Conceptual framework Directory UMM :Data Elmu:jurnal:E:Economics of Education Review:Vol19.Issue1.Feb1999:

64 P. Glick, D.E. Sahn Economics of Education Review 19 2000 63–87 enrollment rates remain below 40, however—among the lowest in the world World Bank, 1995. In addition, in spite of a commitment to improving girls’ access to schooling, the ratio of female to male primary students in 1993 was only 44. This gender disparity in enrollments increases sharply with education level: girls represent only 25 of lower secondary students, 20 of upper secondary students, and just 6 of university students. Thus gender in Guinea is an important determinant both of attending school at all and of the level of schooling achieved. In light of the benefits to investments in education, it is important to identify the factors underlying household decisions regarding the education of children, and especially decisions about girls’ schooling. The edu- cation of parents has been found in many studies to be one of the most important determinants of child school- ing. Of particular interest in the West African context, where incomplete pooling of household resources appears to be the norm and preferences of husbands and wives may diverge sharply, 2 is whether maternal and paternal schooling have equivalent effects on the edu- cation of boys and girls. It might be expected that edu- cated women have both strong preferences for schooling their daughters preferences which may not be shared by their spouses and the ability to ensure that household resources are allocated for this purpose. If as a conse- quence of these factors a mother’s education has a greater impact on girls’ schooling than on boys’, there would be a further rationale for public investments in female schooling: the intergenerational effects of such investments will lead in the future to even greater reductions in the gender gap in schooling and ultimately, in earnings. Boys and girls may also differ with respect to the ways in which household structure, in particular the presence of young children, impinges upon their ability to acquire an education by affecting the burden of household responsibilities. These responsibilities are likely to be imposed on girls more than boys. If this is the case, then policies for example, subsidized childcare that reduce the dependence of households on the domestic labor of girls may increase girls’ enrollments, thus also helping to close the gender gap in schooling. Research on the household determinants of schooling is quite sparse for sub-Saharan Africa, owing in part to 2 There is a sizable anthropological literature for Africa, and West Africa specifically, indicating that men and women within households do not pool income or make expenditure decisions jointly. See, for example, Fapohunda 1988, Munachonga 1988, and Guyer and Peters 1987 and references therein. Complementing these studies is the econometric analysis of household expenditures in Coˆte d’Ivoire by Hoddinott and Had- dad 1995, who find that expenditure patterns differ depending on the share of total family income earned by women. the shortage until recently of comprehensive household level data sets from the region. 3 This study examines schooling choices using household survey data from Conakry, the capital and largest urban area of Guinea. We focus on the impact of parental education, household structure, and income on the schooling of boys and girls. Several schooling outcomes are examined in the empiri- cal work in this paper: years of schooling or grade attain- ment; current enrollment status; and leaving school. We focus on multiple schooling indicators instead of a single one, such as current enrollment, for two reasons. First, each of the three illuminates a different aspect of school- ing choice and thus is of interest in its own right. Second, as described in detail below, each has both advantages and disadvantages, the latter largely reflecting limits in the available data. Since each approach is imperfect, checking for consistency of results with regard to key variables provides a useful informal test of the robust- ness of the findings. The remainder of this paper is organized as follows. Section 2 outlines the conceptual framework underlying the empirical work and Section 3 discusses the econo- metric methodology. The dataset is described and some descriptive results are discussed in Section 4. The econo- metric results are presented in Section 5. The paper con- cludes in Section 6 with a discussion of policy impli- cations of the results.

2. Conceptual framework

Underlying the empirical analysis in this paper is a conceptual model of parental or household decision- making regarding investments in the education of boys and girls. 4 We assume for the time being a “unitary” 3 Much of what has been done has used the World Bank’s Living Standards Measurement Survey data from Ghana and Coˆte d’Ivoire see, e.g. Tansel, 1997; Glewwe Jacoby, 1994. An earlier analysis by Chernichovsky 1985 focused on chil- dren in rural Botswana. 4 The model presented here, like the standard household model, takes as the relevant decision-making unit a two-parent nuclear family. In West Africa, households tend to be larger than this, often including multiple generations as well as mul- tiple wives, and are linked in important ways to other house- holds in the extended family though flows of both resources and people, e.g. through child fostering Lloyd Gage-Bran- don, 1993; Ainsworth, 1992. For many purposes, the extended family consisting of a network of households may, in fact, be the relevant decision-making unit, which in the present context implies that characteristics of a child’s immediate family, such as parental backgrounds and the number and ages of siblings, should be of lesser importance for schooling outcomes. Direct formal empirical analysis of this issue would require detailed data on related households and their linkages particularly rural–urban flows of resources and individuals, data which are rare or non-existent. However, the econometric results 65 P. Glick, D.E. Sahn Economics of Education Review 19 2000 63–87 model of the household such that preferences of the mother and father are identical, or that if they are not, the household nevertheless acts as if it were maximizing a single utility function which would occur if the prefer- ences of only one of the parents counted. Parents are assumed to live for two periods. For a household con- sisting of a mother, father, m daughters and n sons, 5 par- ental preferences are assumed to be represented by a util- ity function U 5 U C t ,C t 1 1 ,S d1 ,...,S dm ,S s1 ,...,S sn 1 where C t and C t + 1 denote household parental consump- tion in the first and second periods and S di and S sj denote the education of the ith daughter and jth son. The second period consumption of the parents depends through remittances on their children’s income, denoted by Y dit + 1 and Y sjt + 1 . Children’s income or wealth in the second per- iod depends in turn on the level of schooling attained in the first period as well as on child-specific variables Z such as sex and natural ability, e.g. Y dit + 1 5 Y dit + 1 S di ,Z di for the ith daughter. Transfers from child to parent in the second period will vary by child characteristics through their effects both on income and on remittance propen- sities; in particular, for cultural reasons daughters may remit a smaller portion of their incomes than sons. Thus we have for second period parental consumption: C t 1 1 5 C t 1 1 Y d1t 1 1 ,...,Y dmt 1 1 ,Y s1t 1 1 ,...,Y snt 1 1 ; 2 Z d1 ,...,Z dm ,Z s1 ,...,Z sn Parents are assumed to care about the wealth or income of their children because of the benefits to their own future consumption through remittances, but they may, of course, also be altruistic and care about their children’s future welfare. In this case children’s wealth would appear as separate arguments in the utility func- tion. In either case the nature of schooling as an invest- ment is clear: greater education expenditures, financed through reductions in current consumption C t or though borrowing, result in higher levels of income for the chil- dren and consumption for the parents in the second per- iod. 6 The education of the children also enters into the presented below, and the fact that enrollments of fostered-in children are lower than those of children living with parents, suggest that even within extended families, the characteristics of the nuclear family retain substantial importance as school- ing determinants. 5 The number of children is taken as predetermined rather than as a choice variable, an assumption that raises concerns for the estimation. These are addressed below. 6 For girls in particular there may also be significant non- pecuniary returns to education investments not represented in this simplified model, such as improved child health, which par- ents may also value. parental utility function Eq. 1 directly, since parents may also enjoy having educated children. The simplified framework above thus brings out the nature of education as both an investment good and a consumption good. 7 Parents in the first period face a full income constraint: F 5 V 1 T m w m 1 T f w f 1 O m 1 T di w d 1 O n 1 T sj w s 3 where V is unearned income, T m , T f , T di , and T sj are the total times available to the mother, father, the ith daughter and the jth son; and w m , w f , w d and w s are their respective wage rates. The contribution of children’s time to household full income should be emphasized. Children in developing countries typically make pro- ductive contributions to household welfare through work in the home, for example by caring for younger children, or in the family farm or business. Where there is a mar- ket for child labor they can also work for others. If such a market exists, w d and w s are the market wage rates for girls’ and boys’ labor; otherwise they represent implicit prices of time determined endogenously by both the demand for household goods and services that use chil- dren’s labor and the home production technology “” is used to indicate that these are potentially endogenous variables. Assuming for convenience that the time of children is divided between such work and schooling and that of parents between work and leisure, the full income con- straint can be expressed in terms of expenditures on leis- ure, goods, and schooling: F 5 V 1 L m w m 1 L f w f 1 P c C t 1 O m 1 P s S di 4 1 w d T S di 1 O n 1 P s S sj 1 w s T S sj where L m and L f are the leisure of mother and father; P s are direct costs per child of schooling; and T di and T sj represent the time of the ith daughter and jth son, respectively, devoted to schooling. Regarding the com- position of P s , direct schooling costs in Guinea do not generally include tuition since virtually all primary stu- dents and the great majority of students in higher levels attend public schools, which are free. 8 However, other 7 The latter aspect of schooling is emphasized in household production models incorporating child quantity and quality Becker Lewis, 1973; Willis, 1973. In this framework, par- ents derive utility from both the number of children they have and their quality, an important dimension of which is their schooling, and both quantity and quality are choice variables. 8 Private schooling was banned until 1984 and has yet to make major inroads on general education: less than 5 of pri- mary school students are enrolled in private schools Educational Development Center, 1994. 66 P. Glick, D.E. Sahn Economics of Education Review 19 2000 63–87 direct private costs such as books, uniforms, and trans- portation appear to be considerable: they were cited in interviews of parents and students as a major barrier to school attendance World Bank, 1995. w d T S d i and w s T S s j , the hours of the daughter and son spent in school or schoolwork multiplied by the price of their time, rep- resent the foregone contributions to household or market production of having the daughter and son attend school. Each term in brackets combines this opportunity cost with direct costs P s S di or P s S sj and thus represents the total cost of schooling for each boy or girl. Parents maximize utility subject to the full income constraint and the constraints relating earnings to school- ing and parental consumption to child earnings, resulting in reduced-form demand equations for boys’ and girls’ quantity of schooling as well as for other goods and leisure as functions of all prices and wages, vectors of individual factors Z and household and community fac- tors H, and maternal and paternal education S m and S f : S di 5 S di w m ,w f ,V,P c ,P s ,S m ,S f ,Z di ,...,Z dm ,Z s1 ,...,Z sn ,H 5 S sj 5 S sj w m ,w f ,V,P c ,P s ,S m ,S f ,Z di ,...,Z dm ,Z s1 ,...,Z sn ,H As indicated in the introduction, parents may have dif- ferent preference orderings, including different prefer- ences for schooling, in which case the assumption of a unique parental utility function such as that in Eq. 1 is invalid. A variety of theoretical models of the household that incorporate differing preferences of household mem- bers e.g. husbands and wives, mothers and fathers have been developed and are referred to as “collective” house- hold models. We briefly describe the approach here rather than present it formally because the reduced-form demand functions for schooling derived in such a frame- work are for our dataset identical to those just derived within the common preference framework. Collective models posit separate utility functions for the husband and wife. When preferences differ there must be some mechanism for determining how to allocate household resources; the most common assumption is that the household engages in a cooperative bargaining game leading to Nash equilibrium demands for commodities McElroy Horney, 1981. Each partner’s power in bar- gaining is a function of the income under his or her direct control and, ultimately, of the utility he or she would be able to achieve outside of the marriage—the “threat point”—since this represents the fallback position. 9 The 9 That is, with a stronger fallback position, a partner can thre- aten more credibly to dissolve the partnership if allocations are not sufficiently in line with her or his preferences. More realisti- cally for smaller allocation decisions, the partner can threaten to retreat to a non-cooperative solution, where budgets are com- pletely separate, while maintaining the partnership see Alder- man, Chiappori, Haddad, Hoddintot Kanbur, 1995. threat point, hence bargaining power, is a function of “extrahousehold environmental factors” McElroy, 1990 that affect opportunities outside the partnership. These include unearned income accruing to the individual, sex- and education-specific wage and employment rates, the legal framework as regards, for example, child-support, and the partner’s possibilities for remarriage or financial support from relatives. Reduced-form demand equations for schooling derived from a bargaining framework would include such factors in addition to those already in Eq. 5. Test- ing the unitary household model against the more general collective model essentially involves testing for the sig- nificance of extrahousehold environmental variables that would not affect schooling under common preferences. A variant of this approach involves comparing the effects of unearned that is, exogenous income received by each spouse. 10 Unfortunately, our dataset lacks information on or variation in these factors, so we cannot augment our reduced-form equations to conduct such a test. Still, the notion of non-unified preferences and bargaining over resources within the household has substantial appeal in the present context and will be helpful in interpreting some of the empirical results of this study. The direction of the effects of a number of the vari- ables in the schooling demand function can be predicted from theory. 11 In particular, factors that raise perceived returns or lower the costs of education will raise invest- ments in education. The schooling of the parents S m and S f is one such factor and is expected to be positively associated with children’s schooling. Educated parents are more able to assist in their children’s learning, raising the returns relative to less educated parents, and are also 10 The unitary model assumes that income is completely pooled so the source of income should not matter for allo- cations; hence common preferences are rejected if the effects of husband’s and wife’s unearned income differ. The equality of male and female income effects has been rejected for out- comes such as own labor supply Schultz, 1990, child health Thomas, 1990, and household expenditure shares on health, education and housing Thomas, 1993. Rao and Greene 1991 find that a woman’s bargaining power, proxied by female employment rates and sex ratios reflecting re-marriage possibilities significantly affects fertility decisions. Handa 1996 uses female headship, treated as endogenous, as a proxy for maternal bargaining power within the household in school- ing demand equations and finds a positive effect of this variable on schooling. However, some of the exclusion restrictions imposed for identification of the structural schooling equations are questionable; it is assumed, for example, that age and edu- cation of the head has no direct impact on child schooling. 11 Our discussion of these effects is informal. For rigorous theoretical models of human capital accumulation treating the schooling demand effects of some of the factors we consider, see, for example, Barros and Lam 1992 and Behrman, Pollak and Taubman 1995. 67 P. Glick, D.E. Sahn Economics of Education Review 19 2000 63–87 more likely to recognize the benefits of schooling. Posi- tive parental schooling impacts are also expected from a schooling as a consumption good perspective, since bet- ter-educated parents are likely to enjoy educated children more than less-educated parents; thus mother and father education will act as taste-shifters in the schooling demand functions. Like parents’ education, household income will, under plausible assumptions for developing countries, posi- tively influence the demand for children’s education. Poor households may be unable to afford the direct or indirect costs of schooling and be constrained in their ability to borrow to cover the costs. Since wealthier households are likely to be able to pay for schooling out of current income or savings and have easier access to credit, children from such households are expected to be more likely to enroll and to stay in school longer. Income will also have a positive effect on schooling if education is a “normal” consumption good. 12 Gender differences in schooling—namely, greater schooling for boys—can come about because the returns to educating boys are greater than for girls or because the costs are lower, or because parents simply prefer edu- cating sons. 13 With regard to returns, if women are dis- criminated against in the labor market in terms of access to employment or in earnings, the monetary benefits to 12 Another potentially important determinant of the demand for schooling is school or teacher quality, which affects the labor market productivity benefits of schooling and possibly, through impacts on grade repetition, the costs of attaining a given grade level. Data on school quality are not available for this study; even if they were, it is uncertain whether sufficient cross-sectional variation would exist within the region surveyed urban and peri-urban Conakry to enable estimation of the effects of changes in quality on schooling demand. The same considerations apply to direct costs of schooling books, uni- forms, etc.. Thus we focus on the socio-economic character- istics of households. 13 More precisely, if direct impacts on parental utility of child schooling are ignored, so that parents care only to maximize their consumption, the first order conditions of the model imply that the relative marginal effects of female and male education on parents’ second period consumption will equal the relative marginal costs of educating girls and boys the ratio of opport- unity costs if direct costs are zero, i.e. ∂ C t 1 1 ∂ S d ∂ C t 1 1 ∂ S s 5 w d w s With diminishing marginal second period consumption returns to schooling, a reduction in the marginal benefit to schooling girls relative to boys the ratio to the left of the equality implies a lower optimal level of investment in girls’ education. Similarly, an increase in the relative opportunity cost for girls implies a reduction in girls’ schooling relative to that of boys. investing in their education will be lower than for boys. 14 As noted, there may be substantial returns to female schooling in non-market production, but parents may not be aware of these non-pecuniary benefits or may value them less than monetary ones. Even if educated girls go on to work and receive earnings on a par with men, income remittances to parents from married adult daugh- ters, who join their spouses’ families, may be lower than from adult sons. Finally, the returns to parents from edu- cating girls could be low because the quality of the schooling that girls receive is poor, reflecting school and teacher attitudes or interruptions in attendance or school- work resulting from girls’ household obligations. The last factor mentioned points to the possibility that the opportunity cost of educating girls is higher than for boys. Girls in developing countries are typically called on to perform more household work than boys, reflecting cultural or social attitudes toward the proper economic roles of women and girls. Given these attitudes, the mar- ginal cost of girls’ time will be higher than boys’ w d w s and consequently the demand for their school- ing will be lower. 15,16 For the same reasons, it would 14 More precisely, the incremental effects of schooling on expected earnings, determined by employment entry prob- abilities and wages, must be lower for women for this to be the case. Previous work on Conakry using this dataset Glick Sahn, 1997 was unable to reject the hypothesis that the returns to schooling in either wage employment or self-employment were the same for men and women, but did find a non-linear effect of schooling on women’s employment probabilities: rela- tive to no education, the likelihood of working fell with primary education and rose with secondary schooling and college levels which very few women in Guinea have attained. There appear to be relatively few opportunities outside of small-scale self- employment for women with just a primary education. 15 This may not be the case if there is also a strong demand for boys’ labor on the family farm or in a family enterprise, but neither of these possibilities is very relevant in the current context. Agriculture is not a major activity in the urban and peri-urban setting of this study, and participation of adolescent boys and girls in household enterprises and for that matter, in wage employment appears to be quite low Glick Sahn, 1997. 16 The distinction between “cultural attitudes” that shape con- straints and “household preferences” regarding the sexual division of labor in the household is not a sharp one, since the latter are shaped by the former. Consequently, the distinction maintained by economists between preferences and exogenous constraints is also blurred: what appears to be a manifestation of household constraints differential opportunity costs of boys and girls may instead ultimately reflect preferences about the allocation of time within the household. There is, however, one situation in which opportunity costs will appear to be higher for girls for reasons having nothing to do with household prefer- ences. Lower benefits especially in employment probabilities to female schooling may induce even parents who have no gen- der bias to have their daughters specialize early in household 68 P. Glick, D.E. Sahn Economics of Education Review 19 2000 63–87 also be expected that certain changes in household struc- ture will affect girls’ schooling more strongly than boys’. An increase assumed exogenous in the number of very young children, for example, may raise the demand for the labor of girls in childcare in the home. Given the time constraint, this will reduce their schooling relative to that of boys. By the same logic, additional older sib- lings or adult women may reduce the opportunity cost of a girl’s time by providing substitutes for household work or through economies of scale in household pro- duction, thereby raising the likelihood of enrollment or the average level of schooling among girls in the house- hold. 17 In the empirical work, we do not attempt to esti- mate endogenous shadow prices of time for girls and boys. Instead, the reduced-form schooling demand func- tions include family composition and other household factors represented by H in Eq. 5 that are expected to influence, possibly differentially by gender, the value of children’s time. Would the collective household model with individual preferences generate any different expectations about the impacts of the factors discussed above on girls’ and boys’ schooling? The model predicts that factors that raise the bargaining power of the wife should increase allocations to goods she prefers. The mother’s education stands out as one such factor for which the dataset pro- vides information. Women with more schooling are able to earn more, improving their fallback position and if they are actually working the level of income under their direct control. Thus, if women value the schooling of their children more than men do, maternal schooling will have a stronger impact than paternal schooling on children’s education. Further, mothers may prefer to allo- cate resources including for human capital to daughters while fathers prefer sons, as suggested by evidence for work, in preparation for their future likely roles as homemakers. As a result of this initial specialization and investment, girl’s home productivity will be higher than boys in later periods, hence their opportunity costs will be higher. This is akin to the “efficient specialization” hypothesis of Becker 1981. Thus the gender difference in the price of time is a manifestation of lower returns to schooling for girls, not parental preferences. 17 Note that this conflicts with a major implication of the quantity–quality model described in note 7: that child quantity and quality average schooling should be inversely related. An inverse relationship exists because having more children raises the cost of providing a given amount of education resources to each child, and conversely, a higher level of quality raises the implicit price of child quantity. However, in a context where child work is important, so that the cost of schooling includes the opportunity cost of foregone labor in the home or family farmenterprise, the relation of the number of children to aver- age educational investment is less obvious. As noted in the text, increases in the number of children in particular age or sex categories can lower the marginal value of an individual child’s time, encouraging school investments. child height in the US, Ghana and Brazil presented in Thomas 1992. Then increases in mother’s schooling would have a larger beneficial effect on daughters’ edu- cation than on sons’, and father’s schooling would favor sons’ education. The former is particularly plausible because the mother’s bargaining power and her prefer- ences for daughters’ schooling are both likely to rise with her own education. Note, however, that these relationships of maternal and child schooling are also compatible with unified household preferences. For example, a larger maternal education impact on girls’ education than boys’ may reflect maternal preferences for schooling girls in a bar- gaining framework or that households in which the mother has an education also have strong common pref- erences for girls’ schooling. The latter will arise through marital sorting if men who choose educated wives also want educated daughters or if educated women choose spouses who also have a preference for educating daughters, resulting in heterogeneity in preferences between households rather than within them. The prob- lem, as stated above, is that the dataset does not provide a measure of individual bargaining positions that would not also be a determinant of schooling under common household preferences. 18

3. Empirical approaches