Policy, Research, and External Affairs e * * 2g-*e-V.4. LWomen In Development Population and Human Resources Department The World Bank August 1990 WPS 468 Does the Structure of Production Affect Demand for Schooling in Peru? Indermit Gill The more important the services sector, the more likely girls are to get more education. The more important industry, the more likely boys are to get more education. Both sexes get more schooling as the supply price of schooling falls, but girls gain more than boys do. The Policy. Rescarch, and Extemal Affain Complex distributes PRE Working Papers todisseminate the findings of wosk in progress and to enourage the exchange of ideas among Bank staff and all others interested in development issues. These papers carry the names of the authors. reflect only their views. and should be used and cited accordingly. The findings. interpretations. and conclusions asm the authors' own. They should not be attributed to the World Bank. its Board of Directors, its management, or any of its member countries. Polley, Research, and External Affairs Women In Development WPS 468 This paper - a product of the Women in Development Division, Population and Human Resources Department - is part of a larger effort in PRE to dtermine if and how women's productivity (and thus family welfare) are improved when women are given more access to education, training, credit,'health care, and otherpublic resources. Copies are available free from the World Bank, 1818 H Street NW, Washington DC 20433. Please contact Maria Abundo, room S9-123, extension 36820 (55 pages, including tables). Analyses of gender differences in investments in education more than agriculture does. Parents human capital typically emphasize family form expectations about the sector their children resources as the determining factor. These are likely to work in as adults and choose levels; studies usually find that investments in male of schooling accordingly.) offspring are greater, that these differences narrow as the level of household wealth in- - As services' share in GDP increases com- creases, and that equity is also affected by the pared to agriculture (holding industry's share composition of household wealth (proxied by the constant), girls' demand for schooling increases amount the mother earns and/or her educational more than boys' demand for schooling. level). An increase in industry's share in GDP Gill addresses one drawback of these analy- relative to agriculture (holding services' share ses: they do not explicitly consider the factors constant) is more closely associated with an that determine the demand for schooling and increase in the demand for schooling of boys health - other than tastes - and why this than of girls. differs for men and women. Gill uses the regional structure of the economy, proxied by the - A decrease in the supply price of schooling shares of services and industry in regional gross increases the level of schooling attained by both domestic product (GDP), as an indicator of the sexes, but the gain is larger for women. demand for educated workers. By examining whethere level of schooling as a function of Increases in wealta, all else being eqZal, are shares of services and industry differs for men associated with increases in both sexes' domand and women, he looks for gender bias in the for schooling. demand for schooling. Gill estimates schooling demand functions for males and females using What are the policy implications of these household data from the Peruvian Living Stan- findings? Some ways to increase educational dards Survey, and provincial data from the levels, especially those of women, include (on Peruvian censis. the supply side) lowering the supply price of schooling - improving access to secondary Gill's primary findings are: schooling, for example - and (on the demand side) expanding the services sector. The As services and industry increase as a share demand- side prescription contradicts the World of GDP, relative to agriculture's share, the Bank and IMF policy advice that developing demand for schooling increases for both boys countries foster the growth of tradables to service and girls. (Both industry and services reward their external debt. rer ky yt work's altcDissesemr and Ceoenl AffairsComplex. ~ ~ ~~ are Anbetvo hsre st e hs idnso to agrky,evnicultre(htolng iesnurysished 'efinins, ntrprtaion, nd onluson inthse aprs oconesarlresnt) girlide and fork scholinicrese )rodmore tha boys deman foreiato scholing The Structure of Production as a Determinant of the Demand for Human Capital in Peru by Indermit Gill Table of Contents 1. Introduction 1 2. The Theoretical Framework 6 3. Some Theoretical Extensions 12 Experience as a Factor of Production 13 Forming of Expectations 14 4. Household Level Empirical Evidence 17 Definition of Dependent Variable: Schooling Shortfalls 17 Definitions of Independent Variables 19 Ree-Its of the Schooling Regressions 23 5. Province Level Empirical Evidence 35 Definitions of Dependent Variable: Illiteracy Rates 35 Definitions of Independent Variables 36 Description of the Data 38 Results of the Schooling Regressions 41 6. Conclusions and Policy Implications 51 References 53 Appendix 55 Analyses of gender differences in investments in human capital typically emphasize family resources as the determining factor. (See Schultz and Rosenweig (1982), Gertler and Alderman (1989) for investment Xn health, and King and Bellew (1989) for investment in schooling.) These studies approach the problem as one of investment by parents in the human capital of male and female children. These studies usually find that investments in male offspring are greater, that these differences narrow as the level of household wealth increases, and that the composition of household wealth (proxied by either the amount earned by the mother and/or her education level) affects equity as well. There are two major drawbacks in these analyses. First, the empirical segments confound the effects (a) of a -ender bias inherent in the utility function of parents, (b) of gender differences in market returns to human capital, and (c) of gender differences in appropriability of returns to investments by parents in children. This paper does not address this issue. Second, these analyses contain no explicit consideration of the factors determining the demand for schooling and health, other than tastes, and why this differs for males and females. It is the second shortcoming Mr Gill is an assistant professor at the State University of New York at Buffalo and is a consultant to the World Bank. The author is grateful to Barbara Herz, Emmanuel Jimenez, Shahid Khandker, Thomas Mroz, John Newman, Pats Pattabiraman, Maurice Schiff, T. Paul Schultz and Jacques van der Gaag for helpful discussions, and to Marcia Schafgans and Ricardo Lago for help with the data. 1 that this paper seeks to rectify. In this paper I use the regional structure of the economy, proxied by the shares of services and industry in regional gross domestic product (GDP), as an indicator of the demand for educated workers. By examining whether the level of schooling as a function of shares of services and industry differs for men and women, we can detect gende bias in the demand for schooling. Based on the theoretical framework developed in Gill and Khandker (1990), I estimate schooling demand functions for males and females using data for Peru in the 1980s. A separate estimate covers households and provinces (called "departments" in Peru). Household data are from the Peruvian Living Standards Survey (PLSS); information on the provinces is based on census data. Findings confirm the results obtained in Gill and Khandker using country-level data for about 100 countries in 1965 and 1987. The primary findings are: o As services and industry increase their shares of GDP, relative to the share of agriculture, the demand for schooling of both males and females increases. * As the share of services in GDP increases compared to agriculture (holding the share of industry constant), the demand for schooling by women increases more than the demand for schooling by men. IIt should be mentioned here that the absence of explicit consideration of demand-side factors in the market for labor is a weakness of much of neoclassical labor economics. 2 o An increase in the share of industry relative to agriculture (. ding the share of services constant), is more closely associated with an increase in the demand for schooling of men than of women. o A decrease in the supply price of schooling increases the level of schooling attained by both sexes, but the gain is larger for women. o Increases in wealth, ceteris paribus, are associated with increases in the demand of both sexes for schooling. The plan of the paper is as follows: Section 2 introduces the basic theory. A representative family is assumed that has an adult couple and one female and one male child. The issue of fertility is thus entirely sidestepped. Parents are assumed to be the decision-makers regarding investments in human cap .tal of children. They do so because the attained utility of their chil..cen matters to them. Attained utility depends upon the income of children as adults and income in turn depends upon the human capital was invested in them by their parents. The demand for schooling of children is focussed upon. Schooling is demanded differentially in different sectors of the economy: industry and services reward education more than agriculture.2 Parents form expectations about the sector of the economy 2Schultz (1975) argues that this demand for education reflects the higher rates of change in industry. Mincer and Higuchi (1988) and Gill (1989) find that sectoral rates of technical change in the U.S. economy between 1960-1985, were related positively to the rates of return to education. Welch (1970) found similar relationships in U.S. agriculture. These results imply that the rates of return to schooling, and therefore the demand for educated workers, would be highest in industry (especially manufacturing), lower in trade and services, and lowest in agriculture. 3 that their children are likely to work in as adults, and choose levels of schooling for each child accordingly. In the basic model, parents use their own time allocation as a proxy for the time-allocation patterns that their children will choose. In section 3 this last assumption is amended. Parents forn Apectations of time-use of children as adults based on both their own work experience and the general pattern in the region of residence, as well as the probability of migration to othr regions. Section 3 also discusses the implications of adding sector-specific work experience (job training) as an additional component of human capital. Section 4 uses household da. a from the PLSS to test the implications of the theoretical framework developed in sections 2 and 3. The advantages of using household data are that the schooling attainment of children is directly observed, and the effects of intrahousehold factors on the demand for schooling of boys and girls can be accounted for by including household information such as the education and occupation of parents. In section 5, I test the implications of the theory using provincial data for 25 departments in Peru in the 1980s. Illiteracy rates are used as a proxy for investment in schooling. I conduct tests to ensure that department illiteracy rates are satisfactory measures of department schooling attainment. The findings confirm the main implications of the theory, and add to the evidence from household analysis. Section 6 discusses the policy implications of the study. The policy implications are of two types: Supply-related and demand-related. 4 Supply-relaced policy prescriptions are those targeted towards lowering the supply price of schooling, such as improving the access to secondary schooling. Derard-related policies aim at increasing the demand for education: The main policy recommendation entails the expansion of the services sector. This contradicts policy advice given by the World Bank and the INF that developing countries foster the growth of tradables to service their external debt. Another policy implication emphasizes the importance of information about the rates of raturn to schooling in the market and the home sector. 5 2. THE THEORETICAL FRAMEWORK Human capital is assumed to consist of two components: schooling and health.3 Parents value only their own .onsumptior and the attainable utility -- that is, the full income -- of their children as adults. Asr. that a couple has only one female and one male child.* The parents' utility function is U - U (C, R,R) . (1) t an where C is the quantity of a general consumption good consumed by the parents, and R and R are the "full" incomes of the girl and b.; respectively when they are adults. R depends on the human capital of children. Human capita' has several observable components, e.g., schooling, health and job tZaining. I assume here that schooling (S) is the only form of human capital. The terms human capital and schooling will be used interchangeably unless otherwise indicated. Thus the returns to human capital functions for the girl and the boy are R - R (Sf) (2a) R - R (S ) (2b) 3The theoretical framework for this paper departs considerably from previous theoretical attempts to analyze gender differences in investments in human capital, such as Gertler and Ald rman's (1989) analysis of investments in health. 'The issue of fertility is thus sidestepped. See Becker and Tomes (1976) for a theoretical discussion of fertility and quality-quantity tradeoffs, and Schafgans (1990) for an empirical treatment using Peruvian data. 6 The budget constraint of the household is Y - C + Ps(St+ SM) (3) where P represents the price of schooling, and the price of the consumption good has been normalized to equal 1. There are two main sectors of employment: home and the market. The rates of return to human capital are sector-specific. Thus overall returns to schooling depend upon the extent to which time is allotted between the household and the market (all nonhousehold) activities. In this paper market activities are subdivided into agriculture, services and industry, identified as follows 0 - Household 1 - Agriculture 2 - Services 3 - Industry Total returns to human capital are a time-weighted sum of returns in each sector. I assume Cobb-Douglas return functions R - t So + ( tSP + tS + t S6 ) (4a) t0 t 1 £ £2 t f3 f R - t.S' + ( tS + tSf + t S' ) (4b) M m0 M Ml a m2 m m3 m where t and t is the fraction of time devoted devoted to activity i by female and male childcen respectively when they are adults,5 and Alternatively, t can be interpreted as the probability of the child being employed in sector i as an adult. 7 3 3 Et E t -1,(5 The following assumptions are made: (I) te to : Everybody gets married and has children, and women spend more time at hom-.. than men. This could be because bearing and rearing children is more demanding of women's time. (See, for related examples, Becker 1985). This.raises the issue of endogeneity, since women may choose not to have children. In that case, from the viewpoirt of this theory, the difference between men and women disappears. Alternatively, it can be explained as a cultural or institutional constraint. In any case, because of time constraint (5), women generally have less time for market activities than men. (ii) 0 < 1 5 6 : The rates of return to schooling are high in the industrial and service sectors and low in the agricultural sector. However, these rates of return do not depend upon the sex of the workers: There is no sex discrimination in the marketplace.7 Since the productivity of schooling at home, a, is not easily or Although I do not have measures of the rates of return to schooling in agriculture compared to the other two sectors, there is strong evidence for Peru that supports this assumption. Schafgans (1990) finds that labor force participation of both men and women in agriculture declines with education, and participation in the nonagricultural wage sector (sectors 2 and 3 in this paper) increases as the education of workers increases. 'The point is not .hat 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 di.,ficult concept to quantify. Since the focus of this paper is empirical, I abstract from assertions that are unverifiable either in principle or in practice. 8 directly observable, nothing is assumed about the magnitude of a relative to #, y, and 6. It is important to remember that t and t are not choice variables for the parents. These are time-allocation decisions by children when they become adults. The only choice variables in the current framework are S , and C. Parents may Impute the values of t and t from their own experiences and expectations of market conditions when thair children will work. This point will be discussed later. Parents maximize their (one-period) utility function given in equation (1) subject to the budget constraint (3). The first order conditions for maximization are tU a- -1+' 6-1 aR . t + t ,j ty S + t S S )3 - AP - 0 (6a) 1 0 t 1 t +2 ( t t 86a 81U a-1 (- 7-i 6-i aR *tea S + ( t M S + t2 S + t 6 S )] - AP - 0 (6b) 8U -A - 0 (6c) BC This assumes that there is no bargaining about transfers from the children to the parents. Suppose instead that contracts bind children to support their parents when they are old. Then parents may in fact decide tef and t as well. $Notice that there is a self-fulfilling nature to the parents'decision. If parents choose the schooling levels under the assumption of a set of tes and t M, and if they are correct about the rates of return to schooling in each sector, children cannot do better than allocate their time exactly as their parents expected. 9 Y - C - P (Sf + S ) - 0 (6d) We can solve for the demand functions for S , S and C. The schooling demand functions will be of the form S - S (Y, P , t,t, t , t ) (7a) S - S (Y, P , t , t , t , t ) (7b) mS 0 1 2 3 where t , i - 0,1,2,3 are combinations of t and t for each i. Note iIf Lm that since both parents have the same utility function, only aggregate t enter the schooling demand functions. These can be written as Sf - S (Y, P, t t, t) (8a) ff S 1 f. 3 S - S (Y, P , t ,t ,t). (8b) M M 8 1 2 3 Equations %3a) and (8b) incorporate the additional constraint faced by women, t f t (that they must spend at least a fixed fraction i of to 0 0 their time at home), in the functional form, S (.). Since females are expected to marry and become mothers, the functional form of schooling demand will differ from that of males. Other than this restriction, however, there is no difference between boys and girls. The signs of S s8P and aS/ BY (and OS/ aP and aSm/ aY ) are predicted by standard consumer theory as being negative and positive respectively, if quality of children is a normal good. Assuming that parents value the hppiness of male and female children equally, the theory 10Following Becker (1981), parents are altruistic toward their children. 10 predicts that these coefficients will be roughly equal in magnitude if the curvature in the utility function is small. That is, if the second derivatives, 82S/ Ops2 and 82S/ 8Y (and 82S/ aP and St/ BY ) are close to zero. If these coefficients are different for males and females, and since we know that mean levels of S t are small compared to S M, this is indicative of curvature in the utility function. However, the theory as it stands contains no predictions regarding the magnitudes of S/ at1, CS/ at, and S/ St relative to 8S*/ at, 3n W aS at and aS*/ at respectively. More structure is needed to determine a 2 m 3 whether, for example, an increase in the time spent by parents in sectoLs 2 and 3 (relative to sector 0 and/or 1) increases or decreases the demand for children's education and whether these magnitudes of response are different for females than for males. 11 3. SOME THEORETICAL EXTENSIONS The theory predicts that within a region, as a sector with a relatively high rate of return to education increases its share in total employment, demand for the education of children in that region will rise.1 The limitations of the framework are: (1) The theory contains no implications for the gender composition of this increase in demand for schooling: It-assumes that if an education- intensive sector (say, industry) increases in importance, the demand for schooling of boys and girls will rise symmetrically. The theory is now extended to allow for systematic differences across sex to such shifts in the demand curve for schooling. (2) The theory assumes that the local structure of the economy (t1, t 2, and t of the parents) determines the demand for education. This raises the question of how parents form expectations regarding their children's future. Issues such as the likelihood and ease of migration may be significant, and these factors may not be gender-neutral. Also, since the rate of return to schooling in the home sector, a, is not observed directly (that is, in terms of wages), perceptions regarding a may be related to household attributes such as the education of the mother. 11 If the industrial sector increases in importance, relative to services and/or agriculture, we expect schooling to rise. Similarly, if the service sector's share rises relative to agriculture, with no change in the share of the industrial sector, then the demand for education will unambiguously increase. However, if the share of industry declines at the same time that the share of services rises, the effect on aggregate schooling demand is ambiguous. 12 Bxperience as a Factor of Production The only distinction between men and women is that, in general 3 3 t t < t c i (9) since t to > t ; 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 on the time spent. That is 6 - 8 (t ), 6' > 0 (10) So the returns to schooling functions (equations (4a) and (4b) ) are rewritten as R - t So + [ tS + tS 7 + t S6(tfs)] (11a) t o 0 ti $1 t 2 f D3 t R - t So + [ tSf + t S7 + t S 6(ts)] (11b) a 6 m am2 m m3 m Combined with restriction (9), equations (Ila) and (1lb) imply that males will allocate more time to the industry sector than equally schooled females because, holding educational attainment constant, women spend less time in market activities than males. This implies that men have a comparative advantage in working in market sectors where the returns to schooling increase with time spent.12 For example, assume there are two levels of education: high and 'aAlternatively, the comparative advantage of women in services may simply be due to the fact that servicet provide more opportunities to work close to home than does industry (Smith and Stelcner 1990). 13 low. Industry and services use only workers with high levels of education, while agriculture uses workers with low levels of education (regardless of sex). Workers with high levels of schooling will work in the other two sectors. Assume that experience is rewarded in industry and not in services (a simplification). More schooled workers with higher amounts of allocable market time (males) will be employed in industry. More schooled females will work in the services sector, where the returns to schooling do not depend on experience. One consequence is that an increase in the industrial sector's share of GDP will raise the demand for education by both males and females, but more for males. Conversely, a rise in the service sector's share in total employment will increase the demand for schooling by both male and females, but more for females. These are testable implications of the theory. Forming of Expectations The model assumes that parents, in deciding how to educate their offspring, base the decision on the amount of time the parents spend on four areas: household activities, agriculture, services, and industry, and the rates of return to schooling in each of these sectors. For example, they expect daughters to allocate time the same way as the mother, and sons to follow their father's patterns. They further expect that the relative rates of return to schooling in each sector will remain unchanged. In the context of Peru in the 1980s this assumption is likely to be incorrect, since a 14 large flow of migrants moves between rural and urban areas. This sub-section modifiies the theory to allow for migration. Parents have some information about potential work opportunities for their children (other than their own occupations) and incorporate this information into the decision on their children's education. So, for example, parents may use the average levels of t and t in the region or fi mi country to determine the probability of employment of female and male children in sector i. Parents may use ti (averaged over both sexes) in deciding how much to educate their children. This may entail migration by children when they are adults; this migration may or may not be gender-neutral. Vector Z is added as an argument in the two schooling equations S - S (Y, P , t , t , t , 2) (12a) f t S 1 2 3 S - S (Y, P , t , t , t , Z). (12b) m a S 1 2 3 where Z represents both "infrastructural" variables (Z) that represent the ability of adults to migrate, and household-specific variables (Z2) that represent the ability of the parents to decode information regarding potential opportunities for children. Z includes both factors that determine the degree to which employment in each sector is perceived as possible (tf and tai, for i - 0,1,2,3) and factors that determine the accuracy with which parents observe the true rates of return to schooling (a, fi, 7 and 6). In the empirical sections, both equations (8) and (12) will be estimated to see whether or not these variables Z have independent and 15 gender-variant effects on investment in schooling. These equations were estimated both by ordinary least squares (OLS) and two-stage least squares (TSLS) techniques to account for possible endogeneity of household income. The results were very similar. In this paper only the TSLS results are reported. 16 4. HOUSEHOLD LEVEL EMPIRICAL EVIDENCE Analysis at the household level has several advantages. First, since the theoretical model is one of household decisions, empirical testing should be done at the household level. Second, it is possible to quantify the effect of intrahousehold distribution of potential or actual earnings (for instance, the education or earnings of the mother compared to the father's) on investment in the human capital of children. Third, the measure of investment is more reliable for the households than the illiteracy rates used in the next section, and the primary and secondary school enrollment rates used in Gill and Khandker (1990). The sample should ideally consist of households in the PLSS that have at least one male and one female child of school age. But when the sample was restricted to such households, the number of observations dropped significantly. Definition of Dependent Variable: Schooling Shortfalls The problems in comparing schooling levels of boys and girls are: o Each boy and girl is likely to have different levels of access to schooling, such as household income, distance from school, quality of schools available, and so on. o The boy and girl being compared are likely to differ in age, and hence on this account alone would differ in years of completed schooling. Thus we need to use a measure of educational attainment adjusted for age. o A related complication is that time or cohort effects will be confused 17 with the true schooling differentials. Suppose, for example, that the government makes primary school attendance compulsory for all children between 5 to 10 years of age in year t. Then comparing the adjusted schooling attainment of a girl who is 10 years old in year t+5 with a boy who is 15 in the same year is likely to understate the actual difference in the schooling of girls. Conversely, comparing a boy who is 10 in year t+5 with a girl who is 15 in the same year will overstate this difference. It is important to weed out these cohort-specific effects. To resolve the first problem, we have included family characteristics in the regression. Regarding the second problem, the comparison is not between attained schooling but shortfalls in schooling attainment of boys and girls. This shortfall, for a child j, is defined as (Age - 5) - (Schooling in Years) - (Schooling Shortfall) , for j - f,m. That is, the shortfall is equal to potential schooling (age-5) minus actual 13 schooling (alternatively, the highest grade completed). The nice thing about this measure is that it is a familiar one: it is identical to Mincer's (1974) definition of potential work experience for adults. The only 1Within the household, it is also possible to compare the quality of schools attended by male children versus the quality of schools attended by female children. For example, in the PLSS, it is possible to determine whe.her the child attended a private or a public school. Indices of quality of schooling of and school-related expenditure on male and female offspring can be incorporatsd to make the schooling differential variable better approximate differential investment in schooling rather than differential schooling attainment. 18 difference is in its application. We deal with the third problem by estimating cohort-specific schooling equations, estimating separately for each of five age groups: 6-15 years, 16-25 years, 26-35 years, 16-45 years and 46-65 years. Definitions of Independent Variables Y: Household income is proxied by using total (food and nontood) expenditures in the household. The advantages of using expenditures rather than income are that: (a) expenditures are less subject to errors in reporting, since they are reported by component. Each component (food, clothing, and so on) is less likely to be systematically under- or over-reported; and (b) total expenditures are a better proxy for permanent income, which is generally the relevant budget constraint. The disadvantages are that: (a) since expenditures on schooling are a component of the total, the issue of endogeneity of a right-hand sidf variable becomes a problem; and (b) schooling levels and 14Ideally, to purge cohort effects from the comparison of male and female schooling levels, the following procedure should be adopted. The average schooling shortfall for male children in each cohort group is first calculated. Then the difference of the schooling shortfall of each child i, male and female, from the average schooling shortfall of male children in the cohort that child i belongs to, is defined as M(t) (Schooling Shortfall) - E (Male Schooling Shortfall) / M(t) - S, J k-1 k j j - m, f; and k - 1, 2 ......, M(t) where j denotes the male children in the cohort, t, that i belongs to, and M(t) is the number of males in cohort t. 19 household income may be jointly determined by other variables, and this results in simultaneous equations bias. (See Schultz 1989 for a discussion of this problem for fertility decisions.) To overcome these problems, the estimation is done in two stages. The first stage consists of estimating household expenditures per adult from the following regression Expenditures/Adult - f + C Age + 4 2Age2 + Schooling 2 + C Schooling + C Training + C Public School? 45 6 + C Landholding + a Unearned Income + CaRural + e (13) Age, Schooling and Training are the age, education, and training of the head of the household and the spouse, Public School? is a binary variable that &aks whether the head and spouse attended a private or a public school, Rural is a region dummy which equals 1 if the region of residence is rural, and 0 otherwise, Landholding is the total area of land sown or rented out by the household, and Unearned Income is the sum of all income other than wage and salaries. The fitted values used are from a regression that excludes Landholding, because the sample was reduced to one-third of its size when landholding was included. The full regression results are in Appendix I. t :The structure of the economy, or the time spent in each of the three market sectors, t1, t2, and t 3, is proxied by their Share in Departmental GDP. 20 ti - Share of agriculture (farming, fishing and #orestry), t- Share of services (personal and business services, health care, hotels, tourism, and so on), t - Share of industry (manufacturing, mining, construction, and so on), averaged over 1979-85. Ideally, I would also have experimented with the share of each sector in department employment, but data were not available. It may be more appropriate to use sector shares in income, since it is the earning power of children that parents are concerned about, not hours worked. Using sectoral shares in departwental GDP as proxies for t assumes that within each sector, the labor intensity of production does not change as sectoral output changes. P : Changes in Ps are proxied by a rural-urban dummy variable. I assume that the price of schooling is lower .n urban than in rural regions. This could be because the average distance to school in rural regions is greater than that in urban regions. Thus within each department, the parameter P5 varies with region of residence. Z: The estimation in this section uses household data from the PLSS for subset Z2 variables, as follows: 1sAlternatively, it could be because rural areas are generally agricultural, and urban areas are more industrial. If children's labor is more valuable in agrarian economies, then the price of schooling would include the higher opportunity cost in rural areas. This effect confounds the structure of the economy with the rural-urban decomposition, and is ignored here. 21 (1) Highest level of education completed by Mother (2) Highest level of education completed by Father (3) Mother's Longest Occupation (4) Father's Longest Occupation The level of education of the pazqnts is measured as follows: -1 - Never attended, 0 - None, 1 - Initial, 2 - Primary, 3 - Regular secondary, 4 - Technical secondary, 5 - Postsecondary Non-university, 6 - University. The occupation of the parents is mdasured as follows: 1 - Did not work, 2 - Missing and not elsewhere classified, 3 - Agriculture, Fishing and Forestry, 4 - Sales vendors, 5 - Service workers, 6 - Production & Transportation, 7 - Clerical, 8 - Professional and Government. The education levels of the mother and father are included to capture information-processing abilities of the household. Educated parents are likely to be better informed about the true rates of return to their children's education. The occupation of each parent is included to test whether parents base their expectations for their children on their own experiences, or whether the crucial determinant is the general structure of production in the region of residence (as proxied by province t ratios).16 16The education of the mother relative to that of the father, when their wages or incomes are not included, may also indicate the extent of female control of the household budget Y. Occupations of the mother and father are more likely to proxy share in actual earned income of the household. Education, when occupation is included, better proxies potential earned income. To the extent that the "bargaining position" of females depends on the income-earning capability of the mother, and not income actually earned, the mother's schooling will have an independent effect on 22 Results of the Schooling Regressions The general form of the estimated equations is S - S (Y, P , t , t , t , Z) for females, (14a) S - S (Y, P , t , t , t , Z) for males. (14b) m m S 1 2 3 Since the t as measured add up to 1, only two of the three shares can be included in a regression. The aim is to examine the relative increases in the demand for schooling as educatr.on-intensive sectors grow in importance, and particularly whether industrial growth raises the demand for boys' education more than girls'. To hold the share of services constant while increasing the share of industry, we must include both the high education sectors in the regression, and omit the low education sector. Another aspect of the problem of multicollinearity is that the share of agriculture in GDP (the omitted class) and the degree of urbanization are highly (negatively) correlated. Since the share of agriculture is equal to 1 minus the sum of the shares of services and industry, this leads to high degree of multicollinearity in the above regressions. To overcome this problem, I use a four-way classification, with government services as the fourth category. The results below are computed with two classes, industry and nongovernment services, and two omitted classes, agriculture and government services. Since the share of gender equity in child investments. Studies have also found that the father's education has significant positive effects on the schooling of children. (See for example, Moock and Leslie 1986.) The education of both parents is hypothesized to have a positive effect on the schooling of both sons ans daughters. 23 government services is about .07, it is not a very important category quantitatively, but it helps to overcome the multicollinearity problem." The regressions estimated for females are: S o 0 1 Household-Income + 2Share-of-Industry + Share-of-Services + 4 Urbanization + Father's Education + 6Mother's Education + 7 Father's Occupation + 8 Mother's Occupation + e (15a) and for males are: S - p + p Household Income + p Share-of-Industry m 0 1 2 + Share-of-Services + p Urbanization + psFather's Education + p Mother's Education + 7Father's Occupation + p8 Mothers Occupation + e (15b) The primary hypotheses to be tested are: Hi. < 0, 1 < 0 : Schooling of the girl and the boy are normal goods. H2. 4 < p1 : Since levels of schooling of girls are lower to begin with, this would imply that equity in human capital investments across sexes is a normal good. H3. #2 < 0, p2 < 0 : Demand for schooling increases as the share of 17To test whether multicollinearity was severe, I used the singular value decomposition technique advocated by Belsley and others (1980). This test is essentially a measure of the sensitivity of coefficients to changes in the matrix of independent variables. This sensitivity is summarized as (square root of) the ratio of the largest eigenvalue of the X'X matrix to the smallest, and is called a "condition index." Condition indices of less than 30 are considered good, and those between 30 to 100 are considered to be indicative of moderate to strong multicollinearity (see Judge and others 1985). The largest condition index for the regressions in this paper was about 26. 24 industry increases at the expense of agriculture, holding the share of services constant. H4. 2 2 : Men have a comparative advantage in industry, since it rewards work experience as well as schooling. So when the share of industry rises holding the share of services constant, the demand for male schooling rises by more than the demand for female schooling. H5. 4' < 0, p < 0 : Demand for schooling increases as the share of 33 services rises at the expense of agriculture, holding the share of industry constant. H6. 4' < 3 : Women have a comparative advantage in services, since the rate of return to schooling in the services sector is independent of the time spent working in this sector. So when the share of services rises, holding the share of industry constant, the demand for female schooling rises by more than the demand for male schooling. H7. 4 < 0, j < 0 : A fall in the supply price of schooling associated with greater urbanization, holding the demand schedule for schooling constant, will raise schooling investments in both girls and boys. H8. 4 < A : The theory has no predictions about the relative magnitudes of the responses of male and female schooling to changes in the supply price of schooling. Table 2 shows the results for the entire sample and separately for each of five age-groups: 6-15 years, 16-25 years, 26-35 years, 36-45 years, and 46-65 years. Schooling decisions of those who are more than 35 years old were made when the structure of the province's economy was different. We 25 expect to find the strongest results for groups that are finishing school (16-25 years) and that have just finished school (26-35 years). We expect to see that an increase in the share of services, holding the share of industry constant, will reduce the schooling shortfall of females by more than that of males. For industry, the effect on male schooling should be stronger than the effect on female schooling. The results indicate that for the sample as a whole, increases in the share of services are significantly and negatively correlated with schooling shortfall for females, but not for males. The share coefficient in the female regressions is greater in magnitude for the services sector than for industry. For males, on the other hand, it is the share of industry that is significant (at the 5 or 10 percent level). This is exactly what our theory predicted. Results by age group show that coefficients for the shares of services and industry are insignificant for the age groups 6-15, 36-45, and 46-65 years. For the group 16-25 years old, the coefficient for the share of services is significant at the 1 or 5 percent level. For the group 26-35 years old, the share of industry and services show strong results. Reassuringly, the correlation between the share of services and schooling seems to be stronger for females than for males. However, the magnitude is greater for the services sector for both sexes -- a contradiction of the theory. For each of the age groups, the coefficient for the share of industry is never significant for females. However, for males in the two groups 26-35 and 36-45 years, the coefficient of the share of industry is 26 significantly negative at the 10 percent level of significance. For the age-specific estimates, the degree of urbanization decreases the schooling shortfall of females, but not necessarily that of males. Surprisingly, urbanization has a strong unfavorable influence on the schooling shortfall for males in the pooled regressions. In the context of our model, where the rural dummy proxies the price of schooling, this implies that lowering the price of schooling increases the schooling of females more than males. The education of both father and mother have significant positive effects on schooling. The education of the mother is more closely associated with the schooling of daughters, and the education of the father with the schooling of sons. The occupation of the mother never matters, while the occupation of the father seldom matters, with higher occupation implying a lower schooling shortfall for both female and male offspring. Household income has a strong beneficial effect on all groups except children aged 6-15 years. A surprising finding was that household income has a stronger effect on schooling of male children than female children. Thus while investment in education is a normal good, the data reject the view that gender equity is a normal good when demand-side factors are included. That is, increases in household income per se will not lower the gender gap in schooling. 27 Table 1 MEANS AND STANDARD DEVIATIONS OF VARIABLES BY HOUSEHOLDS IN PERU: 1985-86 Variables Mean Standard Deviations Department-Level RHS Variables: Share of Industry in Dept.GDP 0.36 0.14 Share of Agriculture in Dept.GDP 0.18 0.13 Share of Services etc.in Dept.GDP 0.39 0.13 Share of Govt. Services in Dept.GDP 0.07 0.03 Degree of Urbanization 66.99 25.49 Household-Level RHS Variables: Predicted Household Income 787.26 385.13 Place of Residence (Dummy: 1-Rural, 0-Urban) 0.43 0.49 Household Size 6.48 2.75 Total Landholding (Acres) 9.21 76.01 Individual-Level RHS Variables: Father's Education Level 1.48 1.76 Mother's Education Level 0.48 1.68 Father's Occupation 4.28 1.67 Mother's Occupation 2.98 1.63 Individual-Level LHS Variables: Female: Education in Years 4.59 4.10 Male : Education in Years 5.49 4.13 Female: Highest Level of Education Completed 1.94 1.69 Male : Highest Level of Education Completed 2.43 1.50 Female: Age in Years 26.70 16.04 Male : Age in Years 26.12 16.13 Female Schooling Shortfall: 6-15 years 2.76 1.92 Male Schooling Shortfall: 6-15 years 2.57 1.73 Female Schooling Shortfall: 16-25 years 7.89 4.35 Male Schooling Shortfall: 16-25 years 7.34 3.75 Female Schooling Shortfall: 26-35 years 18.76 5.84 Male Schooling Shortfall: 26-35 years 16.89 5.28 Female Schooling Shortfall: 36-45 years 31.01 5.67 Male Schooling Shortfall: 36-45 years 29.02 5.76 Female Schooling Shortfall: 46-65 years 46.33 7.03 Male Schooling Shortfall: 46-65 years 44.54 7.13 28 Table 2 HOUSEHOLD-LEVEL SCHOOLING REGRESSIONS: PERU, 1985-86 Independent Equation 1 Equation 2 Equation 3 Variables Female Male Female Male Female Male All Age Groups Constant -5.043 -3.841 5.342 3.665 5.031 3.899 (0.64) (0.62) (0.62) (0.62) (0.64) (0.64) Industry share -1.640 -1.310 -1.002 1.166 -0.943 -1.172 (0.68) (0.67) (0.67) (0.67) (0.67) (0.67) Services share -3.215 -1.281 -2.251 1.112 -2.165 1.216 (0.83) (0.82) (0.81) (0.81) (0.82) (0.82) Rural dummy 0.836 -0.527 0.233 -0.678 0.197 -0.715 (0.20) (0.20) (0.20) (0.21) (0.21) (0.21) Household income -0.007 -0.010 0.006 0.010 0.006 0.010 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Father's education -0.419 0.304 0.388 0.302 (0.06) (0.06) (0.06) (0.06) Mother's education -0.574 0.007 -0.556* 0.005 (0.06) (0.06) (0.06) (0.06) Father's occupation -0.129 0.022 (0.06) (0.06) Mother's occupation -0.015 0.085 (0.05) (0.05) Age group 11.251 11.181 11.091 11.116 11.096 11.120 (0.06) (0.06) (0.07) (0.07) (0.07) (0.07) Adjusted R 2 0.885 0.899 0.892 0.900 0.892 0.900 Sample size 4281 4024 4281 4024 4281 4024 Note: Dependent variable is schooling shortfall - age -schooling - 5 Significant at 10 percent, at 5 percent level, at 1 percent level. Standard errors in parentheses. 29 Table 2 (Continued) HOUSEHOLD-LEVEL SCHOOLING REGRESSIONS: PERU, 1985-86 Independent Equation 1 Equation 2 Equation 3 Variablep: Female Male Female Male Female Male Population Aged 6 to 15 years Constant 0.765 2.768 1.448 3.378 1.595 3.209 (1.40) (1.41) (1.38) (1.00) (1.39) (1.03) Industry share 2.474 1.201 2.047 1.152 2.518 1.106 (1.58) (1.31) (1.52) (1.24) (1.56) (1.24) Services share 1.680 -0.297 3.101 0.179 3.827 0.071 (2.00) (1.65) (1.92) (1.57) (2.05) (1.60) Rural dummy 1.928 0.168 1.383 0.142 1.424 0.125 (0.49) (0.40) (0.49) (0.38) (0.49) (0.38) Household income 0.001 -0.000 0.000 0.000 0.000 -0.000 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Father's education -0.042 -0.309 -0.010 -0.335 (0.14) (0.12) (0.14) (0.13) Mother's education -0.465 0.163 -0.433 0.193 (0.12) (0.10) (0.13) (0.11) Father's occupation -0.129 0.027 (0.11) (0.11) Mother's occupation -0.056 0.109 (0.10) (0.09) Adjusted R 2 0.092 -0.009 0 291 0.102 0.189 0.101 Sample size 156 170 156 170 156 170 Note: Dependent variable is schooling shortfall - age - schooling - 5 Significant at 10 percent, **at 5 percent level, **at 1 percent level. Standard errors in parentheses. 30 Table 2 (Continued) HOUSEHOLD-LEVEL SCHOOLING REGRESSIONS: PERU, 1985-86 Independent Equation 1 Equation 2 Equation 3 Variables Female Male Female Male Female Male Population Aged 16 to 25 years Constant 14.305 1A.296 14.038 14.353 14.689 14.039 (1.13) (1.26) (1.08) (1.24) (1.11) (1.31) Industry share -1.173 -0.866 -0.912 -0.355 -0.813 -0.388 (1.28) (1.57) (1.21) (1.52) (1.21) (1.52) Services share -5.117 -4.939 4.479 4.084 -4.106 4.243 (1.61) (1.70) (1.52) (1.66) (1.53) (1.66) Rural dummy 2.333 1.927 1.828 1.383 1.685 1.336 (0.41) (0.45) (0.39) (0.45) (0.39) (0.45) Household income -0.003 -0.004 0.002 0.004 0.001 0.003 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Father's education -0.404 0.416 0.340 0.420 (0.11) (0.12) (0.11) (0.13) Mother's education -0.616 0.282 -0.573 0.304 (0.10) (0.12) (0.11) (0.12) Father's occupation -0.328 0.031 (0.11) (0.11) Mother's occupation 0.123 0.117 (0.08) (0.09) Adjusted R 2 0.286 0.359 0.366 0.399 0.374 0.399 Sample size 645 445 645 445 645 445 Note: Dependent variable is schooling shortfall - age - schooling - 5 Significant at 10 percent, at 5 percent level, at 1 percent level. Standard errors in parentheses. 31 Table 2 (Continued) HOUSEHOLD-LEVEL SCHOOLING REGRESSIONS: PERU, 1985-86 Independent Equation 1 Equation 2 Equation 3 Variables Female Male Female Male Female Male Population Aged 26 to 35 years Constant 26.685 26.601 25.9926.117 26.66226.230 (0.99) (0.93) (0.95) (0.92) (1.00) (0.97) Industry share -1.384 -1.733 -0.542 -1.590 -0.470 -1.384 (1.12) (1.01) (1.07) (1.00) (1.07) (1.00) Services share -3.845 -2.639 -3.233*2.170 -3.311* 2.124 (1.38) (1.32) (1.30) (1.30) (1.31) (1.31) Rural dummy 1.862 0.525 1.130 0.305 1.029 0.226 (0.35) (0.33) (0.34) (0.33) (0.34) (0.34) Household income -0.006 ***- 0.008 .0.004***.0.007 .0.004***.0.007 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Father's education -0.538* 0.303 0.471* 0.272 (0.10) (0.09) (0.10) (0.09) Mother's education .606***..0262 .0.576***.0.261 (0.10) (0.09) (0.10) (0.09) Father's occupation -0.245* .0.121** (0.10) (0.09) Mother's occupation 0.031 0.093 (0.07) (0.07) Adjusted R 2 0.404 0.500 0.462 0.515 0.464 0.515 Sample size 1148 957 1148 957 1148 957 Note: Dependent variable is schooling shortfall - age - schooling - 5 Significant at 10 percent, **at 5 percent level, **at 1 percent level. Standard errors in parentheses. 32 Table 2 (Continued) HOUSEHOLD-LEVEL SCHOOLING REGRESSIONS: PERU, 1985-86 Independent Equation 1 Equation 2 Equation 3 Variables Female Male Female Male Female Male *Population Aged 36 to 45 years Constant 38.049 39.942 36.471 39.570 36.795 39.437 (0.97) (0.84) (0.93) (0.85) (0.97) (0.89) Industry share -0.730 -1.850 0.073 -1.404 0.074 -1.448 (1.15) (0.94) (1.10) (0.94) (1.10) (0.93) Services share -0.005 -1.650 1.033 -1.360 1.136 -1.407 (1.36) (1.18) (1.29) (1.17) (1.30) (1.18) Rural dummy 0.930 -0.547 0.346 -0.738 0.272 -0.734 (0.35) (0.30) (0.34) (0.31) (0.34) (0.31) Household income -0.009 -0.011 -0.006 * 0.010 -0.006 -0.010 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Father's education -0.382 -0.294 -0.339 -0.303 (0.09) (0.09) (0.09) (0.'09) Mother's education -0.778 0.095 -0.756 0.102 (0.10) (0.09) (0.10) (0.09) Father's occupation -0.174 0.044 (0.10) (0.09) Mother's occupation 0.088 0.010 (0.08) (0.07) Adjusted R 2 0.435 0.610 0.493 0.617 0.494 0.616 Sample size 1064 1030 1064 1030 1064 1030 Note: Dependent variable is schooling shortfall - age - schooling - 5 Significant at 10 percent, **at 5 percent level, **at 1 percent level. Standard errors in parentheses. 33 Table 2 (Concluded) HOUSEHOLD-LEVEL SCHOOLING REGRESSIONS: PERU, 1985-86 Independent Equation 1 Equation 2 Equation 3 Variables Female Male Female Male Female Male Population Aged 46 to 65 years Constant 54.175 54.047 52,687 54.174 52.731 53.386 (1.29) (1.14) (1.28) (1.15) (1.36) (1.22) Ineustry share -3.200 -1.246 -2.321 -1.227 -2.353 -1.282 (1.56) (1.36) (1.55) (1.36) (1.55) (1.36) Services share -4.428 1.215 -2.814 1.153 -2.696 0.910 (1.91) (1.66) (1.89) (1.66) (1.90) (1.67) Rural dummy 0.093 -0.947 -0.498 -0.993 -0.430 -0.996 (0.45) (0.40) (0.45) (0.40) (0.46) (0.40) Household income -0.008 -0.013 0.006* 0.012 0.006* 0.013 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Father's education -0.307 0.233 -0.331 0.257 (0.13) (0.11) (0.13) (0.12) Mother's education -0.636 0.176 -0.647 0.151 (0.16) (0.13) (0.16) (0.13) Father's occupation 0.091 0.191 (0.14) (0.13) Mother's occupation -0.146 0.127 (0.11) (0.10) Adjusted R 20.201 0.381 0.227 0.382 0.227 0.383 Sample size 1264 1418 1264 1418 1264 1418 Note: Dependent variable is schooling shortfall - age - schooling - 5 Significant at 10 percent, **at 5 percent level, at 1 percent level. Standard Errors in parentheses. 34 5. PROVINCE LEVEL EMPIRICAL EVIDENCE Evidence from provincial data confirms the findings of the previous section. Because literacy and income data are based on the census, the results do not reflect any sampling bias in the PLSS. In the last section the focus was on household level variables (the subset Z2). In this section I experiment with various infrastructural parameters (Z ), indicating the degree of development of transport and communications in each province or department. Using data from the department level is tantamount to assuming that households within a department are identical with respect to the parameters they face: income (Y), the rates of return to schooling in the three sectors (a, 0, 1, and 6), and the time-allocation parameters (t ). However, each department has two regions, rural and urban. As in the previous section I assume that the price of schooling (P s) is lower in urban than in rural regions. Definitions of Dependent Variable: Illiteracy Rates S fM: The census data do not provide gender-specific schooling attainment or enrollment by age group for each department. Since illiteracy rates are available at the required level of disaggregation, I use cohort- specific illiteracy rates as a proxy for investment in schooling in the department. Clearly, literacy rates are a gross approximation for schooling levels. There are two defenses for this procedure: First, the study focuses on the differences in human capital between males and females. Even though illiteracy rates do not reflect absolute 35 levels of human capital of each sex, they may provide reliable estimates of relative levels. Second, total illiteracy rates are highly negatively correlated with initial, secondary, postsecondary, and vocational enrollment ratios across departments (see Table 3). University level enrollment ratios are weakly negatively correlated with illiteracy rates, but this is not surprising. Somewhat puzzling is the finding that primary school enrollment ratios are positively correlated with illiteracy rates. This is probably due to the fact that high actual primary school enrollment relative to the number that should be enrolled may indicate that students have fallen behind in the curriculum. Definitions of Independent Variables Y: Departmental gross domestic product per capita is used as the measure for household income. t :The structure of the economy, or the time spent in each of the three market sectors, t , t 2, and t3, is proxied by their Share in departmental GDP. These variables are defined as in the previous section: t- Share of Agriculture (farming, fishing and forestry), t- Share of Services (personal and business services, health care, hotels, tourism, and so on), t - Share of Industry (manufacturing, mining, construction, and so on), 36 averaged over 1979-85. P : Changes in P are proxied by the Degree of Urbanization (percentage of total department population living in urban areas) in 1985. The degree of urbanization is a continuous variable. Therefore P (0 P s 100) is a continuous variable. I also experimented with a rural-urban dummy variable (P. - p,v, where p and v are the price of schooling a child in rural and in urban areas respectively). This is done by estimating male and female schooling regressions separately for rural and urban schooling levels across departments. However, income per capita, Y, and the potential work opportunities by sector, ti, are still department-level: department GDP figures are not available separately for each region. Thus it is assumed that ti do not vary across regions within a department, but Ps does. This assumes 1-hat, within a department, it is costless for an adult to migrate from rural to urban urban regions (or vice versa), but the cost of sending a child to school from rural to urban regions is very high. Z : The vector of department-level infrastructural variables Z include Roads per Square Kilometer, and Post Offices per Square Kilometer. I also experimented with Telegraph Offices per Square Kilometer, Telephone Lines per Capita, and some other variables. Since household specific variables Z2 are only available in a household-level survey, this section only employs department-level infrastructural variables (subset Z) in the schooling regressions. 37 Description of the Data Table 4 shows the means and standard deviations of the dependent and all the independent variables used in the illiteracy regressions. Industry's share is the largest, but is also relatively more variable across departments than the share of services (as evidenced by values of the coefficients of variation of 0.54 and 0.36 respectively). The share of agriculture in GDP (CV- 0.61) is also highly variable across sectors. It seems that the share of industry increases most rapidly at the expense of the share of agriculture. Average illiteracy rates for all ages and for both sexes are higher in rural areas. The aggregate gender differential (female illiteracy divided by male illiteracy) is lower for rural (2.52) than for urban areas (3.54). Given the values of the standard deviations, this difference appears to be statistically significant. A plausible (demand-side) explanation of this finding is that agricultural activities are less education-intensive for women and men. That is, the predominance of a sector with low education intensity in a region will result in low demand for education in that region, but, under reasonable assumptions, in higher equity across sexes in investqent in schooling. Since the supply price of education is probably lower in urban areas, and if the lower price prompts a greater response from girls than boys, the ratio between male and female illiteracy rates should be lower in urban areas. This points out a weakness of studies that focus only on shifts in supply of schooling, and highlights the importance of studying the determinants of shifts in the schooling demand curve. 38 Table 3 CORRELATION BETWEEN ENROLLMENT RATIOS AND ILLITERACY RATES: PERU, 1985-86 ENROLLMENT RATIO TOTAL COHORT GROUP (IN YEARS) (by school g-oup) 15-19 20-24 25-29 30-34 35-39 40+ Initial -.596 -.538 -.552 -.10 -.597 -.616 -.587 Primary .694 .694 .691 .327 .691 .711 .691 Secondary -.815 -.881 -.847 -.639 -.829 -.851 -.812 Postsecondary -.669 -.696 -.672 -.516 -.682 -.714 -.658 University -.402 -.444 -.439 -.332 -.394 -.406 -.412 Other -.725 -.663 -.689 -.483 -.731 -.751 -.736 Notes: 1.All correlations except those marked by an asterisk are significant at the 5 percent level. All correlations are significant at the 10 percent level. The test used to determine statistical significance of the correlation is the t-statistic t - [(n-2r)/(1-r2 0*. n is the number of observations, r is the computed correlation coefficient, and (n-2) is the number of degrees of freedom. 2.The number of observations is 23. Lima and Callao, and Loreto and Ucayali are aggregated to maintain conformity with available GDP accounts. 3.Enrollment Ratios are calculated as ratios of total enrollment in the department to total population. For an idea of how much age- distributions differ across department, see column 5 of Table in the Appendix. Sources: For illiteracy rates, enrollment and population, Peru: Compendio Estadistico, 1987, Systema Estadistico Nacional, Instituto Nacional de Estadistica. 39 Table 4 H EANS AND STANDARD DEVIATIONS OF VARIABLES BY DEPARTMENT, 1981-82 Variables Means and Standard Deviations Mean S.D. RKS Variables: Per Capita GDP (Current Prices) 174.46 148.65 Share of Industry in Dept.GDP 0.37 0.20 Share of Agriculture in Dept.GDP 0.23 0.14 Share of Services etc.in Dept.GDP 0.33 0.12 Share of Govt. Services in Dept.GDP 0.07 0.03 Degree of Urbanization 55.66 21.76 Roads per 1000 Square Kilometers 785.06 2258.45 Post Offices per 1000 Sq. Km. 2.82 1.92 Urban Rural Mean S.D. Mean S.D. LHS Variables: Female illiteracy rate: All ages 18.67 10.84 46.22 16.76 Male illiteracy rate: All ages 5.44 3.12 19.43 9.09 Female/Male illiteracy ratio: All 3.54 0.87 2.52 0.51 Female illiteracy rate: 15-19 years 3.80 2.24 18.42 11.20 Male illiteracy rate: 15-19 years 1.62 0.74 6.96 3.55 Female/Male illiteracy ratio: 15-19 2.45 0.83 2.55 0.63 Female illiteracy rate: 20-24 years 5.85 3.99 26.19 14.42 Male illiteracy rate: 20-24 years 1.62 0.89 7.58 4.47 Female/Male illiteracy ratio: 20-24 3.81 1.47 3.93 1.44 Female illiteracy rate: 25-29 years 8.75 5.67 34.50 17.13 Male illiteracy rate: 15-19 years 1.95 1.09 10.25 5.93 Female/Male illiteracy ratio: 25-29 4.96 2.31 3.84 1.42 Female illiteracy rate: 30-34 years 13.88 9.04 44.94 18.56 Male illiteracy rate: 30-34 years 3.04 1.83 14.40 8.22 Female/Male illiteracy ratio: 30-34 5.01 2.30 3.72 1.77 Female illiteracy rate: 35-39 years 22.08 12.77 55.89 17.56 Male illiteracy rate: 35-39 years 4.52 2.54 19.83 9.93 Female/Male illiteracy ratio: 35-39 5.27 2.07 3.23 1.17 Female illiteracy rate: 40+ years 37.72 17.76 69.69 17.04 Male illiteracy rate: 40+ years 12.00 6.54 34.01 12.95 Female/Male illiteracy ratio: 40+ 3.34 0.88 2.17 0.44 Notes: Means are Unweighted averages. 40 When we examine the patterns across cohorts, we find a small difference for the younger age groups (15 to 29 years) in this ratio across regions, and a significantly larger difference for people 30 years and older. One interpretation of this finding is that over time, the gender differences in urban areas have narrowed more rapidly than in rural areas. Again, a demand side explanation seems plausible. The rapid growth of services in urban areas can account for greater equity even if the supply price of schooling is constant over time, if women have a comparative advantage over men in working in services (see section 3). It is difficult to rule out a supply-side explanation here. It may be that the supply price of schooling has fallen relatively more in urban than in rural areas over time, and that female schooling has a higher price elasticity than male schooling, analogous to Gertler and Alderman's (1989) arguments for health. Results of the Schooling Regressions The general form of the estimated equations for females and males respectively is St r W + # ker-Capita-GDP + 4 Share-of-Industry + i Share-of-Services + 4 Urbanization + e (16a) S - P+ P1Per-Capita-GDP + p2Share-of-Industry + p3Share-of-Services + p.Urbanization + e (16b) "Note that men may still have an absolute advantage in both industry and services. 41 The first three slope coefficients in each equation measure demand shifts, while the last coefficient measures supply price effects. Since S stands for illiteracy rates, a negative coefficient implies a favorable 3ffect on education levels. In this section, I refer to the absolute magnitudes of the coefficients when using the phrases "greater than" or "less than." Table 5 reports the results of the regressions. The first two rows report the results of the regressions (16a) and (16b). The results for the sample as a whole indicate no significant support for hypotheses Hl to H6 listed in section 4. The only significant variable is urbanization, although the signs of the other coefficients conform with the theory. The insignificance of results using a sample of all ages is not surprising. Schooling decisions of age groups 35 years and above were made two decades ago. The structure of the department's economy and hence the demand for schooling is likely to have changed since then. It is more sensible to look at the relationship between the illliteracy of younger cohorts and department income, demand structure, and degree of urbanization. The relationships are likely to be strong for the youngest cohort group, and to diminish as the age of the cohorz increases. Results support this argument : for all but the oldest groups (35-39 years and 40 years and above) there is reasonably strong evidence that a rise in the share of the services sector, holding the share of industry constant, raises the schooling levels of females and males. As predicted by the theory, the coefficient for females is (2.5 to 5 times) 42 larger in magnitude, and intermittently significant at the 10 percent level of significance for a one-tailed test. The coefficient for industry's share is always greater in magnitude for females but never significant. This evidence, combined with the fact that the variance of the share of industry is in fact larger than that of the service sector, is evidence consistent with the view that a rise in the share of services in GDP leads to greater increases in the eOucation of women than an equivalent increase in the share of industry. The coefficient for the degree of urbanization is always negative for both females and males, always greater in magnitude for females, and statistically significant at the 5 percent level for a one-tailed test. Given the high degree of multicollinearity between the share of industry and the rate of urbanization, the high standard errors of the coefficients are not surprising. To increase the degrees of freedom (a common solution for multicollinearity), the female and male schooling equations are estimated separately for rural and urban regions, thus allowing the omission of the urbanization variable. This is roughly equivalent to treating the urbanization variable as a binary variable. It is a test of the alternative view that, within a department, it is costless for an adult to migrate from rural to urban regions (or vice versa), but the cost of sending a child for schooling from rural to urban regions is very high. The estimated equations for urban areas are Urban S. M + 1 Per-Capita-GDP + 2Share-of-Industry + 4 3Share-of-Services + eU (17a) 34 Urban S po + jIPer-Capita-GDP + paShare-of-Industry + p3Share-of-Services + es (178b) and for rural areas Rural St M + #0Per-Capita-GDP + *Share-of-Industry + *3Share-of-Services + e (18a) Rural 8 - p0 +' p Per-Capita-GDP + p Share-of-Industry + p 3Share-of-Services + e (18b) The results for urban areas are rows 3 and 4; for rural areas, rows 5 and 6 in Table 5. Regressions for groups aged 15-19, 20-24, and 25-29 years confirm the theory. The coefficients for the share of services in GDP are significant and larger in magnitude for females than for males, and larger in rural than in urban areas. The coefficients for the share of industry in GDP are generally insignificant for female schooling, and generally significant for urban male schooling. This seems to confirm the hypothesis that industry rewards schooling more than the main omitted class, agriculture, and that men have a comparative advantage in industry. Women, on the other hand, have a comparative advantage in services. The sign and magnitude of the coefficient for GDP indicate that female schooling increases by more than male schooling when income increases, and that the increases are larger for both females and males in rural areas. Regressions that include variables proxying Z, roads per square kilometer, telephone lines per capita, and post offices per square 44 kilometer, were also estimated. The coefficients were insignificant, and the coefficients for per capita GDP, shares of services and industry, and urbanization were left largely unchanged.19 The major limitation of this analysis is that the shares of each sector in a department may be jointly determined with education levels of men and women. For example, services may require more educated women than educated men, so provinces that have relatively more educated women will tend to have a larger share of services in GDP. The issue of causality between share of services and the demand for education of women is left unresolved. This is a crippling limitation for purposes of deciding policy. I address this issue by estimating regressions by cohorts. Assuming that the sectors' shares in GDP are relatively stable across departments, the correlation between schooling and sectoral share should be stronger for younger cohorts if a higher share of services and industry leads to a higher demand for education. On the other hand, if education intensive activities are concentrated in areas with high exogenous education levels, this would imply uniformly strong correlations between education levels and structure of the economy across all age groups. The regressions discussed above indicate that the correlations are weak for older cohorts, implying support for the view that causality runs from economy structure to schooling levels, and not vice versa. The results in this section are similar to those found in Gill and 'eThese results, which are not reported here, are available from the author. 45 Khandker (1990) for a sample of about 100 countries in 1965 and 1987. Migration is likely to be more frequent within a country than across national boundaries. The similarity of results by country and by province implies that the possibility of cross-department migration does not seem to be a significant factor in schooling decisions. This issue needs more examination, though, before any conclusive statement can be made on the effects of migration on the rates of return to human capital. 46 Table 5 CROSS-SECTION SCHOOLING REGRESSIONS: PERU, 1985-86 Dependent Independent Variables Variable CONSTANT PER CAP. INDUSTRY SERVICES DEGREE OF UNADJ. GDP SHARE SHARE URBANIZN. R-SQR. All Ages 1.Female Total 73.713 -0.019 9.196 -13.378 -0.647 .817 Illiteracy Rate (0.02) (17.61) (31.28) (0.15) 2.Male Total 33.479 -0.013 -0.584 -13.979 -0.239 .714 Illiteracy Rate (0.01) (10.42) (18.49) (0.09) 3.Female Urban 50.189 -0.040 -8.892 -60.113 .545 Illiteracy Rate (0.02) (13.72) (17.42) 4.Male Urban 15.010 -0.012 -4.410 -16.739 .587 Illiteracy Rate (0.00) (3.76) (4.77) 5.Female Rural 86.408 -0.049 -3.506 -80.723 .354 Illiteracy Rate (0.03) (25.73) (32.67) 6.Male Rural 42.407 -0.030 -4.380 -44.040 .362 Illiteracy Rate (0.02) (14.61) (18.56) Note: Sample consists of 23 departments in Peru Significant at 10 percent, at 5 percent level, at 1 percent level. Standard errors in parentheses. 47 Table V (Continued) CROSS-SECTION SCHOOLING REGRESSIONS: PERU, 1985-86 Dependent Independent Variables Variable CONSTANT PER CAP. INDUSTRY SERVICES DEGREE OF UNADJ. GDP SHARE SHARE URBANIZN. R-SQR. Population Aged 15-19 Years l.Female Total 38.562 -0.012 -7.420 -26.313 -0.237 .774 Illiteracy Rate (0.01) (10.10) (17.94) (0.09) 2.Male Total 11.364 -0.004 .0.618 -5.171 -0.083 .658 Illiteracy Rate (0.01) (4.04) (7.18) (0,03) 3.Female Urban 10.945 -0.009 -2.472 -13.242 .623 Illiteracy Rate (0.00) (2.65) (3.36) 4.Male Urban 3.687 -0.002 -0.947 -3.786 .280 Illiteracy Rate (0.00) (1.42) (1.80) 5.Female Rural 54.960 -0.040 -14.039 -69.955 .524 Illiteracy Rate (0.02) (15.80) (20.06) 6.Male Rural 15.962 -0.015 0.583 -18.415 .396 Illiteracy Rate (0.01) (6.02) (7.64) Population Aged 20-24 Years l.Female Total 55.657 -0.013 -13.596 -37.260 -0.237 .816 Illiteracy Rate (0.01) (12.65) (22.46) (0.11) 2.Male Total 14.941 -0.006 -2.693 -11.878 -0.073 .691 Illiteracy Rate (0.01) (4.54) (8.07) (0.04) 3.Female Urban 20.624 -0.012 -9.157 - 27.038 .655 Illiteracy Rate (0.01) (4.52) (5.74) 4.Male Urban 4.951 -0.002 -2.873 -5.568 .576 Illiteracy Rate (0.00) (1.14) (1.45) 5.Female Rural 72.187 -0.048 -19.606 -86.341 .511 Illiteracy Rate (0.02) (19.98) (25.36) 6.Male Rural 18.686 -0.019 0.037 -21.900 .387 Illiteracy Rate (0.01) (7.58) (9.63) Note: Sample consists of 23 departments in Peru Significant at 10 percent, at 5 percent level, at 1 percent level. Standard errors in parentheses. 48 T'ole V (Continued) CROS-SECTION SCHOOLING REGRESSIONS: PERU, 1985-86 Dependent Independent Variab. a Vari&ble CONSTANT PER CAP. INDUSTRY SERVICES DEGREE OF UNADJ. GDP SHARE SHARE URBANIZN. R-SQR. Population Aged 25-29 Years l.Female Total 65.174 -0.017 -7.644 -29.348 -0.488 .825 Illiteracy Rate (0.02) (15.14) (26.88) (0.13) 2.Male Total 19.725 .0.010* *2.674 .15.610* -0.096* .718 Illiteracy Rate (0.01) (5.77) (10.24) (0.05) 3.Female Urban 26.485 -0.021 -7.588 -32.347 .585 Illiteracy Rate (0.01) (6.93) (8.79) 4.Male Urban 6.026 -0.004 -2.351 -7.235 .734 Illiteracy Rate (0.00) (1.08) (1.37) 5.Female Rural 85.822 -0.056 -18.634 -97.214 .483 Illiteracy Rate (0.03) (23.83) (30.26) 6.Male Rural 25.492 -0.027 0.314 -29.957 .452 Illiteracy Rate (0.01) (9.25) (11.74) Population Aged 30-34 Years l.Female Total 84.684 -0.021 -18.090 -62.375 -0.451 .724 Illiteracy Rate (0.03) (22.99) (40.83) (0.20) 2.Male Total 28.882 -0.011 -6.516 -20.340 -0.156 .738 Illiteracy Rate (0.01) (8.03) (14.26) (0.07) 3.Female Urban 40.319 -0.031 -10.513 -48.771 .509 Illiteracy Rate (0.01) (12.05) (15.30) 4.Male Urban 10.001 -0.005 -5.422 -11.570 .738 Illiteracy Rate (0.00) (1.78) (2.25) 5.Female Rural 93.902 -0.059 -11.044 -94.348 .397 Illiteracy Rate (0.03) (27.84) (35.35) 6.Male Rural 37.853 -0.033 -6.396 -43.207 .470 Illiteracy Rate (0.01) (12.29) (15.60) Note: Sample consists of 23 departments in Peru Significant at 10 percent, at 5 percent level, at 1 percent level. Standard errors in parentheses. 49 Table 5 (Continued) CROSS-SECTION SCHOOLING REGRESSIONS: PERU, 1985-86 Dependent Independent Variables Variable CONSTANT PER CAP. INDUSTRY SERVICES DEGREE OF UNADJ. GDP SHARE SHARE URBANIZN. R-SQR. Population Aged 35-39 Years l.Female Total 81.797 -0.015 17.634 4.990 -0.845 .855 Illiteracy Rate (0.02) (17.97) (31.91) (0.15) 2.Male Total 36.829 -0.012 -5.359 -19.729 -0.235 .744 Illiteracy Rate (0.01) (10.29) (18.28) (0.09) 3.Female Urban 55.197 -0.046 -8.302 -61.856 .463 Illiteracy Rate (0.02) (17.42) (22.11) 4.Male Urban 12.569 -0.010 -4.472 -13.281 .667 Illiteracy Rate (0.00) (2.71) (3.44) 5.Female Rural 94.333 -0.057 1.534 -74.458 .294 Illiteracy Rate (0.03) (29.64) (37.63) 6.Male Rural 47.133 -0.038 -7.004 -49.919 .416 Illiteracy Rate (0.02) (15.87) (20.15) Population Aged 40 & More Years 1.Female Total 87.159 -0.018 41.269 18.442 -0.885 .792 Illiteracy Rate (0.02) (21.93) (38.94) (0.19) 2.Male Total 51.328 -0.017 8.004 -5.877 -0.442 .698 Illiteracy Rate (0.02) (16.89) (30.00) (0.15) 3.Female Urban 80.244 -0.068 2.073 -86.466 .487 Illiteracy Rate (0.03) (23.09) (29.32) 4.Male Urban 27.812 -0.025 -4.198 -27.254 .459 Illiteracy Rate (0.01) (8.87) (11.27) 5.Female Rural 94.272 -0.056 26.476 -56.789 .265 Illiteracy Rate (0.03) (30.11) (38.23) 6.Male Rural 62.652 -0.044 -0.332 -54.795 .308 Illiteracy Rate (0.02) (21.75) (27.61) Note: Sample consists of 23 departments in Peru Significant at 10 percent, at 5 percent level, at 1 percent level. Standard errors in parentheses. 50 6. CONCLUSIONS AND POLICY IMPLICATIONS The main question addressed in this paper is: Do parents consider future labor activities when making schooling decisions for their childrent? To the extent that current demand for labor and remuneration reflect economic patteras in the future, the answer seems to be that they do. The main new policy implication that emerges from this paper and from Gill and Khandker (1990) is that the expansion of the service sector raises the levels of schooling of both men and women, but has a larger effect on women. That is, both increased human capital and equity between the sexes are associated with an increase in the service sector's share in CDP, at the expense of agriculture. This is probably an intermediate stage. As this structural transformation continues, the share of the industrial sector begins to play the role that services played at the earlier stage. This is in marked contrast to policy that emerges from well-known theories of economic growth, including Rostow (1960), Rosenstein-Rodan (1961), and others. It has been argued that in the process of growth, agriculture is the primary stage, industry the secondary stage, and services the tertiary stage. But two points should be kept in mind. First, the policies recommended here are not aimed at economic growth per se, but economic growth with increases in equity across gender. Second, the nature of the services sector is very different in low-income countries. This policy also contradicts the World Bank's advice that developing countries need to increase the production of tradables. But at least as far as gender equity in earnings potential is concerned, it would 51 be better to encourage the expansion of sectors producing nontradables. Another implication of the findings in this paper is that extending schooling facilities significantly raises investment in schooling, and lowers the gender gap as well. The third implication is that information about the rates of return tc schooling in home activities wll raise the scacoling levels of females by more than those of sales. 52 REFERENCES Becker, Gary S. 1981. A Treatise on the Family. Cambridge: Harvard University Press. Becker, Gary 8. 1985. "Human Capital, Effort and the Sexual Division of Labor." Journal of Labor Economics 3: 533-58. Becker, Gary S. and Nigel Tomes. 1976. "Child Endowments and the Quantity and Quality of Children." Journal of Political Economy 84: S143-62. Belsley, D., E. Kuh, and R. Welsch. 1980. Regression Diagnostics, New York: John Wiley and Sons. Freedman, David, and David Lane. 1983. "A Nonstochastic Interpretation of Reported Significance Levels." Journal of Business and Economic Statistics, Volume 1, Number 4. 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. Gill, Indermit, and Shahidur Khandker. 1990. "The Structure of Production as a Determinant of the Demand for Human Capital: An Application to Country Level Schooling and Health." World Bank, Washington, D.C. Judge, G., W. Griffiths, R. Hill, H. Lutkepohl, and T. Lee. 1985. The Theory and Practice of Econometrics, Second Edition. New York: Joht. Wiley and Sons. King, Elizabeth, and Rosemary Bellow. 1989. "Gains in the Education of Peruvian Women in 1940s-1980s: Patterns and Explanations." World Bank, Washington, D.C. Khandker, Shahidur. 1989. "Returns to Schooling and Male-Female Wage Differences in Peru." Paper presented at the Conference on the Family, Geuder Differences and Development, Yale University, New Haven. Mincer, Jacob. 1974. Schooling, Experience and Earnings. New York: Columbia University Press for the National Bureau of Economic Research. Mincer, Jacob, and Yoshio Higuchi. 1988. "Wage Structures and Labor Turnover in the United States and Japan." Journal of the Japanese and International Economies 2: 97-113. 53 Moock, Peter R., and Joanne Leslie. 1986. "Child Malnutrition and Schooling in the Terai Region of Nepal." Journal of Development Economics 20: 33-52. Rosenstein-Rodan, Paul N. 1961. "Notes on the Theory of the 'Big Push'" in Economic Development for Latin America, Howard Ellis and Henry Wallich, eds. New York: St. Martins Presn. 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. Schafgans, Marcia. 1990. "Fertility Deteirminants in Peru: A Quantity- Quality Analysis." World Bank, Washington, D.C. Schultz, Theodore W. 1975. "The Value of the Ability to Deal with Disequilibrta." Journal of Economic Literature. Schultz, T. Paul. 1989. "The Relationship Between Local Family Planning Expenditures and Fertility in Thailand, 1976-81." New Haven: Yale University Economic Growth Center. Schultz, T. Paul, and Mark Rosenzweig. 1982. "Market Opportunities, Genetic Endowments, and Intrafamily Resource Distribution: Child Survival in Rural India." American Economic Review 72: 803-15. Smith, J. Barry, and Morton Stelcner. 1989. "Modeling Economic Behavior in the Informal Urban Retail Sector of Peru." World Bank, Washington, D.C. Welch, Finis R. 1970. "Education in Production." Journal of Political Economy 78: 35-59. 54 APPENDIX I: RESULTS OF HOUSEHOLD EXPENDITURE REGRESSIONS, 1985-86 Independent Equation 1 Equation 2 Variables Constant 86.043 -179.871 (184.04) (223.50) Age of Household Head 27.226 24.423 (8.25) (13.28) Age2 of Household Head -0.325 -0.276 (0.09) (0.15) Schooling of Household Head -8.889 -29.887 (11.663 (14.13) Schooling2 of Household Head 5.459 4.941 (0.70) (0.82) Training of Household Head 75.001 61.105 (33.51) (36.97) Did Household Head Attend -16.308 35.729 Public School ? (36.20) (41.78) Age of Spouse 10.090 (11.69) Age2 of Spouse -0.141 (0.14) Schooling of Spouse 47.378 (15.14) Schooling2 of Spouse 0.351 (0.97) Training of Spouse 108.219 (38.99) Did Spouse Attend -113.510 Public School? (41.23) Unearned Income 0.008 0.006 (0.00) (0.00) Rural Dummy -153.257 -44.299 (30.07) (35.87) Adjusted R2 0.192 0.244 Sample Size 4377 3325 Note: Dependent variable is total household expenditure per adult Significant at 10 percent, at 5 percent, at 1 percent level. Standard errors in parentheses. 55 PRE Working Paper Series Contact lTle Authott for1ap WPS443 The Inflation-Stabilization Cycles Miguel A. Kiguel in Argentina and Brazil Nissan Liviatan WPS444 The Political Economy of Inflation Stephan Haggard June 1990 A. Oropesa and Stabilization in Middle-income Robel Kaufman 39176 Countries WPS445 Pricing, Cost Recovery, and Rachel E. Kranton June 1990 W. Wright Production Efficiency in Transport: 33744 A Critique WPS446 MEXAGMKTS: A Model of Crop Gerald T. O'Mara July 1990 C. Spooner and Livestock Markets in Mexico Merlinda Ingco 30464 WPS447 Analyzing the Effects of U.S. Gerald 1. O'Mara July 1990 C. Spooner Agricultural Policy on Mexican 30464 Agricultural Markets Using the MEXAGMKTS Model WPS448 A Model of U.S. Corn, Sorghum, Richard E. Just July 1990 C. Spooner and Soybean Markets and the 30464 Role of Government Programs (USAGMKTS) WPS449 Analysis of the Effects of U.S. Richard E. Just July 1990 C. Spooner Macroeconomic Policy on U.S. 30464 Agriculture Using the USAGMKTS Model WPS450 Portfolio Effects of Debt-Equity Daniel Oks June 1990 S. King-Watson Swaps and Debt Exchanges 31047 with Some Applications to Latin America WPS451 Productivity, Imperfect Competition Ann E. Harrison July 1990 S. Fallon and Trade Liberalization in 38009 the C6te d1voire WPS452 Modeling Investment Behavior in Nemat Shafik June 1990 J. Israel Developing Countries: An 31285 Application to Egypt WPS453 Do Steel Prices Move Together? Ying Qian June 1990 S. Lipscomb A Cointegration Test 33718 WPS454 Asset and Liability Management Toshiya Masuoka June 1990 S. Bertelsmeier in the Developing Countries: Modern 33767 Financial Techniques -- A Primer PRE Working Paper Series Contact SAuthoDa for eL WPS455 A Formal Estimation of the Effect Junichi Goto June 1990 M. T. Sanchez of the MFA on Clothing Exports 33731 from LDCs WPS456 Improving the Supply and Use of S. D. Foster June 1990 Z. Vania Essential Drugs in Sub-Saharan Africa 33664 WPS457 Financing Health Services in Africa: Germano Mwabu June 1990 Z. Vania An Assessment of Alternative 33664 Approaches WPS458 Does Japanese Direct Foreign Kenji Takeuchi June 1990 J. Epps Investment Promote Japanese Imports 33710 from Developing Countries? WPS459 Policies for Economic Development Stanley Fischer June 1990 WDR Office Vinod Thomas 31393 WPS460 Does Food Aid Depress Food Victor Lavy July 1990 A. Murphy Production? The Disincentive 33750 Dilemma in the African Context WPS461 Labor Market Participation, Returns Shahidur R. Khandker July 1990 B. Smith to Education, and Male-Female 35108 Wage Differences in Peru WPS462 An Alternative View of Tax Anwar Shah July 1990 A. Bhalla Incidence Analysis for Developing John Whalley 37699 Countries WPS463 Redefining Government's Role in Odin Knudsen August 1990 K. Cabana Agriculture in the Nineties John Nash 37946 WPS464 Does A Woman's Education Affect Shoshana Neuman August 1990 V. Charles Her Husband's Earnings? Results Adrian Ziderman 33651 for Israel in A Dual Labor Market WPS465 How Integrated Are Tropical Timber Panos Varangis August 1990 D. Gustafson Markets? 33714 WPS466 Is There An Intra-Household Lawrence Haddad August 1990 J. Sweeney Kuznets Curve? Ravi Kanbur 31021 WPS467 Structural Adjustment and Living Nanak Kakwani August 1990 B. Rosa Conditions in Developing Countries Elene Makonnen 33751 Jacques van der Gaag WPS468 Does the Structure of Production Indermit Gill August 1990 M. Abundo Affect Demand for Schooling in Peru? 36820