Assessing the productive benefits of nutrition and health: An integrated human capital approach

JOURNAL OF

Econometrics

ELSEVIER

Journal of Econometrics 77 (1997) ! 41 - 158

Assessing the productive benefits of nutrition and health: An integrated human capital approach T. Paul Schultz Deparonent of Economics, Yale University. New Halbert,CT 06520-8269. USA

Abstract Nutrition programs may be evaluated by comparing the productivity of individuals who have benefited from a program to the productivity of similar individuals who have not benefited. To perform such an evaluation a model of the demand for several distinct forms of human capital may be required, of how public agencies and private firms work with households to produce h u m a n capital, and of how these investments increase the productivity of individuals. An integrated wage function with endogenous human capital might then be estimated that provides policym,akers with a tool for simulating the private and social returns to nutrition.

Key words: Nutrition; Health; Program evaluation; Human capital JEL ,,lass!fication: 1! 2; J31; J24

I. Individual productivity effects of human resources Human resource programs are evaluated by comparing the productivity of individuals who have benefited li-om a specific program to the productivity of similar individuals who have not had this benefit. To implement such an evaluation may require a model of the demand of households and individuals for several distinct forms of human capital, of how public agencies and private firms work with households to produce these forms of human capital, and of how these investments increase the lifetime productivity of individuals and thereby contribute to economic growth and socio-demographic development.

An early version of this paper was prepared for the World Bank Human Resources Development and Operations Policy, Washington, DC. Comments on a previous draft are appreciated from A. Bhargava, G. Psacharopoulos, H. Patrinos, R. Chase, A. Judd, G. Ranis, other participants at World Bank workshop, and anonymous referees to this journal. 0304-4076/97/$15.00 i~)~ 1997 Elsevier Science S.A. All rights reserved P11S0304-4076(96)01810-6

L,:.2

T.P. Schult: / Journal of Econometrics 77 (1997) 141-158

This paper proposes to estimate an integrated wage function that provides the policy maker with a basic tool for simulating the private and social returns to human resource stocks, specifically those in schooling, vocational experience, child nutrition, migration, adult health and nutrition, and family planning. To complete the task of program evaluation, information is needed on the value of public sector subsidies and private household resources used to produce the various forms of human capital. Many studies of education, nutrition, health, labor mobility, and training have sought to measure the contribution of these forms of human capital to the productivity of workers and to modern economic growth. The base of knowledge in this field is growing rapidly but is unavoidably qualified because returns to investments in human capital are only realized over a lifetime, in addition, without true social experiments designed to assess how randomized po!icy interventions work through the actions of the family and individuals, statistical estimation of cat:sal relationships may be biased. Regardless, an unusual consensus has emerged between macro- and microeconomists that recent periods of sustained growth in total factor productivity are dependent on improvements in a population's nutrition, health, education, and mobility. Integrated analysis of these various forms of human capital is now needed to measure with greater precision each of their effects on economic development, because stocks of human capital acquired by individuals may not he i~dependent, they potentially interact with each other in t~.~eirimpact on the m~oductivity of the worker, and they are often subject to diminishing (or increasing) returns. Traditional semi-log-linear approximations for wage functions should, thereibre, be extended to more flexible specitications that allow Ior rerun,is to vary and interactions to exist among several forms of human capital in addition to education. Finally, in considering the effects of human capital oq development, we arc primarily interested in how state and family investments influence the formation of reproducible human capital and how it in turn impacts on labor earnings and growth. The exogenous endowments of workers that are essentially unaffected by state and family actions are less central to our objective, except as they mask or modify measured returns to the reproducible component of human capital. ~ This latter source of variation might be called individual heterogeneity. t Do the endogenous components of the variation in human capital exert the same "effect' on labor productivity as do the exogenous endowments? If they differ, it is the productive effects of endogenous reproducible human capital that policymakers need to assess. Speciticaily, much of the variation in height, for example, is due to differences in fixed endowment potential. It is not clear wl,ether differences in this endowment have the same productive effect on workers as do differences caused by variation in childhood health and nutritional investments. Specification tests can be implemented to assess whether different sources of variation in human c.'tpital stocks are exoge~ous or endogenous, and thus whether specific forms of human capital are justifiably treated as exogenous or endogenous variables when estimating integrated wage functions ISchultz, 1994~.

Family of Ori~qn Parent Education Parent Occupation Residence Location Family Heterogeneity

Individual Endowment Heterogeneit~ Stature Frailty Ability

i

Human Resource Stocks Childhood: Nutrition/Stunting (Height) Schooling (Years Completed) Adult: Migration (Residence,birthplace) Nutrition/Health (Body Mass Index) Fertility (unwanted/unplanned)

÷

Hours (per year) Earnings (per year) Inequality and Equity

Wage (hourly rate)

Adult Productivity Household Production Entry into Market Labor Force

4,

Derived Demand for Labor Relative Requirements for Physical Power Educated Labor Female versus Male

Fig. i. Human resource programs and their consequences on productivity.

Local Human Resource Programs Food Prices (Subsidies/Security) Preventative Health ($, type) Orative Health ($, type) School System (distance, fees. quality, $) Job infonrmtion system Credi~ for schooling and migration

Parem .Inputs.and Tmn~ers to Children Endowments Investments in Human Capital Transfers, Loans, Bequests

4,

Regional C~racteristics Cli~JRalafaiyDiseases Transpmtmion Infrdstmeture Mammumomic Conditions for predictable employment oppmtmfities and stable returns to Inmmn resource investments Community Heterogeneity

L~

I

2

144

T.P. Schultz / Journal of Econometrics 77 (1997) 141-158

Fig. 1 illustrates the overall system that constrains parents' demand, given their preferences and resources, for investments in their children's human capital that are expected to affect the children's productivity as adults, approximated in this paper along a single dimension, that of wage rates. Variables could be observed at four levels: the child, the family, the community/region, and possibly the nation. Unobserved variables or heterogeneity could exist at all levels, but for estimating without bias the relationship between human capital stocks and adult productivity, one should be particularly worried about individual, family, and community heterogeneity that might be plausibly correlated with both observed human capital stocks and unobserved productive capacities of the children. The objective of an integrated human resource program evaluation is to estimate three relationships: the social opportunity cost of public subsidies and private household resources needed to produce an increment to a particular stock of human capital, the determinants of the household demand for that human capital, and the wage {i.e., productivity} returns to that human capital stock in the labor market. Many conceptual and statistical issues arise in jointly estimating without bias these behavio..1 and technical linkages in a specific setting, not least of which is the presumption underlying most such studies that the wage structure is stable, allowing the wage pattern across age groups at one point in time to provide a projection for the lifecycle relative returns for a representative child."

2. Individual productivity function Social resources are used .'.o produce stocks of human capital that enllance the productive capacity of the individual over the long run. How individuals allocate their productive capacity ....whether they make further investments in

-'Wage returns to human resource investments should be formulated in a life cycle context. The current discounted value of the stream of future wage increases that occur because of such investments should be compared to the current discounted opportunity costs of the time and goods foregone to create these investments. Without retrospective information on investment costs and wage proliles for birth cohorts who have completed their productive lifetime, empirical studies of hulnan capital have generally substituted cross sectional lacross age groups at ol;,2 time} far longitudinal data lover a lifetimeS.Becker ( 1964} painstakingly describes the assumptions required to justify this empiric~d simplilication. These assumptiot~s are elaborated by Mincer 11974~ to permit him to summarize education/age wage structures by a earnings regression, in which the estimated coellicient on years of education is an estimate of the private rate of return to schooling. Others who follow this path might explore the sensitivity of their findings to alternative assumptions, for example, by estimating wage returns for several narrower age groups, combined with suitable discounting.

T.P. Schultz / Jounlal of Econometrics 77 (1997) 141-158

145

themselves, work in the home, work in the labor market, or engage in leisure activities - should not necessarily influence how we measure the private return to that human capital. Because of the difficulty of assigning a value to the output of labor per hour in self. employment, home production, or leisure, it is a common practice to analyze (Heckman, 1979) the market productivity of persons who work for a wage in the labor market, and then correct for any bias that may arise from analyzing only this selected sample. The critical requirement to correct for sample selection bias is that there is some observed variable that affects the probability of working for a wage but does not affect the market-offered wage. Identification achieved by assumed functional form is not likely to be robust in this case. A natural choice for an exclusion restriction that could identify the sample selection procedure is a variable that affects only the individual's marginal productivity in nonwage production, or the individual's utility from leisure, or both, but is uncorrelated with the market wage. Inherited land, business assets, or wealth are expected to raise the individual's productivity in nonwage activities and tilereby to decrease the probability she or he would work for others as a wage earner. These wealth variables are assumed to exert no effect on market productivity or the wage offer. Individuals who are observed not to work for wages are assumed to value what they produce with their time more highly than the market wages they could receive. It is common to assume tha~ the variables which are observed ,o influence market wage opportunities, such as education, also affect nonwage opportunities, roughly in the same proportion. Without more information on the structure of the nonwage production function and the utility function, the market wage function is a~sumed to signal the marginal productivity of human resource investments in wage or nonwage activities. :~ These productive effects are thea independent of how time is allocated between wage and nonwage activities. This assumption may be particularly important for evaluating how human resource programs enhance the productivity of women because only a small fraction of adult women in many low-income countries work for wages. More research should assess the adequacy of these methods for inferring the productive gains from human resources. The wage function is determined by several types of conditioning variables. First., there are initial endowments, which may be thought of as fixed or at least

•~The productivity of labor in nonwage activities for those who do not participate in wage activities can be inferred only if the functional form of nonwage marginal product function is known, and its identifying argument known and observed, if the underlying nonwage (home)production function is subject to constant returns to scale, as postulated by Becker (1965)~ the inframarginal producer surplus in nonwage production accruing to labor would be zero.

146

T.P. Sehultz / dournal of Economeo'ics 77 (1997) 141-158

not affected by the input of family or public resources.'* Second, there are several dimensions of human capital, which are produced means of production, formed jointly by family and individual private inputs of time and goods, and by public inputs of goods and services, typically provided in the region of residence. The production possibilities to create human resources are constrained by biology, technology, and perhaps, in the short ruin, by organizational limitations. The third type of variable entering the human resource production function is age or post-schooling experience. Experience acquired after completing school is expected to enhance labor productivity, at least initially. Mincer (1974) found a quadratic function in years of potential post-schooling experience {i.e., a g e - years of schooling- seven) approximated reasonably well U.S. male cross-sectional earnings data from the 1960 Census. This quadratic specification could be derived from a model in which the value of a worker's time invested in on-the-job training declined linearly with post-schooling experience. With su~cient data, higher-order polynomials in potential post-schooling experience can also be added to improve the empirical fit of the wage function (e.g., Murphy and Welch, 1990). For other human resource accumulation processes, such as gestation, birthweight, height, weight-for-height, age may be more relevant than post-schooling experience. When education is treated as endogenous, then potential post-schooling experience as defined by Mincer also becomes endogenous. Experience should then be replaced by an unrestricted polynomial in age to avoid introducing another endogenous variable. In a cross-section, it must be remembered that age (or experience) captures both life cycle accumulation (or depreciation) in the rental value of human capital stocks and unobserved differences across cohorts.

3. Types of human capital Five tbrms of human capital are distinguished here that may increase the lifetime productivity of a worker, although the relative returns to different types

'*Griliches ( 1977} summarizes the first decade of evidence on tile bias in estimating wage functions duc to omitting ability (individual heterogeneity in Fig. I). He emphasized that the upward bias in estimated wage returns on observed schooling due to the omission of ability was partially offset by the errors in measurement of schooling Ior other human capital stocks in Fig. I I. in health and nutrition, even the birthweight of the child is not a good measure ofexogenous health endowment or frailty, but exhibits the influence of a mother's prenatal behavior (Roscnzweig and Schullz, 1983}. If individual heterogeneity is an important determinant of human capital investments by parents and society, thea panel data on individuals may be particularly usefid for the estimation of individual fixed effect models lStrauss and Thomas, 1995). These panels may also permit more systematic treatment of errors in measurement in anthropometric measures of health and nutrition (Schultz,

1994),

T.P. Schultz / Journal of Econometrics 77 (1997) 141-158

147

may vary by income and technological stage of the society, and each type may, after a certain point, be subject to diminishing returns. The first is net childhood nutritional status, which is cumulatively measured by early physical growth and development, such as height-for-age and weight-for-age. Some argue that height at the fourth birthday is already a good predictor of an individual's final adult height (Martorell and Habicht, 1986), but there are few contemporary studies that measure the capacity of a malnourished young child to catch up in adult stature when provided an improved diet. Adult height is, moreover, an important determinant of adult productivity, and re-emerges as inversely correlated with chronic health problems among the middle-aged and elderly (Fogel, 1991). Finally, height is inversely related to mortality and consequently length of productive life. In addition to improving adult productivity and health, childhood nutritional status may also enhance the performance of children in school and in early training tasks, and thereby affect the return to other human capital investment activities (Moock and Leslie, 1986; Behrman, 1993). Alternatively, the covariation of nutrition and school achievement among children as observed in the empirical literature might be caused by unobserved heterogeneity of families and children, as would occur if credit constraints or preferences affected families in their investments in both nutrition and schooling. Nutritional status in childhood is the sum of several factors, of which nutritional intake is probably the most important. Another is the exposure to infection and disease, which places extra demands on nutritional inputs and reduces the etficiency with which the body can absorb nutrients, such as through the incidence of diarrhea, or development of immunities. The child's work load can also place different demands on diet. Micronutrient deficiencies may have particularly high benelit-cost ratios, but summary measures of the deficits among aduhs are not yet in common use in survey research on low-income populations (Basta et al., 1979). Poor nutritional status is most commonly revealed in adults by stunti~lg or by their mature height deficit. The second form of human capital is schooling. Children start school at very different ages, from age 5 to 10., and continue for different numbers of years, includil~g repetitions. The length of school terms and the number of hours a day the school provides instruction varies across and within countries. The attendance rate of 'enrolled' pupils during the regular school term is also far from uniform. If there were a consensus on how to measure school quality per hour attended, this would also exhibit substantial variation. Consequently, describing investments in schooling by "years completed' is only a crude firstapproximation. The third form of human capital is miyration, which may occur repeatedly over a lifetime and involve return migration. To simplify, consider any movement of adults from their region of birth as migration, which most often occurs in the first few years after completing school. Migration occurs more frequently for more educated individuals (Sabot, 1982), suggesting again two alternative

148

T.P. Schultz / Journal of Econometrics 77 (1997) 141-158

hypotheses. Education and migration may be complementary forms of human capital, or families that invest more in the education of their children also are more likely to finance their migration, possibly because they face a lower cost of investment funds (Becker, 1967), or because they simply have a stronger taste for child human capital. Much of the return to education for a rural-born individual may be realized only through migration in a country such as Columbia, suggesting a close linkage between these investments (Schultz, 1988a). Migration may not be as closely associated with increasing market-productive opportunities for women as for men, because families often move together, and collectively families may weigh the earnings opportunities of males more heavily than they do those of females (Mincer, 1978). The fourth form of human capital is the capacity to avoid unwanted fertility, which enhances women's market productivity by allowing them to continue their education, to migrate where their skills are most valuable, or to allocate time to their most rewarding work. To the extent that public resources that subsidize family planning and provide associated information about birth control can delay a woman's first birth or reduce the subsequent number of births, a woman's market human capital is likely to increase. Consequently, the impact of family planning subsidies may first help women avoid unwanted childbearing, and second enhancing their market productivity, that will generally lead them to lower their desired fertility. It is not surprising that family planning is viewed as a means to 'empower' women because it is likely to increase their economic opportunities. Little analytical effort has been directed to measuring family planning as a mechanism for enhancing women's human capital. The fifth and final form of human capital is an adult's current health and nutritional status, proxied by weiqht-lbr-heitjht-squared (Body Mass Index = BMI). BMI affects the current productivity of the individual, particularly at low levels of calories and for energy-demanding tasks. This indicator of nutritional status among adults should be treated as simultaneously determined with the choice of productive activity, because increased income can also support increased current expenditures on nutrition and the performance of more demanding jobs (Strauss, 1986; Pitt et al., 1990; James and Ralph, 1994). Unbiased estimates of the one-directional effect of improved adult nutrition on wage or self-employment productivity requires valid instruments that predict current nutritional intakes or BMI and that are uncorrelated with the error in the wage function. The prices of nutrients in the locality or the cost of health care are possible instruments for this purpose (Strauss, 1986; Deolalikar, 1988; Sahn and Alderman, 1988). The functional form of the relationship between height, weight, and adult productivity or wages is unresolved. Waaler (1984) documented that height, weight, and BM! were predictors of mortality in the Norwegian population and reported the specific diseases that were associated with these anthropometric indicators. Fogel (1991) showed in a sample of American Civil War recruits that

T.P. Schultz / Journal of Econometrics 77 (1997) 141-158

149

height was associated with relative risks of morbidity and mortality in much the same absolute manner. Fogel (1994) went on to explain more of the variation in Norwegian mortality by conditioning on both height and BMI, while allowing for their nonlinear effects. An attraction of BMI as a measure of wasting in a malnourished population, instead of weight, is that BMI tends to be nearly orthogonal to height, and this reduces multicollinearity when an outcome such as wages is to be explained by both variables. If, however, the benefits to productivity/wages from HT and BMI are expected to diminish with scale, as Waaler and others find, then expressing HT and BMI in logarithms might be appropriate (Strauss and Thomas, 1995), in which case the linear specification is algebraically indistinguishable from one that simply includes the logarithms of height and weight. Tests for whether height and weight should be combined in explanatory models are provided by Bhargava (1994). The specific connection between childhood nutritional status and health and later adult growth and productivity in low-income countries has only recently begun to receive concerted attention by economists. Human capital is also accumulated in the form of vocationally relevant experience. The actual amount of a workers time invested in the acquisition of such skills or learning-by-doing on a job is difficult to observe and is probably an endogenous decision of the worker and firm. Consequently, this form of human capital is imperfectly measured or controlled in the wage function. As discussed earlier, it is commonly represented by a polynomial either in years experience after leaving school or in age itself (Mincer, 1974).

4. Problems in applying tile framework to program evaluation Only five human capital stocks are considered to simplify exposition, to illustrate the potential for integrating human resource program evaluation~, and to offer one possible specification to guide empirical analysis. There are :many limitations to such a scheme. Health is not separately distinguished as a human resource goal, but the indicators of childhood and adult nutrition are selected because of their high correlation with childhood and adult mortality, s Although these nutritional indicators are less well documented as predictors of morbidity, they nonetheless may summarize parsimoniously the cumulative repercussions of disease and nutrition on adult well-being and the quality of life (Stewart and Ware, 1993; Broome, 1993; James and Ralph, 1994). SChronic health problems measured by clinical investigations among the middle aged can be predicted by adult height IFogel, 1994). Physical limitations on daily activities arc also associated with height (Stewart and Ware, 1993). Body Mass Index is related to reproductive success, health status, and productivity (James and RalplJ, 1994). Days ill and unable to work in the last month are associated with lower wage rates (Schullz and Tansel, 1992).

150

T.P. Schultz / JourJml of Econometrics 77 (1997) 141-158

This approach to evaluating health and nutrition program interventions within an integrated human capital demand and wage framework calls for the coordination of many types of data, some of which are not commonly available today. A household labor-market survey would collect standard wage and labor force activity information in combination with various measures of human capital stocks, such as height and weight for all adults, and nonearned income and assets, to account for who is a wage earner. A local residential community survey for each sample cluster would document current distance to human resource institutions (i.e., schools and clinics), their quality, fees, and when they were initiated, food and medical input prices, climate, and infrastructure that might affect exposure to disease, such as water and sanitation facilities. In addition, retrospective information is needed on the individual's birthplace, migration history, parents education, occupation, wealth, height, and weight. What is lacking in most surveys is information about public community programs in the childhood residence that could account for part of the variation in human capital stocks the individual acquired as children. This childhood community information might be obtained directly from the respondent and indirectly from merged administrative records, an local health and education programs when the migration history of the individual is known. In estimating a wage function that relates human resource stocks to adult productivity, several econometric specification issues are paramount. First, which forms of human capital are exogenous oz"endogenous; in other words, are these stocks uncorrelated or correlated with the unobserved wage ,determinants impounded in the error of the wage function? Individual, family or community heterogeneity could be a plausible reason for human capital slocks to be endogenous. Characteristics of the region of ,'esidence have been used to identify forces that may plausibly influence tile accumulation of human capital stocks. The simultaneous determination of wages and adult weight (BMi) was noted as an instance where the effect of adult weight on wages could only be eslimated by instrumenting weight by local prices of food, availability of health care, or food-for-work programs for unemployed (e.g., Strauss, 1986; Deolalikar, 1988). Although fertility and migration are often viewed as choice variables in an economic household production framework (Schultz, 1974, 1992; Sabot, 1982), few studies attempt to endogenize migration and unwanted fertility in a wage functionP Education is generally treated as exogenous, but the decision to continue in school is occasionally modeled along with the wage function conditional on that educational choice (Willis and Rosen, 1979). Int~'rgenerational studies of the schooling of children could treat childhood access to schools and health care and their quality as public program inputs that may be

~'The estimated effect of one unanlicipated birth is to reduce the wage of Malaysian and U.S. married women by about ten perccu| (Schultz, i~92, Table 5.1).

T.P. Schultz / Journal of Economeo'ics 77 (1997) 141-158

151

expected to influence household investments in human resources but not otherwise impact on the wages of the children when they are adults. IV estimates based on community programs should eliminate bias due to classical errors in measurement of the human capital stocks, and avoid simultaneous equation bias caused by human capital inputs being correlated with the error in the wage function, or being endogenous. But community programs may not be valid instruments because their placement is correlated with omitted variables, such as the healthiness of the environment, or because selective migration of parents to communities with better programs may generate a correlation between unobserved parent preferences and programs (Rosenzweig and Schultz, 1983; Rosenzweig and Wolpin, 1986; Pitt et al., 1993). Community program variables may also be 'poor' instruments in the sense that they explain relatively little of the variation in human capital stocks. There is a typical trade-off of consistency and efficiency. Schooling is supported by the family at origin and is to some degree regulated by the school itself, as it sets standards for completing each year. It is not implausible then to find growing evidence that years of completed schooling appears to be exogenous in a wage function, or in other words that the estimated effect of schooling on wages is approximately the same whether estimated by OLS or IV (Angrist and Krueger, 1991; Schultz, 1994). But forms of human capital that are determined later in an adult's lifetime are more clearly endogenous to the wage function, such as on-the-job training, job search and turnover, job tenure, migration, and, at least for women, fertility. As noted earlier, childhood nutrition and health conditions can be summarized by adult height, whereas adult health and nutrition can be summarized by weight-to-height (BMI). Both of the nutrition-health proxies have been implicitly assumed exogenous by economic historians for the purposes of inferring their effect on labor productivity (Fogel, 1986, 1994; Floud et al., 1990). But development economists have recently considered weight and height as potentially endogenous variables in the labor productivity function (Strauss, 1986; Deolalikar, 1988; Thomas and Strauss, 1996; Schultz, 1994). The challenge is to measure the labor productivity improvements released by short-run and longer-run cumulative advances in nutritional status, as proxied by svch anthropometric indicators as height and weight-to-height. There are other possible econometric approaches to estimating the worker productivity benefits of health and nutrition program interventions (e.g., Basta et al., 1979). The least restrictive is estimating a reduced-form equation for wages, in which the human capital inputs are replaced by the instruments representing the human resource programs or policy interventions. With reduced forms, however, it is not possible to determine how the effect of a program operates thorough changing various types of human capital, for example, how a child health program might affect adult wages through improved adult height

152

T.P. &'huhz / Journal of Econometrics 77 (19971 141-158

or through increased education. It is also not possible to explicitly consider the exogeneity of the human capital inputs by Hausman (1978) specification tests. Another econometric approach to mitigate bias due to omitted variables or heterogeneity introduces fixed effects. These fixed effects at the community level eliminate bias due to persistent regional variables that are correlated with household or individual variables and wages, but they also prevent the estimation of the effect of any community program variables in a single cross-section. With a sequence of repeated cross-sections or a panel on individuals it may be possible to estimate the effect of changing program interventions with fixed community effects or random individual effects (Bhargava and Sargan, 1983). One disadvantage of exploiting only the time series variation in panel data (or relying on fixed effects in this form) is that the errors in measurement tend to be relatively magnified, biasing the effects of human capital inputs in the wage function toward zero tGriliches and Hausman, 1986). Estimating changes in panels using fixed effects, it may still be necessary to treat changes in the human capital variables as endogenous and rely on instrumental variables estimation methods. Pre- and post-intervention outcomes have been traditionally compared in policy evaluation studies at the aggregate level and can be studied as well by this first-differenced methodology at the household level. Control variables can then be allowed to interact with the intervention as they affect wages in a differencein-difference methodology. All of these approaches typically assume that program variation is exogenous to wages. If this null hypothesis is not maintained, the econometrician is challenged to explain the political economy of program placement, and develop an explicit basis for identifying the effects of programs as an endogenous variable in a larger model {Rosenzweig and Wolpin, 1986). These econometric issues of bias and estimation of wage functions have been explored primarily in reference to schooling (Griliches, 1977; Lam and Schoeni, 1993; Angrist and Krueger, 1991; Ashenfelter and Krueger, 1994). Researchers have only recently begun to treat current nutrition as a potentially endogenous input to the wage function (Strauss, 1986; Strauss and Thomas, 1995). In an analysis of surveys in Ghana and Cfte d'Ivoire {Schultz, 1994) the anthropometric indicators of height and BMI appear to be endogenous in the wage function of men and women, exerting a much larger effect on wages according to IV than OLS estimates. Education and migration, on the other hand, did not exhibit a significantly different effect when estimated by IV or OLS methods (Hausman, 1978).

5. Critical aspects of programs to evaluate

Three questions need to be answered in setting public sector priorities to help the family invest optimally in various forms of h,,','.,~n resources and encourage

T.P. Sch~dtz / Journal of Econometrics 77 (1997) 141-158

153

social development. The first question is how to get the most benefit from a given expenditure on allied programs. This measure of efficiency should explicitly allow for the likely effects of one program on the outcomes of all other programs, or cross-program effects. The second question is how program benefits are distributed across types of individuals and families, such as the rich and poor, which may inform us about the equity of the program. The third question is how the cost effectiveness of programs would differ if they were in the private or public sectors. In many spheres it is difficult to achieve the same efficiency in the public sector as in the private sector, but the private sector may not be able to reach the same target groups that the public sector can because of their different organizational structures and need to collect fees for services. Benefits from different functionally-oriented programs are not readily compared, such as nutrition and education. Much work remains to be done before it is possible to compare outcome measures comprehensively, such as a prevented birth, a prevented death, or reduced morbidity, in comparable welfare units. This general approach has two clear limitations. First, the distribution of benefits across groups can be readily evaluated only when the groups are defined in terms of exogenous variables, i.e., where group membership is not related to choices and allocational decisions made by the observed individuals and families. Second, the spatial variation in programs and policies must be assumed random with respect to unobservables, notably the preferences of the population and productivity and healthiness of regional environments. 7 Two programs with a shared objective may strengthen the impact of both programs in achieving their common goal, or one may weaken the independent impact of the other program and vice versa. The former, complementary effects, are most fi'equently documented across different types of human capital investment programs, for example, improvements in child nutrition/health permit children to learn mole at school (Moock and Leslie, 1986; Gomes-Neto et al., 1992), and healthier children can expect to live a longer, healthier life during which to earn market returns frown schooling (Floud et al., 1990). The latter, substitution effects between two social programs, can be expected when the programs are directed to achieving the same end, but through alternative mechanisms or motivations or instruments, such as may be the case when public and private sector programs seek to satisfy the same demands for family planning services. These potential synergies between social programs, either positive or negative, may change with the scale of interventions, possibly reinforcing each other

7To deal with the migration of individuals to regions that provide preferred social programs, or program placement in regions with distinctive populations or welfare problems, raises further identification problems (e.g., Rosenzweig and Schultz, 1983; Rosenzweig and Wolpin. 1986: Pitt et al., 1993).

154

T.P. Schultz / dounzal oJ Sconometrics 77 (1997) 141-158

at low levels and then substituting for each other as saturation is achieved at higher levels. For example, in a child health and nutrition program it may be useful to combine doctors and nurses into different types of programs, some stationary in hospitals and clinics, while others are mobile in outreach teams that educate and promote child health and nutritional practices. Both types of program personnel may be motivated to improve health, but they use different mixes of trained manpower and different organizational delivery systems. In some contexts the clinics and outreach programs may reinforce each other, and in other cases they may substitute for each other. Thus, these cross-program effects must be empirically assessed to improve estimates of how effective public sector program efforts will be in different circumstances. 8 If the income effects associated with the benefits of these types of social programs were negligible, then household demand theory predicts that the estimated (uncompensated) cross-program effects should be symmetric, or of the same sign and equal in magnitude. Estimates of such program interaction effects are more reliable if quadratic terms are also included for the two program effort variables, in which case the specification can be generally interpreted as a second-order Taylor series approximation for any functional relationship between the program input variables and the measure of program output. It is then possible to infer how the returns to each program varies with the scale of program effort, and hence how the marginal returns to program inputs may differ from the average returns. The objective of public policy should be to achieve the same marginal return from equal marginal outlays in various programs, given that both programs have the same objective (Schultz, 1971, 1988b, 1992). Most public human resource programs provide a service that can also be obtained through private markets, although perhaps in a different quality or form. It should be expected that when public subsidies are provided for a good or service that is also available in the private market, some consumers will switch from private to public providers because of the public subsidy, without necessarily changing outcomes or behavior. The only way to assess accurately the effect of a subsidy to either program is to analyze both programs together. Overstating public program effectiveness by only counting demand for the public program is common. '~

SOne simple way to estimate the sign and magnitude of cross-program effects is to add interaction t,ariables between allied htlman resource and social welfare programs to models of output (wages). An early example of this approach applied to the study of fertility is found in Schultz (1971). '~For example, the physical supply of contraceptives distributed freely by a public program is likely to overstate the added contraceptive protection provided to the population by tile program (Scllultz, 1971). Some contraceptors will merely shift their source of supply without improving their contraceptive cflicie,lcy. This is of course only a special case where parallel programs may exist with approximately the same objective.

T.P. Schultz / Journal of Econonwtrics 77 (1997) 141-158

155

To assess how social program effects are distributed, it is convenient to add additional interaction ~ariables between program t,'eatments and exogenous groups. In the earlier illustration of a child nutrition and health program, it might be hypothesized that it is particularly difficult for the less educated women to assess novel nutrition and health technologies, and program inputs are of only little value to the more educated women. Then, public subsidies for these programs should have their greatest impact on the child nutrition and the later productivity of the children of less educated women. An analogy to farm-extension activities is clear where they raise the profits of less educated farmers by a greater proportion than the profits of more educated farmers (Birkhauser et al., 1991).

6. Conclusions for program evaluation

Household survey data from individuals on child health, nutrition, child schooling, anthropometrics, fertility, wage rates, and other sources of income, and household expenditures can be usefully combined with regional data for the residential area on public expenditures on human resource programs. These merged data provide a flexible basis for assessing the success of human resource programs to help individuals increase their health and productivity and families to cope with the challenges of economic and demographic change in the low-income world. Putting modern technologies to effective use by families to protect their health, provide for their nutrition, control their reproduction, and educate their children are closely related achievements lhat do not proceed independently, if public objectives can be achieved by both private and public sector providers of nutrition, schooling, health, or family planning, the simultaneous analysis of both public and private providers is mandatory, in some parts of the world, the public sector may not be the most cost-effective or equitable provider of basic services, even those that are traditionally associated with the public sector, such as health care (Birdsall and James, 1993). The prices and quality of services in the public and private see~ors should be analyzed together with the traditional household demand data on expenditures, time allocation, wages, prices, and nonearned income to isolate the payoff to expanding public sector subsidies. The personal distribution of the wage gains from social welfare programs is rarely estimated but should become an important ingredient in deciding what goods and services the public sector should provide and to what segments of the population they should be subsidized. Subsidies for some public sector services may benefit predominantly the poor and help them overcome ~heir initial economic disadvantages. These subsidies should be associated with families achieving for themselves greater intergenerational mobility, through their improved control of unwanted births and the increased health, nutrition, and

156

T.P. Schultz /Journal of Econometrics 77 (1997) 141-158

education of their children. Other public sector services, such as urban hospital care and university education, may benefit predominantly urban middle and upper classes. These public services may appear to be inequitable income transfers from taxpayers to the rich, without notably affecting average wages, mortality, schooling, or fertility. Identifying which public services should become self-financing by fees-for-service may help sustain and expand government assistance for human resource programs where they remain a cost-effective and equitable human resource investment to increase individual productivity, alleviate poverty, and accelerate growth.

References Angrist, J.D. and A.B. Krueger, 1991, Does compulsory schooling affect schooling and earnings?, Quarterly Journal of Economics !06, 979-1014. Ashenfelter, O. and A. Krueger, 1994, Estimates of the economic returns to schooling from a new sample of twins, American Economic Review 84, 1157-1173. Basta, S.S., M.S. Soekirman, D. Karyadi, and N.S. Scrimshaw, 1979, Iron deficiency anemia and the productivity of adult males in Indonesia, American Journal of Clinical Nutrition 32, 916925. Becket, G.S., 1964. Human capital (Columbia University Press, New York, NY). Becker, G.S., 1965, A theory of the allocation of time, Economic Journal "~5,493-517. Becket, G.S., 1967, Human capital and the personal distribution of income, Woytinsky lecture (University of Michigan, Ann Arbor, MI}. Behrman, J.R., 1993, The economic rationale for investing in nutrition in developing countries, World Development 21, 1749 1771. Bhargava, A,, 1994, Modelling the health of Filipino children, Journal of Royal Statistical Society A 157,415 432. Bhargava, A. and J,D. Sargan, 1983, Estimating dynamic random ell~cts models from panel d~lta covering short time periods, Ecouometrica 51, 1635 1659. Birdsall, N, and E. James, 1993, Health, government and the poor, in: J.N. Gribble and S. Preston, eds,, The epidemiological transition (National Academy Press, Washington, DC). Birkhauser, D,, R.E. Evcnson, and G. Feder, 1991, The economic impact of agricultural extension: A review, Economic Development and Cultural Change 39, 607-650. Broome, J., 1993. Qalys, Journal of Public Economics 50, 150-167. Deolalikar, A.B,, 1988, Nutrition and labor productivity in agriculture, Review of Economics and Statistics, 70. 406 413. Floud, R,, K, Wachter, and A. Gregory, 1990, Height, health, and history (Cambridge University Press, Cambridge) Fogel, R., 1986, Growth and economic well being: 18th and 19th centuries, in: F. Falkqer and J.M. Tanner, eds., Human growth, Vol. 3 (Plenum Press, New York, NY). Fogel, R., 1991, New sources and new techniques for the study of secular trends in nutritional status, health mortality and the process of aging, NBER working paper no. 26. Fogel, R., 1994, Economic growth, population theory and physiology, American Ecor.omic Review 84, 369 395, Gomes-Neto, J.B., E.A, Hanushek, R,H. Leite, and R.C. Frola-Bezzera, 1992, Health and schooling: Evidence attd policy implications for developing countries, Rochester Center for Economic Research working paper no. 306 (University of Rochester, Rochester NY), Griliches, Z,, 1977, Estimating the returns to schooling, Econometrica 45, 122.

T.P. Schultz /Journal of Economeo.ics 77 (1997) 141-158

157

Griliches, Z. and J.A. Hausman, 1986, Errors in variables in panel data, Journal of Econometrics 31, 93-118. Hausman. J.A., 1978, Specification tests in econometrics, Econometrica, 46, 125 I-1271. Heckman, J.J., 1979, Sample bias as a specification error, Econometrica 47, 153- 162. James, W.P.T. and A. Ralph, eds., 1994, The functional significance of low body mass index, European Journal of Clinical Nutrition 48, Suppl. 3. Lain, D. and R.F. Schoeni, 1993, Effects of family background on earnings and returns to schooling, Journal of Political Economy 101,701-740. Martorell, R. and J.P. Habicht, 1986, Growth in early childhood in developing countries, in: F. Falkner and J.M. Tanner, eds., Human growth, Vol. 3 (Plenum Press, New York, NY). Mincer, J., 1974, Schooling, experience, and earnings (Columbia University Press, New York, NY). Mincer, J., 1978, Family migration decisions, Journal of Political Economy 86, 749-773. Moock, P.R. and J. Leslie, 1986, Childhood malnutrition and schooling in the Terai region of Nepal, Journal of Development Economics 120, 33-52. Murphy, K.M. and F. Welch, 1990, Empirical age earnings profiles, Journal of Labor Economics 8, 202-229. Pitt, M.M., M.R. Rosenzweig, and D.M. Gibbons, 1993, The determinants and consequences of placement of government programs in Indonesia, World Bank Economic Review 7, 319-348. Pitt, M.M., M.R. Rosenzweig, and M.D.N. Hassan, 1990, Productivity, health, and inequality in the intrahousehold distribution of food in low income countries, American Economic Review 80, 1139-1156. Rosenzweig, M.R. and T.P. Schultz, 1983, Estimating a household production function, Journal of Political Economy 91,723-746. Rosenzweig, M.R. and K. Wolpin, 1986, Evaluating the effects of optimally distributed public programs: Child health and family planning intervention, American Economic Review 86, 470--482. Sabot, R.H., ed., 1982, Migration and the labor market in developing countries (Westview Press, Boulder, CO). Salm, D.E. and H. Alderman, 1988, The effects of human capital on wages and the determinauts of labor supply in a developing country, Journal of Development Economics 29, 157-183. Schultz, T.P., 1971, A framework for analysis and its application to Taiwau's family planning program, P-4674 (The Rand Corporation, Santa Monica, CA). Schultz, T.P., 1974, Birth rate changes over space and time, in: T.W. Schultz, ed., Economics of the family (University of Chicago, Chicago, IL). Schultz, T.P., 1988a, Education investments and returns, in: H. Chenery and T.N. Srinivasan, eds., Handbook of development economics, Vol. I (North-l-lolland, Amsterdam). Schultz, T.P., 1988b, Population programs: Measuring their impact on fertility and the personal distribution of their effects, Journal of Policy Modeling 10, 113-149. Schultz, T.P., 1992, Assessing family planning cost effectiveness: Applicability of individual demand-program supply framework, in: J.F. Phillips and J.A. Ross, eds,, Family planning programmes and fertility (Oxford University Press, New York, NY). Schultz, T.P., 1994, Human capital and development, Paper pre~cnted at the 22nd International Agricultural Economics Association meeting, August 28, Conference proceedings tl-tarare, Zimbabwe). Schtdtz, T.P, and A. Tansel, 1992, Measurement of returns to adult health, Economic Growth Cenler discussion paper no. 663 (Yale University, New Haven, CT). Stewart, A.L. and J.E. Ware, Jr., eds., 1993, Measuring functioning and well being tDuke University Press, Durham, NC). Strauss, J., 1986, Does better nutrition raise farm productivity?, Journal of Political Economy 94, 297~ 320.

158

T.P. Schultz /Journal of Econometrics 77 (1997) 141-158

Strauss, J. and D. Thomas, 1995, Human resources: Empirical modeling of household and family decisions, in: J. Behrman and T.N. Srinivasan, eds., Handbook in development economics, Vol. 3A (Elsevier, Amsterdam). Thomas, D. and J. Strauss, 1996, Health, wealth and wages of men and women in urban Brazil, Journal of Econometrics, this issue. Waaler, H.T., 1984, Height, weight and mortality: The Norwegian experience (Gruppe for Helsetjenesteforskning, Oslo). Willis, R.J. and S. Rosen, 1979, Education and self-selection, Journal of Political Economy 87, part 2, $7-$36.