LSM112 English i. b-e Tradeoff bof~v2en h-umber cii Children -77 LSMSWorkingPapers No. 42 Glewwe, The Distnhtion of W+ in P a in 198586 No. 43 Vijverberg, PTofitsfmm Self- A CPK Study of U t edPlmire No. 44 Deaton and Benjamin, Thek g Slnndnrdr Su- and Prim Policy R@mc A Study of Cocoa and CoffeePrwducth inCdtcd'looie Nc. 45 Gertler and van der Gaag,h?ass&g .he WillingnesstoPayfir Social M m s in Dewloping Countn'es No. 46 Vijverberg,N~nng~cultural Family Entrrprisls in a t e d'lwirt: A Dextiptiw Analysis No. 47 Glewwe and de Tray,ThePan duringAdjustmolt: A CaseStudy ofated'Iwire No. 48 Glewweand van der Gaag,GmJkding Pourty in kroping CountriesDejinitions, In-tion, and Policies No. 49 Scott and Amenuvegbe,Sump& Dbignsjb thc Living StandanisSurclgrs in Ghanaand Mauritania/Plans &sondagepur &cnquitessur le niwm de vieau Ghanaet en Mmrritanie No. 50 Laraki, Food Subsidies:A CoaStudyofPrice R.@rm in M m w (alsoin French, 50F) NO.51. Straussand Mehra, ChitiAntfPopPmctry in G t e d'lvoire:Estimatesfrom T w Surveys, 1985 and 1986 No. 52 van der Gdag,Stelcner, and Vierberg,Public-Pn'wte Sector WageGnnparisons-and Moonlighting in Developing Corrntris Eoidmafnrm Cdted'lwire and Peru No. 53 Amworth, Sononoecunomic Lktmnh& ofFcrh7ityinCdted'lwire No. 54 Gertlerand Glewwe, The -to Pny& Eduation in Dewloping Countries:Eoidencefrom Rural Penu No. 55 Levy and Newman. R i md6sPtrjsDon& midwnomiques et mcdconomiques sur l'ajustement du march.! du hrolril dmtslevcteur moderne(inFrench only) No. 56 Glewweand de Tray,7hcPmr in Latin America duringAdjustment: A CaseStudy of Peru No. 57 Alderman and GertIer, 'lkSuk&fddity of Publicand Private Heulth Can?& !he Treatment of Childrenin Pukktm No. 58 Rosenhouse, Identifying theP a x Is 'Hendship" a UUsefulConcept? No. 59 Vijverberg, Labor Market Pcqknuna asa Determinant of Migration No. 60 Jimenezand Cox,TheKLioria-ess ofPrimateand Public Sc?uwb: Eoidenc.efrom T w Dmloping Counfries No. 61 Kakwani, htge Sample Dkfdmfh o f S d Inequality Measures: ?GbhApplication toCdte d'lwire No. 62 Kakwani, Testing& SignijZmx ofpoverty D~jimnaes:WithAppliafion to Cdted'Iwim No. 63 Kakwani, Poverty and EwmomkG d WithApplication toCbted'lm'w No. 64 M o d , Musgrove,and Sekner, Education andEarnings in Peru's Infbrmal N:~nfirmFamily EntPrprises No. 65 Aldermanand Kozel, F d a d IltfOrrrml Sector WageDeterminafion in Urb:mLow-Income Neighborhoods in Pakistan No. 66 Vijverberg and van der Gag,Testing& Lobor Market Dualityr ThePrivate lirugeSector in a t e d'loire r I No. 67 King,Does Education PayinLhc LmborMarket? TheLahr Force Participntion, Occupation, and Earnings of Peruvian Womm No. 68 K >z& TheCompositionadDktdmtion ofIncome in Cdted'lwk No. 69 Dea*n, Price ElnsficitkjiumS u n q Data: Extensionsand 1ndonesi.nResu;!s N a 70 G l w ~ eEficient Al- - , c f T t to thePour TheRoblemofUt&zt& Hc3fsehold1 n m No. 71 Glefie, Investigating t k Lkfrmimnts ofHousohoM Welfn?in Cbted'lmire P No. 72 Pitt and Rosenzweig, T kSdatiEity ofFertility and the Determinantsof Human Capital lnvesfments:Parametricmd Scmipunmehic Esti~mtes No. 73 JacobpShndozu Wagesmd hzsmf Family Labor Supply:An Econometric AppI:'cationto the Penivi.an Sim No. 74 Behman, TheAction oflLfuBlpn RCSOI(rcesand P- on OneAnothrc What WeHavf Yet to Learn No. 75 Glewweand Tv~urn-Bash.TheDisfnhfion of Welfrein Ghana,1987-88 No. 76 Glewwe,Schooling,S B , endthePAms to Comment Inwsfment in Education:An Exploration Using DataJclomGhmra (List continueson the insideback cover) The Tradeoff betweenNumber of Children and Child Schooling EvidencefromC6ted'lvoire andGhana Tne Living StandardsMeasurement Study The LivingStandardsMeasurementStudy (LSMS) was establishedby theWorld Bankin 1980to exploreways of improvingthetypeandqualityof household data collectedby statisticalofficesind e v e l ~ pcountries. Itsgoalisto fosterincreased ~ g use ofhousehold data as a basis for policy decisionmaking. Specifically, the ~ s ~ s isworking13developnew methodstomonitorp r o p sinraisinglevelsof living, to identify the corwquences for households of past and proposed government policies, and to improve communicationsbetween s w e y statisticians,analysts, - and policymakes The LSIS Working Paper series was started to disseminate intermediateprd- ucts fromthe EMS. Publicationsin theseriesincludecriticalsurveyscoveringdif- ferent aspects of the ISMS data collection program and reports on improved methodologiesfor using LivingStandardsSurvey (LSS)data. More recent publica- tionsreccnnmend speafic survey,questionnaire, and data processingdesignsa d demonstsatethe breadth of policy analysisthat can be carriedout usingIsS data. ZSMSWorking Paper Number112 TheTradeoff betweenNumber of Children and Child Schooling EvidencefromC8ted'Ivoire andGhana Mark Montgomery Aka Kouam6 RaylynnOliver - TheWorld Bank 8 . Washington, D.C. CopyrightO 1995 Tha InternationalBank for Recorstmction and Development/mWORLDBANK 1818H Street,N.W. Washington,D.C. 20433, U.S.A. All rightsreserved Manufactured in the United Statesof America First printing April 1995 To present the results of the Living StandardsMeasurementStudy with the least possible delay, the typescript of this paper has not been prepared in accordancewith the procedures appropriate to formal printed texts, and the World Bank acceptsno responsibilityfor errors. Some sources cited in this paper may be informal documents that are not readily available. The findings,interpretations,and conclusionsexpressed in thispaper areentirely those of the author(s) and shouldnot be attributed in any manner to the World Bank, to its affiliated organizations,or to membersof its Board of Executive Directorsor the countriesthey represent.TheWorld Bank doesnot guaranteethe accuracy of the data included in thispublicationand acceptsno responsibiity whatsoever for any consequenceof their use. Theboundaries, colors, denominations,and otherinformationshownon any map in thisvolume donot imply onthe part of the World Bank Group anyjudgment on tile legalstatusof any territory or the endorsement or acceptanceof such boundaries. The material in this publication is copyrighted. Requestsfor permission to reproduce portions of it should be sent to the Office of the Publisher at the address shown in the copyright notice above. The World Bank encourages disseminationof its work and will normally give permission promptly and, when the reproduc- tion is for noncommercialpurposes, without askinga fee. Permission to copy portions for classroomw is granted through the CopyrightClearanceCenter, hc., Suite 910,222 Rosewood Drive, Danvers, Massachussns 01923, U.S.A. Thecompletebacklist of publicationsfrom the World Bankis shown in the annual Index of Publications, which contains an alphabetical title list (with full orderinginformation)and indexesof subjects,authors, and countriesand regio-'IS. The latesteditionisavailable free of chargefromthe DistributionUnit, Office of the Publisher, The World Bank, 1818H Street, N.W.,Washington, D.C.20433, U.S.A., or fromPublications, The World Bank, 66, avenue d'I&a, 75116Paris, France. Mark Montgomeryisa research auodateat thePopulationCounaland ani-ssodateprofessorof economicsat the State Universityof New York,StonyBrook Aka Kouameisa lecturerand research coordiiatorat the Institutefor Demographic Trainingand Research (IFORD)inYaounde,Cameroon.RaylynnOliverisaconsultant to theAfrica TechnicalDepartmentof the World Bank. Library of C o a g m Cataloging-in-PublicationDab ~ Montgomery,Mark, 1953- The tradeoff between numberofchildrenandchild schooling: evidence fromC6tedlvoireand Ghana / Mark Montgomery,Aka Kouame, RaylynnOliver. p. cm.-&3dS working paper, ISSN 023-4517;no. 112) Includesb i b l p h i u l references. ISBN0-8213 23-X 1. Fertilily, Human-C6ted'lvoire. 2. F~rtillty,Human-Ghana. 3. Educationdte d'hroire. 4. Education-Ghana. I. Kouam6,Aka, 1959- . 11. O&er,Raylynn,l96& . 111. Title. IV.Series. HB1076.A3M66 3994 304.6'3'096668--dc20 94-237G4 CIP The Tradeoff between Number of Children and Child Schooling: Evidence from C6te dd'Ivoireand Ghana Foreword . ... ........... ........................ vii , Paper Number 1: "Fertilityand ChildSchoolingin CBte d'Ivoire: Is There a Tradeoff?"by Mark Montgomery and A h K~uam6. . 1 Paper Number 2: "Fertilityand ChildSchoolingin Ghana: Evidence of a Quality/-tity Tradeoff"by Raylym Oliver . .. ..71 Foreword A well known featunof W i t y declines in many developing countries is that parents begintohave fker c h i i h but investmore in the health, education, and support of each child. The factors that induceparentsto realize this 'tradeoh is of greatpolicy . interest, sincethq, simultaneouslyencourageslowerpopulationgrowthand higher quality of population. The tm,papers in this wiume examine the detednants of krtility and of child schooling in ated'Ivoin and Ghana to assess evidence of a tradeoff between the number of born and levels of chid schooling. In Cdte d'Ivoire, there is evidence of such a tradeoff in urban areas, btit not in rural areas. Female schooli higher incomr. and improvedchild survival are associatedwith lower W i t y andhigher child schooling. In both urban and nualareas of Ghana, there is evidence of a trade~ff between fertility and child schooling with higher incomes-ad, in rural Ghana, with increases in mothers' schooling. Thesepapers uet w d sevaal products of the Wdd Bank research project on ~ 'The Economic and hiicy DetemhWs of M i t y in SubSaharan Africa', sponsored by the Rwerty and Human W c e s Division of the Afiica 'l'khnical Department (AIWIR) and nlenaqed by Marttra Abwrfh, principal hwsigmr. It is part of a broader research efbn iit the Prmrty and Human Resotlrces Division of the Rlicy Research Department (PRDPH)that ezamims the role of human resources in economic development. The riara used are from the Cdte d'Ivoh znd Ghana Living Standards S u m which aretmof mauy LivingStandardsMeasurementStudy(LSMS) household sirveys implemented in developingcountries with tfre assistanceofthe Mrld Bank. f/h- %w. Kevin Clearer LynSquire Dhcmr Director Africa T ' c a l Depmimmt hlicy Research Department . vii Paper Number 1 Fertility and Child Schcoling in C6te dYIvoire:b There a Tradeoff? - Mark R. Montgomery Aka Kouarne Abstract .............................................. 3 In;roduction ............................................ 5 ' h e Quantity-Qualit-jTradeoE An Overview ....................... 11 Perspectiwson Causarion .............................. 11 The Role of Emgemus Covzriates ......................... 12 AfPican Family Structure and the Tradeoff.................... 15 - Educational blicies mdtheTradeoff ....................... 17 The CBte d'Iwire Setting and Data ............................. 19 Rpulation md Labor Policy ............................ 19 The Educational System ............................... 20 TheLSMSData .................................... 20 result^............................................... 26 Description of the Samp!e .............................. 26 MultivariateFindings ................................. 38 TBe Endogen5ity of Consumption ......................... 40 Rural and Urban Sub-Samples ........................... 41 AiternativeSpecifications .............................. 44 The Quantity-QualityTradeoff Revisited ..................... 48 The Price of Schooling ..................................... 53 Conclusion ............................................ 57 Appendix 1: The Ordered-Probit Model fbr Projected Completed Schooling....59 Appendix 2: Assessing the Exopneity of Consumptionper Adult ........... 63 Abstract This research exploresthe relationshipin Cdte d'Iwire Wween krtility and the iilvestmentsmade by parents ia the schoolingof their children. One expectsthat tamilia with many chi!dren will tend to invest less in each, and that families with fewer children will make sreater human capital investments per child. Tie 'tradeof of quantity b r quality is vividly ilicstxated in the recent economic development of Southeast Asia and MiAmerica. InrespecttoSub-SabaranAfrica, however,theexigtenceofattadeoff has n9t yet bbeen established. The few studies wndccted to date either suggest no particular association between Emily size and schooling in .2frica,or biat at a positive relationship wherein higher W i t y is lin)asdto greater schoolbg per child. This paper weighs the widence ccnceahg ?he quantityqdity tradeoff in Cdted'IvoiP, wing data drawn from thethree rounds of the Cdte&MireLiving Smcdards Measurement Survey (LSMS) conducted from 1985 to 1987. The widence suggests that twvery different relztioilships link fertility and child schoolirg. In the rural areas of Cdte d'Ivoire, there is no adeoff: higher Miity is associzted with higher child schooling. This finding is consisteatwith much of the early research on krtilitj a d scboolibg in Africa. Urban areas, by contrast, ate characterized by the Wideoff that *pears in Southeast Asia and elstwhere in the developing world. F e d e eduetion, which.:owas fwtility 2nd raisw investmen&in child schooling, is an important Extar vroduciqj tbe tradeogin urbss areas, and the results also haply that i x n p ~ dchild survival may b important. One possible explantion br the absence of a adeoiT in rrara) 3rieas is !ess access toM y planning services,which are mi!able from private sou= in urban areas. Acknowledgments This paper was prepared for the research project on "The Econo.nic and Folicy Determinants of Fertility in Sub-Saham Africa." Ths a?ithors are, respectively, Associae Professor of Econoxics, SUNY-Stony Bmk, and Assistant Professor, Institut 22 Formation a de Recherche Demographiques. Mk wuld like to thank Martha Ainsv.ortb, Paul Glewwe, Hannan Jacobs, Tom LeGrmd, Psbert kouty -nd participants in Stony Brook's Applied Workshop in Econ0rnd.c~for hel?ful comments. We wuld dso :iLe to acknowledgewith appreciationthe InstitutNational de la StatIstiqueof Cdte d'Ivoirefir assismcein iinking census variablesto the hogsehold data set. Theopinions expressed in this paper are those of the authors a d do not necessarily refla! policy of B e World Bank or is members. This research explores thl; relationship between fertility in CBte d'1wit-e x.d the imestmznts v d e by parents in trie schoo!;hg of tbeir children. One expectsthat families wid1 nlmy children uili tend to Lwest !ess irr eacb, and rhat fanilies with fewer children will make grez(er investments per chi'd. This aqati* association between fertility and humm capid iavzstment per child, mident in econnrnies as diverse as those of the United States OJlmushek, i992)mi mailad (bodel et al., 1987, 1990), has been termed the 'q:-mtityqodity trajcoff.' Thetradaff is m ubiqui~assto seem one of ~ tSe s~lizedt3cts of xorimic developrne~:t. Yd 3 present, it is rot at all clear that a qumt~~y-qudity adm9exists in S?lb-S&atan Africa. ThefeJv studies conducted to date (DeLancey, 1 M ) either s u ~ e s tno particulv aswciation between fimily ~izeand schoolbg in Africa, or bilrt at a positive relationshipwherein high~rfertility is linked to greater schooling per child. In this paper we will weigh the evidence concerning rhe quantityquality tradeoff in Cdted'Iwire, using data drawn from the three rounds of the C6tedlIvoireLivabgSbndards Measurement Survey (ISMS) conducted in 1985to 1987. V4e should begin by emphasizing the importance cf the qcantityquality tradwff to the prospectsh r long rem economic growth. In Ute d'Iwireas elsewhere iz~Africa, fertilitylevels will likely remain the principal determinant of the Sirmrate of labcr fbrce growth. The human capital invested in childrenwill be the principal determinant of the skills possessed by that labor farce.' Moreover -and this is particularly clm h r wmen -th4 {evels of schooling attained by the cirrect mhcrt of childm wiil shape their future firtility ad intluence tbe health and survimrship of the next generation of children @en& and Schultz, 1992). Aggregate, cross-nationaldata shcw 2clear trade& between quantity and quality, as is evident i3 Figures 1 and 2. The figures present d2ta from all cievclopingcountries with p a capita income levels of $3000 and ~ d in 1939. Total Miity rates are w armyel on the vertical axes; primary anc! secondary gross emllment 1 ~ ; sare placed dong the horimnd. The muatries of SubSaharan Africa (SSA) are distbguished in the diagrams fiom their counterparts elsewhere. A regression line pruvides a descriptive sumnarj of the relatio~hip 1. A small tony of rpsePrth seeks to quantify the ecmomic ba&ts of e c b l k thro~ghthb estimationof-e production hdoos; see GIRvL*~ (1991) wbo cites-mrkofLau,Jarciwn a d but in this regard. This wrk sggests that in J& Anerics d East Rsiu, an k s s e of one y w in svcragc dalt d u c a t i d W-~ ' " ~ ~ Liliitblen in- is j t e of m m 3-5 p e m t in ieal GDP. Curid); ha-, M,such eficisr &crR isds~teblein aggreg& d& 6 r Sub-- -4fSca. Individual-lartl &!a h~Afric~appu'stlcms ussly &DM P Simg association -m ducation md L.: mes in urbaJ ueps (scm cf this m y b dw 60 cdeiztidisn rakr tisarr to real pro6:. 'vip enimammeat) iad d or m i d m~t'6 r i -. in Figure 1 Sour#:1291FCbrJdLkwbpment Indicasom. I I TddFor&iiRotuby kcondory Endlmont RatiosI I *SHoOhuL#r I 1 Figure 2 Source: 1991 WrldDareIopment 1ndicatom. In these figurn, qumtity (fertility) and quality (child schooling) s(iould be -:garded as being endogenous ia mure. That is, both fertility and schoolirg reflect ,oices made by households that are set in economic environments shaped by mark- and by governat policies A set of excgeno~sbackground factors thec&re determines each country's location in the dimensionsof quantityand quality.' The exogenousfBctors include the lwel of income per capita in the country in question, its degree of urbanization, the extent of adclt edmarional attainment, and the probabilities of child survival; they also encompss an =y of country-specificpolicies regarding educational pricing, infkastructureand Emily planning servicedelivery. Thequantity-qudity tciuieoff itself represents a systematic association between tHO endogenous variables, an association that has its origins in fundmefital exogenous determinants of household behavior. As can be seen in Figures 1 and 2, talcing all developing countries under consideration, there exists a strong negative association between kLit2ityand human capital iuwestment in children. But it is also apparent that iu Sub-Saharan Africa, the tradsoff between quantityand quality is weaker than elsewhere. if indeed it existsat all. The slope of the relationship fertilityand primary enrollment rates in the Sub- Saharan countries (Figure 1) is negative but negligible, 2nd although a seeper slope is evident in respect to fertility a d secondary schocl enrullrneot F~gure2) and African countriesseem to clusteraboutth%regressionline, secondaryenrollmentsin most Africm muntries remain low. Consider the devel~pmentprospects facing a country that begins in positicn A of Figurs 1. One poential pathof development is represented in a movement from point A to point R, whereby a red~~clion in future Ib h rfirm gt3wth is accompanied by an increase in human capital per mrko,r. This is the iind of qumtityquality transition that has been in progress in SoutheastAsia (see Knodcl et a]., 1987,1990) and Latin America. Is such a path achievableh r African countries? Or given thedeuelopmanlsof the 1980s, are the dimions indicated by poinls C and D now more likely? Lesthaeghe (1989) has raised the possibility of a 'crisis-led" demographic transition in Africa ir, coll~lectionvith the econariic conditions of the past decade. Incomes per capitahavedeclid in many Afriw countries, and this bas been coupled with increasss in the exogenous costs of child-rearing, m s t notably in regard to the primte costs of schooling (Makha-Adebusoye, 1991). The combination ofincomeand e price effects may result in ertility decline accompanied by a decline in schoo! enrollments, as indicated in m n e n t to point of Figure 1, or may leave krtility essentiallyunchanged while enrollmentsareredeced (point Ato point C). Tbe pmspscts kr !ong term economic p C i .implicit in these development paths izs profiTtlndly diserent from what is iinpliedZn a quantityqualitytransition (A to B). z Cote d'Iwire presents an interesting case study of these issue. After a long post-Independence period of &dy gnnvth, the economy of C(lte d'Iwire wzs unsettled in *e early 1980s by a striwbrutal external shocks, particuIarly in the cocoa and coffee markets. A growingburden of debt and continued high populationgrowth further undermined the situation, such that by 1987 (see Figure 3), GDP per capita had fillen by some 22 percznt from a 1W8high point of 51345. The decade was also w k e d S reductions in per capita government spending in the education and health sectors (also shown in Figure 3). In an analysis of capital budgets, Russell and Stanley warned that as of 1988: It appears that investment in the education sector has been cut about as fir as it can be. In fact, the low level of investment in education risks degradation of the system [Russell and Stanley, 1988:ixl Yet in reameatexpenditureterms, a t e d'lvoire continued to allocate as mud; as 50 percent of its centralgmmment budge to educationand health (WorldBad, 1990b:5), seemingly without a commensurate payoff in educationst perbrmance. Kenya, kr instance,enjoys perhapshalf the GNP per capita of a t e #Iwire, and spends I s than 30 percent of its budget on educaticn and health Wrld Bank, l!f9Ob:6). Y;tKenya achieve;) nearly u t l i d primary enrollment as of 1989 (94 percent), whereas in 1985 a t e d'Ivoids primary earollment ratio stood at only 75 percent (mates 88 pccwt, f.i.a3ales62'). a figure thst may evenhave declined slightly over the decade (see Figure 4). PICapitaCDPadE d u d iandHmalthSpnding (81060) 1400- -100 - 90 1100. - 80 70 -eo - x ba)- - 10 400. ---.-- -30 - 20 zoo- Mh- -10 0 '-0 1se~ ISXI ism i o n two loss Yaw I & u r n A f r h EcOmmic ad Rnancial Data Source: UNESCO Slcrtistical Yearbook, 1991. All this suggests that atpresent, Cbted'Iwire's education system is chascterized by (i) an imbalancebetween cap'd and recurrent expenditures, the farmer being too low relative to the latter; (ii) a level of recurrent expenditures (these being dmted almost entirely to personnel) that may be too high relative to other demands on government budgets; and (iii) sbortcomhgs and inefficiencies in performance thzt may be traced, in part, to hfrastruciural comtmhts and a rationing of student places world 3ank, 1990:6). Clearly Cote d'Iwire has mt succeeded in educating its population to a lwel consistent with tbe resources expended. In such an envimmem, one educaticnd policy option has attracted increasingly serious consideration: the possibilityof shifting someof the recarrent costs of schooling to the primte seaor (Russell ~IKI i988). Th% is, Swlrian hous&oJds may be Stanley, asked to assume a greater share of the fuJ! costs ~f education, particularly at the secondary and higher levels. The higher fees ml!xtd (of lwei subsidies expended) could be employed in turnto ease the infrastructuraland related constraints now afaiding the educational system.' 4 To evaluate such a policy, inbrmation is required on the prics ehsticity of the demaiid fbr schooling, that is, on the extent to which householdsrespond to fee increases by decreasing the enrollment of their childnn. A fu:l evaluation also requires that the spiilaer effecs on fertility be con~iderd. Kdlq and Nobbe (1990) and others have - 2. See Gertler .ndG t m (1989) hr rainsightfbl analysis of this issue in the case cfrural ku. argued that in Kenya, increases in the exogenocs wsts of schooling have given impetus tc fertilitydecliae. Is this a possibility in a t e d'Ivoire as well? m e discussiontbttfollowsis organized in six sections. in SedionIIwe consider the concept d quantityqualityuadeoff in Africa1 settiugs. Section III describes the situation of the C8te d'Iwire, and dkcussesthz LSMS data. The methodological issues involved in the analysis of these data are addressed in Section N. Lq Section V, we examine the mjor determinantsof fitrtilityand child schooling, with special attention to mother's education, household palrmanent income, and urban and rural residence. Section VI takes up the questionof price elasticitiesof demand ikr child schooiing. We estimatethese elasticities and test for spill-wet effects on fertility. Our conclusions are presented in Section VII. The Quantity-Quality 'IkadmfE An Overview Economic development brings about change not only in family sizes, but also in the allocation of resources among family members. Developing societies generally undergo a transition from relatively large family sizes with low levels of human capital investment per chid, to smaller h i l i e s characterized by higher levels of investment per child (Birdsall, 1988). This quantitypudity tradeoff has its rook in the perceived benefits and costs of child schooling. In the course of modernization, labor markets come to display significantdifferentialsin wslings according to schooling level. Parents then begin to view schooling as an avenue to a better li& br their children, and as a human capital k t m e n t which may wer the long term pay dividends to the parents themselves. Yet education is costly both in terms of direct costs and opportunity costs of fbregone child labor; it is generallytoo costly for parents to give each chid schooling and continue to bear the number of children appropriate to traditional circurnsmces. Some element d household expenditures must give way, and typically W i t y falls as household investmentsin educationper child increase. Mpectives on Causation As noted above, the tradeoff behveen fertility and child schooling should be understood to reptesent a systematicassociationbetween twendogmuvariables. High fertilitydoes not in itself causelow school enrollment, nor does higher schoolenrollment cause low fertility. Rather, each of these variables taken individually ~flectsthe full se$ of opportunities abd consmints facing households We term this a "durn-hrm" perspectiveon fertility and schooling.' To und-d the quantityquality tradeoff from this point of view, we should ask what exogenous deknninants of behavior, or what combinations of such determinmts, have the effect of reducing fertility ai the same time as they irrcrease schooling per child. We should recognize that other outcomes are possib!e; indeed, wthing in the underlying economic theory requires the relationshipbetweenM i i t y a d 3. Fertility can exert o c~uselinflwce on child pch0~)tingin the sbortnro, bmwer, especially uadcr mndi!io;is of rapid and unanticipcuwt chpnge in economic cirnrmstaaces such as &cia characteristic of the past d=de ia W& Africa. In such s&ngs, the number of cbildm 50n# by tbe housebolc! may noicorresporjcibtbe numberthatw d d &vc hemch0se-ngim Imawiedp of current economic ci~ums'9n- Thosechildra who are a l d y born, p d a p havingbeen conceived under s prwious economic reghe!, naw place unanticipcuwt constrainb on c m t household choices,including choices regding he&ma(s in the& schooling. H-o might tom thisa 'conditional' crr 'short term1 prspectiwo i t b~lotionsbipbetwm Lxtilityrad dmdiog. Thispaper will adhereto the reduced-hrmpe-tivrt. Wb do sonotout d a belidtht t!x conditionalpeqective* s o d m vM o rw tbeoretidgrounds lbconceptofco~lditid demmds is w.ellsccepted $ mmomicsand some situoticansare c l ~ i best snrlyzedwitb these y d i t i o n a l tools Rather, b e dscisiwissw isstatistical ia nature ccmditimal ur~lyseacamicier child echoolhg as b e i cgbditid cm W t y dccisions. dfatilily iiself is mdopmm in a loqyr-run sense. Hww, cditionnl %nn?lirditionisvdnemb1etoummsudlrovhld tnis bat offcctboth scbolii k dWitychoias. thdPAyobamiqg tbed m k defF& dWty itself on schooling. child schooling to be negatively-s!@ped. In some saings, fir instance, higher household permanent income may be associated with higher fertili? md higher schooling. If sich sign patterns dominate, the fertility-schwli~locus is positi~ly-sloped. In what b l l m we discuss the exogenogs causal mechanisms in more delsil, giving emphasis to the set cf cmiates, includshg lwels cf income, fkmde education, urban residence and child mortality rates, that will figre pmminently ill cur empkical analyses. We shcilld emphasize here, however, one key mechanism that tinderlies ths quantityquality tradmE liquidity consmints affecting the abi!ity of households to - transfer res- 'ces a m s s life-cycleperiods. These constraints come into play in the followip maraer. Consider an exogenous increase in the mnomic benefitz associated with schooling, such as a greater uage premium fir educzted labor. If parents can make claim on soma portion of the earnings of their children, G.. they certainly can in Sub- Sa!!aran Afiica, then this improvement ir. t!!e earnings potential of educated children is equivalent to an increase in ho~lseholdfull -wealth. Indeed, it would seem to enhance the anticipated return to childbearing in general. Why then should fertility fall? The key p i n t is that African parents cannot b o r n on the basis of their full *wealthor anticipated returns to child schooling, so as to finance the required educational investments in children. Their investments must instead be fieanced out of current income, or drawr. from capital that has been amassed at the point when the child is of schwi age. Such binding period-specific budget wnstraints imply that current fl?rtility, consumption or both nay need to be r e d u d if p&-ents are to take advance of the future reedrns to schooling. We shall ream to this point below, in connection with features of sccial organization that affect the budget constraints facing West Africa,l and Iwirian parents. The Role of Fxogerrous C~vsriates Consider the reduced-form relationship between fkrtility and schmling, on the one hand, ad their exogenous determinants, on tSz other. We can represezt the relationships in a general fashion as fbllows: F = AI,E,iT,M) ('9(-1 (-1 (+I S = s(I,E ,U , M ) (+I(+I(+! (3 where F represents fenilit~S is a measure of schooling per child. The set of exogenous covariat6k (I,E,U,M) includes: permanent income, I; -wmen*s educztiop, E; urban residence, U; and child mortality rates. M. The expec!ed signs of the relationships are iridicated below the mri&les in each equation. Rrmanmt inmrne O has an ambigu~useff* on fertility in economic models of the quantity-quality traidff. In the B e c k and Lewis (1973) fbrmutatic.n, ferti!ity Q I and schmling per child (S)are regarded as being subtib~tes,in the economic sense of . @and the term, and if S were to be held constant then ths d for children (F) wuld tend to increase with incorrte. CIuwwer, schooling is i&el a norrnd gcd, the demand fi;r which incrwes with the level of income, and schoclhg expenditurn are an impormt component of overall childnaring expenditures. Thus, as income and demand for schooiing increase, so does tbe 'shadow price" of children. 9ne might therefore expect fertility to increase with resppct to income up to some thre~ho!d income level, but to decrease thereafter as the shdw price effect comm to dominate the relation. That is, F might well behave as a nomid good in ons portion of the income span, and as a1 inferior good at higher levels of incme. As long as incomeremains LIthe range in which both F and S are normal goods, changes in I produce chmges in the same direction in both F and S. Thiit is, F and S are positively associated. If one wzre to plot F and S in a quantityquality diagram analogous to Figures 1 aod 2, the plot wuld show no evidence of a quantityquality tradeoff. When income is in the range in which F i; inferior and S is normal, however, F and S are negatively assc&td and a quantityqualitytrdeoff emerges in response to income pvth. This discussionillustratesonekey point. In seeltingto understand the root causes of a qcantity-qualitytradeoff, we search fbr exogenous covariateshaving effects ca F and S that are of d i f e ~ r sign. i Women's education could produce such a Weoff, alrhough the causal mechanism at work remain a mtter of dispute. One srpeccts that the greater is a wman's educational attainment, the lo1'~ris 3er fertility and the greater a- her investrcents in chiid scSooliog. What is at issue is the iqterpretaticnto be p ! a d on such a relationship Economists hme aka d m attention to the link between education and labor market earnings. The argiient runs as bllows. The higher is a wmafs educational attainment, the gxater is her e i l t i a l wage. If time spent in work and time in child care are rnutudly exclusive, then the wage rate indexes ons of the principal osportunity costs of childrearing. If these oppcrtunity costs dominaz in ktilitj decisions @igherwages also imlly higher full wealth) it bllows that the higher the price of time, the lower should be fertility It is not at all clear that this chain of reawning can be applied to Africa economies. For most Ivoirian and inti& most African wmen, w r k n 4 nat conflict d i m l y with child care. Mere a conflict exists, it can be n ~ & l yresolved through the employment of law-cost srrbstitutes b r the mother's time in child care, such as care pmided by relatives. Therelevance ofthe opportu16ty cost argument in African settings is t h s d r e doubtful, partidarly in rural settings and occupations in which w r k time and child care time u e mutuaIIy compatible. Cddweli (1982) ewisions a larger role b r education than what is m n s i d d in the simple ecoaornic modd. In his kiew, education s~waas a vehicle fjr thz adoption - - of Western ideas r-dkg the hi!y. It encourages a mcre child-centered view of one's parental responsibilities &caan R u d %st Forest Rural West Fo- SavumsS Ethnicity Northern Mande Akan Krou huthem Mande Volt?.is Other . # Survey Year 1985 330 1986 .342 1987 .328 k (IUle 1 conthueson next pugel Enth smpk .%ml Urban n = 4313 n = 2381 n = 1929 Camrmuv or sO(u-- t)4t0 Survivri Rate ' ,807 .777 .a44 P t i n ~ ~ l I r r r y # t o m r u ' Numbet ofSchoob 61.4 Number ofCl~ssrooms 324.9 Numberof T d e r s 339.0 N w n k of Studeata 13341.0 NumbctofGirlsEnrolled 5562.7 I t3merJ41Dora ' 1 M m prices (nondhdper unit wdybt) .P2 Bcef .42 Fish .29 IalprQd Rice .22 Rice .72 Tormbes .64 Prln Oil .12 Mliab . -19 MUu .08 M d o c .09 Buuars .12 MISNuts .36 kulub 57.47 & l P @ e r ~ ) 6818.21 M @ e r ? ) 76;.59 -r@erprir) Tcsbre 1 (continued) -- - -- - Male Daily Agricdtursl Wsge ' 915.1 Primary School ' Not present -152% Distance in Kms. (0 if pieseot) 0.56 Year Built 66.9 Sscondary School Not Preseat Distance in Kss. ' Year Built 68' Schm!ing Problems No moley 64.1 96 High casts 31.0 No teaches 20.5 No school 21.4 No class 26.9 Building class or school 25.1 Building W e r ' s resideace 38.5 No furnib 9.6 No transpottation 35.7 No qwe,other 82.3 r40tcs: 1. A b d w nofsingleyear of primary ~~hooliqg shm: grade 1,496; 2,1.3; 3,1.7; 4,1.6; 534, acd 6.7.9s As for secondary schooling: grade 7,1.9%; 8,23; 9,1.9; 10.3.2; 11,.7; 12,.6; and 13+,1.6% 2. Lcg oE total non-exceptionalexpendituresdividedby n m b r of adultr in tbeM o l d . .' ' 3. Dam sourre: 1988 Census. W o of su~viqgchildren to childreneverborn. 4. Data muse: 1988Census.'Sector Unknm' notincluded. 5. Data source:LSMSpriceqtestionnaire. 6. Primaryschmling !nspecmms Data andSecoodary&had UrbanDeta collocbdby Aka Xmmd. 7. Dztl souiice: LSMS comnunity q u e s c i o ~ 8. No rural LSMS cl-weracootaina sesokdarysckol31. IWt2, Descriptiw Statisticson theS m e of W r e n Entin scvnpb IZlGVl U h n = 8175 n.15067 n=3103 D ~ W D L WV- hdatiw years d shooling 2.78 2.33 3.43 Currait eoloIlmart .370 .283 .512 3NDEPdNDENT %'%RUBUS lndiridrral L a d Dota (=hilCrs Age 5-7 8-9 1.3-11 12-13 14-15 16-17 18-19 20-30 Mother's Educr*ioot .849 .W .695 None .094 -052 .I63 Any p r i m .a .004 .I42 Byond Primmy 12.95 12.78 13.24 @orwr*:onlsault (Log)' aesideace .I70 .446 Abidjan -211 .554 Odrer U b .2&2 .454 ilrdEostF0.e .I26 -204 RunlWForest .212 .?416 SfMMPh I Etbnicity .lo2 .089 .I24 N0.Ihe.n Made .340 358 .310 A h .I20 .i33 .099 Krou .I49 -194 .076 Wkem M a w .lob .!I2 .:CHI Voltaic .I81 .i13 .29? Otba .dB796 F e d e .802 .778 .&dl PurvivPl rate Natrs: fl. Log of: noouceptiod expeaditurrrdivided by ncrnberof dulk in the h~usehdd. 2 DataSource: 1988Ctnsua Ratioofmrvie.rschildrentochildtenevrrborn. L I Figure 5 Ago-Spce;tis FviilityRobs.44 8 end E d d b n f ~ ! . ' m S Y ~ R ~ l h ~ S u w y I rI Z O Y + Y) a m 4 s w m'. 1 Figure 6 Figure 7 SchoolC-rolment Rotes ByS3brAg.r;.n UcVrPmEdveatkr. '.--- --. i 0.1 10 i - Figure 8 -C z r.. - e 0 . - 33 Figures 9 !o 12 explore these urban-rural ditierentials in more a&l. We sepztate Abidjan from other urban areas in thess comparisons. It s e a s chat the important dividing line, however, is between xural area and urban areas in general. As expected, urban fertility falls below fwtility in rural areas, wbereas urban schooling exceeds xural levels. 7 - 1 Figure 11 Figures 12to 16 present breakdowns amrding to the percentile of household consumption pa a t , which serves to mark the position of the household ia the distributior.of pznment income. In some contrast to the figures just presented, here the differentials im &ility are often weak or in unexpected directions Of course permanent income is highly correlzted with the wman's education and residence, and if these have an influence in opposite direction to that ~f income, the effkct of illcome itself may be ma&& in the figuns. In the s&mling comparisons (Figures 15and 16) a strozg incone &ktis evident, although it is again confbunded with edu&~n and resideoce, and ithis case, perhaps exaggerated in size. Filnall~d i h n r a in child schooiingaccordingto ti& sex of the child ares h m ir.Figures I7 and !8. Tale educational gap between male and Emale chid~p,iiis indeed iuge, a finding -istent witb the zggregate enrollment ratio data presen*d in Figure 4. - 8 * Ago-Spriiic FrrtilityRatrs By%mam+ andC ~ * m npPwwntllr ~ ~ Figure 14 CumubCkmEducation UyCNWrAga andCrmmptbn elm Enrdlmni Rotu By~W16a lpaond CMIunrpUonPorcmth 30.7 0.2 a1 5 10 10 2s &tq. Figure 16 1 CumubtivaFducatim eyChna*+ana- - 8 0 s . 10 10 L . . Figure 17 * 4 6 * I. * w I J Figure 18 Multivariate Findings Ta5le3presests the results of a baseline analysk with urban and rural households pooled in the estimation. Ordinaryleast squares techniques are employed to stirnztethe cumulative fertility and cumulative schooling models in columns 1 iwd 3. Frobit regression is employed for the recent fertility and current enrollment models in colums 2 and 4. Column 5 of the table presents the findings from the maximum-likelihood odered-probit estimawr. In all models, age is included as a control variablz. The estimzd coefficients of several variables suggest a tradeoff between child quantity and quality. The coefficients on wman's schooling are negative in the fertility models (3 of 4 coefficients are significantly negative) and positive and highly significant in the schwling models. P,esid&?ce in Abidjan and other urbm a r m is negatively related to fertility, as also indicated in F i ~ r e 9 and 10, but yositbely related to s schooling. Tinecoefficientson consumption per adcit are perhqs surprising. Given controls b r residence and woman's education, this measure of permanent income is now seen t~ be positively d b t e d with fertility and positively associated with schooling as well. ll~us,30th &rtility ad schooling wuld seem to be sormal g d s . The elasticity af cumutative feriillity with respect to consumption, although significant, is rather smalI at .060, which may be compared to Ainsulorth's (1990) estimate of .O8 to -09fbr the households in 1985only. m e s e elasticities are evaluated at the m a . ) The elasticity of recent fertility is similar, bein4.069 at the mean. Cumulativz schcoling is more responsive tha! fwt'diq. to permanent income, with an elasticity or' .195. The current enrollment elasticity is .339. All t!ese estimates suggest that grc*. in pentianen; income uould tend to increase both fertility and schmling per child, zlbeit with very d e s t effects on ertility and a more substantial influenceon schooling. We will return below to discrlssthe weight that should be stached to these interpretations. ' n e woman's schooling and residence effects can be read directly frorll their coefficients in ihe cumulative fertilityand schooling models, as can the effect of sex ir. the cumulative schooling model. The probit estimates of ~lumns2 and 4, hy contrzst, requirs some translation. To consider t5e recent fertility pmbit first, the chance of a b i i in the preceding five years is .666for a w m m without schooling, .623 for a woman with primary schooling, and only .527 for a mman with secondary. (All these are average predicted probabilities calcuiated allwiq other covariates to wry as they do 51 the full simple. They give an idea as to the differentials associated with the m i a t e in question.) Womm who reside in the savannah have a .701 probabi!ity of a recent birth; this probability is .702 for wmen in the rural east forest, ,664 for wmes in the rural west forest, .608 for those in other urban areas. and .547 t%rwmen in Abidjan. The probabilitycf a recent birth is .642 for the Northern Mande, but only .575 t%rSouthern Mande; the other ethnic dummies are insignificant. Similar calculations t%rschool enrollment shw that when ths mother has no schooling, the averageprobability that her chiid wi!l be enrolled is .350. This increases sharply to .448 when the mother has primary schooling, and to .538 when she has been educated to the secondary level. Enrollment is lowest in the s m n a h region, with an ayerageprobabilityof .217. This is less thar,half the enrollment probability in Abidjan, which is -471, or in the remaining urban areas (.464). Males on average have an enrollment probability of .420, as compared with only .318 Br Females. The model of projected completed schooling (see Ap2endix 1) suggests that Ivoirian children will comp!ae an average of 5.26 years (this can be compared to the sample average of 2.78 years, which is misleadingly low). The estimates imply that iilale~will completeon inrerage6.11 years, and females 4.36 years. Childrenof mothers without schooling are themselves projected to complete only 4.89 years of school, as corupved with 7.07 years if the mother has primary school.hg and 8.46 years if shs has any secondarytraining." The Ihdogeneity of Consumption Tine estimatzs just discussed r s t on an assumption that ali cmiates in the analysis are exogenous, that is, unmelated with the error terms ir. the associated fertilityor schoolingequation. Thisassumptisnis much in do~btwhereconmmptionper - adult is concerned. Consider one scenario (and see Ben& and Schultz, 1992, br 13. In the ord&d-probit model with allowance Ibr censoring, the child's age should be inkrpreted d i f f e 4 l y than in tbe cumulative schooling model. In the cumulative schooling model, age is in ptapmxyfbrthespanoftimeoverwhich,xhoolingcanbeaccumulated. In the projected &ling d l , this functionof ap has beemtakenxp by the redefinition of the dependent Muiablasee Appendix l), so that here, age should be interpreted as representingthe effectsof calendnr%me and otherunmeasuredbut trended variables 41 others). \Ve see in Table 3 that consumption is positively related to fertility, and we have interpreted this as the effect of permanent income on fertility. But surely causation in the other direction is equallyplausible. If children make a contributionto Family income, and this contribution is reflected in consumption per adult, the positiae associationseen in Tzble 3 is at least in part the product of raerse causation. To protect the estimates agzinstthis possibility the standard practice is to predict consurnp;ion per adult using a set of exogenous instrumental variables, thereby purging (one hopes) the predicted value of consumption of any conmination f r ~ mendogeneity. The instrumental alariables technique is in principle beyond reproach. But in pnctice it is disturbingly vulnerable to the selection of the instruments and to multicollinearity. In many applications, it is not clear that the method brings one any closer to tl~etruth. That at least is our conclusion fbr C6te d'lvoire; the basis for it cim be fourld in Appendix 2. A series of instrumental variables estimations and formal ex~geneitytests showed that: (i) the urban and rural samples should not be pooled in estimation; and (ii) once the data were partitioned into urban and rural sub-samples, no decisive evidence emerged in favor of the instrument-based results. That is, the tests neither sustained nor clearly rejected the assumption of exogeneity in consu~zption. There remains a possibility of bias in the uncorrected mdels, and our judgement is that if anything, the effects of permanent income on fertility may be upwdly biased in the uncorrected models, whereas the effectson schooling may be downwardly biased. As noted above, the (uncorrected) e f f m d income on fertility are modest in any case. With these cautions in mind, we now tum to the separate results fbr rural and urban areas. Rural and Urban Su?Samples The results are presented for rural areas in Table 4 and urban areas in Table 5. With regard to ruralM i & we find that consumption remains positively associated with cumulativs brtility, with an elasticity at the meaq of .03. No significant effect appears in the recent firtility model, but the positive influenceof consnmption levels on schooling remains smngly significant in the rud sub-sample (elasticities of .249 for cumulative schoolirg and .375 for current enrollment). Woman's education loses sisnificance altogether in terms of rural brtility (recall, however, that only 1.8 percent of rural wmen ha: any secondary schooliq in rural areas). The effects of wmen's ducation on current enrollment, by contrast, remzin positive. The negative schwling dIfFercntials fbr girls evident in the pooled sample reappear in the rural sub-sample. Table 5 %uppliesthe results fir urbzn areas. Here consumption levels fail to reach significance in the cun~ulativefertility regression (while remaining signifi'cad in the recent krtility model, with an estimated elasticity of .I14at the mean) but show a strong inEuenceon curnulztiveschooling md current enrollment, with elasticitiesof. 168 and .297 respectively. Mbrnm's educ~tioz,exerts a coiuistendy negative influence on brtility wittiin urban veas (although the primary school coeffi$ient is not significant in the cumulativefertilityd d ) and shows a consistently positive md significant influence on school~hg.Within ilfbaz~areas, as was thp,case within mral'areas, parental schooling ? investments meal 1pref-ce in Ewr of boys. The differena amount to half a yeax m of schooling in the cimullaive schooling model. Table4. Cumulariw F m i I i and Cwnulatiw Sdtooling ModcIs: heline Results, Rum1 Sample (T-sfatisria inpamnthcses} Any X i G k G Cumulatiw ka5 Cuauhh Cumb Schoolid kdlily School* EnmUncn (Ordered (om) (OW t (Robit) Plobii) 3hn*il)' Ahn Krm SoulSraMudo Vbhric other Wmcdr mr 2&24 15-29 30U 35-39 4 M 4 4549 so* ibuk Ncrcr: a. Wmen aged 50andwcr omintd. b. Oldelcdpmbitd t wirh allanace ;kwr i g k t d ~LiiWd c u l f a h pannvkn a:4.226.4.28C. . ~ 4348.4.444.4.435.4.681,5.207.5.276,1S339,5.445 ,5.9~,5.966.6.121(Scc m i x for expluulioa.) e. M a c d cfikpy: No rhoolirp. * d. Onindcskgov: S m d . e. Ora'medc a w : NodhemMa& f. Omirrcdc a w : .- 15-19 g. Orninedc.rgor). kr cumul~iwand cumaenrdlnvnfmodclr: Ages 5-7. h. Tea g i n s t nuUmodel4th c*~-poidpars* 3andchild'rage. Table.K Cmolatiw Fmilityand Chdatiw Sahooling AIatieL-:- Bareline Results, Wm Sample Gstatistiicr in pmarrhLC#,) - -< htidtncd Abidjan -.324 -.I96 839) (2.73 Ethnicity appears to make orly a modest differenceto rcral fertility levels, with the Southern Mande and k l h i c pwples having slightly lower fertility than other groups. Ethnic differentials in regard to s&ooli=!g, however, ara strikixtg. The Akan, the Krou and the Southern Mande all make higker schooling investments in their children, all else being equal, thvl do the other e W s groups. These ethnicity effects on schooling are evident in both rural and urban area; t!ie ethnic influence on ertility, however, is largely restricted to rural locations. The predicted probabilities fjt &e recent fertility md current enrollment probit models are presented in Table 6. Tables 7 and 8 examine whether consumptionhzs a nodinear effect on behavior, such as could be captured by a squared term, and expl~rethe contributionsmade by child surdiw, rates. We find little evidence !br any important non-linearities in the effects of consumptionon fertilityor schoolingin rural areas Fable 7). Fertility remains a normal good throughout t!!e consumption range (the downward turning point of the quadratic formulation occurs at cocsumptionlevels outsidethe rural range), although the elisticity of cumulative fkrtility at the m a rises scjmewhatto .097. In regard to rural schooling, the effect of consumption is positive throughout the range of consumption (the wefficients are mis!ading in thzt ,t)the squared consumptionterms d d s little and of substantive importanceto our understanding. By contrast, in urban areas Qlble 8) thz addition of a squared tern in consumptiondoes bring out non-linear effects of somesubstantiveimpoace. The effa of consumptionon fertility is posi'h up to the 70thpercentile of the urban consumption distribution, but then n;mnegative at higher lwels of permanent income. A similar effect is evident fb; recent fertiliw although here the downward turning point does not occur uctil the upper 10 percent of the consumption distribution is reached. This is evideilce -if admittedly somewhat fragile evidence -suggesting that fertility behaves as a normal good up to a certain pmanent income thnsho!d, and G5en turns inferior in respect to income. The raultr having to do with child survivai rstes are striking, althocgh a note of caution 3s in order here as well. Recall that the survival ineasrrrc is constructed from 1988 census counts, by commune or sous-p@t?eture, of th= ratio of children surviving to cf,iIdten ever born. Tabje 6 sbaws that within rud areas, the sirviva1'ra:e is negatively associated with cumdati\b fertility, and positihely associated with schooling. The fertiiity elisticity is on the odes of -54 when d u z t e d at the mean. Note too &at no effect is evident wit!!resFect to reccnt fertility The size of the elasticity, the absence cf a strong efTect on ¢ Mil@ and the apparently smng effect cn cumulative fertility, all suggest thaLthe mechzahn linking chiid scr;ival to finliity has to do with h e truncation of brear@eding c a d by a child death (Llqd and Ivanw, 1989). Table 6. N i d e Probabilities cf Rsaurt Fmiliry and C u mEnmlhent: Ruml and * A Sub-Samph 1 RURAL W N H6men:Any aiklren.. Wmcn.. Children.. - births Cumn~ Any birds Current h t 5 yeam EnrPlbncnt krrt 5yeam Enrolbnent -__I Sbmcn's Ed~cation None [.6141 [-nq [.%r] l.4691 Any Primary .688* 3 8 1 .Sf3 -573 BeyondRimary .733* .41P .480 .669 Residence Abidjan .567 .514 9 t h Urban ~ [.a261 [.5091 Rum1 &st kmst .672* 32!5 Rum1 Htrt Forest .658* 326 Savannrh [.696] [-196) Ethnicity Norrl~ernMande 1.7091 [all l.566) [.MI Akm .689* 324 .557 -558 Kmi .694* 3W .612 .541 Southern Mwde .5% 321 .586 .594 Mlkic .6C9* -177. .640 .486* Other .748* -139 .a3 .464* Su Male [321 [.545l FL-male 221 .478 LL Nous: Omittedcatcgoy s h m in b m c b I]. *thecontra& to thc omitled ategoy k mofsignificant. Surce: Average pdidcd pmbabiitiu tram thepmbit nw4& of 'Abla 3 and 4, ca!cuW u: N P, = (1I.N) C + fi1) i j?'k wherc k dmto the covariatt in qutrc#nand thco b r d ~ rubscript i ranges wcr thc Fill n sample (i.e, all rud or all u h m obsav&olu). This fi?mulationall- c-tcs j o&tr *an &tc k (uccepa far those in tlu csPg, . of dummy -la being ~ M ~ , ape rct to zcm) to take on thek ram* duet 3.Thus, when ethnicity dummies ~ICrct to zcm. m abtain the p d i probrbility br the omitted -y of N o h Mmde, to calcub b Akm prediction, we sudl cthniity dummies to zcm accpS br the A h , , dummy; and so on. Note that thh is Q&rrnt Uunr e l e orJy the rub-mnple of N o h Mande or h, that the other c4vrri.lcr arc not rutrided bthe vdwa that they take Sor in - - such sub-runpla . ...3 - m e m T i b 7.Cmuhiw lertility and Cknuhiw Sdmling: Allanrrivc Sprc~clatfom,Rmd RM-Jk (r-tatisiia hpmemhuu..) %AQ) C x l d ? ? Rvfecud Any lnhr Lasi 5 ~ d v r Qurrnt Completed Cbnukfwkr&Cty Yeon w E n r d b Schooling' fas) (-7 fors) -:I (oniemi probir) b. Orded pxM lmdcl mJ ~ w c r ohtitfi~scdrg. c. WuBCIPPI).:NO ~ 4 d i q . C. ~ c u r p c y : ~ . e. 0 m b d m : N c n h n a M I D J a . f. Tul d'ddvi;bcul-poimpn-m o I ud child's yc. -- Tab&8. Cwnu&iw F d l i f yand Cwdariw Sc-: Almmtiw S'~@~cltiom, L'rban SIunpk (T-~iaticticsinpnmhrrts.J Ilbma, awlm h j d clfdnfiw h y l b i r t k k i s amml -d kdig %an shlfiy ~ u d ~ u n t Sdlolr$ -- (om w r (ors. -I (Onicd thrkt) (11 121 111 12) II? R) (0 RJ IlJ RJ Htmcn'r Education* ~ n ~yi - r y -.043 -.a37 -310 -291 -777 .a6 395 317 .4n 310 (0.33) (037) 0.59 (3.17) 3 023) 0-W (433) (6.61) 0-W Beyoad primur -.701 -.a1 -.641 -Jbl 1.0s .m .nr .ao .anti .n9 (5.19 (4.871 (6.58) (6.93) @.MI (4.62) P.07) 0-02! 8-26) (6.69) C o ~ t i o n 4.300 4 3 2.112 2308 4.31 4.22 I . 1.441 1.473 pet e r 0.11)0.16) a.08) (UP 3.42) 0.0~. . I i1.a) 0.01) (6.26) C ~ w - b i a n . -.I57 -.I60 -.On -.m -.IS4 -.I37 .mO -.MI -.W Spmd (1.04) 9.09) (5.94) (2.11) CZ-W (269 (0-71) (0.W 0-67) (4.01) Residence' Abidjan 0 .022 -.Ow -.I 1 1 .a7 I -230 -312 -.033 -.022 (0.19, (0.12) (0.59 @.kc) (D-U) (0.931 039 0.11) (033) [O-26) S J M ~XJ* 4.608 4.490 . -.a70 3.764 3.065 3.937 3.971 2.028 2.206 (2-47) (1-W (0.67) (0-69 V) 0.67) 0-93 0.W 4)-lf) (290) EtEnicity' A h m .a5 -.a30 .W -267 -42 @=I (0s) OM) 0.w c5.m Krr .090 .I42 .9U 112 .624 (032) (I-03) (4.94) rr-91) (5.49 S.Munk 233 .MI .tSO -416 562 3.os) (037) 0.99) (3.49 (4.52) %bit .ICi -231 -32.5 -082 -.a (0-74) 037) :I -67) (0-76) (0.33) O u ~ r .I66 3.59 -332 -.012 -.I95 0.oo) (2.29 CL.13) (0.49) 03) W I M ~ 6 -.617 -318 -224 -343 -3611 - t (6.71) (6.76) 4 (4.41) (6.78) (7.14) R2 4 .4w .474 .4% For? :43.4 103.6 496.2 4213 198.7 1G.G 592.9 1026.5 580.8 5iI.v @-h) (.Wc.m cow) c.m Cow) cm (-w . i-am <.ow) Obrnoticar 1929 1929 1741 1741 3198 31W 3 1 3 1 3I(W fiOI C ~ * r w . n W 8 ~d~ ~ o ~ ~ ' ~ ~ ~ i ~ c b m . - Nofsr: a. Ubmm@fJdcwcroaiiacG, v b. ~ p m b i t t ~ ~ v i t h m . f b n o e o b t ~ ~ . c. OdKkd cakp¶y:No rbcdii. d. im#:r Z ~ m u b . e. ocQi5bd .?-r_y: Nonbem M.ndc f. Ttd ill~IKI&! W& MpOid u&WS Note that in rural areas, child surviwS rates are positively associated with c;mniulative schooling. The elasticity at the mean is 1.12. The meaning of this association, however, is mt entirely clear. As discussed in Section XI, it could represent a greater willkgness among parents to invest in human capital when child survivll is more assured; or it might r e f l a the.strain on b d l y resources produced by largs-than- required numbers of surviviq children inhigh-mortalityenvironments; or the association could simply represeEt ~mittedhctm:-asocizited with child hdth that h m some corre!ational or behavioral I.;& ta sr%c,-';ngdecisions. It is perhaps surprEgf - rar similar findings are in evidence witbin urban areas, as shmn in W l e 8. As in rurai areas, schooling elasticities in the urbvl sector are positive with respect to child survival, and the cumulative fertility elasticities are negative. The elasticity of cumulative W i t y is -1.27 at the mean, which suggests a more-than-compensatingdrop in krtility given an increasein child survival, such thatthe cet repduction rate wuld tend to fall if surviwrship improved. The cumulative schoolingelcsticity is -924atthe mean, and thatof current enrollmentis 2.58; the latter seems implausibly large. The Quantity-Quality 'hdeofl Redsited Let us sum up the results presented abme in Figures 19to 21. We return here to the depiction of the quantityqualitytradeoff in ternsof a locus of tulo endogenous variables. On the vertical axis of Figuies 19to 21, we indicatethe predicted number of children ever born, using coefficieilt estimates fromthe cumulative fwtility models of Tibles 6 and 7 and taking a. given the distribution of exogenous characteristics in the LSMS sample. The predictions based on these sample characteristics are evaluated setting the -rw)man*sage to 40-45, in order that tbe predictions will approximate completed liBime Miity. On the horizontal axes, we dqict the projected years of schooling fbr a rcpreswtatir3.child born ;a,a wman wirh these san~plecharacteristics. Tbepredictionsaredm- From the orderedprobit, p r o j d completed schoolingmodels of Fables6 and 7. One prediction is made if the child in question is &male, and another if the chi!d is male. Tlese predictions are d u m d settiq the child's age to 25 years, so as to approximwe expected completed years of schooling. me results to tkll~w would be very similar if we had examined the predicted proportions of children witb swcdary schooling, fur example.) PredictedFutilityar,dSdrodingPerChild wran~qerD-4%l;b~e25 8 - ,A --@lo C 0,' P- I : f0 n d *:-w_ 0 r --.----- 08 *fa'-.- 0- L: \ L---- 0- -. **b + . - ' - ' ' " " " ' - " " " ' 1 2 1 4 &.dad [s)8 0 10 11 F i t consider Figure 19. We can emrision the full space between the axes as being filled with predicted data points.'4 How should the joint distributionof these two endogenousmriables be summarized to show evidenceof a quantityquality tndeofP1, We suspect that these estimatsdsunrival rate effects may represent fictorsother than sulvival per se, but they are n ~ e l e sintriguing. If taken literally,they suggest s that investments in cbild health and survival may hare a spill+ver influence expressed in fertilityreductions and (perhaps! in schoolhg. B Q * I 14. A single such & pint is ca&m&d as hllws. Considering a giwa \H)L~SMi with exogemw covPriatesX,, we calculate a pair of @ctcC \dues - pi = x,B, -. -2 * 3 = x,B, L 8 a whsnX, mmprises dl covuiateJ i o c l in tbe BKWWSTsbles*d ~ of 7, with being at to 40 and 25 for the W i t y nnd srA~linge q d m s mpectively. W let h(X) repreat h distributionof X in the LSMS samples One summary dwice is indicated in the fertiiity-schoolirg locus cf Figure 19. This gives the expected value of schocrlistg conditiod on the level of Another wj to toummxize the infomarion is to plot the predict& tnan values of fertility and schcoling fir differeat ~cioeconomic Thus, we hne indicated on the graph the mean levels cf M i t y and schooling h r urban residents, b r rural residents, axld for emale and male children." Ths schooling-Gnilitylocus ees a puling and pwrlliar shape. In the upper portion of Figure 19, the locus seems to suggest a positive relationshipbehveen lifetime fertility and child schooling. We see thatthis portion of the curve lies near to the rural mean values for fertility and schooling, and on this basis, we might surmise tbat the positively-sioped pcirtion of the "Weoff"pertains to rural areas. The I m r part of Figure 19 sbows a predominmtly negative association between fertility and schooling; perhaps this is the relaiionship that holds sway in urban areas. Overall, then, we might suspectthatFigure 19has superimposedrwovery different relationshipsof oppositesign. - 15. More formally, represeats the set ofCOWiak \dusX yieldiqg r givca predicted value br kt-tility (in practice we amsm~ctm n p ofWues forF.) 16. This is simply the paitofuutditioanl catan predictions where X, is the covoriatecategory uadacoasiderotion, ag.,all ortipn residents, end the full set ofCavpriates is mitoen u (X,,XJ. L .c 17. In the figunslo Rbllaw, tbe predickd krtilily I d is dwys the same when girls and boys 9 ue co~lporod;only the pdlkted .scbookg levels differ. In effect,we assume thatthe sex ratio * - at birtb is indepenkt of W t w t h e f s Ektility level. Note dso &at the vertical ues ranp frorn r 1- limit d 1to M Limit d 8. ??re ~gressioapredictions dl lie within this range, e* which is plausible 6% ohishigh-Wiity poplatition. In ptdcular, no wmnn w predicted to have O children as of tbe ax! of ba nprodudiw pars. PrddmdFertilityondkhoding: Ruml W o n u n * s r + M X ~ k p ~ # A *--- JLqr I 1 2 3 0 7 h*d4Grndr (Sf Thissuspicionis c o h e d when separateplots areproduced ficrmrd a durban -. Figure 20 sheds light on the relationship bebma child quaritity and quality in rural areas; as suspected, the positive reiationdominates. Tbe male aod M e means aprw in the center of the figure, and sex-specificvalues fbr families in the lawerrnost and uppennost 10petcent of the consumptiondistributionarealso &awn. The general upward slopeof the Mility-schooliq locus, as well as the sequenceof mean values hr boys and girls sqaiiy, suggest the i m of permanent incomes to li-e fkrtility and schooling imstments in rural areas. We shogd also recall that the pemanent income effects may be overstated; it is possible that the true cum is flatter than what is shown here. hdictmd Fmrbiandkhoding: Urban W O m r n A g 4 4 0 4 ~ C N l d r r n ~ z 3 ,A - L E m c *,---- Y as &-. I 7 .a,- B* is 0 &--- -----__ 0 - Mu- c., mrrm cr-nb- 7 O .O f 4 4 ' ' ' . - . ' a . ' - ' 4 5 6 d . d A o , D 10 11 Figure 21 Figure 21 b r urban areas showsthat the conventional, negatively-sloped brtility- schooling locus dominates. The effectsof income are rather mild. For each sex, these incomeeffects arevisible in the slightcurvilinearityas children in the bottom 10percent of the consumption distribution are compared to those at the mean and to those in the upper 10percent. In short, these findings suggest that the urban and rural economic and policy environments of C6te d'Iwire have strikingly different implications fbr thr tradeoff between child quantity an4 quality. The findingstherhre underscore the importance of urbanization to long term development prospects, in which reductions in labor force growth and increases in human capital per worker arefundamental. They also hint at the potentiai role of a policy variable-family planning servicedelivery -thatof necessity has gone unmeasured in our research. Recall that to date, family planning has been availableonly through priwte s-r sources in COted'Iwire, aid lbr the most part these have been restricted to crban areas. Perhaps it is the milability of a means to restrict fertilityin urban areas thst h'asfacilitatedthe tradeoff between quantity and quality. The Price of Schooling We close our analysis of the quantity-qualitytradeoff with an examination of aaother policy variable that may have a direct effect on schooling and, some wuld argue, an important i n d i effect on fertility decisions. This is the exogenous price of child schooling. Figare 22 presents mean values fir school expenditures according to the child's age, calculated from the U M S sample of enrolled resident children. Average school expenditures per enrolled resident child for ali ages amounted to 37,897 CFA." One curve represents total expenditures per child, expressed in thousands of CFA, and the other cunPerepresents the ratio of per child exwitures to consumption per adult. The omes are strikinglysimilar in shape. The right-most curve, representing expenditures relative to consumption, shows that expenditures amount to only 5 percent of adult consumption for children of ages 5-7, but the figure rises to over a third of adult consumptionk r children intheu twenties. ClearIy%esearesignificantamountswhether expressed in absolute or relative terms. Given that school costs are substantial, how much difference do they make to enrollmentand fertility? 'Ib answer this question, we must confront an awkward feature dthe LSMSdata. Thecostsofenrollmentarenotknown for allchildren,butonlyfor children who are enrolled, and Curther, not for all enrolled children, but only for those wbo are resident in the househo8d. To put this in another way, school prices are known only fir a highly selectee sub-sample of childrea. I ExpendituresPerChildonEnrdlsdResidentChildren ToldandR.I& loCommpUon lW c ]a40 I s 1 - Figure 22 * * 18. Or USSl10at the 1986 exchange rate d 346 CFiVdoUer. There is no stcaightbrumrd solution to this selectivity problem. We proceed usinga method outIinediriMaddala (1983)and Heckmar, (1979), in which theprobability that a child is resident and enrolled (i.e., is in the sub-sample for which school expenditures are known) is moddled as a hnction of covariates and instrumental variables. n e results of this first stage are used to fonn a selectivity correction factor, which enables a cost-of-schooling function to be estimated without large-sample selectivitybias. The cost-of-schooling regression includes only exogenouscovariates that vary by sample cluster; we add the child's age to allow for changes in school costs by gradeIml. The idea istoincludeonly the cavatiates that could influencethe exogenous cost of enrollment facing parents. (For instance, we do oot include the sex of the child in tbe body of the regression; it is indudeci only indirectly in the selectivity correction term.) The predicted dues from this regression ace then used to construct a predicted schmling cost variable, which is ins& into the final structural enrollmentregression. As are all instnrmental-variabks approaches, this approach is vulnerable to multicollinearity Ia particular, the striking age pattern in costs evident in Figure 22 is largely retained in the predicted costsvariable. This makes it difficult to distinguish the true effects of school costs on enrollmentfrom afiyunmeasured effects of age tha! might be captured in age-specific dummy variables. The enrollment results we present in m l e 9 below substitutethe child's ageand age squared k r the full set of age dummies used in earlier tables. With this change in functional arm,the effects of the school price variable on rural and urban enrollments appeartobestrongly negative. The elasticities implied are -1.41 for rural areas a d -.76 for urban areas. These can be compared with the range of -.05 to -.61 Ar rural Peru givea in Gertler and Glewwe (1989: Thle 7), and to their more tentr?tivelower bound estimateof -1.17 incorporatingselectivity (see their Table A.4). Elasticities greater than -1 (in absolute -due) imply that a marginal increase in exogenous school costs (e.g., through higher scbool h) result in lower total revenues. would The a t e d'lvoire findings are wn~istentwith a greater anticipated payoff to schooling in urban areas, but the esthates are not especially robust. The introduction of dummy uariables for the child's age removes the significance of the estimated price effects, as does any further separationof the samples into children of primary ages and post-primary ages. We should note here that the education infrastructure variables, which measure aspects of access to schooling, are PEatisticzlly significantin both rural and urban areas. Mowtwr, as also reported by Glewwe (1988), they do not make a greatdeal of difference to the estimated probabilities cf enrollment. R seems unlikely, therefore, that improvements in the access-to-schooiing variables measured here could fuily compensate - - - ! * 55 bousekolds for increases in school costs.'9 (See Gertler and Gleul-we, 1389, for further discussion in the context of Peru.) Do schooling costs have spill-over implications for fertility? We see little widence of this in Cdte d'Iwire. The schooling price variables are individually insignificant; indeed, thesetof schooiingprice and acccss variablesmakes no discernible difference to fertility in either the rural or urban sub-smples, to judge by a chi-square test. We think that it would be premature to dismiss the possibility of such spill-over influences, however. Our test is based >n a schooling price variable that is in essence little more than an age profile; it does not display sufficieirtcross-sectional variation, net of age, to make a smng showing. If the schooling prices variable could be wrrstructed with areal-!we1 or school-level data, as are amilable for Ghana (Oliver, 1992), a great deal more might be learned. - 19. Our e~rperimeobwith mrious quulitydkdmling mewres, coastructed from tbe d community or inspectDrPte data listed in T.Me 1, p.disrppointirg resulb (not shown). Measures such as s ~~ G orSWofschoolmg p~1em.w6scsdby thecommunity t ~ g e n d y seemed to tap .spbctsddemand rsthathrnsupplyamstrPinBorquality. For instance. with dl elseequal. rural communities with r percdvad of teorbersorclessmrns teaded to hve higher-than-typid enrollment levds. Cornmities 1- in primary schaol inspectorateswith high d e n t - h h e r ratiospIso bad higher-th-typical enroUmenh TJble9: Models Idlrding rlte Rice qfschooling (T-statisria in pamheses) M ~ n ' Edusa~ioil' s ~y Rimary .wa (0.62) Beyond Primary 216 (0.96) Consump:ion per Mull .069 258 2.359 .429 (131) 0.60) (t-2@ (0.51) Consumpion. Squarrd -682 -.W Q.12) 8-10) Residencec Abidjan Rcnl Earl ForrsI Runl Ukst Forrsl mnicityl ALn Krou SouthernM a d e %laic NOP r i ~ ~School r y .I23 -.I56 (0.81) (1.34) Year Rimay Built .008 -.OD2 (I -68) (0.63) . * Yur Sccodary Built a 5 -.W .008 -.a17 (1-00) (2.94) (t-W (5.58) - *nt -n's age and chilCrcn'r rp cocfficicnb mt Ihoun. L . Clo*r: ' a. Ulbmcnsgd50 and olderomincd. b. Omitvdcalgor).: No .chodirp. c. Omitrcdcaugoy: k r r v d a i u r Sm& &urban. u h . d. Omintdcakgory: Nonhem M d c e. Edimlcd fmma ~ g r r u i owith dcpsndc~nricbkbcii vhod expcnditurra, exphnabry variables incldirg chid's n sge and a met of c l u ~ ( ~01- hii!acr-~~cI r vlriabks. The cxpendituma f e g r r h include r eoncclion due lo HccLnun (1979) lor sample ulectivity biar The pdW nluca d o c h d expdiwas. which wry aceordig to the age of rhe child. arc enkrcd in thc currcn~enroUmenr rnodelra b e . For the fenility d c l a a b c , the predicrd nlue of schno! expendituua uns cvrlualcd at ichild's) age 12. Conclusion This paper has produced evidencesuggestingthat two very differentrelationships link fertiliq and child schooling in Cote dlwire. 21rural areas, a positive tmdeoff appears, a finding comista with much of the early research on fertility and schooling in Africa. Urban areas, by contrast, are characterizedby the negatively-sloped tradeoff that appears in Soubeast Asia and elsewherein the developing world. With respect to other sectoral and policy issues, the findings of a tradeoff in urban areas of CBte d'Ivoire reaffirms the significance of urbanization to demographic transitions. Yet it is oot entirely clear why urban residence matters. One aspect must have to do with the economic benefits of schooling, qahich are perhaps more clearly visible dad persuasivein urban settings than in rural. Access to schooling, and to family planning, is also facilitatzd by urban residence. It is also interesting that wmen's schooling has a contsxdependent effect on fertility and the schooling of children. There is little evidence that women's schooling makes a difference to krtility in the rural arras of Cdte d'lwire; and even the influence on children's schwling is somewhat tenuous In urban areas, by contrast, the effect on fertility is cleariy negative and the influence on children's schooling is clearly positie The difference between tbese rural and urban resdts inay havc to do with selective out- migration of better-educated women ftom rural areas, such that wmen who remain in these areas have uniformly low levels of education. One wnders whciher investments in wmen's education in rural areas might rhereCore leave rural fertility essentially unchanged, while enmuragiqg greater ~ d - t ~ ? 1 r bmigration and lowcr urban W i t y . m We do nct find any convincing evidence that increases in school costs in Cbte d'I~irewill hme a significant bzajing on levels of fertility. Indeed, if we return to Figure 1of the paper and ask how our findings about price effects would be represented there, the implication is that schooling price increaseswuld be reflected in a movement from point A to point C: littlechange in fertiii& but (considering the elasticityestimates) some considerablereduction in school emilments. However, this last finding is tentative, lbr as discussed abcve, the ccdnsrructed school pricz variables lack the cross-sectional variation needed to make any prec'ke measurements of impact on enrollments. Nevertheless,the large enrollment elasticities are disturbing. They raise some doubts about the wisdom of any further shifting of school costs to the ptirae sector. C!early this b an area in which further work is required. Finally, we are intrigued if not wholly persuaded by the &pknt effects of child surviwrship on schooling. It vauld seep. that in addition to the dim impact of investments in child health on child surviwrsbip, there may weil be an indi-ect benefit exprssed in geater educa!iocd attainment. Thibpossibilityd w e s fiirtherexp:oration. ? Appendix 1 The Ordered-Probit Model for Projected Completed Schooling The model for projected compIetzd schooling combir~esinformation on the number of grades a child has completed as of the time of the LSMS surveys with data on school enrollment at survey. Let S; regresent the level cf schooling that will eventuallybe at?&& by child i. Call P(SIa,XB) the probability distributionof S given covariates X and associated coefficients a and 8. S is of course discrete-valued, taking on values 0,1,2 ...,and P(S)_isa discrete distribution. The difficulty in regard to sthating P(S) is that S itself goes unobswved for a high proportion of childrenin the sample. The Si datafor childrawho are still enrolled as of the survey dates areright-censored, ii~the sensethat eventualeducationalattainment Si is not yet known for such children. All thzt can be said is that, whatever S; may turn out to be, it must either equal or exced the level of zttaiment achieved as of the survey date. Let C, represent the grade compIeted as of the survey axe, and let E, represent enrollment status at survey, where 1indicazs enrolled and 0 not enrolled. E can be then regarded as an indicator of right-censoring. That is, if E=l the level of schmling completed as of the survey date ci provides a lower bound on the level that will eventually be completed, such that SiTc,.Children who are not enrolled at survey can be presumed to have completed their schooling, so that for these children Si=c;. Of course this last assumptioil, whereby non-enrollment is equated with completionof schooling, isproblematic far children of ages 7 or younger. These young childrenmay have Ci=O and q = O simply became they have not yet enteredschool. To guard against this possibilip, we omit Earn the sample all children aged 7 or younger who have no schooling (Ci=r,) and who arenot enrolled &=O), on the grounds that for them, C,=O and E,=O provide no i n f o d o n as to eventual edncaiod attainment. Even fir older children, however, a spell of nonenrollment need not signal the end of the ducational career. A childmay be withdrawnfiom schooltemporarily, under the expectation that he will later resume his studies. The ISMS survey contains a 'question asto whether non-emlled ch3drea expect to return to school in tbe fiture. In principal at least, such information could be used to r e h e the estimatotdeveloped here. But as it happens, this question is addressed only to the children who currently reside in the household. As noted in the text, e*en at ages under 7 some 16 percent of children live apart fiom tbeir mothers; &e figure increases to 28 percent among children aged 12- 13, and to 50 percent among children 16-17. The questionon resamption of schooling is therefore of somewhat limited -due, - - In short, the approach we develop here produces estimates of comple&d schooling that m y be dowtwacdly biased to a degree. It is unclear wh- incorporuing dda on the resumption of schooling would improve the siblation, as t h e data are adable only for an unrepresentative sub-sample of still-resident children. Nevertheless, we intend to investigate the matter in m r e deed in future work. We estimate the discrete distribution ?(S) quing an ordered-?robit specification with aa allowancefor right-censoring- Inthis apprcach, the probabilitythat the level of completed schoolingS=s is defined by areasunder a n o d probability density function between cut-points a*,and a,. The density function in question -call it f(SIXB)-Bas a mean equal to X,@and a variaslcethat is mrmalized to unity. Figure A1 below may help to darify the approach. Supposethat two cbildren are under consideration, one with covariates X, and the other with covariates Xi. Normal densityfunctionsf wit3meansX&3 and X,@aregraphed in the figure, along with cut-pokt pxamtters cq,, a,and or,.* For the child with covariates &, the probability of no schooling (S=O) is represented as the area p, under the curve f(S1X$) to the left of cut-point q.Theprobabity of oneyear of schooling, S=1, is representedby the area under the curvebetween q,and a,;and so on. As the figure is drawn, the child with covariates X,bas a density function that is shiftedto the right relativeto that ofthechild vSithcowiates &. This means that the probabilities associated wi!!~ 0 and 1years of schooling, p,,' :ad pI*,are reduced in magnitude, whereas theprobability associated with ahigher level o,fschooling, p,' in this example, is increased. To put this in acother way, if tte coefficient p, associated with covariate X, is positive, t h a an hcmse in X, increases the likelihood of greater educational attainment. The estimation of P(S 1a,XB) p e e d s by maximum-likelihood methods, with both the cut-points a and the coeffiacnts @ being estimated. To form the likelihood function, the probabilitiesp, are represented in the form +(a, -1 - for S = 0 PI ' - W ) -4i(al,t -X$) fors E (1,2,..,s,J (4) 1 - @a1* -4% ifs > s, where$0 isthe cumulativestandard~onnaldistributionand s, is the level of schooling beyond which we aggregatethe u, pmbabidities." 20. In considerkgthe distribuhof schooling, we make al!owance for some 13 apyameters; feweraredisplayd hae insrder to si~plifythe exposition. 21. In our application, we considersingleyexs I$schooling up to 12 years, and group all higher levels of attairunentimo the category 13+. That is, %=12. Given dl ttis, the probability associated with having C=c completed yezrs o f schooling a d being currency enw!ld (E= 1) is for c r 1. Tie proba5iIitie.s for havicg C=c cornp1e:ed yellrs and not being enrolled (E=O)aregiven by the appropriateexprzssions in equ2tion (1) above ai!h cr,substituted where a,2ppears. Figure A2 shows the results of a baseline stirnation with the chiId's age ectz;A as the sole X covariate. T e figure conpares the actual distribution of years of schoolicg completed as of the L S L survey dates (that is, C), with the predicted disr xtion of ' schooling P(S (a,,Yd) based on C,enrolluent rates E, and age. The allowvlce inade in the estimation for right-cecsoring (cases with E= 1) causes the pledictcd disaibution of schooling P(S) to be shifted to the right, relative to the distributinnof )ears completed as of the survey date. This is precisely what we would expect. The means of the two distributiocs y e 2.78 years and 5.24 years. P th distributions show a substmtial proportion of chzdren with no years of schooling (grade 0),with completed pisnary (grade 6), and wih completed early secondary schooling (grade 10). The predicted schoolingmodel also suggeststhatsome 10percent of children will proceed as iar as 13 years of schooling. &either distribution contains large prcportions of children with incomplete primary or inamplete secondary schooling; this is consistent with what is known about enrollmentpatterns in CCte d'Ivoire (Glzwe, 1988). Finally, note that the predicted distribution of schooling P(S) bears no resemblance at all to a ncnnal distribution. So long as the cut-points rr are allowed to vzqj in the estimation, the assumptionof normaliry in equations (1) and $2) arcounts to little nore than a convenience. ils k the case frere, the ordered-probit method GZI in general produce decidedly n o n - n o d distributions for the dependent variabla being mode!led. I - Figure A1 Figure A2 Appendix 2 Assessing the Exogeneity of Coxxsumption per Adult In this appendix we weigh the evidace regvdig one key variable in the andysis, the log of consurn?tion expendim per adult. is at issue is the degree to which this variable-maybe correlatedwith the error terms in thefertility or schooling equations. If such a conelati~nexists, then we would considerthe consumptionvariable to be statisticallyendogenous. Its presence in the regression equations would impart a bias ?ot only to its own rpgession coefficient,but also, in general, to all other estimated coefficients. There are suF,s$aativegrounds for concern. Benefo and Schultz(1992) provide an excellent discussionof the issues, and we would singleout one possibility for special attention. In a situation in which older children w r k and make a contribution to househdd income in cash or in terms of the value of labor, prior fertility may be an important determirut ofthe level of r a m p t i o n per adult, that is, permment income. What we seek is to measwe the influence3f permanent income on fertility, but the true effect may well 32 clouded by such reverse &on. In her analysis of cumdative f d i t y in the 1985 wave of the Cdte d'Ivoire LSMS,Ainsurorth (1990) appliedthe ZIarrtmat (1978) exogeneitytest, a test based on 'heuse of instrumentalvariables, to mwthe exogeneityof consumptionper adult. She foundthat an assumptionof exogeneitycould notbe rejected atthe95percent confidence levd. Yet Benefo and Schdtz(1992: see&sir text and AppendixD), wing ths 1985-87 waves of these data, a different set of icstm.ne.ts and a slight redefinition of the depwdent k-ariable (amdative fertility:mng women having at least one child), found that exogeneity could be rejected at the same lev4 of confidence consi4ed by Aimworth. The diierences in find- may well be attributableto differences in the set of instrumena variables employed. In selectingthe instruments, the researcher oon.fiozrs a practical cii!emma: the ir~munentsmusr be exogenous, and yet at %e sametime, must be sufiicientlyhighly c o r . e l a with the (potentially) endogenousvariable in questionto produce reliable predicions of that -mable. All often, candidate variabls that perform well in respec. to predictive abiiity must be rejected from the set of instruments because !hey are at least argLtabIyendogenoustbemselvs. Furthermore, the ins-ent set =annGr be dominated by the exogenousvariables &a! ace dresdy represented in tbe structural equation. ?here be additbnal exogmcs Mtiablej adable to sen*=as imrumenis, aad these must make an impcmmt additiod costributionto the $diction of the endogenousvariable (i-e., nnsumption). Othemiseboth the i d e n t i f i eof the structsral-equation and t&e quality of thc structural equation esthates,may be - wnprombed by mdticollinearity. * 4 I We fzcethesegeneric difficultiesin regard to the C6kcl'IvoireLSMSd h . The cluster, somlme and s?us-prefecture dm iisted in Tzble 1 of the text would seem to constitutean unusually ric5 set of instruments. From these we select dummy variables for L5e survey year, :k male agricultural daily -wage, emus-based measures of she composition of the mnle labor force and thz full set of community-level price data available from the LSMS price surveys. We au,ment these with several exogenous variables airzzdy represented in the smctural schooling a d fe~ilityequation., namely, the woman's ag:, residence, and education. When taken together, these instruments are only moderately predictive of consumption per adult: the R2on th2 consunption regression is a modest 25. The R' can be considerably improved, to .35, by the additicn to the instrument set of cluster- level averages of consumption per adult. (The cluster averages are calculated omitting the consumption level of the household in question.) Unfortunately, these cluster consulnption averages are themselves suspect on the grounds of endogeneity. It is very likelythat unmeasured cluster-level characteristicswill appear inthz individualhousehoId consumption levels as well as in the cluster averages of consumption. If we are to maintain confidencehthe exogeneityof the instruments, therefore, it appearsthat an Ra of .25 is about the best that can be achieved. To be sure, Benefo a d ScZuln (1992: Appendix D) do much better in R2terms in their insmmentzl-bariables equation for consumption per adult. But ~e greater predictive power is produced, it seems, by the addition of variousmeasures of household composition, includingthe presence of a husband, his educationand age, and household assets. Tte exogeneity of these variables is a matter absut which reasonable rzsearchers may disagree; we regard them as being sidogenous. Appendix Table 1 sums up oar tests of exogeneiry based on the restricted set of instruments, a d a l s ~pro*rideS tats bzed on the inclusion of cluster consumpt!on averages in the instniment set. The first colurrn of the table presents the uncorrect least-squares or probit estimates of the effect of consumption per adult on fertility and child schooling. We then repon the instrumental--ariables-based(IV)estimate of this coL,sficientand provide the p-value of ?he Hausman exogeneitytat. We also prsent an additional informative criterion by which the simpie OLS and probit results c i a be assessed: do the uncorrected (and potentially bised) estimates lie in a 95% confidence band surrounding the (sserbut less~recisely-esthnat4)instn?mtatalvarihlesesthtes? ?he results are somewhat different fcr fczity a d schoclbg. Irt the case of fertility,,the uncorrected estimates sugges:that consumptionp a adult exerts a positive and significant infl~tnceon fertility overzll. Within rural and urbanweas, the effect varies instrm,d buteach estinztesuggestsapositive influence. When these coefficients are estimated by tile TV method, hcwever, they are uniformly ncgative and insignificant. In a way, and p=icularly for rural areas, chis is wbn one might have suspected if children indeed make a posi,ive contribution to family pernanent income. Cnce the re-rersecausation from f d i t y w incomeis purged via the use of instruments, permanent incomeitself evidently exem no important infldencz on fertility. w ?he picture is more cc?mp:cx than this e?xpJanationsuggests, hwever. k closer inspection of the instnments-based regressions (riot presented here) shows that the negativesign of the IV coeficiects isdue prhcipal!~to the fact that a worn's schooling is positively and strongly associated with the level of consumptjon per adult in her hor~sehold. In the stumra! equation for fertility, prdictd carsumption ievels are 65 therefore highly correla:ed w?h the wm's education, so that the regative direct influence of education on fertility is spread over the consumptioncoefficientsas well as the education coefficients. This is an aspect of the multicollinearityproblem mentioned above. Moreover, Appendix Table 1 shows that once separate fertility equatioa are specified for rural and urban areas, the uncorrected least-squares and probit coefficients generally lie within the IV confidence bands. The single exception occurs when the instruments includeclusteraverages of commgtion, but here the questioaof instruinent exogeneity puts the test in doubt. The lessoas wz extract from these rests, as regards fertility, are (1) that separate urban and mrd equations are required, as would be revonable in any case; and (2)that there is little apparent risk in proceeding with the uncorrected OLS and probit approaches. There is some possibility thp the effects of permanent income may be overstated by the uncorrected approaches, but we see little here to suggest that the N estimates are necessarily any closer to the tnrt'. The results regardingthe schooliig equations reinforce these wncIusions on the whole. T i e uncorrected OLS and probit estimatesgenerally lie within the IV confidence bands (a& exceptions are more likely in thz expanded instrument set) and in some contrast tn the case of fwtity, bere seems here to be no clear pattern of upward or downward bias in the uncorrected OLS or probit coefficients. If one places iaith in the instrument set with cluster consumption averages, the pamm is one of downward bias in the uncorrected coefficients. To sum up, tha,we see in these tests a sufficientjustification for proceediig I:* aut instrumental-yd.riab1es wrrections for the endogeneity of wrrsumption. The assumptionof exogeneityhas not been decisivelysustained, of come, and there remains some risk of bias, but we can find no compelling basis on which to prefer the instrumental-variables alternative. A?pendii Table 1: ExageneityTestson Consumption Per Adult (log), Witbar;dWitbout CiustrrAvemges ofCorlnunptian in the Set ofInstrumental Variable$' I References 1 Fern-iityand 0.ild Schooling in Cdte d'lvoire:Is ?here a W e o n iinsworth, M. 19W. "Socioeconomic Determinants of Fertility in CBte d'lvoire." Living Standards Measwe~entStudy Wrking Paper, No. 53. The World Bank, k h ' i n , D.C. i i m r t b , M. 1992. "Economic Aspects of Child Fostering in ate d'lwire.' Living Standards Measmment Study Hbrking Poper, No. 92. The Mrld Bank, Washington, D.C I dinsworth, M., and J. Munoz. 1986. "The ate d'Iwire Living Standards Survey: Design and Irnpleffientation." Living Stamhis Measmment S . Flbrking Paper, No. 26. The Mrld Bank, Washington D.C. 3ecker, G., and H. Lewis 1973. "On the Interaction Between the Quantity and Quality of Children.' Jownal qfhlitical Economy 81(2)M II:S279-S288. 3en&, K., and T. P.ScbuIk. 1992. "Fertilityand Child Mortality in COte d'lwire aad Ghana." Africa T=hnical Department, The Wrld Bank, Washington, D.C Draft Birdsall, N. 1988. "Economic Approaches to EbpulationGmwth," in H. Chenery and T. N. Sriniman (eds.), Ha.?dbook of DewIopent Economics. Elsevier: North-Holland. Cddwell, J. 1982. "MassEducation as a Determinant of Fertility Decline." Chapter 10 in 3. Cddwell, Theory of RmmIityDecline. London: Academic Ptess. Caidwell, 3. md P. Caldwell. 1987. "The Cultural Context of High Fertility in Sub-Saharan Africa." hpulanon and Dcllelopment Ra3m 13(3):409-437. DeLancey, V. 1990. "SocioeconomicConsequencesof High Fertility for the Family." In G. Acsadi, G. Joh~sn-Acsadi,and R. Bulatao (eds.). Popadation G m h and ReprvduCncnonin SubSahamnAfi,!ca. World Bank,Washington, D.C DjMj6, 0. 1991. Moditt?. S6minairede prbentzltion des rbultats du nxensement de la populatic!~et de I'habitat-1988, Abidjan. Fapohunda, E., and M. 'IWam. 1988. "Family S m t w Implicit Contracls, and the Demand b r Childien in SoilthernNigeria." PopulationandM p m e n t Mew 14(4):571-594. GePrler, P. and P. Glzwwe. 1989. "The Willingness to Pay for Education in Developing Countries: Evidtnce from h r a l Peru." Livirsg Stmdids Measmment Study Wrking Faper, No. 54. Wington, DC:me Nbrld Bank. * Glewwe, P. 1988. "Primary and Secondary School Enrpllment in COte d'lvoire."Draft. Glewwe, P. 1991. 'Schooling, Skiils, and the Rehrm to Government Iwestment in Education: An Exploration Using Data from Ghana.' W ~ nStandrvdr Measrrrement Study Wrkhg Rper, No. 76. g World Bank, Washington D.C. Glewwe, P. and H. Jacoby. 1992. "Student .4chievement and Schmling Choice in Low Income Countries: Evidence from Ghaca." Liking Stariahis Measurement S:udy Wrking Paper, No. 91. The World Bank, Washington, D.C. Gomes, M. 1984. "Fzmily Size and Educational Attaiment in Kenya." A?pdatio~and Dmlopmettt Reviov 10{4):647-660. Grootaert, C. arhdR. Kanbur. 1990. "Analyseo p h t i o ~ e l l ede la pauvr&6 et des dimensions sociales de i'ajustement structurel." Dimensions socides de l'ajusremnt structure1Document de t m i l , No r, Analyse socio4mnomique. \Vashing?on D,C: The Mrld Bank. Handloff, R. (ed.). 1991. CBred'fvoinzA &untry Study. Washington D.C. :Federal Research Division, Library of Congress. Hanushek, E. 1932. "TineTrade-Off Between Child Quantity and Quality." Journal of Political Economy 100(1):84-117. Hausman, J. 1978. "SpecificationTests in Econometrics." Ecommemmca46(6): 1251-71. Heckman, J. 1979. "Sample SelectionBias as t SpecificatiotlError."Ecommetica 47(1): 15341. Kelley, A. and C. Nobbe. 1990. "Kenya at the Demographic Tbming Poi~t?"Wrld Bank Discussion W e r , No. 107. World Bank, Washington, D.C. Knodel, J., A. Chamratrithirong and N. Debadya. 1987. Pa~iland's&producrbv Rewlution: Rapid Fertility Declirie in a aid-Wrld Setting. Madison, Wisconsin: University of Wsconsin Press. Knodel, J., N. Havanon and \V. Sittitra. 1990. "Family S i z and the Education of Children in the Context of Rapid Fertility Decline." Population andDmlopnent Review, 16(1). Kouam6, A. 1987. "De la penurie la sousutilisation de la main-#oeuvre: un essai sur la pmblematique des ressources humaines en CCBte d'Ivoire." Lesthaeghe, R. 1989. "Social Organization, Economic Crises, and the Future of Fertility Control ill Africa." in R. Lesthaeghe (ed.) Reproduction and Sod& Organization in Sub-S&mn AfPica. Berkeley: University of Califkmia Press. . I Lloyd, C.. and S. Ivmov. 11988. 'The Effects of ~ ~ ~ mChild jSurvival on Family Planning Practice v r =d Fertilitya Studies in FmVy Planning 19(3): 141-161. Maddzla, G.S. 1983. Limircd-Dependent and Qditatiw Wiables in Eco~uune~cs. Cambridge: Cambridge University Press. Makinwa-Adebusoye, P. 1931. "C'nanges in the Ccss and Benefits of Children to their hents: The Changing Cost d Educiiting Children." Paper prcentd to %e Seminar on "The Cou* of Fertility Transition in Sub-Saharan Africa," sponsored by the IUSSP Committee on Comparative Analysis of Fertility and the University of Zimbabwe. Harare, Zimbabwe, Nmmber 19-22. McKay. A. 1991. "Estimation of a Regional Cost of Living Index for Cbte d'lvoire 1985-1988.' Mr!d Bank, Africa Technical Department, Rwerty and Social Pblicy Division, Washi?gton, D.C. Oliver,R. 1992. "Family Size a?~dChild Schooling in Ghana: Is There a Tradeoff?' Africa Technical Department, The World Bank, Washington, D.C Draft. Pitt, M., and M. Rosenzweig. 1990."The Selectivity of Fertility and thp,Determinants of Humvl Capital Investments: Parametric and Semi-parametric Estimates." Living Standan& Measmmm Study WrkingAtper, No. 72.World Bank, Washington, D.C. Republique de Cbte d'lvoirz. Direction de la statistique et de la comptabilitt!nationale. i988 Memento chifib tie la CBre d'lvoire 1986-1987.Ministm du Plan, Abidjan. RCpublIque de Cdte d'lvoiie. 1991. DbcIamhamtlonpolin'que de ~orisaziondes resumes ?wnaks. & Chapter 11,Primary and Secondary school enrollment in COte d'lvoire. Abidjan. Russell, S. and W. Stanley. 1988. "Huzan Resources Discussion Paper: RQublique de Cdte d'Ivoire." World Bank, Population and Human Resources Department, Washington, D.C. Scbafgans, M. 1991 "Fzrtility Daerminants in Peru: A Quantity-Quality Analysis* in "Women's Work, Eduuition, and Family V?elfire in Peru," Wrtd Gank Discwsion Paper, No. 116.The World Bank, Washington, D.C. UNESCQ 1991. UNESCO Srdstical Ye~~book, 1991. Vau de Walle, E. and A. Foster. 1W. "Fertility Decline in Africa: Assessment and Prospects." Wrld Bwrk Ethnical W r , No. 125,Africa Techcical Department Series. Washington, D.C. Mrld Bmk. 1988. Zducation in S h S h Afiia: Iblicies jot Adjwtme~~, ~ ~and n Expansion. World Bank, Washington, D.C. World Bank. 1990.AfricanEconomic and =9ricialEata (On diskpa) Washington, D.C. World Bank. 199Clb. "Cdte d'Iwire Human Rescwces Strat=: Process Note." Occidental and Central .Africa Department, Population and H.~rnanResources Division. World Bank. 1991. WrldCewIopmentIFdiicak3rs (On diskette.) Washington, D.C. Paper Number 2 Fertility and Child Schooling in Ghana: Evidence of a QualitylQuantity Tradeoff Raylyan Oliver Contents Abstract ...............................................73 InWuction ............................................ 75 Economic Model of Fertility and Child Schooling Decisions.............. 76 . Fertility and Schooling in Ghana ............................... 79 Estimationand Resttlb ..................................... 83 Baseline Results .....................................87 BaselineResulfs br Urban and Rural Samples ..................89 Estimates Including the Price of Schooling.....................92 Co~lclusbn ............................................ 95 Abstract Raising the quality of children by i n c ~ e i n gschool enrollment and loweringhigh fe~ilityare frequently the goals of public policy in Sub-Saharan Africa. Studies in other parts of the uorld hzve found that at a certain stage of the demographic transition, parents begin to reduce the nunber of ch:Jdren they have in order to increase the investmentsin each child. hrhsps the most obvious invesmei~tin child quality is school enrollment. This paper explores the relation between child schooling and fertility in Ghana, using survey data that includehousebold demographicand economic variab!es as well a mmmrnity-ie\veI data on the access to schooling, to ascertain whether this "tradaffbetween krtility and child schooling isunder way and what policiis will most - likely encou-Oe it. The most striking result is the large impact of mother's schooling L? lowering fertility and raising child schwli4g, and the large predicted impact of female secondary schooling. in partimlar, in rural areas. Incrwsesin houszhold income are also associated with lower fertility and higher child schooling. The results indicate that increases in the fees charged at local schools will raise fixtilily, but by a \wy small amount. However, an increase in fees is also associatedwith higherchildschooling. Thepolicy implications of these rmults are, first, that female schmling can be a ptent instniment b r lowering fertility and raising child school enrollmats, and second, that increasing the cost sf schoolingwuld m t have an important irnpad on fertility. Acknowledgments This paper was prepared fkr be research project on "TheEconomic and Policy Determinants of Fertility in Sub-Saharan Africa," managed by the FWerty and Human Resources Divisicn, Policy ResearchDepartment VRDPH) and sponsored by the Africa Technical Department. The opinionsexpressed in this paper are those of the author and do not necessarily reflect the policy of the World Bank or its members. The author gratefully acknowledges guidance and support from Martha Ainsworth, Julie Anders~n Scbaffner and Mark Montgomery. Helpful commentsfromJohn Pencavel, AnjiniKochar and seminar participants at the World Bank are also appreciated. Introduction Botl schooling and fertility have-an important impact on a couiluy's economic growth and development. Thcs, both have been the object of government acd donor programs, especialiq in dweluping countries. Increases in school adability and enrollment emich a country's human capital. Reductions in M i i t y reduce chiid and maternal mortality risk. In addition to their impdct on economic development, fertility and schooling have an effect on each other. Because high population growth rates hitrder government effor!s to increase the quality and accessibilityof educationalfacilities, family planning programs are apec:ed to complement direct atternyts to increase schooling by slowing the rate of population growth. On the other hand, increases in schooling also complement b i l y plaming program efforts to reduce fertility. Empirical studies show that wmen who have some schooling have fewer children and are more likely to use co~traception.~This complementarity, however, is realized ody after a lag of several years whei! the mrc- schooled women enter their c5ildbear-hgyears. A second, more immediate, complementarity is sugges+dby economic models of fertility decision-making. What has come to be called die quality-quantity tradeoff occurs when a woman decides to produce fewer chzdreu and make a hSgfrer per child investment in hose children. If women do make such a tradeoff, then redu,2icns in the price of schooling, one type of qudity investment, may iaduce a substitiitionaway from quantity and to quality. This idea was presented in Seclrer (1960) to =plain the fertility declines observed in developed countries It may have a more important effect in developingcountrieswhere lack of schooling facilitits o f b constrain the tradeoff. This paper explores tie extent of the complementarity between child szhooling and fertility in Ghana using survey data that include housebold demographic and economic variables as well as community lwel data on the accm- to schooling. Ghana represents an important case because schooling levels there have been relatively high among African couhtriessinceindependenceyet, likemostSub-Saharan Africawuntries, fertilityrates remain high and contraceptive usage rates remain low. The methodology employed i5llm that used by Montgomery and Kouamt? (1993) to explorethe existenceof a quality-quantitytradzoff using similar data from the neighboring country ,r ate dlwire. n e y find a positiv~relationship bmveea enrollment rates and cumulative child schmling and M i i t y ia rural-areas d ate d'Iwikx that contrasts with a negative relation in urban ateas. In spite of the geographical proximity of the tm apuntries, they differ sharply in owall levels of female schooiing and fertility. The data from Ghana alsc 5clude price of schooling variables that are not Yrdilablein the Cdte d'Ivoire data. Thus the tm studies should -- provide an illurninatkg picture of the possible existenceof a tradeofi. SectionTI presents an economic d e l d fixtilityand child schooling decis@us. Section III describes fertilit~whooling IeveIs and the data from Ghana. Section PV discusses the sirnation and results. SectionV contrasts the results to those from h e d'Iwire dother countries, discusses the policy implications and mxludes. f 1. Ainswortb (1990). Oliw ( I N ) and o h m document this r e l . t i d p usiqO&fmm Sub- !hhmm Africa. Economic Model of Fertility and Chiid Schooliug Decisions Microeconomicmodds based on the early uork of Becker (196"' urd Leibmsteh (1557) postul332 a household choosing the nuder of childm to pro .e In order to nrximbe utility subject to its production functioris rind budget md tine constraints. Children are demanded fir their contributionto household incomeand i%rh e utility they provide to their parents. ?l.ese benefits are weighed againstthe costs, in time and g-wds, of raising children. The number of children a household demands is function of the ;I wealth and income of the hous&~ldam!the prim faced by the hous&old. Housrhclds may also choose to ;mest more or less in bod, clothhg, kousing, schooling and health care for a child. The qualiiyquantity models i n d u c e quality as an additiond agumen? in the household's decision, ~ckr~onrlxlgingthat the ccist of children can vary substan4allyaccordingtothe W t m e n t the household choosestomake in the "qualItymof thz child. Ignoring the quality aspect of a commodity can lead to biased estimates of the Jemad fur quantity of that commodity and this is expected to hold true fbr childre3 as it does with other argrments of the household's utility function.' The intenst here is in the demand for children. If children wela normal goocis and if hey requirei. a set Ievd of investment, then rich families would demand more children than poor t - d i e . That this relatiorrship is nr? univenally o b s e d has been expla;3ed by the existenceof a quality-quantitytradwfFwhzrein wealthier fanilia spend more c.n children not by having more children, but by increasing the Iwe: d per child expenditure and often aenhaving h e r children. The woman maximizes her ut:1ity, U, that Is a functio~~of the number of children, C, the quality of children, Q:consumj~tionof market goods, X, and leisure, I,, and hsr tastes, p (ewaticn I). Children require inputs cf the woman's time, T=, and mrsket goods, &. The z!bcation of the wmzn's time to market activities, TM,to child rearing and to Ieisure mzy cot exceed b x time endowment, Sl (equation 2). UtiZty is maximized subjec?tothe time dloatwn consmint and a full incomebudget rorrstraint (equaions 2 and 3). T!ie price of consumer gads, p,, is multiplied by 8 sum of market goods br thz household and those rquirtxi fbr the production of childnn, and t!e price ~f child qcality inpis, pc, is multiplied by the quality b r a& child and &e numbcr of chi!d=n. The sum of expenditures must qud the swn of incomeeamed ftom the sale of t h e in the market, w *TMand exogenous incame, Y. The azmand fir d~ildrenand the demand fbr quality can then be cxprssed as functions of the exogemus vxiabls: prices, p, and p,, wage, w, and exogenous household income, Y- B 2. Tbcil(1952)discusszstbe theorelid biases i n t m d d when qdity isnot taka t b aemnt. Both equations will be estimated ia t h empirid cection below. This reducc.1 ~ %ism ap~mzchr5flects the %ct that both varfables,the number of children and the quality .sC children, arejointly chosen and are a function of the same exop&nousvariables. The .ride: does ;lot yostulaz a causal relatioxhip between quantity a ~ quality-households d d:, not choose feaer ch'ldren b m s e tiey spend more on schwiing nor do they choose more schooling because :+ey have h e r childrtl. Ratl.er k t h are chosen to maxi-nue utri2y as constrained b. income in face of txisting pikes. Comibcr the expcctd impact of the exog?cc,~~s variables on the dctmd tbr numbers and qgality of children. An increase in the pxice of other goods will raise the demand fir children and quality as urornzn ellbstituteauay from purchased consumption goods. Houwr, it will also i ~ e!asincome to Se speut on all inputs to the utilit'j : fuqction. Cernaqd fbr chjld12a will also be d i i d j reduced by the increase in the cost of pmducirtg ciildren. The presumption is ~rsuallymade that the suktitutio~effect is smdler Cr, 4the combined negatiw inr~meedect wxl owil-price e k t and tha: the det dffecto, : - .-.-sing price cf other goodson the denvnd ibr children and quality will &a be negat~r.. 'The price of q u d ; ~is expected t~ reduce demand for quality; the income and substicution effects wrk in the same direction. The eifect of p, rm the demand br children is :rrdetefi3kact. The income effect will be negative, as will tho m-prke effect, but there will also be .u!offsettiqg positiw substitution2ffect. The effect of a woraasis wgi~on demand iLr childi.en ;a theoretidly indeterrniaaat. It car- also be decompose3 into an incone and a substitation effect. I?! increxse ia th3woman's Rage increasesthe costtrfproducing :hi!:',ren. The 'crwn~rice" siibstitrrti~neflec:nn the demand b r cbildren is uaambig~ouslynegiiw. Hoverer, an increasi: in maen's wages :au dso raise her i;,coms. The effcct of the increase m iriwrne on the de=d b e children will depend on wklher or uot :!i;-ldm ate normal goods. Ths net effect is uually fiunl.to be n@,*., m e presumed effect of vl>nan's wage ol:quatlrj, ~n ?he,orherhard, iq pasithe; botir ib? k::;;;: -ad substitutionefu'ects are e x p d l-.be ~siiiw. Holding xomen's wiig* a d pripxs corntaut, an inmse iF exogcwi inconre is a*d toraisetho,denaud fcrchildrwraid quality. Brn~iricalwrork, bmwsr, often sh& that increases !n income lw.ues d e d %r children. 'ibs ex~ec.Weffects cif the exqgem>usvariablesoa ihdemand b r children and qcdit-yare s u m in the eigri - - ab@e equations 6 and 7. - 8 Montgomery and 'huarnt (IF,! suggest that if b e estimated effect @fthe individual exogenous mriables on the demand x children is opposite to that of its estimated effect on the demand fur quality, then fiat could be taken 2s evidence of a qndityqaantity traJcdF. This reflects tbz Becker view of quality and quantrty as substitutes. In this scenuio, fertilitymay be higher for families who face a higher price ofschooling. HG-mer, as ths discussion of equations S and 7 shws, only the signs of wmen's wage ue cxpected to be opposite, For example, if child schooling and number of cbildren are complments, and schooliag is considered to be an essential input into childming, then an increase in the price of a h o o l i ~might decrease the demand b r cbildtea so that schoolkg can be purchased fbr each child. This view of &c importance df the quality/quaritity tradeoff in red-icing firtility is supprte3 by the wid-ce from Kenya. Kelley and Nobbe (1990) &A that increases in the cost of schooling were partially responsible b r the fertilityieductions in there. Several aspectsof h~useholddecision-n&i in Sub-Sahardn Africa may weaken the r-xdictive pow= of the neo-classical model presented above. In her analysis of krtility in Kenya, Gomes (1984) finds the zrqmption of equal per child arpenditute to be i d i d . She rep.$ that it is common practice f~r eldest childrexi to be fmred with resource rn attain an education; punger children wili have access tc W r resources until tfe ddest is able to cmtribute. In addition to the differences across birth orders, &erd are substantial differences in the treatment of boys and girls. Primary school enrollmentratios fbr girls average 15percent below those for boys in Sub4zhan.n Africa CAbrld Baak, 1B3a). The types of invesmcnts availa5le to the majority of hcuseholds in Sub-Saharan Afica also differ from thase in dwdoped countries N'here the level oftotal expenditure is tow, a significant portion of t!!e expendibre on children will be spent to achieve a suhsistencelevel of ! h i q Pi,r the chiidren that is closely tied to that of Be parents. Finally, the economic model discussed above doeslnot allw for the possibility of achieving the demanded quantity or quality of children except by &.ering fertility behavior and expenditure. Ir. ,Cub-Sahm Africa child hstering is very wmmcn (Ainsworth, 1992). Chi!dren arc ofier-cent tolive with relatives to contributetheir labor to eha relative's h~usehold,at gain iccess to schools and other opportunities, and, wnetimes, to reduce the uxt to the parents of misitig the childrzn. l3is efficient and ubiquitous cost- and benefit-sharing qchanism will dso attenuatethe predictions of the ~ d dbut the signsshou:d not be -4. , 1 Fertility and Wwling in Ghana In order to nrotivate the empirical spsificatioas and interpret the results presented in Section IV, it is useful to review the motext within which fertility and schooling decisions are being made. This section o~itltlinesthe current levels of fertility and describes the history of popuIation a d education polirj in Ghana. G h m was one of -he first SubSahuan Africa countriss to adopt a comprehensiwplicy designed toreducepopulationgm& rates(Ghana, 1969). As part of the populationpolicy, a National Family Planning Programme was launched in 1370. It became the focus of the country's efforts b reduce population growth rates. In 1989, Owusu et.al., found 'no evidenceuf any really seriousoppositionto family planning on political or religious grounds withi? the mtry'. The adop~onof such a policy in Ghana contrasts sharply with the strongly pro-natalist official policies pre~iliqguntil only recently in most African countries As a result, availability and haowledge of modem conmeptives are r2.i-ively high. Contraceptivesuewidely mailable in pharmacies and health facilitiesthrwgbout Ghana. Family p l d n g serviccs are o I I d at the nearest health facility to 55 percent of the population and at the nt*anst pharmacy bor 81p e m t of the households (Ghana Living Standards Survey, 1s 38-89). Pills can be purchased witkout a doctor's prescription. Thequa- of women live within 5 miles d a source of d m contraceptives and 82 percent of the women kncnv of at leasf one modem method of contraception (Oliver, 1994). Abortionsare legal br abroad range of medical, juridical and socioeconomic reasons. 'Ihe only restrictions are that the approval of the husbazd is required and that the abortion must be pe&~nnedby a physicizn (Saibner, 1W). In spite of this, M y size and total feitility in Ghana remdn high aad contraceptiveuserats re& lm W 1 e 1presents recent G h d a n tad W i t y rates [WR) calculated by current residence aad wman*sschooling. There is sisbsfantial mistion across groups. Wbmen with somesecondary schooling in urban areas have a Ti% of 3.3 and thosewith no scho~ling51rival areasaverage7.4. TDe averageacmss all Ghanaian uomea is 6.3 iir higherthan tte averages of2.7, 3.1 and 4.2 in East Ash, I atin Amerh ;sad South Asia, feqiECcJvely (Wrld Bank, 1993a). Twenty4mn percent of the w m who h a cohabited have used a modern method at leastonce, 33pwat areeumdy usingsomekindofcontraception, but only 6 percent arecurrentlyusing a d e r n method ofm&ac@on (Oliver, 1994). This is fu below the oontraceptive prwalence rates in Westan ccuntcies and Asian co~ntries, 70percent and 40-oercentrespectkey, but higher than those prevaila% elsesv5erc in - Sub-Saharan Africa. Except for Kenya, Botnwanrand Zimbabwe, use of a~ metbodof ; Eunily p!- is below IQ percent50f the SubSabanaAfrican counaizs . 6 -5 - Ghana bedhe indm- in 1957. After r decade of sommie prosperityMd 5 anemphasis on eduktion, Ghambecamea leaderamong SubSaharanAfrican wuntriss GhaM A-i 0 1 h c r U r h Rural KOEducation 7.02 5.42 6.14 7.36 1-10 years 5.88 4.23 5.05 6.65 Over 10 ytaus 3.24 3.28 3.70 3.05 Note: Ibt.lFc&y R.bu the numberdchildrena wmmwcld have if sl~urcrc to survive allof her e child beuingyeam and beuchildrn at cub rgc at the prm'Xq agwpccifio m y rate. Sample TFR wemalcuhtcd basedonthe numhcr of dildccaborn in lbeLst fiw yeam to wmcn divided into five year cobo*. Sourre: GhanaLiving S t a d d s Suncy,1987-88 and 19SSS9. in boa enrollment ad quality of education by the late 1960s' An wnornic downturn and a dramatic decrease in portion of ONP spent on ducation caused a significant reversal. By the early 1980% W s fbr textbooks were anavailable, the quality of teachershad declined, Pnd a largemajorityof studentscompletingprimary schoolin rural areas were completely illikr~te. Since the low point in 1984, quality and e~rollmeilt bave begun to impma In the Ghanaian school sytem, aAer six years of primary schooling, students continue to middle school (grade3 7 through 10) or pass an examination that qualifies them to commence secondary school. P r i i and middle schools ixe compulsory, though enrollment is well belw 100p e n t of school age children. Secondary school =mists d five yea's of study, a f k which CGE C level exams are tzken, and two additional m j e a r s ,aflerwhich CGEA level exams are taken. Students who pass A level exams u e qualified D enter cae of Ghana's three universities. There are also commercial, mxtional ad teachertraining schools. School enrolln;~; ratios in Ghana areamcry the highest in Africa, though they are low wmparcd to other wuntries at similarlevels of GNP. 'Ibtalgross primary and secondary enrollmentratesare73percent id 39percent in 1988comparedto67percent and 18percent br all Sub-SaharanAfrica.' This is after the drop in prima$enrollnent rats from 3 high of M3 percent in 1980that fbllcwed the Ghanaian economic crisis and decrme in percent of GNP devoted to education @asel, 1992). Primmj schools are located in most towns throughout the country; 9G percent of the hous&olds are within 032 mile of a primary s&ool, Mugh the quality of the facilities, sgpplies and teaching - .I 3. G l m (1994) desxik the Gbtnrinne d u & d a m and the qunlity of inputs in his akasive analysis of tbowmomia of scbaol qoality inve&ments 4. Grossprimaryuvollment ratiosarctbedieofpupilsto &% popu!atim ofsc£iml-iige. children. l%egrossearollmt ratiosmy acted 100p m t incountrieswiUlwiwrsal e~roilmentbecause some pupils rut younger or older i&a country's s t d a d p r i m school age. Table 2: Complered Sc- of Mmen in Ghanu -- - n Mean Yuas cfschaohg Pimnt withNo Sdwoling Age 15-19 788 5.2 3296 Age 20-24 932 5.6 35 Age 25-29 931 5.3 38 Age 30-34 777 5.3 38 Age 35-39 521 4.3 51 Age 40-44 369 3.0 65 Age 45-50 401 1.7 77 All Women 4719 4.7 4396 Source; Ghana Living Standards Sumey, 1987-88 d 1988-89. varies greatly. Seventy percent of the househdds are within 2 miles of a middle school but only 25 percent are that close to a s e c o d q s&ool. Tables 2 and 3 describe the levels dsdmoling reported for women and children in the Ghana Living Standards Survey. 'ibt ~oean years cf completed schooling fbr wumen age 15to 50 is 4.7 years. The averageis highest h r women age 20 to 24 at 5.6 y m . The percent of women with no s c h d q decreases from 77 perceat h r wmen Gver age 44 to 32 percent for women agz ISo 19. For children age 4 to 30 the mean pas of schooling completed is 3.4 years. Among those who have had some schooliq br;t are no loqec in school the mean is ?.O years. The gercent currently eanlled is higbat tor children age 19 and 11,76percent, anA lowest mpr;gchildren age 20 to 30, 11 percent. Tbe percent never enrdkd is highest fbr children age 4to 7. Many do not enter the schools in Ghana until age 7or 8. The Ghanaiaa school system diffes f h m that of CBte d'lwh in several respscts. Average years of srkooiioq and percemage currently enro!:ed are bth higher in Ghma. Unlike the situation in Cdte d'ImU)2g there is very little tepcating of grades in Ghana. W~ththe exception of the exan t h t is rquired fbr entry into secondary school, students phceed throilgh-the@es regardless of academic prOgieSS and gnde completed ri-Bed not reffed skill aapisitios. - ~ h e b a t aused in this paper arefmm taoylrsofthe Ghana Living Standards Survey (GLSS). Tbe G U S wi carried o;alqthe Gbma Statistid Service in 1987-88 a 1988-1989with assistance ficm the Wodd Ed. The Ghana Living I Standards &as urement Study data, sample d d o n and strategy anrl d i colIection are . described i n World Bank (1993b). The smq gathered utensive i n f o d o n on household members' health, education, m e n i , and income. Data were also collected on migration, selfem;,lq;nent, arnr;Jsr?pticn znd farming. One wman was selected at random from among m e women mthe househoid aged 15-50to respoild to Am 4 7 Age 8-9 A p 10-11 Age 12-13 A p 14-15 Ags 16-17 Age 18-19 Age 20.30 All Child- . the fettilitysection of tbe questiomak Birth hiswry, contraceptionknowledge and use and the circumstancesof the woman's last birth were cavered. Over 3,000 households from 2 N randomly selected clusters represent@ all pacts of Ghana were suneyed in each year. The 198849survey wls supplemented by a school survey that included the primary and middleschools in haif of the clusters. Inbrmation ms gathsred ori health facilities, school facilities, teacher qualification and, especially useful for this survey, annual and other sciiool fees. Estimation and Results The mrthodolw that wil! be empIqd is described in detail in Moutgomery and buame (1993). Fertility and child schooling decisions are presumed to be made simultaxmsly, subject to the same set of exogenous variables. The fbllowing set of reduced-form expations will be estimated: Si = XD1, + ZlI2, + Wi33s+ Vfi3F+ )t, h r all i children in the household F is the masure of fertilicjh r an individualuloman, S, is the schooling of the ith child, X, Z and V are characteristics of the wman, her household, and the community, respectively, and W, are characteristicsspecifictothe individual child. The first equation is estimated fbr one woman agz 15to 50from each household. The complete dab set, including economic, employment and expendituredata in addition to household composition, krtility ai;d community information is available fbr 4,625 momen.$ m e S equati~nsare estimated fir all of the uomen's survivirg cbildren 5 to 30 yearaof age at the time of the Tuo measures of krtility will be estimated. The &zt, Aildren ever born, measures cumulativefertility, the number dchildren the wloman has given birth to in her lifetime, irrcluding those that died shortly a% birth. Beaus- women of all ages are included, children ever born does not measurecompleted W i t y . Age must be included to control fir how far from completed k d i t y the wmen are. And because age must be includd h r that remn, birth cohort e f f m cannot.be measured separately. Children ever born is a discrete non-negative variable. Ordinary Least Squares (OLS)'estimates are not efficient but the resulis are easily interpretable and do wt differ substa~tially from the predictions of an ordered probit model. The second measure of Mility measures recent W i t y arid assumes a value of one if the woman has given birth in the last five years, tegvdless of whether oi not the 5. Of tbe 6,328hou&oids surveyed kthe l987-88and 1988-89muorbof the GLSS, 25 pcrceat (1,582)did not containany wmen between tteages of 15 and 50. In addidon, 121 b o d l d s were excluded due to irreconcilablediscrepancies rad missing data. e 6. Of the 4,625mmen, 964have naw givenbirth. Thesurveynotes birthsof 14,774childm. nr Of these, 12,279 living and 9,517were age4 b30 at tbe tim of tbe suntey. 1,473childm could not be linked to the schooling data and 164were e x c k i due bother missing data. Tbe -- sample includes the wmen's children who are Evbg away from the ho&cld and excludes - childreniivbg in thehouseboldw& havecr differentmother. Thered* -le kIudes7,880 d children. *2 - 7. There are ao schoolin~equati- lor wmm who have Lud no cbildm. Tbo po'2ntial 0 schooling h r cbildren of w o r n who curreotly bPve no children is not taken bbaccount. Pi# and Rmwweig (1991) discussthe selectionp d m nrrd poteotial selectivitybiasassociatedwith these tva equations Thg find midace ef selectiviiybut littleindicationd selectivitybhs No correction tbr se!ectivlty has been d e here. child survived. Again, age must be included to ar~trolfor changes in the birth rate of childbearing throughout the life cycle. It is however, a rneasurz of fertility decisions made during a specific historical period fbr all ;%men. Probit estimates of the binary variable will be compared to OLS estimates of ihr. number of children ever born to the woman. Similarly, two measures of child scl~volingare estimated: the years of completed scnooliagof each individual child and a binaryvariable for whether the child is currently enrolled in school. Children 4 to 30 years old are included in the sample. As with cumulative fenilitv, ultimate schooling levels are not ho-kn for younger children. Age of the child nust be inciuded to control for incoaplete schooling. Cumulativeschooling will be estimated using OLS and compared to the probit estimates of current school enrollment. Wiabie definitions are provided in Table 4 for the sample of women and Table 5 fbr the sample d children. The household characteristics include the woman's schoolhg, household consumption expenditure, language of the household head (a proxy fir ethnicity) and the woman's age. No data on marital status or spouses are inciuded because marriage decisions are considered to bejointly endogenous to fertilitydecisions. Location of current residence is divided into five areas: AccraKumasi, Other Urban, Rural Coastal, Rural Forest and Rural Savannah. There are inporhnt differences in the cost of living and access to social services across areas. Later specifications introduce the availability and cost of schooiing in the community. Child specific characteristics include the age and sex of the child. Household consumption expenditure serves as a proxy for permanent incomz. It includes all expenditure, an imputed d u e of rent, the use vdae of durable goods and the value of consumption of household production.' The log of total household expenditure (in cedis) divided by the number of adults (ages 16 and ova) in the household, and the square of per adult expenditure are iccluded as explanatory variables? A purely exogenous measure of permanent income is not adable. Because of the eodopneity of the consumption, household expenditure and i~ square ate escirJatedusing years of completed schmlirg and occupational categcr- of the household h@A. The estimates presented below use the predicted per add: consumption expendimre. . * 3. Tbe housebold expenditurn varinble includes expenditu.~oo schooBng. lKis in~roducsa ptentid cornlation. Hcrwlcvtr, schooling expdihmrr average only 6.2 percent of total buuddd expnditura Further, becruse expenditure is iuolud+to control fbr the level of reuwrrces svriiableb tl.hauseholds. it would be incorrect to exclude apenditure on schooling. ~ 9. Ih,GLSSdefinesahousebaldto kiude 'all thepeopiewho nornr;rllyliveandeattheir mwls togetber ia the dudling b r at least t&w of thz last helve months'. lhble 4. tbn'ableM e m and Slzmiiud Dmmm-om: Samlile of Mbrnen All Mbnun Rmd M e n Urbanh n n = 4,625 n = 3.054 n = 1.571 Mean 3 d . D ~ . Mean Sd.Dtv. Mean -Dm. - -- - - -dent variables CHILDREN EVER BORN 3.203 BIRM, LAST 5 YEARS 0.621 Wman's SdlooIing NO SCHOOLWG ' 0.433 rn4ARY 0.160 MIDD1.B 0.357 SECONaARY 0.050 Ln@XPEN DITURE)" 11.599 Ln@XP)SQUARED 135.099 YEAR1 0.508 YEAR2 0.492 C u m t Residesce ACCRNKUMMI 0.101 OTHER URBAN 0.2'8 RURAL C W 0.151 RURAL FOREFT 0.321 RURAL SAVANNAH 0.189 Languaged housebold head AKAN 0.486 EWE 0.159 GA 0.075 DACBANl 0.033 HAUSA 0.021 NZEAJA 0.011 OniER 0.215 %man's Age AGEIS-19 0.166 AG-24 0.198 AGE2.5-29 0.197 AGE)O-34 0.165 AGE35-39 0.114 AGE40-44 0.078 AGE45-50 0.084 - --- a. The variabk used in thceuimationu thc log of p a dclt h o d l d expendim Themnnkvcl oftotal hourchold utpcndilulc k 313,600 dii.adpmximdy $1,363. T&e 5. W a M eMeav and Standad Daiatiom: Sm@e of Child~n All W e n Rvml GiHren Urban Chikiren n=3SSO n = 5,:W n = 2,496 biable Mean Sd.Dev. k8ean :.'td.Dev. Mean Sld.Dcv. 7 Drprndcnt w l e s GRADES COMPLETED 3.373 3.958 3.124 3.8% 3.910 4.035 ENROLLEDIN SCHOOL 0.532 0.499 0.483 0.50,' 0.638 0.481 Mbrlicrk Schooling NO SCHOOLLNG 0.569 0.495 0.627 0.484 0.443 0.497 PRIMARY 0.142 0.349 0.140 0.347 0.146 0.353 MIDDLE 0.261 0.439 0.218 0.413 0.353 0.478 SECONMRY 0.029 0.167 0.015 0.121 0.058 0.234 Ln@XPENDITURW 11.606 0.264 11.569 0.263 11.686 0.249 LD(EXP) SQUARED 135 269 6.034 134.379 5.971 137.189 5.716 YEAR1 0.498 0.500 0.482 0.500 0.534 0.499 YEAR2 0.502 0.500 0.518 0.5W 0.466 0.499 C m n t Residence ACCRAIKUMXSI 0,081 0.274 .- 0.257 6.437 'XliER 'JRBAN 0.235 0.424 0.743 0.437 RURAL COASI' 0.147 0.355 0.216 0.411 .. RURALFOREST 0.349 o . m 0.511 0.500 ... . .. RURAL SAVANNAH 0.187 0.390 0.273 0.446 .. .. LmpDvqgc tfhoy~ehokikd AKAN 0.483 0.5M) 0.508 0.530 0.428 0.495 EWE 0.160 0.367 . 0.152 0.359 0.178 0.383 GA 0.063 0.244 0.039 0.194 0.1s 0.320 DAGWI 0.030 0.170 0.030 0.171 0.W8 0.i66 HAUSA 0.033 0.170 0.008 0.088 0 . m 0.267 . NZEMA 0.011 0.103 0.011 0.106 0.010 0.098 a E R 0.224 0.417 0.252 0.434 0.163 0.369 C%ikikAge a d S a FEh4m-E 0.479 0.500 0.473 0.499 0.492 0.500 MAiE 0.521 0.500 0.527 0.499 0.508 0.500 MiE5-7 0.247 0.431 0.249 0.433 0.'242 0.428 AGE 8-9 0.148 0.355 0.149 0.356 3.147 0.355 ~\ce10-11 0.114 0.318 0.111 0.314 0.121 0.326 AGE 12-!3 0.112 0.317 0.112 0.316 0.112 0.316 AGE 14-15 0.089 0.285 0.086 0.~330 0.097 0.297 M E16-17 0 . m 0.261 0.075 0.263 0.071 0.256 AGE 18-19 0.061 0.240 0.053 0.235 0.067 0.250 &ZZ 20-30 0.155 0.362 0.160 0.367 0.143 0.350 a. T k nrisbkwed intbtdmatioaischc log ofperd u g householdupcndiiuxe. Thcmeanleveloftotal household~padiiturcb3 1 3 , a adis, appmximetdy 51.363. k Theobsemtionisofleamadethathous olddecisionsbary substantiallybetween urban and rural areas. nis was h n d to be e in Ghana fiPr mntriiceptive rrse and W i t y decisions (Olivw, 1994) and by Montgomery and Ko1iarn6 (lW3) in their analysis of the QualitytQuuuitytradmff in %d'lwire. Estimationsarc presented for the entire sample and then urbanmd nrtal subamples are estimated s~pp-uately.Tests fix stzitistically significantdiffereaces vaiilbemade. Meam aqd standad deviations fbr the urban and rural sub-samplesare also provided in Tables 03 and 5. Baseline Results Thz estimation results b r the full sarnple of wmen and their childrec are presented in Table 6. Schwling of the wman has a significmt impact on both brtility and child schooling. Women with secondary schooling have an estimated 1.1 fewer children than wmen with no schooling. Children whose mothers have some secondary schooling have an estimated 0.9 yeas niore schooling than thosewhose mothershave no schooling. These zre large differenceswhen compared to samgle mzans of 3.2 children and 3.4 years of school. The iropact of any middle school isjust under half as large as the impact of secondvy scbool in the cuauiative fertility estimate. The relative eff't is smaller fir cumulative schooling but nearly eq~alb r current enrollment. Primary schooling has a much smaIler impacton cumulativekrtility and has no significant impact on current fertility. Children whose mothers hiwe some primary schooling are significantly more likely to be enrolled 31school b-atthere is no significant differencein cumulativeschoolingbetween childrenwhose mothers hav* someprimary schoolingand tbose whose mothers have no schooling. Predicted annualhousehold per addt expenditureand itssquarehave a significvlt effect on both cumulative and current fertiiity. The estimated impact on cumulative fertility is positive except fbr households iii the highest three expenditure deciles. The estimated impact on current fertility is positive except fi>rhousehold whose expenditure Mls among the richqt 15percent of households The expenditurechid schooling relation is also not straightfbrward. The coefficientson expenditure and expznditure squared are significant in estimates of both cumulative and current schooling. 'I'he combined impact is positiv6 h r ail households for both dependent \ariab!es. Coefidencs oc current residence reveal significant differencesin both fertility and ch2d schooling across areas. Tbe reference graup is wrnen, or children, in the mral stvannab. 'Nomen in AccraIKumiwi have 0.4 fewer predicted children, wmen in other : r iarea have 0.3 fe~erchildren. There is no predicted differzncs across the three Tural regions Recent fertility is significantly lover br all urban, semi-urban, costal and forest wmm ti!a lbr wmen in the rural savann3h. Predicted child schoolirg is lmtr Bor children in Accra/Kumasithan fbr chzdren in h e r u d savannah. It is also lower in the m d &rest and coasAdregions. The estimated coefficients on carrent enrollmentb l l w &e sane P a m . a Akm, the omitted languge group, acwunts fi;r nearly h f f of the smgle. Cumulative fertility is Iawr for all other language groups except one; Hausa wmen have predicted children ever born 0.3 higher thm Alran wmen. llle predicted coe%cients on the language group variables in the currect fertility equation are insignXcarit. It appears that wmm in all groups hme children at appiexhztely the same rate, Hausa wmen mathue longer than other wonen and, as a result, have higher cumulative krti!ity. I n t e a q g H a w with the age etegories relnoyes the significance from the Hausa cgefficient a d replaces it with a significant coefficient on age 40-44 and age over 44 b r Hausa wrnen. Table 5. Cwnulananw CurrentEnility and Schoafing Mod&: and Baseline P?ksd&,Full Sample M e n C h U e n Chikiren Bir1 hrl Y r m4 Current Ever Born 5y e m Wmaling Enmknt OLS Mil CLS &obit WIG P 8 P I B I P I Hbmon's Schojliing PlZlMARY -0.139 -2.055 -0.096 -1.361 -0.086 -0.633 0.100 1.732 MltDLE -0.498 - 8 . w -0.178 -2.909 0.243 1.995 0.366 6.817 SECOHWRY -1.083 -9.788 -0.662 -5.372 0.915 2.859 0.512 3.090 Ln(EXPEND) 28.938 3.569 42983 4.888 49.820 3.085 26.724 3.769 Ln(EXP) SQ - 1 . 3 -3.529 -1.W 4.771 -1.934 -2.839 -1.M7 -3.508 YEAR1 0.513 3.711 0.868 5.782 1.952 4.785 1.138 6.037 Currrnr Residence ACCRCVKUMASI-0.415 -3.694 -0.868 -7.500 -0.814 -2.135 -0.299 -1.548 OTHERURBAN -0.267 -2.721 -0.- -6.052 -0.195 0.674 0.137 4.936 RURAL COAST O.M4 0.249 -0.451 4.287 -1.282 -3.493 -0.571 -3.252 RURAL KlTaEST 0.100 1.073 -0.365 -3.810 -0.595 -1.854 -0.478 -3.111 torqpmge tfhowehold head EWE -0.194 -2.988 0.024 0.361 -0.439 -2.312 -0.257 -3.587 GA -0.2% -?.555 0.113 1.177 - 0 . ~ 1 -1.625 -0.330 -3.887 DAGBAh'I 4.344 -2.509 0.139 0.966 -3.008 -7.978 -1.359 -6.198 HAUSA 0.278 1.494 -0.037 -0.250 -0.126 26.402 0.141 0.964 NZEMA 4.104 -0.638 -0.155 -0.722 -0.381 -1.304 -0.262 -0.949 OTHER 0.065 0.917 0.354 4.636 -0.661 -3.495 4 . 1 ~ -2.139 W7man.s Age AGE-24 1.046 12.892 1.275 12.547 AGE25-29 2.287 19.917 1 3 3 11.742 AGE30-34 3.628 27.511 1.351 9.624 AGE3S-39 4.736 35.633 1.049 8.452 AGE40-4G 5.963 42.460 0.588 5.754 AGE45-50 6.650 43.%7 0.135 2.253 ChildbAge andSor MALE 1.077 13.851 0.34~1 9.443 AGE 8-9 0.531 5.756 0.700 13.832 AGE 10-1 1 1.471 14.074 0.837 13.630 hGE 12-13 2.438 16.631 0.669 9.813 AGE 14-1s 3.872 19.881 0.965 8.664 t AGE 16-17 5237 20.427 0.521 4.505 AGE 18-19 5.973 20.140 0.130 1 . 1 ~ 3 ~ AGE 20-39 6.511 22.847 -0.751 -7.482 Constant -i68.5W -3.592 -255.i03 -5.009 -316.576 -3.317 -169.163 -4.a4 R' OM4 0.258 0.351 0.251 LiILclihaA -2321.55 -4.69 n 4625 4625 7880 7880 . 1. E%cludedcakgoriea: No schooling, Rum1 S a ~ Alun lurpuqe, Wrlpn'r a s 15-19, ChiWa age ~ , 5-7. Child's scr fmk The estimations of child schoo1.q reveal striking differences aaoss languzge gmups. Dagbani children have 3.0 ).ears fewer predictai schooling than Akan children, an important difference on a variablewitb mean 3.4. Ewe and Ga children have 0.4 and 0.3 less predicted schooling, respstively. Current schooling estimates are sim:lar, children from all nonhkan speakinghouseholds are less likely to be currently enrolled in school. Male children hme predicted schooling 1.1 years higher than their femde counterparts and are more likely to be currently enrolled. Predicted cumulativeschooling increases st a decreasicigrate as the child's age increases. This is sexpected. Predicted current enro!lment increases through age 15and then declines. The estimationspresented in Table 6 provide some evidenceof a quality-qcantity tradeoff. This is especially trcle of the woman's schooling variables that exert a significimt positive impact on child schooling and a significant negative impact on fertility. Household expendiwre has a positive predic:ed tnpact on sc3ooling for all households and a negative impzct on f d i t y for the wealthiesthouseholds The negative fertility-expenditure relation among the wealthy households is not seen in the CBte d'Iwireanalysis. 'To this point it is possible to say only that sose of the results aie consistent with a tradeoff but b e widace is not civerwl?ei&g. Baseline Results for Urban and Rural Samples Theresults presented abcr~epmvide a basic daderstandingof the.relation between the independmt variables and fertilityand child schooling. However, Fiests indicate that the hypotheses that the estimated relaticins are the same for urban a ~ mraI nomen can d be rej&ted for all ibur dependent variJb!es at the 1 percent Itvel. Table 7 presents the estimation resu!ts b r the same set of wiables for rural wmen on!? Rtsnlts fbr urbw women are presented in Table 8. For rural vmmen, primary schoolingf5r the wman has no significant impact or fertility. The predicted impact of middle schooling on cumulztive fertility is -0.4 for rud women and slightly bigher, -0.6, for urban women. The impact of secor9uy schooling is similar b r both sub-sarnp!~: -1.2 znd -1.1 b r rural md urban wumen, respective1y. The predicted effect of wmatis secondary schoo!ing on j e m of child schoolirg is much strongzr for rurai child~n,1.6 compared to 0.5 ibr urban wmen. Primary school has GO significant impact on cumalative child schooling fbr either group of wmen. All levels of schooling increase significantly the robability of cncceat e:uollmens cif children in runl arm; oaly middle schoolilg has a significant impact in urban arm.; I 90 Tile 7. Cuntulatiw 04 C3m-t Fertilify and Schoding Models: Boseline Resul. Kurd Ssmple %me. Children Children Birrh lart Earsof Currcrf E~erBorn 5 yean Schooling EatvlbnerJ 0.5' h b l OLS Probit krizble -3 I B t B t B t %ran 5 Schooling PRIMARY -0.112 -1.378 -0.055 -0.635 -0.170 -1.208 0.168 2.247 MIDDLE -0.429 4.306 -0.144 -1.875 0.108 0.757 0.323 4.752 SECONDARY -?.I67 4.306 -0.856 -3.9 7 1.622 3.328 9.5i9 1.790 LzmPEND) 47.257 3.988 50.939 3.914 3.882 1.545 14.227 !s*.- Ln(EX?) SG -1.930 -3.893 -2.138 -3 J06 -0.986 -1.398 -0.336 -1 .' \EAR1 0.920 4.776 0.959 4.756 0.573 1.270 0.869 3 1 CurrentResidenc? RURAL COAST 0.274 -2 495 0.608 4.962 -0.784 -2.194 -0.620 -2.806 RURAL FOREST0.219 -2.W -0.497 4.403 -0.168 -0.503 -0.617 -3.122 L5rgrroge of horrsrholi head EWE -0234 -2.8d0 0.116 1.309 -0.3-3 -1.671 -0.254 -2.470 GA -0.3! 1 -2.543 U.1?9 1.229 -0.412 -1.781 -0.462 -3.310 DAGBAN: 3.444 -2.S69 0.035 0.190 -2.102 -7.672 -1.479 4.713 HAUSA 0.115 -0.309 -0589 -1.679 -0.102 -0.16 0.213 2.715 NZEMA 6.118 -0.6i7 0.042 0.156 0.232 0.841 -0.222 -0.584 mHER 0.058 0.654 0.336 3.477 -0691 -3.317 -0.355 -2.946 Wmank Age AGE20-24 0.910 8.817 1.288 9.877 AGE.25-29 2.087 14.0% 1.397 8.219 AGEqO-34 3.532 2r Xq3 1.125 6.200 AGE35-39 4.819 28.544 1.072 6.339 AGE40-44 5 :a 35.Y23 0.625 4.939 AGE45-50 6.892 38.552 0.187 1.452 Child's A p and S a MALE 0318 8.808 0.309 6.- AGE 8-9 0.505 4.719 0.764 17.234 AGE I t 11 1.352 1C.?31 0.931 12.032 AGE 12-13 2.245 13.335 0.832 10.088 AGE 14-1 5 3.456) 14.627 0.862 8.315 l.GE 16- 17 4.641 14.037 0.652 4.339 AGE 18-19 5.194 13.701 0.303 2.G87 AGE 20-30 5.543 l3.4E2 5.503 -3.702 a * I Conslant -278.650 4.067 -3C2.228 -4.0?2 -166.326 -1.674 -97.121 -1.880 R' 0.677 0.261 0.312 6.197 L i l i ! -1487.48 -2W8.06 n 3054 30% 5384 5384 1. Exfludrd c~kgoria:No rclloo;irg, Rural ~ a v r d kh, n !a?guap, Xbfixufs age 15-19. Child's agc 5-7. Child's sex fanalc t 91 T i8. ~ a i aidwh n tArtit@and Schooling M&: Baseline ILsultr, UrbanSample wmun QliLfm Childra B i r l h h Y C M g cmnt Ever h 5years schooht EnroIbnent o u wit OLS h b i t &rid& B t B t B t B I -- mlma'ss9lcmbg PRIMARY 0.258 -9109 - 0.231 -1.809 0.051 0.230 3.003 0.027 MW L E 0.603 4.287 0.251 -2.428 0.213 1.512 0.426 4.025 SECONMRY -1.066 -7.360 0523 -3.W 0.544 1.- 0.161 0.731 WEND) ii.718 1.078 3.128 2.829 15.148 0,739 m.84~ 2.415 L,(MP)SQ 4-9 -1.172 -1.42 -2.7~4 0.543 o.ao - 1 2 -2.246 YEAR1 0.W 0.013 0.727 3.283 0 . ~ 3 0.101, 1.m 5.029 C~nar Residemx OTHER URBAN 0.203 2.481 0.289 3.243 0.264 1.344 0330 2.239 Lolgvppe o f h e h d ihead EHE 0.118 -1.093 0.128 -1.231 0.468 -2.536 -0.349 -2.860 CA 0.117 -1.066 0.073 0.547 0.263 -1.472 -0.071 0.494 DAGBM'I 0.280 -1.203 0.276 1.189 -1.624 -2.420 -1.070 -3.841 HhZM 0.268 1333 0.005 0.030 0.183 0.618 0.292 1.808 N P M A 0.045 0.158 0.186 -1.3811 -1.076 -2.946 -0.814 -2.622 CYlHE2 0.105 0.881 0.409 3.231 0.444 -1.861 0.153 1.197 Mhaa'sAge AGW-24 1.201 9.464 1.278 7.477 AGm5-29 2517 14.751 1.Wl 8.527 AGEXI-34 3.815 18.360 1.743 7.599 AGE35-39 4.616 21.887 1.167 5.708 AGMO-44 -'.600 22.482 0523 2.862 AGl35-50 5.988 21.214 0376 1.958 Crn'r&:&&?Sa MALE 0.654 5.525 0.221 2.988 AGE 8-9 0.813 5.127 0.830 8.907 AGE 10-1 1 1.816 10.190 0.904 9.044 AGE 12-13 3.027 11.622 0.641 5.2Q9 AGE 14-15 4.242 12.410 0.696 5 430 AGE 16-17 5.283 12.336 0.655 3.228 AGE 18-19 6.162 13.948 0.095 0.501 AGE20-30 7.251 15.427 -0.792 -4.983 1- E~~ludcd No Ichoolirlg. O t k urhn.A~UI hnguclgc. U b w 9 sage 15-19. CMd" age 5-7, C h i l s s # x h a k - Household expenditurebas a predicted negativeimpact on emulative fertilityfor all wrne.1 in both sub-samples The predicted impact on recent fertility is positive for d l but the wealthiest households Predicted cumulative schooling and curreratenrollment lncrease with household expenditure fbr rural and urban children. The predicted difference in rumulative schoolkg between the richest and poorest households is 4.4 years in rural households and 3.3 years tbr urban households The male-female direrentid is larger in rural areas; boys have 0.8 years higher predicted cumulative schooling. Both the rural and urbn regressions, pmvide evidence of a tradeoff between child schooling and fertility is supported by ttz coefficients on women's schooling 2nd household expenditure. The results of the P d i t estimations are more easily inteiireted in nble 9. Average predicted probdilities are calculated fbr each schooling, resideace, and language of household head catwry using the recent fertility and current enrollment resulk k r rural and urban sub-s3mples. Estimates Including tbe Price d Schooling The baseline resclts provide some evidence of a qualityquantity tradeoff. To explore the issue firther, specific measures of the cost of schcoling are included in the estimations presented in Dhle 10. The first is a binary variable b r the presence of a primary school in the community, the second is the mean annual schml fkb r the third year of primary school among primary schools in the community, and the third is a binay variable tbr whether or not there are textbook in the community primary school. The school data were collected fbr ~ n l ya sub-anple of the rural clusters. The average annual fee charge differs from the measure used in the Montgomery and Kouame analysis. They use actual expenditure per child to compute a price of schooling. The measure included in this analysis is the actual fee chqed by schools as collected in the school ficility svvey. If there is a quality-qvmtity tradeoff occurring in fwtility and schooling decisions, the price of schoolirg should be positively related to fertility and negativeIy reIated to schooling, the prrspace of a school should be lower fertility and raise schooling, and the quality shwld lowe. fetility and raise schooling. The quantityfquality hyc->thesisis paiially borne out by the estimations. The annual fee does h m a signifiWr ~ s i t i vimpact on h t h cumulativeand recent fertility e but it aieo has a significant posithe impact on schooling. The quality has no significant predicted eff& on fertility but has a significant positiw effect on child schooling. Tabe 9. hdicted RvbobilirS qfP#p%nt Rrtilw and h n ! hIImnt: Rvml t J r h Erth kxsf Crmnl Birth hu Cumnt 5y e a EaokU 5 - h l b n c n t Ubman's Schooling NONE 0.676 0.448 PRIMARY 0.661' OSOO MIDDLE 0.636 3549 SECONMRY 0.416 0.23 Current Ruldcnce ACCRAlKOMAS1 .. OTHER URBAN .. RURAL CQAST 0.752 RURAL KUESI' 0.589 RURAL SAVANNAH 0.622 Lungvcq~eof hoysehokih c d AKAN 0.626 0538 EWE 0.459' 0.458 GA 0.676 0394 DAGWI 0.636' 0.138 HAUSA 0.444- 0.602' NZEMA 0.638' 0.46s- OTHER 0.717 0.427 Chikl's S a FEMALE .. MALE .. Nora: Prcdiictcdprob.bltk uc clllcukbd using the c u m k+iGty and current umIlmart regressions p ~ ~ i n ~ b l c r 7 a n d 8 d t b t r o r m u h(1flJ)=Ck,,,@(CI,k%& +&)-krtfenbthe : ~ cmarhb in question. M idutr uc ahrhbd fir all in cach sub-5~pk.uift h y wlut all in each ofthe diff'utnt catgoriu, in twn. .odcbe istbcnalcubtedover all aomtn. Contmst to &e omitted category is not s i p d i c l d Tdle 10. Ciunulnnnnw Curn?ntfirtilily and Schooling Models: and Including the Rice of schooling; Ruml sample Hbnvn W r e n Children Birrh lasf Ycm 4 Current Ewr Born 5 y e m -ling ENollmcnr OLS Robit OLS Rvbir \biob& P t B t B t P t \Ymum's SchooIing YEARS -3.060 -3.728 -0.036 -2.211 -0.032 -0.563 0.041 2.858 LnGXPEND) 73.298 3.026 83.994 3.915 90.292 1.558 42.967 2.880 Ln(EXP) SQ -3.108 -2.980 -3.547 -3.852 -3.W 1.524 -1.739 -2.791 Rice 4 S c h o o r i PRIMARYSCH -0.033 -0.200 -0.076 -0.432 -0.208 -0.459 -0.061 -0.459 ! ANNUAL FEE 0.004 3.546 0.004 3.230 0.004 1.558 0.002 2.523 ~ O O K S -0.157 -0.411 -0.047 -0.057 2.749 3.371 1.665 2.781 c ' w c t u Residence RURALCOAST -0.289 -0.951 -0.897 -3.217 -2.259 -1.6% -1.068 -3.349 RURAL FOREST -0.345 -1.240 -0.654 -1.989 -0.877 -0.668 -0.635 -2.161 L a g w e ofhoyfehold head EWE -0.269 -1.089 0.229 0.892 -0.753 -1.191 -0.190 -1.233 GA -0.878 -2.417 -0.359 -1.042 -0.249 -0.215 -0.541 -1.840 DAGBPNl 0.098 0.218 0.806 0.861 .. .. .. .. HAUSA -2.035 -1.556 -1.574 -2.043 0.141 0.089 -0.342 4.536 (3MER 0.723 2.805 0.903 2.956 -0.179 -0.273 0.003 0.015 Ubman'sAge AGE20-24 1.016 4.402 1.642 5.400 AGEZS-29 2.054 5.377 1.376 3.326 AGUO-34 3.638 8.128 1.218 2.823 AGES-39 4.766 10.666 1.194 3.620 AGE40-44 6.095 14.381 0.679 2.729 AGE45-50 6.632 14.690 0.264 0.920 Child's Age andSa MALE 1.436 i.5.9 0.228 2.452 AGE 8-9 0.826 2.565 0.876 6.220 AGE 10-11 1.565 4.235 0.927 5.786 ACE 12-13 1.449 4121 0.529 3.465 AGE 14-15 2.855 4.965 0.678 3.602 AGE 16-17 3.ZOO 4.179 0.676 2.450 AGE 18-19 3.741 4.032 -9.065 -0.211 AGE 10-30 4.097 4.970 -0.566 -2.217 Con~tnnt 430.318 -3.067 496.219 -3.980 -557.264 -1.592 -265.96 -2.980 c R' 0.645 0.299 , 0.193 0.254 Liilihood -262.10 -522.99 n 592 592 1012 1012 1. Excludcd cstcpia: No s c h w h . Rum1Savannah. Akan language. Wman's age 15-19, Chid's age - 5-7. Child's sex female -4 * I Conclusion Raising the quality of children by increasing school enrollment and decreasing high fertility are frequentlythe gods of public policy in Sub-Saharan Africa. This paper has explored the possible interaction behveen household fertility and child schooling decisions suggested by the bypothesi of a q~alityquantitytradeoff. Some evidence of the existence of a qualityquantity tradeoff in fectiiity and cLild,c .booling decisions in Ghana is pmided by the empirical results. The most striking result is the statistically significant and large impact the schoolingof the mother has on both fertilityand child schooling. Women with secondary schoolinghave 1.1 fewer children than wmen with no schoolirg and children of wmen with secondary schooiing have 0.9 years more schooling than children of wmen who have no schooling. Increasing femaleenrollmentcan be expected to raise child schoaling and lower fertility. Beyond the contrasting signs on the mother's schooling and location of current residence; the support of the qualityquantity tradeoff is limited. The predicted impact of schoolingisseen in both the urban and rural sub-samples with some subtle differenfa. Availability of primary scl~ooldoes not have a significant impact in rural areas on either krtility or child schooling. This result may be due to lack of variation as primary schools are widely accessible. On the other hand, secondary schoolingof the mother has a much larger predicted impact on child's schooling in rural areas. The malefemale schoolingdifferential is somewh.at more marked in rural areas. The price of schooling variable does exert a statistically significant impact on fertility, but the magnitude of the effect is small. A ten percent decrease in the annual fee from the mean reduces the predicted children ever born by only 0.06 children. The impact of the annual fee on child schooling is also positiw. This may be because fees are higher in communities where demand fbr schooling is higher. The presence of a primary school does not significantly affect fwt'iity or schooling. Quality of local schools has a signifcant positive effect. In their analysis of CBte d'Ivoire, Montgomery and Kouamk find sharp differences across urban and rural areas that are not present in Ghana. They also tentatively assert that there is no evidence that increases in school costs would raise fertility. While the predicted impact in Ghana is statisti-ally significant, it is small. The policy implications are turo-fbld. The first is that &male schooling can be a p a n t instrumentfbr lowering futurefertilityand raising futureschool enrollment. The second is that, wen though the* is some evidence of a tradeoff at w r k between the I number of children and the schooling of those childrzn, increasing the cost of schooling would not be expected to induce a large increase in fertility. Indeed, changing school costs wuld not have an important impact on fertility. - -0 2 References Fertility and Child Schooling in Ghana: Evidence of a Quality/Quantity T i e o f Ainsbvorth, M. 1992. "Economic Aspects of Child Fostering in C6te d'Iwire.' Living Standards Measurement Study Ubrking Paper, No. 92. Washington: World Bank. Ainsworth, M., 1990. "Socioeconomic Determinants of Fertility in Cote d'Iwirem. Living Standards Measurement Study Wrking Paper, No. 53. Washington: World Bank. Becker, G.S., 1960. 'An Economic Analysis of Fertility' in National Bureau of Economic Research, Demographic and Econiwic Change in Developed Countries. Princeton: Princeton University Prss. Benefo, Yafi and T.P. Schultz. 1993. "Determinantsof Fertility and Child Mortality in C6te d'Iwire and Ghana." World Bank, Poverty and Human Resources Division, Policy Research Department, Washington, D.C. Glewwe, Paul. 1994. m e Economics of School Quality Imstmenrs in Developing Countries: An Empirical Study of Ghana. f,rthcoming. Washington, DC: The World Bank. Gomes, M. 1984. "FainilySize and Educational Attainment in Kenya." Population and Developcent Review, lO(4): 6 4 7 4 . Kelley, Allen C. and Charles Nobbe. 1990. "Kenya at the lbming Point?" Ubrld Bank Discussion Paper, No. 107. World Bank, Washington, D.C. Leibenstein, Harvey. 1957. Economic B a c ~ n e s ad Economic G m h . New York: John Wiley and s Sons, Inc. Montgomery, Mark and Kouame, Aka. 1993. 'Fertility and Child Schooling in C6te d'Iwire: Is There a Tradeoft?," World Bank, Poverty and Human Res~urcesDivision, Policy Research Department, Washington, D.C. Oliver, R. 1994. "The Effect of the Quality, Price aqd Availability of Family Planning on Contraceptive Use in Ghana." Warld Bank, Poverty and Human Resources Division, Policy Research Department, Washington, D.C. Owusu, J.Y. and Z.M.Kofi Bake. 1991. "Family Plannirg Services in Ghana." Ip Ghana Population Policy: Future Challenges, Report of National Population Conference held 28-30 June, 1989 to commemotate the Twentieth Anniversary of Ghana Population Policy. Accra: Ministry of Finance and Economic Planning. Pitt, Mark and Mark Rosenzweig. 1990. "The Selectivity of Fertility and the Determinants of Human Capital Investments: Parametric and Semi-parametricEstimates." Living Standards Measurement Study Wrking Paper, No. 72. World Bank, Washington, D.C. . R2public of Ghana. 1969. Population planning for Natiorz l Progress and Prosperity. Ghana Publishing Corporation: Accra-Tema, Ghana. Scribner, Susan. 1994. 'hlicies Affecting Fertility and Contraceptive Use: An Assessment of 12 Sub- Saharan African Countries' Fblicy Research Department, Fbverty and Human Resources Division, The World Bank. Tansel, Aysit. 1992. "School Attaiient, Parental Education and Gender in a t e d'Ivoire and Ghana." Northeast Uni\ersities Development Consortium Conference, 1992, Boston University. Theil, Henri. 1952. "Qualities,Prices and Budget Inquiries." Mew of Economic Studies, 19: 129-47. World Bank. 1993a. WddDewloprnent Repon. New York: Oxbrd University Pms. . 1993b. "Ghana Living Standards Survey (GLSS), 1987-88 and 1988-89: Basic Inhnnation.' Mimeo. Pbverty and Human Resources Division. Distributors of WorldBank Publications ARGENTINA KOREA, W U COF SOUTHA R I U BOTSWANA Culor HinchSRL Pan Yau BooiC a p u a ~ n .=wSm* W k Cllcru Cvcmn P O Bor 101.K w m g r w odadUnir.mity Rro nand.165,4th FhrOfc 453/*- SeaJ SovaMAfrio U33Butr.nAhes P 0.Bm 1141 lLnranSeodrw(jnm GpTorn6900 O~%N del LbroInamaaoorul P O Borw -40 Ynxido fmsubl#.l adm !082 8- A i m SeaJ ~ t i O l u l S u k s r i p r i o n ~ FRANCE P.O. bn41095 AVSTRAUA. PAPUA NEW CU141EA Wald BrnLPubboobra MAUISlA FIJI. SOWMON ISLANDS, 6b. avenw d'ku UN~rnitydhhhya q I b v + VANUAN,AND WESTUN SAMOA 75116Pam BdshDp,IiInitd D A hfcaution L w ~ m PO. ~ ~ ~ ~ ~ , J . L ~ P u U J I B U U 648WhikhomeRmd GERMANY 5 9 i u I K ~ l r v a p f Militcium3U2 LXC-Vnhg \i.d P o p p M d a ALk J MD(IC0 53115lbnn (NFOTEC A,ZaaAOPoQI ad60 GHANA IrodorWpucMcPeo DF. t r r c n ~ i d l U J ~ M d ~ RivmI*whHal. w POIb. 011B IL LindebxuvlnOI-Publihres 0 5 U - A N . P.O.Ib. 102 7 a A f I h b b q e n GREECE P a p m l i l h U N E W Z E A W 35.Smumm9. E E S c O ~ L I d -2 10667. Admu R i r a t e W Bag99914 P*rMuLc !WEDM BELGIUM HONG KONG, MACAO AvdLnd F r i u e a F ~ w o m B p ( Jean D.lmnoy &b20(1)LEb ~ p w u ~ l a 5 6 Av. du Rd -XI2 46-v5wpiJumSar m c o u s 1 m47stockkh 1Db0 mnds w-carm Udvari* RP.LiPLit.d 7th- ThnrGourmsrr3dinllsidp \\'VWW M cmrnlI4BqXong WvabMAiImgYRS P . Q h l 3 M Ib.d.n b l A D scans HUNGARY 01409S.oPa&, SP ~ e m f o r ~ u t n ~ y NORWAY S w n 7 x U u N D -ui Ut l7-19 h ' u v ~ ~ c i o n ~ m4rkPa* CANADA H-1117 -apumac pmbh3212 LchMunv PO.Bor6125EmnUd C H I m t L u v n n - rsuno+scMolugN WDlA N Q b O l W 6 -,pu(kc Allid Publirtm Rirapc Ltd Van Di- EditimuT-. ADECO J4B5Eb 751M a t Rod PAKISTAN PO.kl.465 Madrrr -6fnOJT . ~ s a a t ~ CH1211Cauva16 RedFbbWlin&Ca a,-+- 1 T W A l ~ R o r d INDONESIA P.O. rh729 TAYUNIA Oruwl. onhtio R~~ btwm510Q) CdadUdr-crwy Rru K183WI JlLnBaobdur3 uILt.b.sw P a B o r 1 8 1 PERU P O . B a S a 9 CHNA JWrn l EdtWRvnolbSA DuaS.L.m O\irufirunddk Ecommic A p u p Q M 4 PusWlingHarw M Liaul THARAND 8.Da FoSiDonghe KDvLubPublam cmrmmputsrmtyur @-*s P O Bn 19575511 m u m 3QbSJornRold Tchnn InmrudBoDkCrnta W k COWMBIA Suite17?,,GyM 10 Wcml.aLMr IRELAPrrP -rower1 r n I D A D k TOBAGO Apuadokno312fO tovamrrmSuppkAgenq Ayah Avenue. KV.deb +-tia s- unit ao&ea D.E 4-5HurrrunRad CorPtmaion . P W a m S a r Dublin2 MkabMccm- COlE WNOIRB E L W e s t l n d i a CenrndTdi6oaetdehWm ISRAEL POUND YOPM L i t r m Mbi ~ ~ ~ g S a v i o rUGANDA PO.BoxXQSj ULPahu31/37 GvrtroLrd. Td Aviv61560 M W u n r r . l n ~ , R m m r . ~ a w b c n P a Box9991 CYPRUS R0.Y. tnnnuoaul Aa(69)Wlmprl. -aa dAppliedResearch P0.81- c?-P-wrgc T d Anv 611X UNITEDKINGDOM 6. D;oens%Engomi ).Gu&fo Lrd. P.0. Ear TY)6 ITALY P Q h 3 NUpy SAUDIARABIA. QATAR !uLBookShm QEMREPUBUC P.O. Bor 31% Sarion4hfcaudon Cmkr Riyadh IlCn U M B l A P O . b r k S Czu'vrmityd Ibokhp cs?13S7PnguC1 SLOVAKREWBUC GvnFistRoadGmpw SwartG1.C Ud. P.O. h 3 2 3 7 P DNMARK Krupimtr4 -Lnkl SunfwdrLittrnIw r.0.mu ~ i 2 r RwnoenuAU&ll 5 3 99 Br~tirkva5 OK-lmO F-rg C JAPAN Easternw-virr DO.UINICAN REPOCUC Hongo G, mLk,&u 113 ECirar Taller. C pur A. Tokyo FkuundbncLube1laGt6lia30P Aputadode cone^^ 21W 2-1 KWYA SlntoDoningo Mriu BookSar-icc(-1 Lrd. b a n I-louse,Mlm~aro u mS EGYPT,iUUB REPuBUCOF P.O. or rnlj Al Ahram Nairobi Al GI.. SIT& G i r o LSMS Working Papers(continued) No. 77 Newman,Jorgensen,and Pradhan, Works' Beqitsfrom Bolivia%Emergency Social Fund No.78 Vijverberg,Dual Selection CriteriawithMultiple Altrmnti~cs:Migration, WorkS k u , a d Wages No. 79 Thomas, Gender Di@enoes in Housrhold Rrsourre Allocations No. 80 Grosh, TheHousehold Surwy as a Tool& Policy Change Lcssonsfrorn theJamaicanSurwy of Living Conditions No. 81 Deaton and Pamn, Patternsof Aging in Thrriland and Gled'lwirr PJo.82 Ravallion,Does UndernutritionRrspondto Incomes ant Prices? Dominance Tests& Indonesia No. 83 Ravallion and Datt, Growthand W t n h t i o n Gmponmtsof Changesin Poverty Measure: A Decomposition with Applicntions to B d and India in the 1980s No. 84 Vijverberg,Measuring Incomcfnrm Family Enterpriseswith Household S u r i s No. 85 Deaton and Grimard, Demand Analysisand Tm Rrfmm in Pakistan No. 86 Glewweand Hall, PoccQ a d Inequnlity during UnorthodoxAdjustment: TiCaseof Peru, 1985QO No. 87 Newman and Gertler,Family Produdi'mty, Lnm Supply,and Welforrin a Lout1nw.m Country No. 88 ILvallion,PoccQGnnpariwns: A Guidcto Conceptsand Mefhods No. 89 Thomas, Law, and Strauss, Public Policy and Anthvmctric Outcomesin G t ed'lmire No. 90 Ainsworthand others, M m r i n g the Impact of Fatal Adult Clnessin Sub-Sahamn Aficn: An Annotated Household Qucstionnaim No. 91 GlewweandJacoby,Estimating the Drtmninants of CognitiveAchieoement in LoutInwme Countries:TheCuseofGhana No. 92 Ainsworth, EconomicAspccfs of UdkiFostering in G t ed'Im're No. 93 Lavy, lnoestment in Humm Capital.Schooling Supply Gnstraints in Rural Ghana No. 94 Lavy and Quiglq, Willingnessto Pay& the Quality and Intensity ofMEdical Care: Lnu-Income Households in Ghana No. 95 %ulk and Tansel, Mcusumnent of REturnstoAdult Health:Morbidity Eficu on WageRates in G t e d'lwire and Ghana No. % Louat, Grcuh,andvander Gaag, kYdjk Implimfions #Female Hendship in,'nmaiwr Holmholds No. 97 Coulombeand Demery, Household Size in Gte&Itmire: Sampling Bias in the ClLSS No. 98 GlewweandJacoby,Delayed Primmy School Enrollment and ChildhoodMalnutrition in Ghana:An EconomicAnalysis No. 99 Baker andGrosh, Poverty Reduction thrvughGeographicTargeting:How WellDoes It Work? No. 100 Datt and Ravallion,Income Gains& thePoorfnrmPublic WorksEmployment: Evidencefrom TrvoIndian Villages No. 101 Kostermans,Asswing t h Quality ofhthropometric Data: Backpnd and Illustrated Guidelines& Surwy Managers No. 102 van de Walle, Ravallion,andGautam,How WellDoes the Social Saf* Net Work?:The Incidence of Cash Benefits in Hungmy, 1987-89 No. 103 EknefoandSchultz,D e t m i m t s ofFerh7ityand Child Mortality in Gte d'lwire and Ghana No. 104 Behrmanand Lavy,@ildren's Health and Achievement in Schwl , I No. 105 Lavy and Gemin, Quality and Cmt in EfealthCareChoicein Dmloping Countries No. 106 Lavy, Strauss,Thomas,andDe Vreyer,TheImpact of the Qualityof Health Careon Children's Nutrffionand St~mivalin Ghana No. 107 HanushekandLavy, S c h l Quality, A~zievementBias, and Dmpout Be 4. r in Egypt - No. 108 Feyisetan andAhworth, Contraceph Useand the Quality, Pcce, and Amzhhlity of Family Planning tn N&e& No.A09 Thomasand Maluccio, ContraceptimGm'ce,Fertility, and Public Policy inambabwe No. ¶I0 Ainsworth, Beegle, and Nyamete, TheImpact of Female Schooling on Ferfilifyand Contraceptive Use:A Study $Fourteen SubSaharan Guntries No. 111 aliver,ContraceptiveUsein G l i -The Role of S-ce Awilabilify, Quality, and Price T'~',E \SORL.D G.ixi; A ~ai:nerin oscni?.~r;.ar;=.:, 2nd s:reng:htxnirg e;onorn::>> :o i ~ . F x2 k5.c z.;z:itv. cl! L.!:, . \ 2 n d e\par,li rrbs;t:itv FOR I'EOPLE E\.E~;Y\~ fiER!I