based on a parameterhatwas significantlydifferent romzeroonly atthe 10percent evel. While thepositive mpactof Arabic24. Intuitively,while it is plausible hat a sicklychild will increaseparents'healthknowledge, here sless reason o think hatparents btaingreateriteracyandnumeracykillsinresponseo boutsof sicknessin theirchildren.Thisis consistentwith the findings n SectionIVB;the Hausmanests clearlyrejectedtheexogeneityof healthknowledgebut couldnotreject hejoint exogeneityof mathematics,ArabicandFrench kills.Finally,as a practicalmattert is verydifficult o find nstrumentalariablesor these threeskills in the 1990-91 ENNVMthatdo not also affecttheacquisition f healthknowledge.25. Thisstatements basedondiscussionswithWorldBankstaffwho haveworked n healthandeducationissues in Morocco.

  • 8/6/2019 risna

    29/37

    Glewwe 151

    literacy s plausible literatewomencanacquirehealthknowledgebereading ariouslnpwrittenmaterials),hepositive mpactof numeracys less intuitive. t is probablythe casethatmathematicalkillshelpmothersmonitor heirchildren's llnesses andmoreaccurately pplymedicinesand treatments. n addition, t maybe thatmathe-matical killsdevelopmothers'abstracteasoningabilities,which n turnhelpsthemto organizeand refine the healthknowledge hey acquire.Giventhe high colinearitybetweenschooling, iteracyandnumeracy,as shownin Table2, morepreciseestimatesmightbe obtainedby droppingnsignificant ari-ables. Thus n the secondregression eportedn Table 7 theFrenchiteracyvariablewas dropped.The resultingdecompositions shownin the secondline of Table 8.Overall, he resultsdo notchangeverymuch.Inparticular,hesignificantly egativeimpactof years n schooldoes not "go away." Dropping earsof schoolingas well,whichis hard o justify econometrically,was done in the thirdcolumn of Table7,and he associateddecompositions re shown n the third ow of Table8. Theoverallresult s notvery satisfyingbecausenow Arabicskills haveno significant ffect onhealthknowledge,andtheestimateof 32shown n thelastcolumnof Table8 (0.273)is muchlarger han the estimategiven in Table 3 (0.165).Thebottomhalf of Table8 examinesdecompositions asedonEquation ', whichincludesthe impactof income.The firstrow (that s, the fourthrow of Table8) isbasedon the firstregression n Table7. The totalimpactof education s estimatedto be 0.285. This is also muchhigherthanthe estimateof 0.165 in Table3, so thedecompositions nderlyingtmaynot be very precise.Thatbeingsaid,thedecompo-sitionindicates hatthe effect of educationvia its effect on incomeis 0.089, whichis aboutone thirdof the total effect. The rest of the decomposition thatis, thedifferentmpactsvia healthknowledge) s very similar o the case where ncome issubstituted ut.Thus the previousdiscussionapplieshere.Thefindingsof thissectioncanbe summarized s follows. The evidencesuggeststhateducation mproveschild healthprimarilyby increasinghealthknowledge.Italso has an impactby raisinghousehold ncome,butroughestimates ndicate hatthis incomeeffect is only aboutone thirdof the total effect. This is similarto thefindingsof Thomas,Strauss,andHenriques1991),who found ittleimpact hroughimprovedhousehold ncome.There is no direct effect of either Arabicor Frenchliteracyskills,nor of numeracy kills,on childhealth.Neither s thereevidencethatotheraspectsof schooling,particularlyhanges n mothers'values,have anydirecteffect.Thequestion hen arisesas to how mothersobtainhealthknowledge.School-ing alone hasno contribution, ndmayeven havea negativeeffect (dueto reducedtime spentat home by girls in school). The lack of a positiveeffect is consistentwiththe fact that Moroccan choolsdo not teach healthknowledge o students. n-stead,childrenacquirehealthknowledgeby acquiring iteracyandnumeracy killsin school, whichthey thenuse to attainhealthknowledgeoutside of school. OnlyArabic iteracyappears o matter;French iteracyhas no significant ffect.Overall, hesefindingsarequiteinteresting nd havesomeimmediatepolicy im-plications.However,further esearch o confirm,or possiblyrefute,these findingsis in order.The decompositions ere are basedon simplefunctional orms andthepointestimatesare notparticularly recise.Theyshouldbe treatedas suggestivebutnot definitive.

  • 8/6/2019 risna

    30/37

    Table 7Direct and Indirect Impacts of Schooling on Health KnowledgeDeterm

    Determinants f HealthKnowledge MathematicsConstantArabicFrenchMathMother's choolingMother'sheightMother's ather'sschooling

    0.001(0.00)0.088(3.19)-0.052

    (-1.77)0.184(5.53)

    -0.185(-3.35)0.004(1.44)0.036(0.48)

    0.014(0.03)0.077(2.78)

    0.175(5.38)

    -0.220(-4.38)0.004(1.42)0.035(0.45)

    0.078(0.17)0.011(0.47)

    0.133(4.39)

    0.004(1.29)-0.020

    (-0.26)

    0.031(0.07)

    1.268(34.41)0.002(0.67)0.054(0.47)

  • 8/6/2019 risna

    31/37

    Mother'smother's choolingFatherbornhereNumberof radiosNumberof televisionsAvailabilityof newspapersMother'smarried istersFather'smarried istersFather'smarried istersmissingR2Samplesize

    -0.146(-1.36)0.067

    (1.66)-0.028(-0.83)0.086

    (2.28)0.184(2.91)0.015(1.42)-0.006

    (-0.58)-0.005(-0.09)0.468884

    -0.150(-1.42)0.065(1.58)-0.028

    (-0.82)0.084(2.22)0.191(3.03)0.014(1.36)-0.007

    (-0.62)-0.009(-0.17)0.466884

    -0.205(-2.05)0.072

    (1.76)-0.029(-0.87)0.091

    (2.37)0.168(2.65)0.016(1.48)-0.005(-0.44)0.010(0.18)0.458884

    -0.230(-0.91)-0.042(-0.89)0.003

    (0.09)0.121(2.99)0.058(0.72)0.003(0.24)0.022(1.66)0.114(2.14)0.843904

    Notes:1. All regressionsncorporateommunity ixedeffects.2. Asymptotic-statisticshownin parentheses.

  • 8/6/2019 risna

    32/37

    154 The Journal of Human Resources

    Table 8Decompositions of Impact of Mother's Schooling on Child Health

    ThroughHealthKnowledgeVia Via Via ThroughDirect Arabic French Numeracy Income Total

    (aHK) (T1L L) (ILaOL) (rINaN) (66o 4) EffectNot conditioning n income -0.268 0.221 -0.095 0.338 - 0.196-0.319 0.194 - 0.322 - 0.1970.028 - 0.245 - 0.273Conditioning n income -0.267 0.220 -0.095 0.337 0.090 0.285

    -0.317 0.193 - 0.320 0.090 0.2860.028 - 0.243 0.090 0.361

    VI. Summary and ConclusionThree major conclusions can be drawn from the empirical work inthis paper.First,healthknowledge appearsto be the most importantskill that mothers(indirectly) obtain from their schooling that prepares them to provide for their chil-dren's health. Second, estimating the impact of health knowledge on child health

    could suffer from substantialendogeneity bias that can underestimateits true impactif one does not instrument the healthknowledge variable.Third,the analysis suggeststhat schooling contributes to mother's health knowledge in Morocco only indi-rectly-health knowledge is not directly learned in school but instead is learnedusing literacy and numeracy skills acquired in school.The above conclusions have direct policy implications for Morocco. First, theysuggest that health knowledge should be directly taught in Moroccan schools.26Theyshould be taught at an early age because girls who drop out early may never acquiresufficient numeracy and literacy skills to allow them to acquire health knowledgeon their own. Many girls in Morocco leave school at a very early age; only 28 percentof women aged 18-20 in the 1990-91 ENNVM continued their schooling beyondprimary school. Even more disturbing is that 51 percent of these women never at-tended school at all-this latter fact suggests the need for a major effort to teachyoung women basic health knowledge in adult education programs.These results and their policy implications may well apply to other developingcountries. If the finding that health knowledge is the key skill for improving childhealth is confirmed in other countries (an important task for future research), anycountry where a large proportionof women do not go beyond primaryschool should26. Oneobjection o thispolicyrecommendations that healthknowledge aught n schoolwill merelydisplacehealthknowledgeobtained lsewhere, o thateventualhealthknowledge ttainedwill notchange.Yet if literacyandnumeracy kills acquiredn schoolarelow, theremaybe little acquisition f healthknowledge ater n life. Moreover, or women who leave school withgood literacyandnumeracy kills,a higher"initialstock" of healthknowledgewill allowthemto reacha higher evel of healthknowledgethanthey would havereachedwith a low ornonexistentnitialstock.

  • 8/6/2019 risna

    33/37

    Glewwe 155

    add basichealth education o its primary chool curriculum.n addition, f a largeproportion f womendo not even attendprimary chool,healtheducationprogramsfor womenof child-bearing ge should also receivehigh priority.Finally,the find-ings here support wo generalpolicy recommendationsor developingcountries:1. Educationof girls should be a high priority;and 2. School qualitymust not beneglected,since women will not be able to raise their level of healthknowledgeaftertheirschoolingis completed f they leave school without basic literacyandnumeracy kills.

    Appendix 1Descriptionof Health KnowledgeTestThe healthknowledgetest used in the 1990-91 MoroccanEnqueteNationaledesNiveauxde Vie des Menages(ENNVM)consistedof the followingfive questions,given to the respondentn his or hermaternalanguage:1. Is it possible to get vaccinations or childrenwithoutpaying money? If yes,where?

    (Answer:Yes, at public hospitals,Red-Crosscenters,or visits by nursestovillages).2. Whatshouldone do to a woundto avoid infection?(Answer:Wash it well with soap,applyalcohol and cover it).3. What s the best way to preventchildren romgetting polio?(Answer:Vaccination).4. If a child developsdiarrhea,whatshould one do if no doctor s available?(Answer:Use boiledwater, eedriceorcarrots, ive salts,avoid milkandfats).5. In places wherethe water s not safe to drink,what shouldone do to it beforedrinking?(Answer:Boil it or adddropsof "javel").

  • 8/6/2019 risna

    34/37

    Appendix 2SupplementaryTablesTable AlFirstStageRegressionsor Test Score Variables

    HealthKnowledge Expenditures erCIncluding Excluding Including ExArabic, Arabic, Arabic,French,and French,and French,and FreMathematics Mathematics Mathematics Ma

    .~ q I-_ lC4 1 tx t n Ix xi n "tiatmers scnooiingMother's choolingArabicFrenchMathematicsSexAgeAge2Mother'sheightFather'sheight

    U.U3Z)(3.22)-0.195(-4.17)0.997(4.07)-0.073

    (-2.65)0.186(6.80)-0.015(-0.60)-0.002(-0.66)0.000(0.44)0.005(2.28)0.001(0.51)

    U.UOu(3.27)0.125(6.43)

    -0.013(-0.51)-0.001(-0.45)0.000(0.23)0.006(2.43)0.001

    (0.53)

    (U.U11(5.11)0.153(3.51)-0.053

    (-2.45)0.044(1.75)0.039(1.54)-0.029(-1.40)-0.002(-0.84)0.000(0.74)0.009(4.84)0.002(1.00)

    -(-1-(-

  • 8/6/2019 risna

    35/37

    Father'sheightmissing -0.046 -0.041 0.029(-1.33) (-1.15) (1.05)Rental ncome 0.007 0.003 0.010(1.12) (0.53) (2.15)Childrenoverseas -0.133 -0.001 0.072(-1.69) (-0.01) (0.61)Irrigated ropland 0.001 0.001 0.006(0.62) (0.76) (6.89)Unirrigatedropland -0.000 -0.000 0.000(-1.82) (-2.01) (1.95)Treecropland -0.033 -0.034 0.015(-2.13) (-2.35) (1.57)Mother's ather'sschooling 0.053 0.092 0.136(0.64) (1.06) (1.40)Mother'smother'sschooling -0.228 -0.315 -0.059 -(-2.49) (-3.37) (-0.55) (-Fatherbor here 0.059 0.069 -0.082 -

    (1.72) (2.24) (-2.69) (-Mother'smarried isters 0.015 0.014 0.003(1.75) (1.66) (0.45)Father'smarried isters -0.018 -0.013 0.017(-1.93) (-1.31) (1.99)Father'smarried istersmissing -0.038 -0.009 -0.027 -(-0.84) (-0.19) (-0.75) (-Numberof radios -0.023 -0.014 0.118(-0.85) (-0.51) (5.36)Numberof televisions 0.057 0.069 0.211(1.93) (2.24) (7.69)Availabilityof newspapers 0.207 0.195 0.035(3.78) (3.56) (0.65)R2 0.469 0.424 0.653Notes:1. Asymptotic -statistics n parentheses.2. All regressionsncorporateommunityixedeffects.

  • 8/6/2019 risna

    36/37

    158 The Journal of Human Resources

    Table A2Partial R2 Statistics for Instrumented Variables

    InstrumentSetsNine Health

    Nine Health KnowledgeNine Health Knowledge InstrumentsKnowledge InstrumentsPlus Plus Five IncomeInstrumentedVariable Instruments Mother's Schooling Instruments

    Health knowledge 0.143 0.179 0.158Mathematics 0.273 0.773 0.283French 0.218 0.788 0.224Arabic 0.200 0.808 0.203Income - -0.394Reading a medicine bottle 0.231 0.682Notes:1. The nine healthknowledge nstruments re:mother'smarried isters, ather'smarried isters, ather'smarried istersmissing, atherbornhere,number f televisions,number f radios,availability f newspa-pers,mother's ather'sschooling,and mother'smother's chooling.2. Thefiveincome nstruments re:rental ncome,number f childrenivingoverseas,rrigated rop and,unirrigated ropland,and treecropland.

    ReferencesBarrera, Albino. 1990. "The Interactive Effects of Mother's Schooling and Unsupple-mented Breastfeeding on Child Health." Journal of Development Economics 34(1):81-98.Behrman, Jere. 1990. "The Action of Human Resources and Poverty on One Another:What We Have Yet to Learn." Living StandardsMeasurement Study Working PaperNo. 74. The World Bank. Washington, D.C.Behrman, Jere, and Anil Deolalikar. 1988. "Health and Nutrition." In Handbook of Devel-opment Economics, Vol. I, ed. Hollis Chenery and T.N. Srinivasan, 631-711. Amster-dam: North Holland.Bound, John, David Jaeger and Regina Baker. 1993. "The Cure Can Be Worse than theDisease: A Cautionary Tale Regarding InstrumentalVariables." Technical Paper No.137. Cambridge, Mass.: National Bureau of Economic Research.Cash, Richard A. 1983. "Oral Rehydration in the Treatment of Diarrhea: Issues in the Im-

    plementation of Diarrhea Treatment Programs." In Diarrhea and Malnutrition: Interac-tions, Mechanisms, and Interventions, ed. Lincoln Chen and Neville Scrimshaw. NewYork: Plenum Press.Cebu Study Team. 1991. "Underlying and Proximate Determinants of Child Health: TheCebu Longitudinal Health and Nutrition Study." American Journal of Epidemiology

    133(2): 185-201.. 1992. "A Child Health Production Function Estimated from Longitudinal Data."Journal of Development Economics 38(2):323-51.

  • 8/6/2019 risna

    37/37

    Glewwe 159

    Davidson,Russell,and JamesMacKinnon.1993.Estimation ndInference n Economet-rics. OxfordUniversityPress.Gibson,Rosalind. 1990.Principlesof NutritionalAssessment.OxfordUniversityPress.Glewwe,Paul. 1997. "How Does Schoolingof Mothers mproveChild Health?EvidencefromMorocco."LivingStandardsMeasurement tudyWorkingPaperNo. 128. Wash-ington,D.C.:TheWorldBank.Schultz,T. Paul. 1984. "Studying he Impactof HouseholdEconomicandCommunityVariableson ChildMortality."PopulationandDevelopmentReview10(suppl.):215-35.Strauss,John. 1990. "Households,Communities nd PreschoolChildNutritionOutcomes:Evidencefrom Cote d'Ivoire." EconomicDevelopment nd CulturalChange38(2):231-61.Strauss,John,and DuncanThomas. 1995. "HumanResources:HouseholdDecisions andMarkets." n Handbookof DevelopmentEconomics,ed. Jere BehrmanandT.N. Sriniva-san,Volume3.Thomas,Duncan,JohnStrauss,and MariaHelenaHenriques.1991. "How Does Mother'sEducationAffect ChildHeight?"TheJournalof HumanResources26(2):183-211.Wolfe,Barbara, nd JereBehrman.1987. "Women'sSchoolingand Children'sHealth:Are the Effects Robustwith AdultSiblingControl or the Women's ChildhoodBack-ground?"Journalof HealthEconomics6:239-54.WorldBank. 1994. "Kingdomof Morocco,PublicExpenditure:ssues and Outlook."Washington,D.C.: MiddleEast and NorthAfricaRegion,WorldBank.