Policy, Research, and External Affairs L WORKING PAPERS ! 1 ~~~~~~~~~~~~~~~~~b Women in Developn;snt Population and Human Resources Department The World Bank July 1991 WPS 725 How Structure of Production Determines the Demand for Human Capital Indermit S. Gill and Shahidur R. Khandker To promote gender equity, expansion of the services sector should be encouraged. But this runs counter to the World Bank and IMF policy of encouraging the production of tradable goods (produced mainly in agriculture and less so in industry) to service debt. So direct government intervention is needed to promote investment in women's human capital. ThePolicy,Rcscarch, andExternalAffais ('omplex distnbuies l'R WorkingPdperstodissermnatethefitidutgsof work Lnt rrogrcss and to encourage the exchange of ideds among l3ank staff and all othcrs -itrested in dCVclopMCnt issuCs. 'hcsc papers carry he names of the authors, reflect only thcir vicws, anid should be used and cited accordingly. Thc findings, iiiterpretaLions, and conclusionis are the authors' own. They should not be attributed to the World 1Bank, its lBoard of Difcctors, its management, or any of its member countnes Policy, Research, and External Affairs Women In Development WPS 725 This paper -a product of the Womeni in Development Division, Population and Human Resources Department -is part of a larger effort in PRE to detennine if and how women's productivity (and thus family welfare) are improved when women are given more access to education, training, credit, health care, andotherpublic resources. Copies are available free from the World Bank, 1818 H StreetNW, Washington DC 20433. Please contact Audrey Sloan, room S9-121, extension 35108 (43 pages, with tables). Explanations of lower investments in female Gill and Khandker test these propositions schooling and health than in male assume that using primary and secondary school enrollment demand for these components of human capital ratios and life expectancy levels (as proxies for somehow exists - and they concentrate on the investments and schooling and health) for about supply of human capital by the household. 90 countries in 1965 and 1987. The data tor 1965 appear to be broadly supportive of the proposi- Gill and Khandker try to remedy the neglect tions; data f)r 1987 support tlhem only weakly. of demand-side factors by examining exogenous dimensions of development. They include the The cmpirical analysis cannot determine structure of production - represented by the whether changes in the economic structure cause shares of agriculture, services, and industry in increases in the demand for education, or national employment or income - as an identi- whether improved education facilitates a largely fying variable for the demand for human capital. exogenous transition from an agrarian to an industrial/service economy. If issues of causal- Their reasoning is that the production ity are resolved in favor of the views Gill and functions of these three sectors differ in their Khandker express in this paper, interesting requirements for skills. Industry and services policy implications emerge. require more educated workers than agriculture does - and industry requires more full-time Most important, expansion of the services educated workers than services does. sector would greatly help reduce gender inequity at the same time as fostering growth. The authors assume that women have a comparative advantage over men in the home This finding highlights the problem with sector, so womeni spend more time at home. But relying purely on economic growth to reduce the industry favors males over females more than gender gap in human capital. If income growth the services sector does. If the importance of is accompanied by structural transformation of industry increases at the expense of agriculture, an economy from agrarian to industrial and then the demand for schooling will increase for both to domination by the services sector, there is no men and women, but especially for men. In- assurance that the economic status of women creases in the importance of services will will improve in the early stages of this transfor- similarly increase the demand for schooling mation. more for women than for men. Because the human capital of women has If health is equally valued by all sectors, but significant externalities - that is, because social health and schooling are complementary inputs returns to women's education and health are to production, changes in production that en- higher than private retums - the case is strong courage more schooling for men (or women) for direct goveniment intervention in invest- will also encourage more investments in health ments in women's human capital. for men (or women). The PRE Working Paper Serics disseminates the findings of work under way in thc Bank's P'olicy, Rcscarch, and Extemal AffairsComplex. Anobjectiveofthescries is to geithesefindingsout quickly, cven if presentations arlc ess than fully polished. The findings, interpretations, and conclusions in these papers do not necessarily represent official Bank policy. Produced by the PRE Dissemination Ccnter TABLE OF CONTENTS 1. Introduction 1 2. Analytical Framework 5 The Basic Structure 5 First-Order Conditions 11 Comparative Statics 13 - Price Elasticities of Schooling and Health 14 - Income Elasticities of Schooling and Health 15 - Sector-Share Elasticities of Schooling and Health 15 Introducing Gender Differences 17 - Differences between Industry and Services 17 - Health Investments, Rates of Return and Longevity 19 3. Estimation Results 23 Definitions of Variables 23 Disclission of Regression Results 26 - Primary School Enrollment Regressions, 1965 and 1987 28 - Secondary School Enrollment Regressions, 1965 and 1987 28 - Life Expectancy (Health) Regressions, 1965 and 1987 29 - Fixed Effects Regressions (1965-1987) 30 Su-mmary Statistics 33 Correlations Between Important Variables 34 Primary School Enrollment Cross-Section Results 35 Secondary School Enrollment Cross-Section Results 36 Life Expectancy (Health) Cross-Section Results 37 Secondary School Fixed-Effect Results 38 Life Expectancy (Health) Fixed-Effect Results 39 4. Policy Implications and Conclusions 40 References 42 Helpful discussions with Ken Chapman, Govind Hariharan, Barbara Herz, Ravi Kanbur, Randy Lutter and T. Paul Schultz are gratefully acknowledged. We also benefited from comments by participants at a World Bank seminar. 1. INTRODUCTION It is well known that women are generally less schooled than men. There is also scattered evidence that people invest less in girls' health, though this is generally reflected not in lower life expectancy but in the higher incidence of illness among women. Explanations of lower investment in human capital of women assume that demand somehow exists, and focus on determinants of supply of human capital by the household.1 We argue here that this neglect of demand side factors results in serious gaps in our understanding of the process of accumulation of human capital, and hence of economic development. We attempt to remedy this neglect by including "exogenous technological dimensions to the development process" (Schultz 1988) that differ over gender, time, and countries. Specifically, the structure of production, represented by shares of agriculture, industry and services in total employment or in gross domestic product (GDP), is included as an identifying variable for demand for human capital. We first develop an analytical framework within which some of the determinants of differences in the human capital of males and females in developing countries can be comprehensively examined. Then, some of the implications of the theory are tested for about 90 countries in 1965 and 1987. In future work, we plan to put the implications of this and competing models See Schultz (1989) for a survey. Empirical a lyses of gender differences in human capital confound the effects of (1) gender biases inherent in the utility function of parents, (2) gender differences in appropriability of returns to investments in children by their parents, and (3) gender differences in market returns to human capital. 1 to test using household, district, and provincial data.2 Following Schultz (1981), human capital can be assumed to consist of three components: bealth, schooling and on-the-job training. There are complementarities between these components. The most well documented, for the U.S. labor market, is that between schooling and on-the-job training. Complementarities between health and the other two components have not been empirically explored in detail. At the high levels of health observed in developed countries, investments in health do not add substantially to overall human capital. Since neoclassical labor economics until recently has been primarily directed towards developed economies, the relative neglect of health as a component of human capital reflects these low returns to incremental health. In poorer countries, given the high incidence of ill-health, increments in health substantially increase in the effectiveness of the other components of human capital. The lack of emphasis on health results in serious gaps in our understanding of the process of human capital accumulation, and hence of economic development. Just as investigating the complementarity between schooling and 2See Gill (1990) for household and province level evidence on these issues for Peru in the mid-1980s. 3This assertion relies on the following argument: Given the relatively high "levels" of health (as reflected, for example, in life expectancy, mortality and morbidity rates) in developed countries, marginal investments in health will not substantially effect the productivity of other forms of human capital. Hence the returns to investment in schooling and on-the-job training are not strongly correlated with changes in health levels. In less developed countries, given the low health levels, marginal investments in health will substantially increase the returns to schooling and on-the-job training. 2 on-the-job training requires the introduction of the firm as a decision maker, a formal study of the relationship between health and returns to schooling requires the introduction of household production. A larger fraction of health is produced within the household than either schooling or job-specific skills, Understanding the determinants of investments in health necessitates the study of household production. Since women are the primary agents in household (or nonmarket) production, the role of women becomes central to the study of accumulation of human capital. The issue of complementarity between health and schooling within the context of a household is thus a richer one than the study of this relationship at the level of an individual. Studies have documented that the health of children, as reflected in anthropometric measures such as height-by-age, weight-by-age and mortality rates, is positively correlated with the education of the parents (See Behrman 1990 for a survey). Less well documented but also significant is the finding that differences in schooling and health of sons and daughters are negatively correlated with the economic status (occupation and/or education) of women in the household. This paper incorporates decisions by parents regarding both the education and health of children into a model of the household that highlights the status of women. On this issue of women's status, the aims of this paper are: First, to make precise some of the claims and allegations made regarding the See, for example, Rosenmweig and Schultz (1982) and Gertler and Alderman (1989) for investments in health, and Gill (1990) for investments in schooling. 3 existence of bias against females in the allocation of resources within the household. The idea is to formulate these questions explicitly, so that it is possible to identify whether and to what degree there is evidence of this bias. Second, to identify causes of this bias, with the objective of isolating key factors that can be used for policy. In contrast to earlier studies that attempt to account for male- female differences in human capital, we do not assume any discrimination against females either at home (in the parents' utility function) or in the market (in the returns to human capital). All that is assumed in this paper is that women have a comparative advantage in working in some sectors of the economy. Thus increases in the shares of these sectors will increase the demand for female human capital.5 This explicit attention to factors that can be used as policy instruments -- and the relative neglect of factors reflecting gender bias in tastes -- is the point of departure from the earlier literature such as Gertler and Alderman (1989). The paper is organized as follows. Section 2 develops the theory. Section 3 tests these hypotheses using data from World Bank (1990), United Nations (1990), and from Summers and Heston (1988) for 1965 and 1985-87. Section 4 concludes the paper with a discussion of the policy implications. 5Differences in marginal effects of sector shares on investments in schooling and health are also indicative of segmentation of the market for labor. For example, if an increase of one percent in the share of services (at the expense of agriculture) raises the demand for schooling by more than a one percent increase in the share of industry (again at the expense of agriculture), then labor market segmentation cannot be ruled out. 4 2. ANALYTICAL FRAMEWORK A complete theoretical treatment of gender differences in human caiital would incorporate all three forms referred to above: health, schooling and job-specific training. This paper explicitly incorporates only investments in health and schooling. This is done because of two reasons: First, an effective treatment of on-the-job training requires a multiperiod model; and second, labor market studies for developed countries have already documented the complementarities between schooling and job training. The focus here is on complementarities between components of human capital accumulated at the level of the household, i.e., health and schooling. Because of these two reasons, investment in job-specific human capital is almost completely ignored. The Basic Structure A representative household consisting of an adult couple, one female child and one male child (indexed by subscript j - f, m respectively) is the unit of study. Fertility is therefore exogenous. Parents value only their own consumption and the welfare of their children as adults, which is assumed to depend upon each child's income. The parents' utility function is U - U (C, R , Rf; Z) . (1) where C is the quantity of a general consumption good, and R and R are the an f "full" or potential incomes of the male and female child respectively when they are adults, and Z represents household characteristics (tastes, education, location, etc.) that affect the utility obtained by consuming C, R , and R Assume for the sake of simplicity that the utility function has a 5 Cobb-Douglas form U - Z.RO.R7.C' (2) a f Full income depends upon human capital. Assume that human capital has two observable components: healtb and schooling. The returns to human capital functions are: R - R (S , H,) (3a) R - R f(Sf, Hf) (3b) where S and H represent schooling and health capital respectively. The budget constraint of the household is: Y - C + P s(S£ SO) + Pa(Hf+ Ha) (4) where Y is the exogenously decided income of the household, P and PH represent the price of schooling and health respectively, and the price of the consumption good has been normalized to equal 1. There are two major sectors of employment: home and market. The returns to human capital are sector-specific. Overall returns to schooling depend on the time allotted to household and to market (nonhousehold) activities. The market sector is divided into three subsectors: agriculture, services, and industry. Sectors are identified as follows: O - household activities 1 - agriculture 2 - industry 3 - services. Assume a Cobb-Douglas form for human capital earnings functiona: (Ei t ) (Ea t ) R- S H (5a) 6 (Fla tiL ) (Ea ftfL Rf - S f H £tL f H f (Sb) where t M and tfl is the fraction of time devoted to activity i by female and male children respectively when they are adults, and I t L - T (6.) itft - Tf (6b) In the simplest case, T, the total working life of a person, is assumed to be the same for males and females, i.e. T -T T, and T can be normalized to equal 1. The earnings functlons d:.splay the following properties: (i) The marginal returns to schooling and health are decreasing functions of the level of schooling and health respectively, i.e., aR i/aS > 0, aR/aH > O, and aBR/82S < 0, a2R/aH < O, J -m,f. (ii) Schooling and health are complements in the earnings function, i.e., a2R/lSBaH > 0, j - m,f. It is important to remember that t L and tL are not chosen by the parents. These are tiLe allocation decisions of the children when they become adults. The choice variables in the current framework are S , Sf, H , Hf, and C. Parents may lmputo the values of t and tfL on the basis of their own experiences and expectations of market conditions in the future when their children will work. There is, however, a self-fulfilling nature of this decision. If parents choose the schooling levels of the girl and the boy under the assumption of a set of t at nd tfL, then, if they are correct 7 about the returns to schooling in each sector, children cannot do better than allocate their time exactly as their parents expect-d or intended them to. The following assumptions are made: (i) fo t (ii) a - a - a - a - a, for both j - m,f. First, everybody gets married and has children, and women must spend more time at home than men. This could be because bearing and rearing children requires more women's time. Thus to could be the time required to bear and rear two children, since in our framework each household has one boy and cne girl.6 Even if there is no other difference between men and wowen, small differences in the productivity of males in child rearing compared to that of females can lead to large differences, even complete specialization, in work allocation patterns. (See, for example, Becker 1985). Alternatively, it can be explained as a cultural or institutional constraint. In either case, because of the time constraint (6), this implies that women generally have less time to spend in market activities. The argument that women must spend more time at home because of their comparative advantage in childbearing and rearing raises the issue of endogeneity of fertility, since women may choose not to marry or have children. In that case, from the viewpoint of this theory, the difference between men and women disappears. Incorporating fertility decisions by 6Changes in fertility would therefore change t . 0 8 determining family size within the model will change some~ of the results derived below. The second assumption is that return coefficients for health are the same across sectors. That is, health is equally necessary at home and in the market for effective completion of tasks. This can be tested very crudely by quantifying the extent to which ill-health prevents normal work activity in the market compared to activity at home.7 This assumption allows us to write equations (5a) and (5b) as follows: (Ea t ) a R -S ML H (7a) m m m (Za t ) a R u S £ 1i H (7b) f f f since a - a - a - a - a.s 0 1 2 3 On the basis of stylized facts, it is assumed that the returns to 'The returns to sector-specific work experience are difficult to measure, again because of the lack of "earnings" data in the home sector. Anilogous to the argument for schooling, higher technical change rates in the market may result in the returns to accumulated work experience being lower (due to obsolescence of previously learned skills) in the market than in household production. There is an important qualification: Technical change may require continulty of participation, so that work experience needs to be augmented by a measure of uninterruptedness of work participation. aThe private economic rate of return to schooling (according to the Mincerian approach) is thus dS S . E aJt . ,for- m, f, S ii0 i which is not a constant, but depends on the level of schooling S and sectoral time shares t 9 schooling are high in industry and services and low in agriculture. However, these may or may not differ by the sex of the workers: That is, there may or may not be sex discrimination in the market.9 Since the return coefficient of schooling at home, a 0, is not easily or directly observed, nothing is assumed about the magnitude of 0 relative to al, a2, and a. It is likely that returns to schooling at home and in the market are the same at low (primary) levels, and begin to diverge at higher levels of education (secondary and postsecondary). It is difficult to test this proposition. One way is to relate characteristics of sectors with the demand for education. Schultz (1975) suggested that education may be related to the ability to deal with job disequilibria. A well documented fitding in this context is the complementarity between technical change and education. Education may facilitate the implementation of new technology. (See, for example, Welch 1970 and Gill 1989). Then, if measures of technical change are available for the market sector and the home sector, we may be able to Infer relative rates of return to schooling in the two sectors.;° 9When it is assumed that these coefficients are the same for men and women, the point is not that there is In fact no discrimination against women. The rationale for this assumption is simply that discrimination (either at home or in the market) is a very difficult concept to quantify. Since the goal of this paper is to derive testable implications, we abstract from assertions that are unverifiable either in principle or in practice. 10The assertion that a decrease in labor market discrimination against women results in an increase in schooling levels of women assumes that returns to schooling are higher in the market than at home. As pointed out earlier, this assumption is not directly verifiable. However, the assertion that an increase in the returns to schooling in jobs where men have an "intrinsic" comparative advantage (for instance, mining) increases male-female schooling differentials is, within the above framework, not dependent upon any assumption regarding human capital earnings functions. It is claims like the 10 First-Order Conditions for Maximization Parents maximize their (one-period) utility function given in equation (2) subject to the budget constraint (4). The first order conditions for maximization are : Z.C .R .0.R- E a t S )H AP 0 (ea) f 3 LZ ML ML t -i8 Z.C'.R'8.-y.R" 1. z ttS f t)s AP 0(8b) 6 R- mi mL- Z.C .RP.a.R1.S H::ftf_ Ap _ 0 (SC) m £ CD. £ Z. C'.R8 y R'-' S ffLa.H &- A 0 (8c) z.c6 (f1t1 a- A?5- (d Z.R0.R0.6.C61 - A - 0 (Se) f m Y - C - P(S+ S) -P(H+ ) - 0 (8f) It is assumed that the second-order conditions for maximization are satisfied. That is, the utility function and returns to human capital functions have the requisite curvature. Using the first four optimality conditions, we obtain above that this paper seeks to make precise. 11 s (sa tL M- ML ML )&- R.Za t S H' R .S H £ L0 aL ML m f a 3 (S( t t-y) 7/j9, and (Et-) - W', R . E At S fL .Ha R .S fL fL HaI m -0 fL fLi f £ 5 Simple manipulation yields 3*S E a t HR S t-0 mL Mt m * - - , where asterisks denote optimized values. 3** a t .H S L-0 aft fLf Sf This equality tells us that the male-female schooling ratio is in general not equal to the male-female health ratio. If the time-weighted return to schooling is greater (smaller) for males, then the male-female health ratio will be smaller (greater) than the male-female schooling ratio. That is, if 3 3**** E eg t > (<) , a t then H /H < (>) S /S. L 0 mi mL Li0 ft ft m f m f We can solve for the demand functions for S , Sf, H , H and C. Schooling and health demand functions will be S* K pTK (9a) S - -.Kh (9b) f 6P S 8P H*- #-'a .K (9C) H H - _ .aK (9d) f 6P 12 where K - Y.6 6 - 8(a + Ea Mt ) - 7(a + EafLtf) is normalized income. Note here that 6 represents the response elasticity in utility with respect to consumption C, P(a + Ea t ) measures the response elasticity of utility with respect to investment in male children (in terms of schooling and health) and 7(a + Ea ftf) measures the response elasticity of utility with respect to investment in female children. If parents value their own consumption C more (less) than investment in their offspring, then 6 is greater (smaller) than (#(a + Em Mt ) + 7(a + EfLtf )]. Comparative Statics Comparative statics with respect to the prices Ps and P yield: aS '.Za t Y.6 m m mL (.lOa) 8ps 6.P: (6 - #(a + m itM) - 7(a + ZEaftf)] aS 2.> t Y.6 , _i fL. (10b) OP 6.P2 (6 - p(a + Zm t 7) - (a + ZEdt )] $ S mi MLf fii dH* P.a Y.6 - - (10c) aP 6.PH [6 - (a + Ea t ) - y(a + Ea t )J * H ~~~~~MLiML fi ft * OH - .a Y.6 f2* (10d) aP 6.P2 [6 P (a + Ed t ) - 7(a + Ea t )] a H MLiML fLifL If standard consumer theory applies, then the own-price effect 13 must be negative. This implies that 6 > [1(a + EaMt ) + ) (a + EtfJ, i.e., parents value consumption more than investment in their offspring. This does not, however, mean that parents must invest equally in male and female children. If parents value their son's human capital more than their daughter's, then the demand for education and health of boys will be less own-price elastic than that for girls. This would mean that | 5 8mJ | < JUL S8ap t and |P.dH | |P.03:11 | % H ape f P However, computing these own-price elasticlties with the functional forms assumed above yields P .aS* P .aS* S aPS S £ < -1, and P .01 P .8H so apD Bsf Ps H ae 3 H ' H OP HOP8 That is, the demand for education and health are equally price elastic for boys and for girls.11 Given the homothetic preference structure and return functions, parental investment in male and female children are equally price elastic. No gender bias exists as long as preferences are homothetic. 11The cross-price elasticities all equal 0: PS. a* P.aH* P.as* P.Oas pS Hf apS s H f H 14 Again, if parents value sons more than daughters, then the demand for girls' human capital (schooling and health) will be more Income elastic than that for boys. But in the framework above, Y .as Y .S _ --m_ 1, and Y .aH Y aH 0n f H aY ayH BY '- Again, no gender bias is revealed for investments in children's human capital. Comparative statics with respect to sector shares in time allocation (t L) yield: * ML Y Emtt t + +(6 - p(a + Ea Lt ) - y(a + Em ti)3 = - mImI mimIf2 U (02e) at 6L 6P [6 - p(a + Ea Et) - 7(a + Ea Ltf) aS SaY yEs t +t6 °(+ 't )- v(a + Et tL)l £ - - ~~~fltU t.d 15 (10f) at t 6P [6 - p(a + Em tM) - r(a + Ea f tf) aH BaY ID _ , ~~~~~~~~~~~2 (108) atML 6PH [6 - 0(a + Ea et) - 7(a + E t )td] OH5 7aY aI aH - -a S f~~~~ (10h) at f 61 16 - (a + Eamti) - y(a + Em Lt,)O1 For a rlse in tima allocation to an activity i, holdlng constant the time allocatlons to other sectors, to raise the demand for schooling and health, aS /Ot > 0 and aH /at > 0, respectlvely for i- 1,2,3; j -u,f. The condition that parents value their own consumption more than investment 15 in their offsprings' human capital, i.e., 6 - N(a + E it ) - (a + af tf ) > 0, is sufficient but not necessary to assure this. Thus an increase in expected working life (due to, say, an increase in longevity 12) will always raise the investment in schooling and health. The more interesting case arises if the increase in t is at the expense of time allocated to some other sector. The results above suggest that the implications for investment in schooling and health would depend only on the relative values of the schooling coefficient a in the two sectors. Thus if the schooling coefficient is larger in the sector whose share in time has increased than in the sector whose share has decreased, this would increase both investment in schooling and health. Differences in these coefficients across sexes may exist: * > *2 > 2 as /at < a s/at according as P a .Ea t < ay .aEa t ,and * ML. f f1 * L L ML mL fL Lfi fi * > *2 > 2 3H /dt < aHl/ftfi according as L a i < y afL' If women work more at home (sector 0), and if a0 is smaller than (or is perceived to be smaller than) aI, a2 ,and a3, then parents will invest less in the schooling of girls than of boys, even if there is no discrimination against female offspring either at home (i.e., P - y) or in the labor market (i.e., a M- fL' for all i). However, under the functional forms assumed above, investments in health will be the same for boys and girls. 12The next subsection examines this in greater detail. 16 Introducing Gender Differences The following issues are taken up in turn in this section: (1) So far, we have assumed that the demand for human capital in each of the three market sectors is equally gender-intensive. Thus if an education intensive sector (say, industry) increases in importance, this raises the demand for schooling of boys and girls equally. The framework is now extended to allow for systematic differences across sex to such shifts in the demand curve for schooling. (2) We have ignored a crucial difference between the effects of investment in schooling and those of investment in health. While investment in schooling increases the potential or "full" earnings of a worker in any period -- the quality of life -- investing in health results both in an increase in the quality of life as well as adding to the number of periods that a person is likely to live -- the quantity of life (See Ehrlich and Chuma 1990). (1) Differences Between Industry and Services The only distinction between m.a and women is that, in general 3 3 L-1 L-1 since tfo > t0; women must spend more time at home than men. Suppose now that experience in sector 3 (time allocated to sector 3 activities) adds to the returns to schooling in that sector, but the other sectors' rates of return to schooling are not dependent upon the time spent. That is: 17 a Cs (t C)s > 0 for j m,f (13) Restriction (12) implies that males will allocate more time to the industry sector than equally schooled females. This is because, holding educational attainment constant, women spend less time in market activities than their male counterparts. This implies that men have a comparative advantage in working in market sectors where the returns to schooling increase with time spent, since they have more time to spend in market activities. A simple example illustrates this. Suppose there are only two levels of education, high and low. Industry and services use only educated workers, while agriculture uses workers with low education levels. Experience is rewarded in industry and rat in services (a simplification). Then educated workers with higher amounts of allocable market time, males in the present setup, will be employed in industry. Educated females will be sorted into the service sector, where the returns to schooling do not depend upon time spent working (i.e., the returns to work experience per se are zero). Alternatively, we could assume some form of complementarity between the home and services sector. Suppose that time spent at home increases the productivity of a worker in the service sector but not in industry. Women will have an advantage in working in services since they spend more time at home than men. This advantage may be offset to the extent that sector-specific experience is important in the services sector, since women spend less time than men in nonhousehold activities. Yet another way to model this complementarity is to assume that it is easier to split time between home and services than between home and industry. This could be because it is easier 18 for service workers to work at or close to home. A plausible cause is that industry generally requires greater installed physical capital than the service sector, which is relatively labor intensive. This results in women having a comparative advantage in services. (See Smith and Stelcner 1990). Schultz (1989) reports the percentage of women in the labor force by sector. The relevant findings are reported in Table 1 below, both for sectoral employment and wage earners. These numbers support the assumption that a larger fraction of females is employed in services and commerce than in industry. This difference exists both in high income and in low income countries. So the assumption that women have a comparative advantage in the services sector seems to be robust for the entire spectrum of countries.13 The consequence of any of these assumptions is that an increase in the share of the industry sector will raise the demand for education of both males and females, but more for males. Conversely, an increase in the share of the services sector in total employment will increase the schooling demand of females by more than males. These implications of the theory can be tested. (2) Health Investments, Rates of Return and Longevity Incorporating effects of investment in health on the length of life can best be done by making total time available (T and Tf) a function of health levels: T - T (H), T' > 0, T" < 0 (14a) M £ T, - T (H f), T' > 0, T" < 0 (14b) isSchafgans (1990) finds similar patterns of female labor force participation in Peru during the mid-1980's. 19 Table 1 PERCENTAGE OF FEMALES IN THE LABOR FORCE, BY SECTOR BETUKE 1950-1982 (For the Four Main Sectors of Employment) Sector of Employment Percentage of Females In Labor Force Among Wage Earners High Low High Low Income Income Income Income Agriculture 32.7 28.4 18.9 31.5 Manufacture 29.8 25.8 29.7 20.7 Commerce 43.8 27.9 45.1 22.1 Services 52.0 31.8 53.5 31.8 Total 35.6 24.9 36.6 22.5 Source: Schultz (1989), Table 4. 20 Thus, investments in health increase the total amount of allocable time as well as (uniformly) increasing the returns in each sector. Both the level and the duration of returns are increased by investments in health. The interesting issue here is whether lifting the time constraint will affect the demand for human capital of males and females uniformly. To address this issue, assume that: (a) women do not allocate any of the additional time to household activities, so the choices facing both men and women concern only the market sector, (b) industry rewards work experience more than the service sector, and (c) all factors of production (health, schooling and work experience) are subject to diminishing returns. The effects of an increase in total time available T depend on the relative rates of return to work experience in each sector, and the curvature of sectoral production functions. To begin with, women allocate less time to market activities than men. Since experience is subject to diminishing returns in each sector, a one year increase in time available will raise the returns of women by more than men in each sector. In the market, health investments will therefore benefit women more than men. If industry rewards experience more than services, a larger fraction of the increase in time available will be allotted to industry. Thus, incorporating the effects of health on longevity results in: (a) larger increases in the market participation of women than men; and (b) larger increases in time allotted to industry than services for both sexes.14 14Here the service sector should be thought of as all nonagricultural market activities in which experience is not a factor of production. 21 There is also an interaction between health and schooling. A larger T can be regarded as a longer payoff period to investments in human capital. Under reasonable conditions, this increase in the payoff period will increase the schooling levels of both men and women. Since returns to schooling are perceived to be greater in the market than in household tasks, and since market participation of women increases by more when life expectancy increases, the schooling levels of women will rise by more. From a policy evaluation perspective, some of the gains in gender equity that are attributed to education policy should in fact be credited to health extension services. 22 S. ESTIMATION RESULTS Definitions of VarLables 8f Investment in schooling of children is represented alternatively by primary and secondary school enrollment as a percentage of the population In the age group that should be enrolled in primary school (aged 6 to 11 years) and secondary school (aged 12 to 17 years) respectively. For example, if male primary school enrollment in a country is 50,000, and the total number of boys aged 6 to 11 years is 100,000, then S equals 50 percent. Notice that this percentage could be m greater than 100 if a significant fraction of the population older than 11 years is enrolled in primary school. Therefore, high primary school enrollment ratios are not necessarily a good thing. The problem is less severe for secondary school, but it may nonetheless exist. To overcome this problem, we report results using ratios of female to male schooling as a dependent variable (Sf/S ). Since the denominators ln schooling enrollment ratios (the number of girls and boys in the age groups 6-11 and 12-17 years) are likely to be very similar in magnitude across sexes, female-male schooling ratios are likely to be more rellable than Sf and S singly. H £ Health investments are proxied by life expectancy at birth in years. The problems with treating life expectancy as a measure of investment in health include: (a) Life expectancy at birth ofteG depends on a different set of factors than life expectancy at older ages, due to infant mortality. (b) Gender differences in life 23 expectancy depend significantly on biological factors. (c) Variation in life expectancy rates across countries is very small compared to international differences in health levels as measured by morbidity, anthropometric indices, nutritional intake, and so on. (d) Life expectancy levels will at best capture investments in health aimed at increasing the "quantity of life" and not "quality of life". For these reasons, health regressions must be interpreted differently and more narrowly than schooling regressions. Y Per capita income is used to represent household income. For cross-section estimations for the years 1965 and 1987, these data were from World Bank (1990). For the pooled cross-section time-series estimations, per capita income data were from Summers and Heston (1988), for the years 1965 and 1985. This was done because pooled estimations require per capita figures in constant prices, and Summers and Heston report deflated per capita income (using three types of deflators) that are generally considered superior to price data in World Bank (1990). t The time allotted to agriculture, industry, and services (t1, t2, and t3 respectively) are proxied by their shares in total employment. The data for employment shares are from United Nations (1990). i 1sWe also experimented with sectoral shares in national income, using data for GDP shares from World Bank (1990). Technically, since t is interpreted as the probability of being employed in sector i, the share of each sector in total employment is the appropriate measure of t . One rationale for using shares in total Income (GDP) is that this measure' reflects both the probability and returns from employment in each sector. However, GDP share data are less reliable. Only the results of regressions using employment 24 PS PH The prices of schooling and health are proxied by the degree of urbanization: the percentage of population living in urban areas. The degree of urbanization is expected to be negatively correlated with both Ps and P . Schooling and health services are generally more easily accessible and of better quality in urban areas. Therefore, the larger the fraction of population living in rural areas, the higher are the average costs of obtaining educational or health services.16 Since Ps and PH are not available separately, the effects of changes in the relative price of schooling (health) on health (schooling) investments cannot be measured. As a result, the differences in price elasticities of these components of human capital cannot be studied.17 shares are reported in this paper. 16The price of schooling may be higher in rural areas for another reason. Rural areas are generally agricultural. If children are employed in greater numbers in agriculture (than in industry or services) because of the nature of tasks, the total price of schooling may be higher in rural areas due to the higher opportunity costs faced by children of schoolgoing age. 17Another difference between the prices of health and schooling arises because schooling requires time input of children while investments in health do not. We do not address this issue. 25 Discussion of Regression Results The equations for schooling to be estimated are: Sf - 4) + 4)Per Capita GDP + 42Industry Chare + 43Services Share + 4Urbanization + e f (15a) 8 mJA 0+ JAPer Capita CDP + p2Industry Share + 3Services Share + 4Urbanization + e (15b) Sf/S * pO + pIPer Capita GDP + p 2Industry Share + p Services Share + p4Urbanization + e (15c) The equations for health are: H - a- + oaPer Capita GDP + a2Industry Share + a 3Services Share + o4Urbanization + f (16a) H - r + r Per Capita GDP + r 2Industry Share + r Services Share + 4Urbanization + ehm (16b) H-/H - + vuPer Capita GDP + U2 Industry Share + v Servicea Share + v Urbanization + 6h (16c) To examine the correspondence between estimated schooling equations (15a) and (15b) and the schooling demand functions (9a) and (9b) respectively, rewrite any of the latter in its logarithmic form, e.g., Log(S) - Log(K) + Log(#Ea t ) - Log(6P ) (17) It can then be seen that, in the general forms estimated below, coefficients for sector shares (t1) capsule the magnitudes a and a. Relative values of these coefficients therefore depend only on a , the sectoral coefficients of schooling in the human capital earnings functions. Similarly, in the correspondence between estimated health equations 26 (16a) and (16b) and the demand functions (9c) and (9d) respectively, the coefficients for sector shares (t ) capsule the magnitudes P and aL. Relative values of coefficients for tL therefore depend only on aL the sectoral coefficients of health in the human capital earnings functions. Since we have assumed that a - a, for all i, the differences in these coefficients reflect only second-order effects (which are supressed by the assumption of a Cobb- Douglas functional form for the human capital earnings functions, but which are allowed for in the general formulation of the estimated demand functions). Differences in these coefficients could also reflect the impact of health investments on the time available (T), due to the longevity-increasing effects of these investments. This is discussed further in subsection (4) below. The two education intensive sectors (industry and services) are included in the regression, and the low education sector (agriculture) is excluded. If industry sector increases in importance relative to agriculture, holding the share of services constant, schooling demand increases. Similarly, if the share of services in employment rises at the expense of agriculture. with no change in the share of the Industrial sector, then the demand for education will unambiguously increase. However, if industry's share declines at the same time that the share of services rises, the effect on aggregate schooling demand is ambiguous. Hence agriculture must be the omitted class. Table 2 reports summary statistics for a sample of 91 countries in 1965, and 87 countries in 1987. Table 3 lists correlations between the main variables used in the above regressions. Table 4 provides results for equations using primary school enrollment ratios as the dependent variable, 27 Table 5 for those using secondary school enrollment, and Table 6 reports life expectancy regressions. Samples consist of all countries for which the data were reported. Table 7 and 8 report results for fixed-effects regressions for secondary school enrollment and life expectancy respectively. From the matrix in Table 3, we can see that the raw correlation between industry share and enrollment is higher for males than for females. The raw correlation betweeen services share and enrollment, on the other hand, is higher for females than for males. The correlations for health (last two columns) are remarkably gender neutral. The share of services in total labor force is, however, highly collinear with time (0.82), and the degree of urbanization (0.72), while the share of industry is not. Including time and the degree of urbanization will undoubtedly weaken the observed relationship between services share, school enrollment and life expectancy. Diagnostic checks for the regressions confirmed the existence of multicollinearity problems with regressions that include urbanization. Condition indices are greater than 100 (See Belsley, Kuh and Welsh 1980). When urbanization is not included, the problem of multicollinearity disappears; the condition indices drop to below 30. Since urbanization (representing Ps or PB) is an important variable, and must be included in regressions, we report both regressions with and without the degree of urbanization. (1) Primary School Enrollment Regressions: 1965 and 1987 The results for regressions that use primary school enrollment ratios as measures of investment in schooling of girls and boys provide some support for the theory. (See Table 4 below, especially the coefficients in 28 bold letters). While primary school enrollment numbers are relatively poor indicators of investments in education (see section on Definitions of Variables above), female-to-male ratios are better indicators of gender differences in these investments. The results of these equations for 1965 indicate that while increases in services share reduces the gender gap in primary school enrollment, the effect of industry share is insignificant. In 1987, however, both sectors become beneficial to gender equity, possibly reflecting a worldwide decrease in market segmentation during these 23 years. Given the degree of multicollinearity between the independent variables, high t-statistics for are especially reassuring. (2) Secondary School Enrollment Regressions: 1965 and 1987 The results of secondary school regressions (Table 5) below are identical in spirit to the primary school regressions. There are again structural differences in the equations for the two years, with industry becoming the more important equalizer of schooling investments of boys and girls in 1987. Per capita income also becomes an important influence in 1987, which is somewhat puzzling, since one would expect budget constraints to be more binding in 1965. But due to high multicollinearity, the estimated coe,"icients for per capita income and degree of urbanization are unreliable. (3) Life Expectancy (Health) Regressions: 1965 and 1987 Health regressions must be interpreted differently from schooling regressions, since effects of investments in health increase both the returns to human capital within a period and the duration of these returns. Life expectancy at birth is assumed to represent the probability of surviving a 29 given perlod, which depends ln turn on the amount of health lnvestment. Regressions using female-to-male ratios of life expectancy yield very poor fits (unadjusted coeffLcients of determination are less than 0.15), so only the female and male life expectancy regressions are discussed here. The results are remarkably gender-neutral for both 1965 and 1987. In 1987 even the intersectoral differences (between services and lndustry) become statistically insignLficant, although coefficients for females are usually somewhat larger. This is consistent with a decrease over time ln segmentation of the market for human capital.18 (4) Flxed-Effects Regresslons (1965-1987) To weed out country effects that may be present in yearly cross- section data, we adopt the following procedure: The mean of each variable for 1965 and 1987 was calculated for each country. Regressions employed deviations Regressions using shares of industry and services in total GDP as instruments for the structure of production are similar to results using employment shares. The main difference is that the coefficient for per capita income in these regressions is always positive and significant. We also estlmated regressions using both GDP and employment shares. These are not reported here for the sake of brevity. Employment shares pick up more of the variation in human capital than do GDP shares, though this is more the case for services than for industry. A likely explanation is that parents can observe the difference between probability of employment in services and agriculture (the omitted sector), but not the expected income in the services sector, since it generally consists of informal services. On the other hand, formal sector (industry) income possibilities are easier to observe. Hence, we should observe the GDP share for industry and employment share for services to pick up the effects of production structure on demand for human capital. The results provide some evidence consistent with these conjectures. 30 from the mean.19 Tables 7 and 8 list the results for schooling -- only secondary school regressions are reported for the sake of brevity -- and health regressions using pooled data for 1965 and 1987 (1985 for per capita income). The sample size was smaller (72) since it consists of countries that have the required data for both 1965 and 1987. Secondary school enrollment regressions provide strong support for the predictions of the theory. The services sector's share has a larger 19For details, see Binswanger, Khandker and Rosenzweig (1989). Lot D it be the set of variables that proxy the production structure in country j at time t (sectoral shares of industry and services in employment/GDP), and Dt - Dt ( t) (20) where U is the effect of unmeasured country factors, and e is a time- specific error term. Let C be the level of the ith human capital variable in country j at time t. Then jt jt it it 9t J S t) (21) where Y and Pjt are per capita income and degree of urbanization in country j at time t The simultaneity between C and Dj arising from their joint dependence on unobserved country-level variables U can be overcome if an additive model is estimated C c0 + c Y + c D + c P + c U + e (22) it 0 1 it 2 it e s t 4 5 5t The relationship for country means over time is C - c + cIY + c2DJ + cPJ +c4uJ +iJ (23) 5. 0 1 5. 2 i. S 5. 4 5 5. (3 Subtracting equation (23) from (22) we get (C -a ) -c (Y - t) +c(D - 6) + c(P ~+ (e -) (24) t J. 1 i+t J. 2 jt J. S jt S. jt J. If cj is random and uncorrelated with Dj , then this relationship can be e.timated using ordinary least squares. 31 positive effect on female secondary schooling than does the share of industry in employment. Industry share in fact has a significant unfavorable effect on investments in female schooling, and an insignificant effect for males. Per capita income has a higher coefficient for female school enrollment, though the coefficient is statistically significant only at the 10 percent level. Urbanization exerts a positive gender-neutral influence for both boys and girls. The time trend is statistically significant, and the regressions that use female-to-male enrollment ratio as the dependent variable indicate some equalization over time. Health regressions display relatively gender neutral results. Industry share is more important than that of services in determining health investments. The result is not surprising when the longevity enhancing effects of health investments are considered. (See discussion in section 2). Since industry is the sector where experience is valued more than in other sectors, increases in life expectancy (available time T) will increase the time devoted by both men and, to a greater extent, women to industry at the expense of services. So while increases in both industry and services increase the investments in health, the longevity enhancing effects of these investments are more important than the (within period) productivity of health as an argument in human capital earnings functions. The time trend in life expectancy dwarfs other influences, implying important omitted variables. Urbanization exerts a positive, gender-neutral influence on life expectancy. Regressions using female-male life expectancy ratios yield poor fits: Coefficients of determination are smaller than 0.06. 32 Table 2 MEAN & STANDARD DEVIATIONS OF VARIABLES, 1965 AND 1987 Variable 1965 1987 Mean S.D. Mean S.D. RPS Variables: Per Capita GDP (Current Prices) 547.15 760.22 4291.63 5860.05 Per Capita GDP (Constant Prices) 2526.00 4968.00 3428.00 3780.00 Share of Industry in GDP 27.63 13.18 29.73 10.79 Share of Services in GDP 44.22 10.46 49.49 10.66 Share of Agriculture in GDP 28.18 18.48 20.81 16.94 Share of Industry in Employment 18.00 13.80 14.30 8.20 Share of Services in Employment 26.00 16.00 43.20 23.60 Share of Agriculture in Employment 55.70 29.30 42.40 29.30 Degree of Urbanization 36.28 24.59 48.99 25.60 LNS Variables: Female Primary School Enrollment 68.36 35.24 86.00 30.66 Male Primary School Enrollment 84.87 29.80 95.68 24.43 Female/Hale Primary School Ratio 0.76 0.26 0.88 0.18 Female Secondary School Enrollment 19.76 21.71 46.68 34.02 Male Secondary School Enrollment 26.10 22.19 50.47 29.67 Female/Male Secondary School Ratio 0.60 0.32 0.83 0.33 Female Life Expectancy at Birth 55.16 11.97 64.94 11.35 Male Life Expectancy at Birth 51.82 10.81 60.66 9.99 Female/Male Life Expectancy Ratio 1.06 0.03 1.07 0.03 Notes: Enrollment percentages are ratlos of enrollment In school to the population Ln the relevant age group. 33 Table 3 CORRELATIONS asT Wi VARIADLES, 1965 AND 1987 y t t p s P itt P R S SEC sSE H H 2 3 t a a f m Time .24 -.34 .82 .84 .68 .51 .85 .87 .94 .93 Per Capita -.15 .27 .11 .09 .13 .28 .26 .13 .14 Income (Y) I.F Share of -.49 -.19 .07 .16 -.46 -.37 -.18 -.17 Industry (t 2) 1F Share of .72 .49 .37 .85 .81 .76 .75 Services (t3) Urbanization .68 .52 .75 .79 .85 .85 (P) Female Primary .RZ 87 .48 .60 .78 .80 School Enrollment (S ) f Male Primary PRI .32 .43 .59 .62 School Enrollment (S ) Female Secondary sac .95 .80 .78 School Enrollment (S ) Male Secondary sac .85 .84 School Enrollment (S ) Female Life 99 Expectancy (H ) Male Life Expectancy (H) 34 Table 4 CROSS-SECTION PRIMARY SCHOOLING REGRESSIONS, 1965 AND 1987 Dependent Independent Variables Variable PER CAP. SHARE IN EMPLOYMNT DEGREE UNADJ. SAMPLE GDP INDUSTRY SERVICES URBAN R-SQR. SIZE 1965 Female Primary -0.0007 0.9498 0.9810 .534 91 Enrollment (-0.90) (2.14) (2.37) Male Primary -0.0002 0.6477 0.7225 .404 91 Enrollment (-0.22) (1.52) (1.82) Female/Male -0.0000 0.0049 0.0073 .356 91 Primary Ratio (-1.35) (1.26) (2.02) Female Primary -0.0007 1.0574 1.1360 -0.1583 .536 91 Enrollment (-0.93) (2.12) (2.18) (-0.49) Male Primary 0.0002 0.6772 0.7650 -0.0434 .405 91 Enrollment (0.23) (1.41) (1.52) (-0.14) Female/Male -0.0000 0.0063 0.0094 -0.0021 .362 91 Primary Ratio (-1.40) (1.45) (2.05) (-0.73) 1987 Female Primary -0.0015 1.0020 0.7516 .409 87 Enrollment (-1.21) (2.24) (3.32) Male Primary -0.0016 0.4338 0.5827 .245 87 Enrollment (-1.42) (1.09) (2.88) Female/Male -0.0000 0.0074 0.0034 .443 87 Primary Ratio (-0.39) (2.93) (2.63) Female Primary -0.0019 0.9323 0.6996 0.1294 .411 87 Enrollment (-1.32) (1.99) (2.84) (0.55) Male Primary -0.0022 0.3087 0.4894 0.2319 .259 87 Enrollment (-1.77) (0.75) (2.23) (1.11) Female/Male -0.0000 0.0079 0.0037 -0.0008 .445 87 Primary Ratio (-0.09) (2.96) (2.64) (-0.58) * Intercept term not reported; t-statlstlcs In parentheses. Employment Shares are percentages. 35 Table 5 CROSS-SECTION SECONDARY SCHOOLING REGRESSIONS, 1965 AND 1987 Dependent Independent Variables Variable PER CAP. SHARE IN EMPLOYMNT DEGREE UNADJ. SAMPLE GDP INDUSTRY SERVICES URBAN R-SQR. SIZE 1965 Female Secondary -0.0001 0.8939 0.5109 .750 91 Enrollment (-0.23) (4.35) (2.67) Male Secondary 0.0001 0.9142 0.4768 .730 91 Enrollment (0.30) (4.17) (2.34) Female/Male -0.0000 0.0041 0.0138 .544 91 Secondary Ratio (-1.83) (1.03) (3.67) Female Secondary -0.0001 1.0138 0.6835 -0.1763 .755 91 Enrollment (-0.32) (4.44) (2.85) (-1.19) Male Secondary 0.0001 1.0033 0.6052 -0.1312 .732 91 Enrollment (0.24) (4.10) (2.36) (-0.83) Female/Male -0.0000 0.0069 0.0177 -0.0040 .557 91 Secondary Ratio (-1.95) (1.54) (3.78) (-1.39) 1987 Female Secondary 0.0035 0.7463 0.5940 .821 87 Enrollment (4.67) (2.81) (4.42) Male Secondary 0.0030 0.3780 0.5524 .757 87 Enrollment (3.90) (1.40) (4.04) Female/Male -0.0000 0.0156 0.0059 .468 87 Secondary Ratio (-0.63) (3.50) (2.63) Female Secondary 0.0034 0.7111 0.5678 0.0652 .822 87 Enrollment (3.95) (2.56) (3.87) (0.47) Male Secondary 0.0028 0.3473 0.5294 0.0569 .757 87 Enrollment (3.29) (1.23) (3.55) (0.40) Female/Male -0.0000 0.0154 0.0057 0.0005 .468 87 Secondary Ratio (-0.64) (3.29) (2.34) (0.19) * Intercept term not reported; t-statistics In parentheses. Employment Shares are percentages. 36 Table 6 CROSS-SECTION HEALTH (LIFE EXPECTANCY) REGRESSIONS, 1965 AND 1987 Dependent Independent Variables Variable PER CAP. SHARE IN EMPLOYMNT DEGREE UNADJ. SAMPLE GDP INDUSTRY SERVICES URBAN R-SQR. SIZE 1965 Female Life -0.0002 0.5685 0.2849 .848 91 Expectancy (-1.63) (6.48) (3.49) Male Life -0.0002 0.4698 0.2878 .828 91 Expectancy (-1.44) (5.55) (3.66) Female/Male Life -0.0000 0.0010 -0.0003 .089 91 Expectancy Ratio (-0.39) (1.79) (-0.52) Female Life -0.0002 0.6159 0.3532 -0.0698 .851 91 Expectancy (-1.72) (6.30) (3.45) (-1.10) Male Life -0.0002 0.5332 0.3792 -0.0933 .834 91 Expectancy (-1.57) (5.71) (3.87) (-1.54) Female/Hale Life -0.0000 0.0006 -0.0008 -0.0006 .113 91 Expectancy Ratio (-0.29) (1.02) (-1.23) (-1.35) 1987 Female Life 0.0009 0.3678 0.2209 .839 87 Expectancy (3.58) (3.58) (5.16) Male Life 0.0007 0.2823 0.2151 .&24 87 Expectancy (3.11) (3.60) (5.42) Female/Male Life 0.0000 0.0010 -0.0001 .126 87 Expectancy Ratio (1.12) (2.00) (-0.51) Female Life 0.0008 0.3474 0.2056 0.0379 .840 87 Expectancy Ratio (2.82) (3.94) (4.42) (0.85) Male Life 0.0006 0.2719 0.2073 0.0193 .825 87 Expectancy Ratio (2.56) (3.31) (4.79) (0.46) Female/Male Life 0.0000 0.0009 -0.0003 0.0003 .147 87 Expectancy Ratio (0.44) (1.59) (-0.96) (1.28) * Intercept term not reported; t-statlstlcs In parentheses. Employment Shares are percentages. 37 Table 7 FIXED EFFECTS SCHOOLING (SECONDARY SCHOOL ENROLLMENT) REGRESSIONS, 1965-87 (Sample consists of 72 countries: LDC8, Middle Income & Industrialised) Dependent Independent Variables Variable TIME PER CAP. SHARE IN EMPLOYMNT DEGREE ADJ. DUMMY GDP INDUSTRY SERVICES URBAN R-SQR. Female Secondary 14.3753 0.0004 -0.2694 0.5371 .787 Enrollment (7.04) (1.27) (-2.14) (5.11) Male Secondary 16.5344 0.0002 -0.0544 0.3469 .784 Enrollment (9.02) (0.77) (-0.48) (3.68) Female/Male 0.2062 -4.2e-6 0.0069 0.0027 .627 Secondary Ratio (7.72) (-1.05) (4.21) (1.95) Female Secondary 11.1035 0.0005 -0.3337 0.4910 0.2826 .792 Enrollment (4.31) (1.62) (-2.59) (4.61) (2.04) Male Secondary 12.6151 0.0003 -0.1315 0.2916 0.3385 .794 Enrollment (5.52) (1.25) (-1.15) (3.09) (2.76) Female/Male 0.1462 -2.3e-6 0.0058 0.0018 0.0052 .646 Secondary Ratio (4.40) (-0.57) (3.48) (1.33) (2.90) Notes: t-statlstlcs ln parentheses. 38 Table 8 FIXED EFFECTS HEALTH (LIFE EXPECTANCY) aEORESSIONS, 1965-87 (Sample consists of 72 countries: LDCs, Middle Income & Industrialized) Dependent Independent Variables Variable TIME PER CAP. SHARE IN EMPLOYMNT DEGREE ADJ. DUMMY CDP INDUSTRY SERVICES URBAN R-SQR. Female Life 9.5670 -0.0002 0.1683 0.0480 .905 Expectancy (20.27) (-3.50) (5.78) (1.98) Male Life 8.5593 -0.0002 0.1523 0.0404 .883 Expectancy (18.06) (-2.61) (5.21) (1.66) Female/Male 0.0087 -2.9e-' 0.0001 -0.0000 .060 Expectancy Ratio (2.34) (-0.51) (0.44) (-0.22) Female Life 8.4088 -0.0002 0.1455 0.0316 0.1000 .911 Expectancy (14.41) (-3.01) (5.00) (1.31) (3.19) Male Life 7.4572 -0.0002 0.1307 0.0248 0.0951 .889 Expectancy (12.68) (-2.13) (4.46) (1.02) (3.02) Female/Male 0.1113 -3.7e-7 0.0002 -7.3e-6 -0.0002 .058 Expectancy Ratio (2.36) (-0.65) (0.63) (-0.04) (-0.86) Notes: e-statlstlcs In parentheses. 39 4. POLICY IrPLICATIONS AND CONCLUSIONS For the purposes of policy making, the main weakness of this paper is that it does not forcefully address the issue of causality between the structure of production (shares of agriculture, industry, and services) and education. That is, the empirical analysis cannot determine whether changes in the structure of the economy cause increases in the demand for education, or whether increases in education levels facilitate a largely exogenous transition from an agrarian to an industrial/service economy. To determine policy, strong correlations between education and production structure are not enough; we need to know the direction of causality. This paper tries to finesse this problem by using enrollment rates in primary and secondary school, and not education levels of adults. In the case of health investments, similarly, we use life expectancy at birth and not health stocks of working adults. Gill (1990) also addresses this issue of causality by estimating the demand for education in Peru during the 1980s by age group. The schooling of children (younger groups) is strongly correlated with structure of production, but this correlation is found to be weak for older workers. It is therefore likely that the causality runs from production structure to schooling demand, and not the other way round. If this issue of causality is resolved in favor of the views expressed in this paper, some interesting policy implications emerge. The main implication is that service sector expansion helps to reduce gender inequity at the same time as fostering growth. This runs counter to the policy advocated by the World Bank and the International Monetary Fund that 40 developing countries encourage production of tradable goods (produced mainly in agriculture, and to a smaller extent, in industry) to service debt. For the purposes of promoting gender equity, the production of nontradables (service sector products) should be strongly encouraged. This finding also highlights the problem with relying purely on economic growth to reduce the gender gap in human capital. Theories of economic growth proposed by Rostow (1960) and Rosenstein-Rodan (1961) imply that income growth is accompanied by structural transformation of an economy from agrarian to industrial and then to one dominated by the services sector. There is no assurance that during the early stages of this transformation, the economic status of women will improve. 20 If the human capital of women has significant externalities (that is, if social returns to women's education and health are higher than private returns), this strengthens the case for direct government intervention in the process of investment in women's human capital. 20Though endogenous fertility, as mentioned in an earlier footnote, could change this result. 41 REFERENCES Becker, Gary S. "Human Capital, Effort and the Sexual Division of Labor." Journal of Labor EconomIcs 3 (1985): 533-558. Becker, Gary S. and Nigel Tomes. "Child Endowments and the Quantity and Quality of Children." Journal of Politlcal Economy 84 (1976): S143-S162. Behrman, Jere R. "Women's Schooling and Nonmarket Productivity: A Survey and a Reappraisal." World Bank manuscript, September 1990. Belsley, D.A., E. Kuh and R. Welsh. Regression Diagnostics, New York: John Wiley and Sons, 1980. Binswanger, Hans P., Shahidur Rhandker, and Mark Rosenzweig. 1989. "How Infrastructure and Financial Institutions Affect Agricultural Output and Investment in India." Policy, Planning and Research Working Paper No. 163, World Bank, Washington, D.C. Ehrlich, Isaac, and Hiroyuki Chuma. 1990. "A Model of the Demand for Longevity and the Value of Life Extension." Forthcoming, Journal of Political Economy. Gertler, Paul, and Harold Alderman. 1989. "Family Resources and Gender Differences in Human Capital Investments: The Demand for Children's Medical Care in Pakistan." Paper presented at the Conference on the Family, Gender Differences and Development, Yale University, New Haven. Gill, Indermit. 1989. "Technological Change, Education, and Obsolescence of Human Capital: Some Evidence for the United States." Unpublished Ph.D. dissertation, University of Chicago, Chicago. Gill, Indermit. 1990. "Does the Structure of Production Affect Demand for Schooling in Peru?" Policy, Planning and Research Working Paper No. 468, World Bank, Washington, D.C. Mincer, Jacob. 1974. Schoollng, Experlence and Earnings. New York: Columbia University Press for the National Bureau of Economic Research. Pitt, Mark M., Mark R. Rosenzweig, and Md. Nazmul Hassan. 1989. "Productivity, Health and Inequality in the Intrahousehold Distribution of Food in Low-Income Countries." Working Paper, University of Minnesota, Minneapolis. Pitt, Mark M. and Mark R. Rosenzweig. 1988. "Estimating the Intrafamily Incidence of Health: Child Illness and Gender Inequality in Indonesian Households." Working Paper, University of Minnesota. 42 Rosenstein-Rodan, Paul N. 1961. "Notes on the Theory of the 'Big Push'," in Economic Development for Latin Amerlca. Howard Ellis and Henry Wallich, eds. New York: St. Martins Press. Mark Rosenzweig and T. Paul Schultz. 1982. "Market Opportunities, Genetic Endowments, and Intrafamily Resource Distribution: Child Survival in Rural India.' American Economlc Review 72: 803-15. Rostow, Walt W. 1960. The Stages of Economic Growth: A Non-Communist Manifesto. London: Cambridge University Press. Schafgans, Marcia. 1990. "The Extent and Impact of Women's Contribution in Peru: A Descriptive Analysis," World Bank, Washington, D.C. Schultz, Theodore W. 1981. Investing in People: The Economics of Populatlon Quality. Berkeley: University of California Press. Schultz, Theodore W. 1975. "The Value of the Ability to Deal with Disequilibria." Journal of Economic Literature: 827-46. Schultz, T. Paul. 1988. "Education Investment and Returns." Handbook of Development Economics: Volume I. Hollis Chenery and T.N. Srinivasan, eds. Elsevier Science Publishers. Schultz, T. Paul. 1989. "Women's Changing Participation in the Labor Force,' Policy, Planning and Research Working Paper No. 272, World Bank, Washington, D.C. Smith, J. Barry, and Morton Stelcner. 1990. "Modeling Economic Behavior in the Informal Urban Retail Sector of Peru." Policy, Planning and Research Working Paper No. 469, World Bank, Washington, D.C. Strauss, John. 1986. "Does Better Nutrition Raise Farm Productivity ?" Journal of Polltical Economy 94: 297-320. Summers, Robert, and Alan Heston. 1988. "A New Set of International Comparisons of Real Product and Price Levels for 130 Countries, 1950-85." Review of Income and Wealth, Income and Wealth Series 30: 1-25. United Nations Development Pr4 razmme. 1990. t Development Report. Oxford University Press, New York. Welch, Finis R. 1970. "Education in Production." Journal of Political Economy 78: pages 35-59. World Bank. 1990. World Development Report. Oxford University Press, Washington, D.C. 43 PRE Working Paper Series Contact E e D~~~~~~~~~~~~~~1aLe foQr pSe WPS702 Should Price Reform Proceed Sweder van Wijnbergen June 1991 M. Stroude Gradually or in a "Big Bang?" 38831 WPS703 The Political Economy of Fiscal Sebastian Edwards June 1991 A. Bhalla Policy and Inflation in Developing Guido Tabellini 37699 Countries: An Empirical Analysis WPS704 Costs and Finance of Higher Rosemary Bellew June 1991 C. Cristobal Education in Pakistan Joseph DeStefano 33640 WPS705 What Causes Differences in Abby Rubin Riddell June 1991 C. Cristobal Achievement in Zimbabwe's Levi Martin Nyagura 33640 Secondary Schools? WPS706 Successful Nutrition Programs in Eileen Kennedy June 1991 0. Nadora Africa: What Makes Them Work? 31091 WPS707 Population, Health, and Nutrition: Population, Health, June 1991 0. Nadora Fiscal 1990 Sector Review and Nutrition Division, 31091 Population and Human Resources Department WPS708 Nongovernmental Organizations and Jocelyn DeJong June 1991 0. Nadora Health Delivery in Sub-Saharan Africa 31091 WPS709 An Empirical Macroeconomic Model Luis Serven June 1991 S. Jonnakuty for Policy Design: The Case of Chile Andres Solimano 39074 WPS710 Urban Property Tax Reform: William Dillinger June 1991 V. David Guidelines and Recommendations 33734 WPS711 Financial Reform in Socialist Millard Long June 1991 M. Raggambi Economies in Transition Silvia B. Sagari 37657 WPS712 Foreign Direct Investment in Thomas L. Brewer June 1991 S. King-Watson Developing Countries: Patterns, 31047 Policies, and Prospects WPS713 The Determination of Wages in Simon Commander June 1991 0. Del Cid Socialist Economies: Some Karsten Staehr 39050 Microfoundations WPS714 Women in Forestry in India Ravinder Kaur July 1991 A. Sloan 35108 WPS715 Promoting Girl's and Women's Rosemary Bellew July 1991 C. Cristobal Education: Lessons from the Past Elizabeth M. King 33640 WPS716 Financing Training: Issues and Christopher Dougherty July 1991 C. Cristobal Options Jee-Peng Tan 33640 PRE Working Paper Series Contact AJnhobf o Date WLr WPS717 Does Financial Liberalization Really Jacques Morisset July 1991 S. King-Watson Improve Private Investment in 31047 Developing Countries? WPS718 Impact of Investment Policies on Andrea Gubitz July 1991 S. King-Watson German Direct Investment in Developing 31047 Countries: An Empirical Investigation WPS719 How Trade and Economic Policies Ramon Lopez July 1991 M. Gunasekara Affect Agriculture: A Framework for Ridwan Ali 32261 Analysis Applied to Tanzania and Bjorn Larsen Malawi WPS720 The Outlook for Commercial Bank Ellen Johnson Sirleaf July 1991 S. King-Watson Lending to Sub-Saharan Africa Francis Nyirjesy 31047 WPS721 The Demand for Money in Developing Patricio Arrau July 1991 S. King-Watson Countries: Assossing the Role Jose De Gregorio 31047 of Financial Innovation Carmen Reinhart Peter Wickham WPS722 Is Rice Becoming an Inferior Good? Merlinda D. Ingco July 1991 P. Kokila Food Demand in the Philippines 33716 WPS723 Improving Women's and Children's Olayinka Abosede July 1991 0. Nadora Nutrition in Sub-Saharan Africa: Judith S. McGuire 31091 An Issues Paper WPS724 Fiscal Issues in Adjustment: Riccardo Faini July 1991 D. Ballantyne An Introduction Jaime de Melo 37947 WPS725 How Structure of Production Indermit S. Gill July 1991 A. Sloan Determines the Demand for Human Shahidur R. Khandker 35108 Capital WPS726 Perspectives on the Design of Anwar Shah July 1991 A. Bhalla Intergovernmental Fiscal Relations 37699 WPS727 The Effects of Debt Subsidies on Mansoor Dailami July 1991 A. Bruce-Konuah Corporate Investment Behavior E. Han Kim 80356