Short- and long-term impacts of economic policies on child labor and schooling in Ghana

SP DISCUSSION PAPER NO.0212 25527 Short re el el MM 1 ; of Economic Policies on fmg Labor ind Schooling in simm Niels-HIugo Blunch, 9U 3 Sudharshan Canagarajah >_ and Sangeeta Goyal May 2002 rot LABOR MARKETS, PENSIONS, ASSISTANCE T H E W O R L D B A N K   Short-and Long-term Impacts of Economic Policies on Child Labor and Schooling in Ghana Niels-Hugo Blunch, Sudharshan Canagarajah and Sangeeta Goyal May 2002   Abstract While the issue of child labor in developing countries has received increased attention in recent years, most of the empirical analysis has been based on one-time cross sectional samples. While this may give an idea of the incidence and determinants of child labor at one point in time, it is silent about the dynamics of child labor over time and sometimes may even be influencing policy choices against child labor adversely. This paper attempts to fill this void, analyzing the dynamics of child labor and schooling in Ghana, aiming at investigating the impact of broad economic reforms on child labor and schooling in the short, medium and long run. Starting from a premise that the simple - direct - relationship between poverty and child labor, which has often been seen as the feature of child labor, may not adequately capture the multi facetted nature of child labor, we find evidence of asymmetries in the child labor-poverty link, as well as quite complex dynamics in the evolution of child labor and schooling and their determinants over time. Most notably, child labor is found to be responsive to poverty in the short run, but not in the long run, while child schooling is unaffected by poverty in short run but responds in the medium- to long run. These results suggest that child labor acts as an economic buffer of the household in the short run, regardless of changes in the economic environment or perceptions of the latter following economic reforms, thus supporting - and refining - the poverty explanation of child labor. JEL classification: I21, J24, 015 Keywords: Child labor, Ghana, human capital accumulation, poverty   Short- and Long-term Impacts of Economic Policies on Child Labor and Schooling in Ghanat Niels-Hugo Blunch, Sudharshan Canagarajah, Sangeeta Goyal* 1. Introduction Child labor is a pervasive phenomenon in developing countries. For example, ILO estimates that there are more than 250 million child laborers between the ages of 7 and 14 in the world of whom at least 120 million are involved in work full time, ILO (1997), while UNICEF estimates suggest that as much as 400 million children are working, UNICEF (1999). At the same time, human capital has been shown to be an essential element in a sustainable growth strategy for any economy. Hence, arguably, child labor which adversely affects the accumulation of human capital is highly costly in terms of the future growth and development of a country. For equity reasons, too, it is considered highly detrimental: child labor forecloses options arising from increasing educational levels to the individual concerned, perpetuating and even widening existing inequalities in society at large. For these reasons and for the ethical repugnance the phenomenon of child labor generates, solutions are increasingly being sought by policymakers, international organizations and aid agencies, to reduce and ultimately eliminate it. For all of these reasons, the analysis of child labor has received increased attention in recent years. While these attempts to grapple with the economics of child labor theoretically and empirically have led to some definitional and conceptual clarity, however, the existing literature suffers from various weaknesses. First, the nature of the data sometimes makes the analysis less applicable to child labor than could be desired, simply because the analysis of child labor was not on the agenda of the designers of the survey. Thus, researchers have often had to merely use a survey for analyzing child labor, which in most cases did not try to gather information specifically for * The views expressed here are those of the authors and should not be attributed to the Ghana Statistical Service or the World Bank or any of its member countries. Contacts: nblunch()worldbank.or , scanagaraiah(a)worldbank.org, sp264acolumbia.edu. t We would like to thank Amit Dar and Helena Skyt Nielsen for helpful comments and suggestions and the Danish CTF for financial support for this work. All remaining errors are our own. The data were collected by the Ghana Statistical Service in collaboration with the World Bank. I  the analysis of child labor. Clearly, policy advice from surveys plagued with such data problems are not useful. Indeed, it may even be harmful to the understanding of the nature of child labor and its eradication. Second, related to this is the fact that most analyses of child labor are based on a single cross- section. However, child labor is inherently dynamic, its structure and composition being likely to change over time in developing countries as a response to reforms promoting higher growth and increased economic and human development. There are significant differences in the nature and characteristics of different kinds of child labor across space and time - their causes and consequences also differ.' It is crucial to recognize this essential characteristic of the phenomena from the point of view of correct policy and the efficient use of scarce public resources for amelioration of the problem. However, little work has been done so far on analyzing the dynamics of child labor. This study is a preliminary undertaking to understand child labor as a multi-dimensional dynamic phenomenon, linked to child schooling, in the context of broad economic policies undertaken by developing countries focusing on the experience of Ghana. In particular, we conjecture that the popular belief of a simple, possibly linear, relationship between the welfare or poverty of the household and child labor is far too simplistic. Child labor is a multi-dimensional phenomenon, in particular, it cuts through time, as well as through space. Also, child labor is likely to be influenced by the overall (macro) economic conditions. In particular, the incidence of and determinants of child labor may be expected to be affected by economic growth and, thus, indirectly, by economic policies aimed at fostering growth. With this in mind, we propose the following three hypotheses to be tested in the analysis of incidence and determinants of child labor and schooling in Ghana over time: (1) the short-, medium- and long-run responses of child labor and schooling from economic policies are likely to differ. In particular, in the short run, the policies may have no impact and/or not be credible. Hence, child labor and schooling may be unaffected in the short run, but may respond in the medium- to long run. Specifically, we suggest that the incidence of child labor will decline over time in response to economic policies, aimed at economic growth and eradication of child labor. On the other hand, such policies - coupled with policies to XSee Parsons and Goldin (1989) for a study of late l9t' century US. 2  strengthen education - are likely to affect schooling in the long run, though possibly not in the short run. Related to this first hypothesis is (2) the relative importance of the determinants of child labor are likely to change over time. In particular, child labor would seem to be more responsive to poverty in the short run, while non-responsive in the long run, due to the fact that in the short run, households would want to use child labor as a buffer against economic shocks to the households, whereas in the long run, once the changes in the economic environment from the economic reforms have materialized, the household will be less prone to use child labor as a buffer against economic shocks. Similarly, related to this is the conjecture that (3) the relative importance of the determinants of child schooling are likely to change over time, and in opposite directions of those of child labor: in the short run, child schooling would seem to be unresponsive to changes in the poverty status of the household, due to the use of child labor as a buffer against economic shocks to the households in the short run, implying that children will be withheld from school in the short run, despite changes in the economic environment or in the perceptions of the economic environment. The changed poverty status of the household has to fully materialize, after which child schooling will then adjust. Note that child labor and child schooling are not merely each others mirror images. Most notably, children may be idle, doing neither2 (so-called "loafing" children) or do both. Also, the determinants of child labor and child schooling - as well as their relative magnitudes - may be different. While most notably due to the reasons discussed in hypotheses (2) and (3), such differences could also come about due to attitudes and tastes towards child labor and schooling differing across as well as within regions and/or urban/rural locations. The paper is organized as follows: the next section motivates the paper by discussing inter-linkages between economic growth, economic reforms and child labor. Section 3 reviews issues in child labor and household decision making, which in turn serves to formalize the issues involved in capturing them in a theoretical economic model as well as its estimable, econometric counterpart in section 4. Section 5 briefly discusses the evolution of the Ghanaian economy and .presents a preliminary descriptive analysis of child labor incidence, determinants and dynamics in Ghana. The econometric analysis of child labor In case of insufficient nutrition, indeed, "doing nothing" may be the optimal response. 3  determinants and dynamics in Ghana follows in section 6. Finally, section 7 summarizes and concludes as well as provides avenues for further research. 2. Economic Growth, Economic Reforms and Child Labor Many developing countries in the last two decades have undertaken major economic reforms - especially stabilization and structural adjustment policies - aimed at making the structure of the economy more market oriented and conducive to sustained economic growth. Many of these reforms have been pursued in the wake of macroeconomic crises - declining output and rising inflation leading to severe setbacks economy wide. A typical package of macroeconomic reforms comprises of what has come to be known as the so called 'Washington Consensus', Broad and Cavanagh (1999). Macroeconomic crises and even subsequent reforms do not affect all households identically. Rather their effects tend to vary across sectors of employment, income/wealth levels, geographic location, gender and age and other factors. The analysis of household responses to crises and to stabilization and structural adjustment policies has been pursued in its own right, especially the supply response of rural peasant households to changes in the prices and incomes, Binswanger (1989). Households combine the characteristics of producers, consumers and suppliers of factors of production (especially labor power) simultaneously (which makes their behavior a complex task to analyze). Thus, changes in market prices will influence the behavior of households as producers, consumers and suppliers of factors of production simultaneously. The net effect of economic crises and reforms on household behavior may be ambiguous. It will depend upon the magnitude and direction of price/substitution and income effects. It will also depend on other factors such as the stage of development the economy is in.and the nature of the markets - their imperfection and/or their absence - in which the households participate - especially labor markets and their characteristics.3 Labor markets are one of the most important channels through which macroeconomic policies are mediated. Contractions in aggregate demand have short-run impacts on households as demand for labor declines. People may become unemployed, work fewer hours, find work in the informal sector and/or 3For example, imperfect and/or missing markets will impose additional constraints on household optimization. Market signals may be weak and noisy due to market imperfections and incompleteness attenuating the linkages between macroeconomic changes and households' behavioral responses. 4  increasingly become self-employed. If incomes are falling due to falling wages and/or higher relative prices of goods, the labor force participation of households may increase - especially those of secondary workers such as spouses and children - to simply maintain household budgets, ), Eele, Hay and Hoddinot (1993), Skoufias (1994). These short run effects can have adverse long term impacts - for example due to a permanent reduction in the human capital accumulation of children who may have joined the labor market as a rational response to short term adversities but remain permanent workers even when economic recovery gets underway. Furthermore, to assess the impact of economic reforms on the incidence and characteristics of child labor in particular, we will need to enter the realm of intra-household allocation processes given the complex interdependence of time use of adults and children in a household. On a more general note, the effect of growth in reducing poverty and increasing household and individual well-being is more complicated than the simple 'trickle-down' story. It is now recognized that these effects are determined by the complex interplay between the nature and composition of growth, household behavior and government policies and expenditures.4 Economic growth in all developing countries has been accompanied by increasing rates of industrialization and urbanization. Not only the formal sector but the informal sector has been growing and at a more rapid pace than the former. Most children (and other members of poor and many non-poor households) find wage work in the urban and rural informal sectors in a range of activities. Rapid growth of the informal sector presents increasing income generating activities for children. It is relatively easier to enter the informal sector and does not necessarily require fixed costs such as those for entering the formal sector, Hill (1983). Given that agriculture continues to be the main occupation of the majority of people in all developing countries, and that the poor are found in greater concentration in rural areas and in agricultural activities, economic reforms that improve the terms of trade in favor of agriculture may result in a greater incidence of child labor with the increase in available economic opportunities.S This can be considered a short term behavioral response to changes 4Of particular importance for poverty alleviation and increased household welfare is how growth affects households in the lowest income categories. Most growth episodes in developing countries have tended to leave the bottom twenty percent of the population in terms of income distribution untouched. 5The impact will be greater, the lower are the expected returns from a low quality, irrelevant education system.  in relative prices (substitution effect). If, on the other hand, as a result of reforms, the labor market becomes more responsive to human capital attainment, this will increase incentives for households to invest in the education of their children and reduce (at least some) child labor. This is a longer term response being an investment decision (income effect). The net result of economic reforms on the incidence of child labor can thus vary depending on the length of time being considered and intensity of each effect on the specific context. Child labor itself comprises a whole range of activities, with usually specialization along gender and age lines. The distribution of child labor by gender and age and type of activity is also likely to change as a result of reforms - it may make one category of child laborers better off (usually boys) and another worse off (usually girls). After this brief discussion of the links between economic growth, economic reforms and child labor we now go on to present some important issues related to the analysis of child labor from the previous literature. 3. Issues in Child Labor and Household Decision-Making In order to "set the stage" for the empirical analysis, it would seem fruitful to briefly review some important issues that arise in the analysis of child labor6. This will both serve as a general theoretical background of the empirical analysis as well as highlight issues that needs to be incorporated in the latter. Households are settings for important productive activities, many of which are done by children, Skoufias (1994). Child labor seems to play many roles in a household's economy: provide labor input in the family farm or enterprise, work in the market for wages; do household chores to complement adults' efforts or substitute for them to release them for wage work; as consumption and/or income smoothing device. From a purely economic point of view, a complex of demand side and supply side and short term and long termn issues seem to be at work in child labor outcomes. In this section, we will locate the child labor decision in the context of household decision-making as well as the other factors that impinge upon it and mediate it. We will also evaluate current theoretical-analytical and empirical-descriptive literature that tries to understand the child labor issue but with an eye for insights this 6 For an excellent survey of economic theories of household behavior, see Bergstrom (1996) and Strauss and Thomas (1995). 6  literature might provide to our two assertions above - that child labor is a multi-dimensional and a dynamic issue. ChUd Labor and Poverty: Among the causes of child labor, poverty is often cited as the most important one, and in the same vein, economic growth considered a necessary condition for its reduction. It may be conjectured that most child laborers belong to poor households who need their children's income to supplement their household budget and who would experience a substantial loss in welfare if children were sent to school, not only because of direct costs of schooling, but also due to losses from foregone income, Bonnet (1993). Theories of the child labor-poverty interaction include those that base their explanation of its existence on the low incomes of households with or without credit constraints. A recent paper by Basu and Van (1998), for example, makes the assumption that the non-work of children is a 'luxury good' for households. Households offer the labor of their children if income is below a certain threshold level and not if it is above it. The extent of child labor is determined by demand conditions and equilibrium in the labor market. They show the possibility of multiple equilibria: a) high equilibrium wages and no child labor and b) low equilibrium wages and positive child labor. Ranjan (2000) shows how child labor can arise due to a combination of poverty and credit constraints. Altruistic rational parents care for the quality of their children and want to invest in their education as long as the returns from education are high enough. However, faced with credit constraints (due to their poverty) they settle for the second best choice of sending their children to work. If the primary determinant of child labor is low household income, then any measure to deal with it will have to be household income-based. However, whether poverty is the most important reason for child labor is still a matter of considerable debate. Another variation on the postulated relationship between income levels and child labor is the inverted U for many countries in West and Sub-Saharan Africa, Andvig, (1999). The evidence from existing empirical studies is inconclusive oil this question, Canagarajah and Nielsen (1999), Chandrasekhar (1997). Nevertheless, the assertion that child labor and poverty are correlated phenomena is indisputable. There is also little disputing that economic growth is necessary both for poverty reduction and for generating the required resources for implementing government policies that target child labor eradication. The point being emphasized here is 7  that economic growth per se may not be sufficient to reduce the incidence of child labor, Swaminathan (1998). Additionally, economic growth may not even be necessary for reducing child labor, see Weiner (1991). Child Labor and Credit Constraints: Related to - although not identical to - the poverty explanation of child labor discussed above is the possible existence of credit constraints. Education is a commodity like many others. It is an investment good - current resources have to be foregone for future returns. Education may be privately or publicly provided. In developing countries, public provision of education which is free or subsidized is the norm.7 There are private costs and benefits and social costs and benefits from educational attainment (or lack of it) for members of a society. Research indicates that both social and private returns to investing in education are extremely high, and the returns are the highest for public expenditure on education compared to other categories of public expenditure, Psacharapoulous (1994). Given high private returns, the question arises as to why parents do not - more or less automatically - send their children to school? It is argued that lack of access to credit is an important factor that can prevent households from investing optimally in the education of their children, Ranjan (2000). Credit and insurance markets are beset by pervasive market failures everywhere in the developing world. These market failures are especially severe for low-income households who are unable to provide the requisite collateral to procure loans. Not only in developing countries but also in developed countries, it is difficult to secure loans for human capital investment purposes. When parents are unable to send their children to school, they may put them to work as a second best alternative for the use of a child's time. Imperfect or absence of credit and insurance markets in general have been shown to result in the misallocation of resources at the household level, Rosenzweig and Wolpin (1993). Apart from underinvestment in children's schooling, imperfect credit and insurance markets can have other adverse impacts on the incidence of child labor. Child labor is one of the devices households use to manage risk if other forms of insurance are not available. Short 7In most developing countries, the private schooling sector accounts on average for 10 percent or less of student enrollment. 8  term child labor may be used to smooth consumption in bad times.8 The literature analyzing intra-household allocation has found that the latter depends on endowments and preferences of individual members of the household. Similarly, some studies find that risks are not shared perfectly within households and impacts on different household members differently, Doss (1996), Dercon and Krishnan (1997). Sawada (1997) finds that in rural Pakistan, negative shocks experienced by households, are smoothed through the use of child labor, but more for girls than for boys. These short term measures used by households have long term impacts, particularly through their adverse influence on the effective schooling attainment of children, Hanan and Skoufias (1997). Some children may drop out of the schooling system completely and become full time child laborers. There are few empirical studies that test the effects of access to credit on the household's schooling-work decision. In general, there is no consensus whether the effect is positive or negative - it can go either way under different circumstances, especially depending on labor market factors and the degree of substitutability between hired labor and family labor (Wydick, 1999). Alternatively, we can think of short term child labor as an ex-post behavioral response and long term child labor as an ex-ante household strategy. This distinction in terms of the time horizon may be useful for judging the welfare properties of child labor and for policy purposes. It also adds to the list of reasons as to why it is important to realize that there are different kinds of child labor with different causes and consequences. 9 Child Labor and Schooling - Quality and Subsidies: Another important dimension of child labor is the fact that given the central importance of education and schooling in modern societies, it is difficult to discuss child labor in isolation from schooling. Childhood is the time deemed for the acquisition of education. Work competes (at least to some extent) for a child's time with the alternatives of 8 Long term child labor, on the other hand, may be due to the absolute poverty of the household, a portfolio diversification strategy or a response to the relative returns to a child's time. 9 Note that this discussion is based on the underlying assumption that the household tries to minimize consumption and/or income fluctuations in the same manner as an individual does. The unitary model of household decision making has proved very powerful in terms of analyzing household behavior as will be discussed later. However, it has often been rejected in empirical work as the true model of the household. Cooperative and non-cooperative models of household decision-making seem to be able to capture more of the ways in which resources and risks are allocated and/or shared within households (Doss, 1996). 9  schooling and leisure. In the neo-classical framework of the economics of the household (Becker, 1965) rational optimizing households allocate their children's time (as well as adults' time) where they perceive the marginal returns to be the highest depending upon available economic opportunities and endowments. Moreover, as has been discussed above, education is an investment good where current costs have to be borne for expected (but uncertain) returns in the future. Many studies document and estimate both the private and social returns to education. It has generally been found that both kinds of returns to education are very high in developing countries, and of all levels of education, returns to primary education are the highest, Psacharopoulos (1994). Given such high returns why, then, is the demand for education and/or participation in education low in developing countries? If the poor lack sufficient funds to bear the opportunity cost of education of their children on the part of poor households and have little or no access to credit, then even if returns to education are high, they may be constrained to choose work over the schooling of their children. We also have to look towards the available economic opportunities and the quantity and quality of educational services that the poor have access to. As is generally the case, if opportunities for the poor are mostly found in the informal sector - in kinds of activities that require little or no formal schooling - and if schooling services are either not available or are of poor quality, the returns from education may be so low that parents will rationally want to send their children to work where relative returns may be higher. When household resources are scarce, there may be intra-household competition for resources among siblings which will be influenced by the demographic structure of households. For example, in many countries a child with more older siblings, especially elder sisters has a greater chance of going to school and lower probability of working. The presence of older siblings relieves the child from work responsibilities, especially in the unpaid household sector. Again, as has been discussed before, household risk attitudes may also play an important part in selective discrimination in the intra-household allocation of educational resources. Some poor households may send some children to work and others to school as part of household portfolio diversification. The labor force participation of children in developing countries is not incompatible with schooling (though it may affect the quantity and quality of learning). Many child 10  laborers also attend school and the category of children who combine work and schooling is in many cases a very important one. Not only because often for children who combine work with schooling, work is necessary for the children to pay for their own and/or siblings' education. To what extent children are able to combine work with schooling - without undue adverse effects on human capital accumulation - will depend upon the types of activities they are engaged in. The combination of work and schooling is now being advocated by scholars and activists as an intermediate step in the policy program of reducing and ultimately eliminating child labor, Grootaert (1998). Educational expansion, by itself, has been considered an antidote to child labor, especially by policymakers. However, it has not been as effective in meeting this goal although increasing greater child participation in schooling activities is an important goal in its own right. Ravallion and Wodon (1999) show in the case of Bangladesh that school subsidies lead to increased school enrollments and schooling duration. However, there is only a small and less than proportionate reduction in child labor. The implication of this observation is that increased time devoted to schooling may come out of leisure instead of labor. It may also lead to increased labor participation of other children in the household to make up for lost labor time. Much more analysis needs to be done to understand the trade-offs as perceived by the households themselves in making the schooling-work decisions on behalf of their children. Child Labor and Social and Cultural Norms: Economic incentives (or disincentives, as case may be) apart, attitudes towards work and schooling vary from context to context and influence the relative valuation of different uses of a child's time which varies by gender and age. There are socially prescribed roles for boys and girls of different ages - children working long hours in the household or the farm may be considered by the household, the childiren themselves and the community as normal and part of 'growing up"0, see Andvig, Canagarajah and Kielland (2001) for a discussion in an African context. 10 There are interesting exceptions - which may vary by income and wealth levels of households. For example, very poor households may not be able to afford to follow social norms. In societies where women working outside the household are disapproved of, women from the poorest households may do wage work to maintain family subsistence. As incomes increase, there will be a move towards the conservative social norms where women are secluded within the threshold of the household. 11  Baland and Robinson (2000) demonstrate that even when parents are altruistic, child labor can arise because children cannot write credible and enforceable contracts with their parents to transfer resources to them in the future. Parents are unable to capture the full returns from their investment in children's education and therefore underinvest in education, relative to what is socially optimal. As labor and schooling exhaust a child's time endowment, less schooling means more labor. Child Labor and Demand Side Issues: When discussing child labor, it should be noted that only a small percentage of all child laborers work for wages in the market - most children are unpaid domestic workers. Nevertheless, we can conceptualize a 'child labor market' in either case; for the unpaid domestic sector, we can consider households as employers of their own children." The demand side of the market when children are employed for wages has not been as actively or deeply researched as has been the supply-side (household) factors that explain the existence of child labor. Children in wage work are found mostly in the informal sector in manufacturing and services, in petty trading and other self-employment activities (and as domestics in households) in the urban areas. In rural areas, children may work as hired laborers for wages or in-kind compensation. However, in the rural areas of most developing countries, children work as unpaid workers on family farms or in household chores. Assumptions about the labor market in general and the child labor market in particular are of crucial importance for both a better understanding of the phenomenon and the ensuing policy-implications. For example, the existence of well behaved (child) labor markets may be a more appropriate assumption in some contexts than in others. Also, in some cases, the child labor market may be monopsonistic or otherwise segmented for demand-driven reasons, Chandrasekhar (1997), Grootaert and Kanbur (1995). The structure of the demand for labor in general and for child labor in particular will also depend on the nature of the technology being used in the industry/market in question. In some industries such as carpet weaving, bangle-making, matches and mining, it is deemed that children are employed in large numbers because of their greater suitability for the tasks at hand, the so called 'nimble fingers' argument. The argument is that children have 1 Practically, the latter is hard to operationalize because children contribute to domestic chores as part of their 'socialization' and effort in this sphere will be considered as child labor if it crosses some critical limit. 12  irreplaceable skills and cannot be substituted by adult labor as far as these industries are concerned.12 On the other hand, it has been shown for some industries where a heavy concentration of child labor is found that employing children is only marginally profitable for firms, Burra (1994). They can easily switch to adult labor and pass on the increased cost to customers without hurting profits. Why, then, is there a demand for child labor in these industries? Firms employ children not only for their better suitability for certain tasks but also for their lower reservation wages, and for non-pecuniary reasons such as their lack of organized bargaining power. Finns incur certain fixed costs in hiring and training a given set of workers apart from the variable costs paid out as wages, Oi (1996). By hiring child laborers, firms economize on these fixed costs.)3 However, it may also be the case that by employing child laborers, employers increase their bargaining power over adult laborers. So, again, whether child labor is a substitute or a complement to adult labor is open for debate. Plausibly, children and adults are substitutes in some types of activities and not in others. Nevertheless, the child labor market cannot be discussed in isolation from the adult labor market. In fact, the nature of the interrelationship is important for analyzing the impact of macroeconomic reforms on the nature and incidence of child labor. To the contention that child labor displaces adult labor and/or child labor is associated with high adult unemployment and underemployment must be juxtaposed on the observation that a high child participation rate is found both in high and low adult participation rates scenarios. The combination varies from context to context. Child and adult labor interact in complex ways both from the demand and supply sides. For example, Skoufias (1993) finds that in India, a rise in female wage rates increases a child's laboring time. Levy (1985) finds the opposite in Egypt. In the first case, the substitution effect outweighs the income effect and vice versa in the second case. The difference in outcomes in the two contexts requires deeper analysis of underlying conditions for better understanding the dynamics of incentives, institutions, resources and values at work. The child labor market is embedded in larger social realities which may have significant influence on its nature. In developing economies, very often social institutions interact with market institutions to produce different kinds of labor market segments. For example, gender casting of work, which exist in most societies. A different 12 These industries are also often important in the country's exports. 13 In fact, in some industries, in the first few months when children are undergoing 'apprenticeship', they are not even paid, Burra (1994). 13  example is the caste system in India, which has been a basis for occupational choice and division of labor both within and outside the household. Externally, it may be used as a rationing device in a demand constrained child labor market.14 In the case of children working in the unpaid domestic sector, we can think of households as employers of their own children. The demand for labor depends upon the household production function both for household consumption and for sale in the market. If an external market for child labor exists, then the opportunity cost of a child's time will be the market child wage. If not, but if instead the child substitutes for adult members in household chores, then the opportunity cost of a child's time will be the market wages for adults. It is unlikely that perfect markets exist for all household goods demanded and supplied by households so separation of production decisions from preference parameters is implausible. The household maximization problem endogenously determines the shadow prices of household goods and shadow wages of household labor along with household equilibrium in the demand and supply of household commodities and factors of production.15 On a cautious note, the analogy of the presence of implicit markets within the household is useful but has to be interpreted with caution. These markets are far from the perfectly competitive markets we find in theory and not even close to the kinds of markets found in the real world. What is exchanged and at what price when children do household work is not at all clear. Work within households have elements of both coercion and cooperation and it is not easy (if not impossible) to disentangle the two. Child Labor and Household Decision Making: Anthropological and sociological studies document household structures found in various parts of the world which are significantly different from the common preferences model, with or without production. The common preferences, or unitary, model assumes the existence of an aggregate utility function for the household and members pool income and labor supply. It does not take account of conflicts of interests between individual members of the same household that may have an important bearing on how resources and responsibilities are allocated within the household. Hoddinott and Haddad (1995) show how 14 This may be true even in the allocation of scarce schooling resources. is This too may be problematic to disentangle as the production technology for household goods is unobservable. 14  incomes earned by women have a different effect on the allocation of household expenditures compared to incomes earned by men. Given the shortcomings of the common preferences/unitary model, researchers have used cooperative and non-cooperative bargaining models of households in recent years to analyze the issues of intra-household resource allocation, see, for example, McElroy and Homey (1981), Lundberg and Pollak (1994). In many cases, the income pooling assumption of the common preferences model has been decisively rejected. Furthermore, these analyses show that the identity of members inside households and the dynamics between them matter for the household's economic decisions. Therefore, without adequate regard for the appropriateness of the household model being used in a given context, inappropriate and consequently harmful policy actions may be taken based on an incomplete analysis. Collective models overcome important shortcomings of the common preferences and unitary models of the household and their econometric counterparts prove that individual preferences and endowments matter in intra-household processes. However, they too leave some questions unanswered. These models generally assume that whatever be the sharing rules within households, outcomes are Pareto-efficient. Work by for example Udry (1994), on the other hand, show that Pareto-efficiency cannot be assumed but needs to be tested. This work finds evidence against Pareto-efficiency in intra-household allocation processes. In order to analyze the welfare economics of child labor, especially in domestic work, a possible approach could be to test for the efficiency properties of allocation of labor and other resources between adults and children of different age and gender. Existing studies, moreover, do not so far include children as potential decision-makers even though as child laborers they supply labor and earn income. Children are assumed to be public goods for the households and only adult decision-makers are analyzed. There is some evidence suggesting children who contribute to household income acquire some influence in household decisions and in the allocation of expenditures/resources. 4. Economic Model and Econometric Methodology In this section, we formalize the ideas laid out in the previous section by presenting the economic model underlying the econometric analysis. Then follows a discussion of the econometric methodology applied in this paper, i.e., how to move from the theoretical economic model to an econometrically estimable model. 15  Economic Model: We will now describe the economic model of the household decision process regarding children's time allocation that underlies the econometric analysis in the next section. First, the objective of the household decision maker'6 needs to be made explicit. It seems reasonable that the objective is to maximize utility of adults and children in the household jointly and, further, that the utility of the household depends on aggregate household consumption'7, as well as the child's schooling and leisure. This gives rise to the following household utility function (for simplicity in the presentation, we assume the household only has one child but the results can readily be extended to the general case)): ui = u'(Ci ,Li,Si ,Zi) (4.1) where C is the aggregate aggregate consumption, and L and S denotes leisure and schooling of the child, all of which are for household i. It is assumed that u(.) is strictly quasi-concave in C, S and L. Inclusion of the vector of exogenous household and community variables Z allows for heterogeneity across households. For example, the quality of and/or the distance to local schools will influence the parents' expected utility from sending their children to school. Also, it allows for differences in tastes due to ethnicity and/or religion of the household. The maximization of (4.1) is subject to the following constraint on the child's time: T7= Si +Li +Hi (4.2) Where T2, total time available to the child of household i, may be devoted to schooling, 5, leisure, L, or wage labor, H. The maximization of (4.1) is also subject to the household budget constraint: Cj = Wj Hi + ]Y, (Zj (4.3) 16 Or decision makers: we allow for more than one decision maker - in the extreme case, we could think of a collective decision process. However, smce our emphasis is on the empirical (reduced form) results, we will not explicitly model this further. 11 In this model formulation, we operate with one cornposite good, C. However, one may think about this as a basket of all possible consumption goods. 16  where C, W and H denotes consumption, (child) wages, and (child's) hours worked, respectively. Y denotes the income from all other sources than child labor, and is assumed to be a function of Z, also (which will include, for example, the parents' education, occupation and landholdings). The household decision maker maximizes (4.1) subject to (4.2) and (4.3) with respect to C, S, L and H, given W, Z and T. We may combine the constraints (4.2) and (4.3) into the composite constraint: C; + WjS; + W; Hi = WjT; + Y; (Z) (4.4) This is also denoted the "full income" budget constraint. From (4.4) it follows explicitly that W; is the price of schooling to parents, since this is the rate at which they could "sell" their children's time at the (child) labor market. With the composite constraint (4.4), the optimization problem of the household decision maker is to maximize (4.1) with respect to C, L and S subject to (4.4). This leads to the first-order conditions: u (-) _ 1 (4.5) U` (. W; uc,(.) _1 (4.6) uI.(. W, which simply state that the household decision maker will equalize the marginal rates of substitution between consumption and schooling and consumption and leisure with the relative prices (where the price of consumption has been normalized to one). From the first order conditions we may then subsequently obtain the derived demand functions (as perceived by the household decision maker) for the child's schooling and leisure: 17  Si = S(W,, WVTi + Y1(Zi), Zi) (4.7) Hi = I(Wi, WiTi + Yi(Z1), Z,) (4.8) Where Si and Hi are the decision variables (schooling or work of child in household i), and the other variables are as before. Econometric Methodology: In moving to an estimable model, essentially we are after an empirical counterpart of the economic model as given by equations (4.7) and (4.8) above (noting that child labor supply is merely the "dual" of the demand for leisure function, (4.8)). There are several possibilities here. Previously, the literature that has regarded schooling and child labor a joint decision has applied either (1) a bivariate probit, thus simultaneously estimating a probit for the schooling decision and one for the work decision in a seemingly unrelated regression structure; see, e.g., Canagarajah and Coulombe (1997) and Nielsen (1998)), or (2) a multinomial logit for the four possible outcomes (school, no work), (school, work), (no school, work), (no school, no work); see, e.g., Grootaert (1998). Both models may be justified in terms of our objectives, since both models are applicable in cases, where there are multiple outcomes (although the bivariate probit requires exactly four outcomes to be applicable, which, however we have here, namely the four different activities of children), which do not lend themselves to a natural ranking, Greene (2000). However, we apply a multinomial logit model due to the marginal probabilities being computationally easier to obtain, see Greene (2000) for details. Our dependent variable is the categorical indicator variable: 1 if childi neither attends school nor works 2 if child i only attends school Yi 3 if child i both attends school and works 4 if child i only works 18  The log-likelihood for the multinomial logit model is: n I lnL=Z= E Ed,ln Prob(Y; =j) (4.9) i=I j=O where e 'x' Prob(Y= j) = - , (4.10) 1+Ze XI k=1 where Y is the dependent variable (activity of the child), subscript j denotes the possible outcomes of the dependent variable (school, work, both or neither) and subscript i denotes the individual. xi is the vector of explanatory variables for individual i, while fi is the vector of parameters. An important issue in analyzing the link between poverty and child labor by means of estimating the economic model, as given by the equations (4.7) and (4.8) above which needs to be addressed, is the problem of potential endogeneity/simultaneity. This problem arises whenever one or more of the explanatory (i.e., "right-hand-side") variables are not exogenous to the dependent (i.e., "left-hand-side") variable. In the present case, household expenditure may not be exogenous to the activity of the child, in the sense that, for example, a relatively high household expenditure per capita, which would seem to imply that the household is wealthy, could be expected to induce the household not to send their children to work. However, it could also be that households are able to sustain a relatively high household expenditure per capita because they send their children to work. Whenever this problem is present, it may lead to biased estimates also on the estimates of the other parameters of the model, i.e. for variables which are themselves exogenous to the dependent variable. One solution of this is to apply instrument variable estimation techniques. However, it is hard - if not impossible - to come up with an instrument for household expenditure that is not correlated with the child activities (which is required for the instrument to be valid). However, since one of the worst consequences of endogeneity is invalid paramneter estimates for the other variables, i.e. other than the potentially endogenous explanatory variable, one 19  way of checking whether severe endogeneity bias is present is to run the models without the potentially endogenous explanatory variable (here, per capita expenditure). The results from this exercise reveals that the parameter estimates (and, most importantly, their signs) are fairly robust to omitting (or, similarly, including) this variable, in turn indicating the absence of serious endogeneity bias from the expenditure variable. In addition, one may claim that due to the expenditure variable being defined as expenditures for the past year, while the labor- and schooling variables are defined as activities during the past week, there is such a marginal overlap that the expenditure variable may be said to be endogenous to the labor supply/schooling choice by construction - or, at the very least, pre- determined. In any case, combining this argument with the somewhat informal test of the practical consequences or otherwise of the potential endogeneity/simultaneity discussed above, it appears valid to claim that there are no severe endogeneity/ simultaneity problems, at least for practicalpurposes. 5. Descriptive Analysis and Presentation of the Ghanaian Economy'8 Ghana has undergone wide-ranging economic reforms from the mid 1980s onwards. The core of the reforms have been to gear the economy towards greater market orientation. Transitions are mixed blessings for the poorer households in any country as some of them may experience short-term welfare losses though ultimately benefit in the long run. The naYve view that secular changes in the economy will equally affect all population groups and sub-groups has been discarded, and it has been recognized that structural adjustment programs may benefit some groups and hurt others, at least in the short run, Demery and Addison (1993), Demery and Squire (1995). Hence, it is important to study the impact of reforms, in particular whether and how it affects different groups differently. Such analysis will enable policymakers to design and implement more effective programs in the future. For our present study we use three rounds of the Ghana Living Standards Survey (GLSS). The Living Standards Surveys are multipurpose household surveys, that are nationally representative, and collect information on a vast array of socioeconomic variables characterizing households and their welfare. The rounds used in this study are the Ghana Living Standards Survey 1987/88 (GLSS1), 1991/92 (GLSS3), and 1998/99 (GLSS4), ' The macro economnic indicators quoted in this section are from Alderman (1994). See also World Bank (1993, 1995 a,b). 20  respectively. Economic reforms were undertaken in Ghana, starting with the ERP in 1983. At a crude level, then, the GLSS1 will reflect short term effects of economic reforms and the GLSS3 the medium to longer term effects, while the GLSS4 will reflect long term effects of economic reforms. Ghana's Macroeconomy and Economic Reforms: By 1983, Ghana's economy had more or less collapsed. Between 1970-80, Ghana's growth rate had fallen by an average of 1.5 percent per annum. Inflation averaged 66 percent, with wide year to year fluctuations. Cocoa production, which dominated both export earnings and government revenues, continued to decline due to distortions induced by trade and exchange rate restrictions that had been practiced for decades. Domestically also, basic institutions had stopped functioning as had most of the public infrastructure. Moreover, 1983 was also the year of multiple exogenous shocks - the country experienced one of the worst droughts in its history, the sharp decline in world cocoa prices continued, and a million Ghanaians and their families were forced to return from neighboring Nigeria adding to the country's welfare burdens. The wide ranging Economic Reform Program (ERP) was undertaken in 1983 in Ghana with the major aims of restoring economic growth and making the economy more market oriented. Ghana initiated its economic reforms with stabilization and trade liberalization, including fiscal and monetary restraints, phasing out of quantitative restrictions and tariffs on imports and foreign exchange liberalization and removal of domestic price distortions. Agricultural sector reforms were undertaken towards the end of the 1980s with the major aim of reducing and removing the biases and distortions in the rural sector. Ghana had seen significant and continuing decline in the production of cocoa from the mid 1950s onwards when it produced more than 30 percent of the world's cocoa. The failure of the state Cocoa marketing board in giving a higher share of the market value of cocoa to farmers resulted in smuggling to neighboring Cote d'Ivoire and subsequently lack of incentive to produce leading to a declining share in the world markets. Most of the reform in this sector focused on removing price distortions in order to improve production incentives for farmers. It was expected that as a result of the latter, rural farm and non-farm growth would increase, leading to an overall reduction in poverty. Ghana implemented market determined exchange rates, reductions in import tariffs and export taxes, divestiture of state 21  enterprises and removal of agricultural input subsidies to enhance incentives for greater production and incomes in the rural sector. The Impact of Economic Reforms on Child Labor: Economic reforms in Ghana were initiated in 1983, and reform of the agricultural sector was undertaken later towards the end of the 1980s. By many counts, economic reforms in Ghana have been successful. The growth rates of GNP and GNP per capita have been positive every year between 1983-1993. The agriculture, industry and service sectors, which had registered declining growth prior to reforms, grew at positive rates from 1984 onwards. These macroeconomic developments just described - as well as their possible effects on micro-level economic behavior - are likely to be visible at the micro level, too, as captured by the household surveys applied in this study. Tables Bl and B2 in Appendix B describe the typical characteristics of male and female child laborers in the years 1987/88 and 1991/92 respectively. 19 Tables B3 and B4 in Appendix B describe schooling and labor force participation decisions by gender, age and income quintile of the households of the children in the two survey years, respectively. The age group being considered is children between 7 and 14 years. Although a child may start work from when he or she is 4-5 years old, the Ghana Living Standards Survey uses age 7 as the cutoff for recording work activities. The mean ages of male and female child laborers in both rural and urban areas in the two years have remained more or less constant around 11 years. The proportion of rural child workers, both male and female, have increased substantially between 1988 and 1992, from 33.4 percent to 49.1 percent for boys and from 25.2 percent to 41.2 percent for girls, respectively. More girls have entered the labor market than boys in the interim time period. The proportion of child workers in the labor force, both male and female, too have increased although the average number of hours spent in the labor market have decreased substantially. When we compare and contrast the schooling activities of male and female children in the two years, we can see from Tables BI and B3, that school participation has increased for both boys and girls, though less for girls than for boys. Nevertheless, on average, 28 percent 19 We do not present figures from the 1997 survey due to missing observations on key variables, leading to extremely small samples which, in turn, would lead to unreliable conclusions. In particular, the final effective sample of children between the ages 7-14 is only 443 - as comnpared to 1732 and 3516 children in the two earlier samples - of which only around 1 percent are in the two categories "only work" and "neither attends school, nor works". In turn, this renders application of a multinomial logit invalid and we will apply a binomial probit model instead (we return to this issue in more detail later, in the econometric analysis section). 22  of children between 7-14 years of age are out of the school system. From Tables B2 and B4, we find that the percentage of both boys and girls in the work only category have declined in 1991/92 from 1987/88, and that in the categories school only and work and school have increased. While this is generally true for all income categories, the largest increases have come in the two highest - 4h and 5h - quintiles. The incidence of child labor and child schooling is believed to be strongly positively related to the level of household income - this is the so called 'poverty hypothesis' for child labor discussed above. Simply put, a decline in poverty can be expected to be associated with a decline in child labor and an increase in child schooling. Between 1988 and 1992, poverty in Ghana declined from 37 percent to 31 percent as implied by the measures of depth and severity of poverty. This decline in poverty came entirely from the rural areas and some from urban areas other than the capital city Accra. Accra registered a triple fold increase in poverty - from 8.6 percent to 23 percent. Comparing the composition of child workers between rural and urban areas in the two years from Tables Bl and B4, we find child labor in urban areas being more or less constant, while we see a substantial increase in rural child labor in 1991/92. From the point of view of the poverty hypothesis as it relates to child labor, this is puzzling given that rural.poverty declined and urban poverty increased in the years under consideration. Also we must be mindful of the fact that these inferences are being drawn from simple cross-tabulations. Given the joint causation of different household behavioral outcomes, identification of cause and effect is not possible without a regression analysis with multiple controls. We will do this in section 5. Comparing the results from the household surveys with the macroeconomic indicators20 reveals that the (macroeconomic) growth period from 1983 onwards was punctuated with decline in 1987 and 1988. Real minimum wages and average real monthly income from agriculture also fell substantially. World cocoa prices fell substantially during the adjustment period in 1986. Agriculture has experienced very slow growth in spite of reform measures being taken. We can expect these circumstances to strengthen incentives for greater use of child labor in the years 1987/88 by households. In contrast, child schooling decisions are more appropriately based on long tenn/permanent income factors of households, due to the nature of the investment, being a long-term (human capital) 20 From Alderman (1994). 23  investment. Between 1987/88 and 1991/92, among all child laborers, those employed in agriculture increased substantially - from 67.5 percent to 96.3 percent for boys and from 67.9 percent to 88.5 percent for girls. It should also be noted that the agricultural sector continues to account for more than 40 percent of Ghana's GDP. The majority of the poor in Ghana, around 70 percent, live in rural areas with agriculture and allied activities as their main occupation. In Ghana, as in much of Sub-Saharan Africa, most agriculture is based on production in small family farms. Generally, households are large, while landholdings are small. In 1988, small landholders constituted 94 percent of the rural population. Agricultural wage labor is a small minority and a seasonal phenomenon. Moreover, child labor in Sub- Saharan Africa is a rural and family-based phenomenon. This is in contrast with the kind of child labor found in Latin America, where most child labor is found in wage work in urban areas. Generally, these preliminary results lends some support to the hypotheses carved out in the introduction of this paper, child labor and schooling responding primarily in the medium to long run. Less children engage primarily in child labor and more children primarily in schooling in the later sample (medium- to long run) than in the earlier sample (short run). More children are combining the two activities in the later sample than in the earlier. This is true of both genders, all ages and income quintiles. There seems to be a complex of demand and supply side factors at work here. Certainly, the relation between income levels and child labor is neither simple nor straightforward. However, these tentative conclusions needs to be substantiated in a multivariate regression analysis, where we take multiple factors into account. 6. Econometric Analysis of Child Labor and Schooling Determinants and Dynamics We now move to the econometric analysis. Due to the amount of estimates, we will discuss only the major results - and focus mainly on the activities "school only" and "work only" - and refer to the tables in Appendix C for the details. The results from the full sample for GLSS1 and 3 in Tables Cl and C2 in Appendix C reveal substantial differences between the two periods. In the earlier sample, older children tend to be less likely to go to school and more likely to work, ceteris paribus (the marginal effects are around 11 and 16 percent, respectively). This result is in line with a perception that the marginal product, and hence, the labor-value of the child, rises with age, hence 24  leading to older children being relatively more likely to work and less likely to attend school. The latter indicates the existence of inter-generational spill-over effects in education, or, similarly, a great deal of persistence in education at the household level. The next striking result is that children that are biologically related to the household are more likely to attend school, while at the same time less likely to work, pointing to the possible importance of so called "placed children", a phenomenon which is well-known in the next-to-neighboring country of Benin, Kielland (1999).21 Interestingly, the mother's education does not seem to be affecting the activity of children, while the father's education has a consistently adverse effect on child labor across all levels of education (no education is the reference category), while education of the father promotes the education of the child, as well. Household composition proves to be an important factor. In particular, the presence of brothers and sisters between 7 and 14 years of age enhances the likelihood of attending school (by around 4 and 6 percent, respectively), in line with previous research, for example Nielsen (1998). Traditions seem important, as revealed by the statistical significance of some of the religious variables. Coming from a Catholic or Muslim background increases the likelihood of attending school (by around 17 and 25 percent, respectively), while Catholics are both less likely to only work and more likely to combine work and school (by around 16 and 5 percent, respectively). The existing evidence on the effect of the gender of household head is mixed. In the present context, male headed households seern to be more prone to sending their children working and less likely to send children to school (by around 10 and 5 percent, respectively) relative to female headed households. Canagarajah and Nielsen (1999) suggests that it may work both ways, since it may be conjectured that either female heads may care more about the children and therefore use less child labor. On the other hand, it may equally well be the case that female heads may be forced to make use of child-labor more often because they are under economic pressure due to being the sole (adult) breadwinner in the household due to the husband being dead or having left either permanently or temporarily. 21 The number of children, which have a non-biological relationship to the household head is 444, or a little more than 25 percent of the sample. Hence, since this group is a substantial part of the sample, this issue seems to warrant further analysis. 25  Our poverty measure, (log of) household expenditures per capita, does not seem to be important except in the work only equation. Furthermore, the impact is small, with an estimated elasticity of about 0.05, implying that a 1 percent income increase increases child labor with 5 percent. This apparently counterintuitive result may be explained by it not necessarily implying causality from poverty to child labor. Indeed, it may be the case that the households are less poor, i.e. can sustain a higher per capita consumption level because they engage in child labor. Being engaged in raising livestock adversely affects the probability of only attending school, while it increases the probability of only engaging in child labor (by around 14 and 12 percent, respectively). This points to the importance of child labor as a rural phenomenon, as does also the fact that the regional variables are significant and positive in the schooling only equation for Accra and urban areas outside Accra, while negative for the rural coastal region. Also, in the work only equation the regional variables for Accra and urban areas outside Accra are statistically significant and negative. Moving to the later sample from GLSS3 (1991/92), the results from which are shown in Table C2 in Appendix C, the first remarkable thing is that age no longer seems to be important for the activity choice, since now there is an age effect only in the equation for combined work and schooling. However, this may reflect the fact that the older, the children are, the more likely it is that they can engage in both activities, since engaging in physical work and at the same time attending school demands a certain maturity of the physique (this variable was borderline significant in the GLSS1 sample). The father's education is, just as was the case for GLSS1, very important for children's activities: the father's educational attainrent consistently positively affects the tendency for a child to only attend school, while negatively affecting the tendency of a child to only engage in child labor. However, the mother's education now also comes out significant in the case of middle secondary education. Household composition again proves important. Children with a Catholic, Protestant or other Christian background are more likely to only attend school relative to children with and Animist/Traditional religious background. However, only children with a Catholic (and, marginally, with a Muslim) background are less likely to engage in child labor (by 5 and 4 percent, respectively). 26  Children from male headed household are, as was also the case for GLSS1, less likely to attend school only and more likely to only engage in child labor. The influence of labor demand variables is now massive, as compared to GLSS1. Both owning land and owning livestock are now statistically significant variables in the household child labor/schooling decision process, both affecting the probability of only attending school negatively, while affecting the probability of only working positively. This contrasts to GLSS1, where only the livestock was important in the household decision process. This points towards the existence of asymmetries in the determinants of child labor over time, something which is consistent with out main hypothesis of there not being a linear relationship between poverty and child labor - indeed, other variables - including the household's demand for labor - are part of and jointly determines the household's child labor/schooling decision. These results indicates the increased importance of labor demand of the household as a determinant for child labor and, thus, points towards the importance of the main conjecture of this paper of child labor being a multi-dimensional concept, components of which include - but are not limited to - the household's labor demand and economic activities as well as poverty/welfare of the household. In particular, the labor demand of the household (or similarly, the economic activities of the household) appear to have become more important in the medium term (as revealed by GLLS3) than they appeared in the short term (as revealed by GLSS 1). As was the case with GLSS1, the importance of child labor of mainly a rural phenomenon is sustained by the results from GLSS3. People from urban areas outside Accra are more likely to only attend school, and less likely to work, and vice versa for children from rural coastal and rural forest areas. Lastly, the importance of household welfare as measured by log expenditure per capita has changed. In the GLSS1 sample, there was only an effect on the "work only" decision, whereas with the GLSS3 sample, the importance of household welfare has shifted to the schooling only decision. The estimated marginal impact is 0.06, implying that a 1 percent increase in household expenditure per capita increase the probability of a child only attending school by 6 percent. Again, this indicates the existence of substantial non-linearities in the relationship between poverty and child labor. In particular, this result suggests that macro economic circumstances - including the massive structural adjustment policies which have taken place between the two samples - is an integral part of this relationship and a major determinant of the underlying process governing 27  the evolution of child labor supply and child labor demand, even though the impact is modest. Moving to the GLSS4 sample, the first thing to note is that this part of the analysis has to be done applying a binomial response model rather than the multinomial response model applied previously. It is essentially a data problem, in the sense that there are just too few observations in the "neither work, nor school" and the "only work" categories to validly estimate the model by means of a multinomial logit model. To be specific, as earlier mentioned, there are only five observations in each of these groups, corresponding to around 1 percent of the sample, in effect leaving us with only two groups (i.e., "only school" and "school and work"). Hence, with only two classes, a binomial probit/logit is more appropriate. In addition, the total number of observations is extremely low, owing to a high dropout of the sample on the question of whether the child has been working the past seven days or not (implying that the obtained result, discussed in the following, should be taken with a grain of salt). The results from the binomial probit are shown in Table C3 in Appendix C. Noting that the reference category is children that attends school only, the results overall appear fairly consistent with the previous results in several respects. To begin with, being a child of the household negatively affects the probability of combining work and school, relatively to only attending school. In accordance with the earlier results, having a sister between the ages 7 and 14 also decreases the likelihood of having to work in addition to attending school, implying some substitutability in child labor within the household. Similarly, the locational variables confirm that child labor is still very much a rural phenomenon, also in this more recent sample. However, now smaller siblings proves important in the sense that they increase the likelihood of combining work and school, relative to only attending school, while children from both religious and non-religious backgrounds are less likely to combine work and school (hence more likely to only attend school) than are children from households with a traditional (animist) religious background. This points to the importance of traditions and customs as an important determinant of child labor. There seems to be a poverty-child labor link present in this more recent sample, also. In particular, its importance has switched in this later sample, such that it now affects the decision of combination of school and child labor activities relative to only attending school. 28  The parameter estimate is -0.02, implying that a 1 percentage point increase in per capita expenditure will lead to a decrease in the probability of engaging in child labor (combined with school) of 2 percent, ceteris paribus. With the caveat of the effective (sub-) samples being quite small in mind, we now turn to the predictions of the models, which are shown in Tables DI and D2 in Appendix D. First, we note that there has been a dramatic decrease in the predicted probabilities of only working and an increase in the proportion of children that combine school and work both for the full sample as well as across the various sub-samples, controlling simultaneously for the various factors hypothesized too influence this decision while the proportion that only attends school has remained stable and/or decreased, depending on the (sub-) sample and year. These developments support the tentative conclusions based on simple cross-tabulations in the descriptive analysis. Hence, even when taking into account all possible influences simultaneously may we conclude that some support exist for the conjectures made in the introduction of this paper, namely that child labor and schooling is likely not to respond in the short run, while it will respond in the long run - and with opposite sign: child labor will decrease and schooling will increase, once the effects form economic reforms materialize and/or become credible. Further, again, the importance of parents' education on children's activities is very significant. Children, whose mother and/or father have no education are less likely to only attend school and more likely to only engage in child labor activities than is the sample as a whole. This is true for both the GLSS1 and the GLSS3 (we do not make predictions for GLSS4 due to the low number of observations, as earlier mentioned). Lastly, while there generally is no strong pattem in the impact from poverty as measured by the expenditure quintile of the household, there is a strong positive link between attending school only and poverty of the household, such that the poorer the household, the less likely it is that a child only attends school. These results are largely consistent with the "real" developments, as discussed in the presentation of the descriptive statistics in section 4, and, thus, indicates that the estimated models capture the features of the data well. So, our main hypotheses are supported in this more rigorous multivariate (regression) framework. 29  7. Summary and Conclusions We analyze the determinants of child labor in Ghana, focusing at the dynamics of child labor in the short-, medium and long term, in particular the response to growth and economic policy. Previous studies of child labor and its determinants have mainly been based on a single cross-section. While this may provide a useful "snap-shot" of child labor incidence and determinants at one point in time, we suggest that this approach does not satisfactorily capture the multidimensional character of child labor. In particular, the question of asymmetries in the incidence and determinants of child labor across time seems to have been seriously underdeveloped in the existing literature on child labor, thus leading to imprecise conclusions on factors influencing child labor and, in turn, offering policy choices, sometimes leading in the wrong direction as far as the eradication of child labor is concerned. To fill this void, we analyze and compare the incidence, determinants and dynamics of child labor in Ghana after a brief review of the inter-linkages between economic growth, economic policy and child labor and the literature on child labor and household decision theory as it relates to this paper. Starting from a premise that the simple - direct - relationship between poverty and child labor, which has often been seen as the feature of child labor does not adequately capture the multi facetted nature of child labor, we find evidence of asymmetries in the child labor- poverty link, as well as quite complex dynamics in the evolution of child labor and its determinants over time. In particular, we find evidence of the poverty-child labor link - while clearly evident in the data - changes over time in the case of Ghana, in the short run affecting only the work decision, while in the longer run affecting the schooling decision, only. For the most recent sample, GLSS4, the importance of the child-labor poverty link has switched, such that it now adversely affects the decision of combination of school and child labor activities relative to only attending school, indicating long-run responsiveness of child labor to policy interventions. Due to the low number of usable observations in the GLSS4, however, the importance of the latter result should not be over-emphasized. While the results of the dynamic responses of child labor and schooling from policy interventions, thus, are not clear-cut, they do suggest that the view of a simple, linear relation is not valid (or at least, only as a static phenomenon) and, in particular, do not adequately 30  capture the dynamic evolution of child labor, as it responds to changes in the economic and political climate of the economy. In addition to these very dynamic findings related to poverty, most of our other main results are quite robust over time. In particular, we confirm evidence from earlier studies of a positive link between parents' education and the likelihood of a child attending school only, and similarly a negative link between parents' education and the likelihood of a child only working. Household composition also proves important, in the sense of the results revealing substitutability between the siblings within the household. A main result is the existence of structural differences related to rural areas in Ghana, both coming through the increased (child) labor demand from households engaged in agriculture and livestock raising, as well as regional differences among Accra, other urban areas and rural Coastal and rural Forestal regions, the latter two generally exhibiting higher demand for child labor than the former. Similarly, the likelihood of only attending school is generally found to be larger for urban areas than for rural areas. Traditions and customs, as captured by the religious background of the child proves important as a determinant of child labor and schooling, since children with a Christian, Muslim or other religious background are generally more likely to only attend school, while less likely to only engage in child labor, relative to children from a traditional/animist religious background. This results points towards the importance of other than purely economic factors being crucial to increase our understanding of child labor, its determinants and its prevalence. A key result is that of the incidence of child labor - both actual and as predicted by our models, introducing multiple influences of child labor and schooling - falling dramatically over time. Additionally, schooling increases over time. These two results - being largely consistent with the "real" developments, as discussed in the presentation of the descriptive statistics in section 4 - thus indicates that the estimated models capture the features of the data well. Hence, even when taking into account all possible influences simultaneously may we conclude that some support exist for the key conjectures of this paper, namely that child labor and schooling is likely not to respond in the short run, while it will respond in the long run - and with opposite signs: child labor will decrease and schooling will increase, once the effects from economic reforms materialize and/or become credible. In the short run, child 31  labor policies may be ineffective, possibly even counterproductive, since it takes some time for the policy to have its full effect. A very clear example of this is the result that poverty has no affect on the schooling decision in the short run, while it positively affects the probability of only attending school in the long run. We may interpret this result as a certain degree of "sluggishness" being inherent in the household decision process, since the household decision maker first has to believe that the proposed policy interventions will truly be made before she feels certain enough to act on them. Due to the high costs of schooling, both direct and in the form of forgone earnings of the child, human capital investments seem particularly sensitive to policy announcements and to their credibility. While this study attempted to shed some light on the dynamics of child labor and its determinants, it is a very modest first attempt to do so. Much more remains to be done. In particular, we suggest that more attention be directed to the poverty-child labor link, as the present study raises serious doubts about the validity of this link as merely a simple, linear relationship. Also, it would be interesting to analyze whether similar results hold for other countries who have experienced marked economic reforms. Further, other factors than the traditional economic factors needs to be analyzed more in order to increase our understanding about child labor and its determinants. These include, but are by no means limited to, intergenerational spill-over effects in education as well as traditions and customs as revealed by, for example, children's religious background. These are all factors, the inclusion of which is crucial to increase our understanding of child labor as a dynamic phenomenon, an understanding which is absolutely crucial to the design and implementation of sound policy in this area in the future. 32  References Alderman Harold (1994), "Ghana: Adjustment's Star Pupil?" in David Sahn (ed.) Adjusting to Policy Failure in African Economies, Cornell University Press, Ithaca and London. Andvig, Jens (1999), "Child Labor in Sub-Saharan Africa: An Exploration", Norwegian Institute of International Affairs. Andvig, Jens, Sudharshan Canagarajah and Anne Kielland (2001), "Child Labor in Africa: The Issues", Mimeo, World Bank, Washington, D.C. Baland, J-M and James Robinson (2000), "Is Child Libor Inefficient?", Journal of Political Economy, v108, n4: 663-79. Basu, Kaushik and Pham Hoang Van (1998), "The Economics of Child Labor", American Economic Review, Vol 88:3, 412-427. Becker, G. S. (1965), "A Theory of the Allocation of Time", Economic Journal, 75, 493-517. Bergstrom, T. C. (1996), "Economics in a Family Way", Journal of Economic Literature 34, 1903-34. Binswanger, H. (1989), "The Policy Response of Agriculture", Proceedings of the World Bank Annual Conference on Development Economics 1989, Washington, D.C.: World Bank. Bonnet, M. (1993), "Child Labour in Africa", International Labour Review, v132, n3: 371-389. Broad, Robin and John Cavanagh (1999), "The Death of the Washington Consensus?", World Policy Journal, 16, No. 3:79-88. Burra, Neera (1994), Born to Work, Oxford University Press, New Delhi. Canagarajah, Sudharshan and Harold Coulombe (1997), "Child Labor and Schooling in Ghana" Policy Research Working Paper 1844, The World Bank, Washington D.C. Canagarajah, Sudharshan and Helena Skyt Nielsen (1999), "Child Labor and Schooling in Africa: A Comparative Study", Social Protection Working Paper No. 9916, Human Development Network Social Protection, World Bank, Washington, D.C. 33  Chandrasekhar, C. P (1997), "The Economic Consequences of the Abolition of Child Labour: An Indian Case Study", The Journal of Peasant Studies, Vol 24: 23, Frank Cass, London. Demery, Lionel and Lyn Squire (1996), "Macroeconomic Adjustment and Poverty in Africa: an Emerging Picture", World Bank Research Observer, 11:39-59. Demery, Lionel and Tony Addison (1993), "Impact of Macroeconomic Adjustment on Poverty in the Presence of Wage Rigidities", Journal of Development Economics, 40:331-48. Dercon, Stefan and Pramila Krishnan (1997), "In Sickness and in Health: Risk-Sharing Within Households in Rural Ethiopia", Centre for the Study of African Economies Working Paper Series No. WPS/97-12. Doss, Cheryl R. (1996), "Testing Among Models of Intrahousehold Resource Allocation", World Development, 24:1597-1609. Ebdon, Rosamund (1999), "Working Children's Futures: Child Labour, Poverty and Education", Mimeo, Save the Children UK. Eele, Graham, Roger Hay and John Hoddinot (1993), "Household Food Security and Nutrition", in: Demery, Lionel, Marco Ferroni and Christiaan Grootaert, eds. Understanding the Social Effects of Policy Reform, pp. 155-72. Washington, D.C.: World Bank Fafchamps, Marcel (1998), "Intra-household Access to Land and Sources of Inefficiency: Theory and Concepts", Mimeo, Department of Economics, Stanford University. Fallon, Peter and Zafiris Tzannatos (1998), Child Labor: Issues and Directions for the World Bank, Social Protection Human Development Network, World Bank, Washington, D.C. Glewwe, P. (1996), "The Relevance of Standard Estimates of Rates of Return to Schooling for Education Policy: A Critical Assessment", Journal of Development Economics, 5, 267-290. Greene, William H. (2000), "Econometric Analysis", Prentice-Hall. Grimsrud, Bjoeme (1999), "Measuring and Analyzing Child Labor Issues", Mimeo, World Bank, Washington, D.C.' Grootaert, Christiaan (1998), "Child Labor in Cote d'Ivoire", World Bank Policy Working Paper 1905, World Bank, Washington, D.C. 34  Grootaert and Kanbur (1995), "Child labor: A Review", Background Paper for World Development Report 1995, Office of the Vice President, Development Economics, World Bank, Washington, D.C. Hanan and Skoufias (1997), "Risk, Financial Markets, and Human Capital in a Developing Country", Review of Economic Studies, v64, n3: 311-35. Hill, Anne M. (1983), "Female Labor Force Participation in Developing and Developed Countries - Consideration of the Informal Sector", Review of Economics and Statistics, v65, n3: 459-68 Hoddinott, John and Lawrence Haddad (1995), "Does Female Income Share Influence Household Expenditures? Evidence from Cote d'Ivoire", Oxford Bulletin of Economics and Statistics v57, nl: 77-96. Intemational Labour Office (1997), "IPEC at a Glance", ILO: Geneva. Kielland, Anne (1999), "Child Labor in Benin", Mimeo, Africa Technical Families Human Development 3, World Bank, Washington, D.C. Levy, Victor (1985), "Cropping Pattern, Mechanization, Child Labor, and Fertility Behavior in a Farming Economy: Rural Egypt", Economic Development and Cultural Change, Vol. 33 (4): 777-91. Lundberg, S. and R. Pollak, "Noncooperative Bargaining Models of Marriage," American Economic Review--Proceedings, 84:2, 132-7. McElroy, M. B. and M. J. Homey. "Nash Bargained Household Decisions: Toward a Generalization of the Theory of Demand," International Economic Review, 22, 333- 49. Moock, Peter R., Harry Anthony Patrinos and Meera Venkataraman (1998), "Education and Earnings in a Transition Economy", Policy Research Working Paper No. 1920, World Bank, Washington, D.C. Nielsen, H. S. (1998), "Child Labor and School Attendance in Zamnbia: Two Joint Decisions", CLS WP 98-15, Centre for Labour Market and Social Research, Aarhus, Denmark. Oi, Walter Y. (1996), "Labor as a Quasi-fixed Factor", in Orley C. Ashenfelter and Robert J. LaLonde, (eds.): The Economics of Training Vol. 1. Theory and Measurement Elgar Reference Collection. International Library of Critical Writings in Economics, vol. 65. Cheltenham, U.K.: Elgar. 35  Parsons, Donald and Claudia Goldin (1989), "Parental Altruism and Self-interest: Child Labor among Late Nineteenth-century American Families", Economic Enquiry, v26, pp. 637-659. Psacharopoulos, George (1994), "Returns to Investment in Education: A Global Update", World Development, 22:1325-1343. Ranjan, Ray (2000), "Child Labor, Child Schooling, and Their Interaction with Adult Labor: Empirical Evidence for Peru and Pakistan", The World Bank Economic Review, v14, n2: 347-367. Ravallion, Martin and Quentin Wodon (1999), "Does Child Labor Displace Schooling?: Evidence on Behavioral Responses to an Enrollment Subsidy", World Bank Policy Working Paper 2116, World Bank, Washington, D.C. Rosenzweig, Mark R. and Kenneth I. Wolpin (1993), "Intergenerational Support and the Life-cycle Incomes of Young Men and Their Parents: Human Capital Investments, Coresidence, and Intergenerational Financial Transfers", Journal of Labor Economics; 11:84-112. Sawada, Yasuyuki (1997), "Human Capital Investments in Pakistan: Implications of Micro Evidence from Rural Households", Pakistan Development Review, v36, n4: 695-710. Skoufias, Emmanuel (1993), "Seasonal Labor Utilization in Agriculture: Theory and Evidence from Agrarian Households in India", American Journal of Agricultural Economics, v75, nl: 20-32. Skoufias, Emmanuel (1994), "Market Wages, Family Composition and the Time Allocation of Children in Agricultural Households", Journal of Development Studies, v30, n2: 335-60. Smith, Lisa C. (1994), "Structural Adjustment and Welfare in Rural Africa: The Role of Resource Control in Households", Department of Agricultural Economics Staff Paper Series, University of Wisconsin-Madison. Strauss, John and Duncan Thomas (1995), "Human Resources: Empirical Modeling of Household and Family Decisions", in: Behrman, Jere; Srinivasan, T. N. (eds.) Handbook of Development Economics, Vol. 3A, Amsterdam; New York and Oxford: Elsevier Science, North Holland. Swaminathan, M. (1998), "Economic Growth and the Persistence of Child Labor: Evidence From an Indian City", World Development, 26:1513-28. 36  Udry (1994), "Risk and Insurance in a Rural Credit Market: An Empirical Investigation in Northern Nigeria" Review of Economic Studies, v61, n3: 495-526. UNICEF (1999), The State of the World's Children 1999, Oxford: Oxford University Press. Weiner, Myron (1991), The Child and the State in India: Child Labor and Education Policy in Comparative Perspective, Princeton University Press: Princeton. White, H. (1980), "A Heteroscedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroscedasticity", Econometrica, 48, pp. 817-838. World Bank (1993), Ghana 200 and Beyond: Setting the Stage for Accelerated Growth and Poverty Reduction, Africa Regional Office, Western Africa Department, World bank, Washington, D.C. World Bank (1995 a), Ghana: Growth, Private Sector, and Poverty Reduction - A Country Economic Memorandum, Country Operations Division, West Central Africa Department, World bank, Washington, :D.C. World Bank (1995b), Ghana: Poverty Past, Present and Future, Population and Human Resources Division, West Central Africa Department, Africa Region, World Bank, Washington, D.C. Wydick, Bruce (1999), "The Effect of Microenterprise Lending on Child Schooling in Guatemala", Economic Development and Cultural Change, Vol. 74:4, 853-870. 37  Appendix A: Definition of Variables Age Age of individual Age squared Age of individual squared Female 1 if female, 0 if male Child of household 1 if child of household, 0 otherwise Mother lives in household I if mother lives in household, 0 otherwise Father lives in household 1 if father lives in household, 0 otherwise Mother: None 1 if mother completed no education (reference group) Mother: Primary 1 if mother completed primary education Mother: Middle secondary 1 if mother completed middle secondary education, 0 otherwise Mother: Post middle secondary 1 if mother completed some post middle secondary education, 0 otherwise Father: None 1 if father completed no education (reference group) Father: Primary 1 if father completed primary education Father: Middle secondary 1 if father completed middle secondary education, 0 otherwise Father: Post middle secondary I if father completed some post middle secondary education, 0 otherwise Children 0-6 years Number of children 0-6 years of age Brothers 7-14 years Number of brothers 7-14 years of age Sister 7-14 years Number of sisters 7-14 years of age Males 15-59 years Number of males 15-59 years of age Females 15-59 years Number of females 15-59 years of age Older than 60 Number of individuals older than 60 years of age Anumist 1 if Animist/Traditional religion, 0 otherwise (reference group) Muslim 1 if Muslim, 0 otherwise Catholic 1 if Catholic, 0 otherwise Protestant 1 if Protestant, 0 otherwise Other Christian 1 if other Christian, 0 otherwise (Log) per capita expenditure (Log) per capita expenditure Male head of household 1 if male headed household, 0 otherwise Age of head of household Age of head of household Age of head of household squared Age of head of household squared Household owns livestock 1 if household owns livestock, 0 otherwise Household owns land I if household owns land, 0 otherwise Schooling expenditure Total schooling expenditure Distance (in minutes) Distance to school (in minutes) Rural Savannah 1 if individual is from Rural Savannah, 0 otherwise (reference group) Accra I if individual is from Accra, 0 otherwise Urban areas outside Accra 1 if individual is from urban areas outside Accra, 0 otherwise Rural Coastal 1 if individual is from Rural Coastal Region, 0 otherwise Rural Forest 1 if individual is from Rural Forestal Region, 0 otherwise 38  Appendix B: Descriptive Statistics Table B1. Typical Profile of a Child Worker in Ghana 1987/88 Variable Male Female Average Age: Urban 11.5 11.2 Rural 10.9 10.9 Child Worker Composition: Urban 5.0 4.5 Rural 33.4 25.2 Proportion of child workers in labor force 11.6 9.0 Child Labor Force Participation Rate 17.6 13.6 Participation in Farming (percent) 67.5 67.9 Current School Participation (percent) 55.3 47.5 Contribution to total hours of participation 11.0 10.7 nationally Source: Calculations based on Ghana Living Standards Survey 1987/88 Table B2. Joint Labor Force and School Participation Rate (last 7 days) in Ghana 1987/88, by gender, age and expenditure quintiles ,. ~ -, .*Wdork'Only SchoolfOily Work & None Gender Male 19.6 40.9 14.4 25.1 Female 18.6 38.1 9.4 33.9 Age 7 9.0 44.3 5.9 40.6 8 15.6 44.0 5.7 34.7 9 15.7 41.9 11.0 31.8 10 19.8 37.8 14.4 28.0 11 19.9 42.0 14.1 23.9 12 24.7 32.6 14.3 28.4 13 25.1 37.3 17.2 20.4 14 25.8 36.1 16.4 21.8 Expenditure Quintile 1 19.9 36.3 11.6 32.2 2 20.2 38.1 13.7 28.0 3 21.2 39.8 10.4 28.6 4 17.2 40.4 12.7 29.7 5 16.0 45.1 11.3 27.7 Source: Calculations based on Ghana Living Standards Survey 1987/88 39  Table B3. Typical Profile of a Child Worker in Ghana 1991/92 Category Male Female Average Age: Urban 11.8 11.3 Rural 11.0 11.0 Child Worker Composition: Urban 4.5 5.2 Rural 49.1 41.2 Proportion of child workers in labor force 14.0 10.0 Child Labor Force Participation Rate 29.3 26.7 Participation in Farming (percent) 96.3 88.5 Current School Participation 76.7 68.3 Contribution to total hours of participation nationally 5.4 5.3 Source: Calculations based on Ghana Living Standards Survey 1991/92 Table B4. Joint Labor Force and School Participation Rate (last 7 days) in Ghana 1991/92, by gender, age and expenditure quintiles : Work Oniv School Only Work & None __________ __________ ____ __ - -School Gender Male 9.2 56.6 20.1 14.1 Female 9.4 51.0 17.3 22.3 Age 7 4.7 56.2 7.4 31.7 8 6.7 59.1 11.4 22.8 9 6.1 57.5 17.0 19.4 10 8.8 55.2 20.2 15.9 11 8.2 56.3 23.6 11.9 12 11.5 51.3 22.6 14.6 13 14.0 46.1 28.3 11.6 14 16.4 47.1 24.3 12.2 Expenditure Quintile 1 13.1 46.4 15.5 24.9 2 6.8 54.1 21.7 17.3 3 10.5 53.8 18.6 17.1 4 8.7 55.2 19.2 17.0 5 5.7 64.6 19.1 10.6 Source: Calculations based on Ghana Living Standards Survey 1991/92 40  Appendix C: Determinants of Child Labor and Schooling Table Cl. Determinants of Children's' Activities in Ghana, GLSS1 '_ *t;,- , .Sho'o'onIy'l' * X -SchooVand aWork - Work only -off.., Std.error e Sd Coeff. Std.error Individual characteristics: Age -0.113 0.060 0.057 0.036 0.162 0.048 Age squared 0.004 0.003 -0.002 0.002 -0.006 0.002 Female -0.035 0.026 -0.043 0.015 -0.016 0.020 Child of household 0.126 0.055 0.014 0.029 -0.074 0.042 Parent characteristics: Mother lives in household -0.046 0.037 -0.042 0.020 -0.007 0.028 Father lives in household -0.034 0.046 -0.014 0.026 0.047 0.037 Mother: Primary -0.041 0.050 -0.019 0.028 -0.015 0.046 Mother: Middle secondary 0.033 0.043 -0.042 0.027 -0.016 0.037 Mother: Post riddle secondary 0.114 0.125 -0.120 0.112 0.212 0.097 Father: Primary 0.277 0.051 -0.022 0.037 -0.087 0.047 Father: Middle secondary 0.147 0.034 0.039 0.020 -0.091 0.028 Father: Post middle secondary 0.127 0.055 0.049 0.029 -0.125 0.047 Household Characteristics: Children 0-6 years -0.005 0.009 0.006 0.005 0.011 0.008 Brothers 7-14 years 0.043 0.017 -0.002 0.009 -0.008 0.013 Sister 7-14 years 0.058 0.020 0.010 0.011 -0.011 0.017 Males 15-59 years 0.004 0.012 -0.022 0.007 -0.009 0.011 Females 15-59 years 0.012 0.013 0.018 0.007 -0.003 0.010 Older than 60 -0.022 0.030 -0.036 0.016 -0.023 0.022 Muslim 0.169 0.050 -0.014 0.039 -0.001 0.047 Catholic 0.254 0.054 0.048 0.027 -0.158 0.056 Protestant -0.048 0.054 0.111 0.024 -0.022 0.045 Other Christian -0.016 0.059 0.056 0.029 -0.090 0.056 (Log) per capita expenditure -0.014 0.026 0.001 0.016 0.048 0.020 Male head of household -0.096 0.032 0.046 0.019 0.047 0.027 Age of head of household -0.012 0.005 -0.003 0.003 0.003 0.004 Age of head of household squared 0.000 5.71E-05 5.09E-05 0.000 -2.9E-05 4.64E-05 Household owns livestock -0.140 0.089 -0.045 0.053 0.118 0.049 Household owns land -0.024 0.039 -0.017 0.021 -0.003 0.032 Cost of Schooling: Schooling expenditure -3.69E-07 7.61E..06 9.39E-06 4.04E-06 -7.45E-06 6.23E-06 Distance (in minutes) -5.lE-05 0.001 0.001 0.001 -0.001 0.001 Location: Accra 4.628 0.222 -2.679 0.204 -5.456 0.238 Urban areas outside Accra 0.088 0.042 -0.016 0.029 -0.195 0.038 Rural Coastal -0.090 0.044 -0.005 0.028 0.045 0.032 Rural Forest 0.005 0.035 0.004 0.020 -0.034 0.026 Constant 1.205 0.431 -0.495 0.257 -1.476 0.347 Notes: The coefficients are marginal probabilities. Number of observations = 1732; Wald )?(102) = 7679.67; Prob > x2 = 0.000; Log likelihood = -2073.911; Pseudo-R2 = 0.107. Bold: Statistically significant at 5 percent; underline: Statistically significant at 10 percent; cursive: Statistically significant at 15 percent; standard errors are robust Huber-White Sandwich standard errors, White (1980). 41  Table C2. Determinants of Children's' Activities in Ghana, GLSS3 - -, - -* - - School only , School and Work Work only Coeff.- Sld.error Coeff. Std.error Coeff. Std.error Individual characteristics: Age -0.052 0.041 0.165 0.030 0.050 0.034 Age squared 0.002 0.002 -0.007 0.001 -0.001 0.002 Female -0.038 0.018 -0.012 0.012 -0.020 0.014 Child of household -0.029 0.042 0.013 0.026 0.092 0.034 Parent characteristics: Mother lives in household -0.091 0.029 -0.020 0.019 0.050 0.025 Father lives in household 0.145 0.037 -0.045 0.023 -0.174 0.028 Mother: Primary 0.0S0 0.032 -0.001 0.021 0.021 0.028 Mother: Middle secondary 0.064 0.030 0.027 0.020 2.19E-05 0.028 Mother Post rniddle secondary -0.019 0.093 -0.075 0.072 -0.126 0.104 Father: Primary 0.090 0.036 0.032 0.023 -0.059 0.031 Father: Middle secondary 0.066 0.025 0.018 0.016 -0.048 0.020 Father: Post middle secondary 0.147 0.038 -0.031 0.028 -0.133 0.040 Household Characteristics: Children 0-6 years -0.009 0.008 -0.001 0.005 0.002 0.006 Brothers 7-14 years 0.023 0.014 0.004 0.008 0.002 0.010 Sister 7-14 years 0.025 0.016 -6.6E-05 0.010 -0.019 0.012 Males 15-59 years -0.020 0.009 0.024 0.005 0.003 0.007 Females 15-59 years 0.010 0.010 -0.008 0.007 -0.001 0.007 Older than 60 0.010 0.020 0.004 0.014 -0.014 0.016 Muslim 0.039 0.027 0.055 0.019 -0.040 0.021 Catholic 0.120 0.031 0.021 0.022 -0.050 0.025 Protestant 0.140 0.031 0.052 0.019 -0.035 0.024 Other Christian 0.111 0.030 0.009 0.019 -0.031 0.023 (Log) per capita expenditure 0.060 0.018 -0.001 0.012 -0.014 0.014 Male head of household -0.105 0.026 0.006 0.018 0.136 0.022 Age of head of household -0.000 0.004 0.002 0.003 -0.005 0.003 Ageofheadofhouseholdsquared -1.7E-05 0.000 -1.2E-05 2.74E-05 6.07E-05 2.85E-05 Household owns livestock -0.084 0.043 -0.064 0.039 0.065 0.028 Household owns land -0.054 0.019 0.003 0.013 0.035 0.016 Cost of Schooling: Schooling expenditure 1.07E-05 2.74E-06 8.75E-06 1.76E-06 2.08E-06 2.20E-06 Distance (in minutes) -0.000 0.000 -0.001 0.000 0.000 0.000 Location: Accra -5.032 . -2.141 . -2.835 Urban areas outside Accra 0.153 0.036 -0.146 0.033 -0.055 0.036 Rural Coastal -0.088 0.032 0.075 0.022 0.070 0.022 Rural Forest -0.007 0.025 0.150 0.017 0.034 0.019 Constant -0.2447 0.317 -1.180 0.235 -0.313 0.259 Notes: The coefficients are marginal probabilities. Number of observations = 3516; Wald X2(99) = 1082.58; Prob > %2 = 0.000; Log likelihood = 4088.398; Pseudo-R2 = 0.135. Bold: Statistically significant at 5 percent; underline: Statistically significant at 10 percent; cursive: Statistically significant at 15 percent; standard errors are robust Huber-White Sandwich standard errors, White (1980). 42  Table C3. Determinants of Children's' Activities in Ghana, GLSS4 -. ' : ' ' : CoefT. .Std.error Z-statistic P-value Individual characteristics: Age 0.020 0.020 1.04 0.296 Age squared -0.001 0.001 -0.78 0.433 Female -0.002 0.008 -0.26 0.794 Child of household -0.790 0.163 -4.77 0.000 Parent Characteristics: Mother lives in household 0.065 0.024 4.16 0.000 Father lives in household 0.115 0.064 3.24 0.001 Household Characteristics: Children 0-6 years 0.017 0.008 3.73 0.000 Brothers 7-14 years -0.010 0.007 -1.56 0.120 Sister 7-14 years -0.003 0.006 -0.50 0.618 Males 15-59 years -0.003 0.006 -0.52 0.606 Females 15-59 years -0.000 0.005 -0.04 0.967 Older than 60 0.005 0.016 0.28 0.777 Muslim -0.020 0.009 -3.60 0.000 Christian -0.423 0.178 -4.05 0.000 No religion -0.016 0.008 -3.09 0.002 (Log) per capita expenditure -0.021 0.010 -2.92 0.003 Male head of household -0.102 0.067 -2.78 0.005 Age of household head -0.004 0.002 -1.80 0.071 Age of household head squared 4.25E-5 2.03E-5 1.97 0.048 Land -0.001 0.008 -0.13 0.895 Cost of Schooling: Schooling expenditure 3.27E-7 1.43E-7 3.19 0.001 Distance (in minutes) -0.000 0.000 -0.78 0.437 Location: Other urban 0.527 0.118 5.70 0.000 Rural Coastal 0.996 0.004 8.77 0.000 Rural Forest 0.295 0.070 5.56 0.000 Notes: The coefficients are marginal probabilities. The locational dunmny for Accra as well as the dummy for livestock ownership, "other religion" and parents' education had to be omitted due to collinearity, number of observations = 443; Wald X2(25) = 125.38; Prob > %2 = 0.000; Log likelihood = -73.259; Pseudo- R2= 0.618; standard errors are robust Huber-White Sandwich standard errors, White (1980). 43  Appendix D: Predictions Table Dl. Predictions, GLSSI Varijble* - _School only Schooland. Work only. Neither - ~~~work V Full sample 34.1 13.7 25.1 27.1 Girls 33.2 11.0 23.8 32.0 Boys 35.0 16.3 26.2 22.6 Child of the household 35.3 13.0 25.2 26.5 Not child of the household 30.9 15.8 24.6 28.8 Mother no education 31.8 13.7 27.2 27.4 Mother primary 38.6 17.0 15.7 28.8 Mother middle secondary 44.0 12.9 17.8 25.3 Mother some post-secondary 45.5 4.6 36.4 13.6 Father no education 28.2 12.1 30.0 29.7 Father primary 53.0 8.7 20.0 18.3 Father middle secondary 40.0 17.3 18.2 24.5 Father some post-secondary 40.8 17.8 17.1 24.3 Across expenditure quintiles: Lowest 35.9 12.1 23.4 28.7 Next-to-lowest 33.0 14.1 25.0 27.8 Middle 34.5 13.6 24.9 27.1 Next-to-highest 33.8 15.1 25.2 25.9 Highest 33.0 14.3 28.1 24.6 Table D2. Predictions, GLSS3 Variable School only . School and Work only Neither -work Full sample 31.6 17.6 18.1 32.8 Girls 30.1 16.8 16.6 36.5 Boys 32.8 18.3 19.4 29.5 Childofthehousehold 31.3 16.3 18.1 34.3 Not child of the household 32.8 22.9 18.1 26.2 Mother no education 28.0 16.2 19.2 36.6 Mother primary 43.2 21.5 16.8 18.5 Mother middle secondary 46.0 24.5 12.6 16.8 Mother some post-secondary 45.7 11.4 5.7 37.1 Fatherno education 25.1 13.9 20.7 40.3 Father primary 38.6 24.0 15.0 22.4 Father middle secondary 39.5 24.5 15.1 21.0 Father some post-secondary 51.3 16.7 9.2 22.8 Across expenditure quintiles: Lowest 27.6 15.4 18.4 38.6 Next-to-lowest 30.9 17.5 17.8 33.9 Middle 31.7 19.2 18.4 30.6 Next-to-highest 32.4 18.6 18.4 30.7 Highest 37.7 17.9 17.2 27.2 44  Social Protection Discussion Paper Series Titles No. Title 0214 Income Support Systems for the Unemployed: Issues and Options Milan Vodopivec and Dhushyanth Raju 0213 Social Protection @ Your Fingertips. Using Information & Communications Technologies in Social Protection Knut Leipold 0212 Short-and Long-term Impacts of Economic Policies on Child Labor and Schooling in Ghana Niels-Hugo Blunch, Sudharshan Canagarajah and Sangeeta Goyal 0211 Supporting and Expanding Community-Based HIV/AIDS Prevention and Care Responses: A Report on Save the Children (US) Malawi COPE Project by Susan S. Hunter 0210 World Vision's Experience Working with HIV/AIDS Orphans in Uganda - 1990-1995 by Joe Muwonge 0209 The Reformed Pension Systems in Latin America by Jose E. Devesa-Carpio and Carlos Vidal-Melia 0208 Mandatory Annuity Design in Developing Economies by Suzanne Doyle and John Piggott 0207 Long-Term Welfare and Investment Impact of AIDS-Related Changes in Family Composition: Evidence from Uganda by Klaus Deininger, Anja Crommnelynck and Gloria Kempaka 0206 Child Labor Handbook by Alessandro Cigno, Furio C. Rosati and Zafiris Tzannatos 0205 An Overview of Labor Markets World-Wide: Key Trends and Major Policy Issues by Gordon Betcherman 0204 Options of Public Income Support for the Unemployed in the Philippines and Social Protection by Jude H. Esguerra, Makoto Ogawa, and Milan Vodopivec 0203 Unemployrnent Insurance and Unemployment Assistance: A Comparison by Wayne Vroman  Social Protection Discussion Paper Series Titles continued No. Title 0202 Does Eurosclerosis Matter? Institutional Reform and Labor Market Performance in Central and Eastern European Countries in the 1990s. by Michelle Riboud, Carolina Sanchez-Paramo and Carlos Silva-Jauregui 0201 Pension Reform and Capital Markets: Are There Any (Hard) Links? byEduardo Walker and Femando Lefort 0131 Child Labor, Nutrition and Education in Rural India: An Economic Analysis of Parental Choice and Policy Options by Alessandro Cigno, Furio Camillo Rosati and Zafiris Tzannatos 0130 Social Protection and the Informal Sector in Developing Countries: Challenges and Opportunities by Sudharshan Canagarajah and S.V. Sethuraman 0129 Chile's Pension Reform After 20 Years by Rodrigo Acufla R. and Augusto Iglesias P. 0128 Labor Market Regulation: International Experience in Promoting Employment and Social Protection by Gordon Betcherman, Amy Luinstra, and Makoto Ogawa 0127 Generational Accounting and Hungarian Pension Reform by R6bert I. Gal, Andras Simonovits and Geza Tarcali 0126 Orphans and Other Vulnerable Children: What role for social protection? edited by Anthony Levine 0125 Child Farm Labour: The Wealth Paradox by Sonia Bhalotra 0124 What Can Be Done about Child Labor? An overview of resent research and its implications for designing programs to reduce child labor by Bjorne Grimsrud 0123 Measuring and Analyzing Child Labor: Methodological Issues by Bj0rne Grimsrud 0122 Family-Controlled Child Labor in Sub-Saharan Africa-A Survey of Research by Jens Christopher Andvig  Social Protection Discussion Paper Series Titles continued No. Title 0121 Is Child Work Necessary? by Sonia Bhalotra 0120 The Cost and Benefits of Collective Bargaining: A Survey by Toke Aidt and Zafiris Tzannatos 0119 The Informal Sector Revisited: A Synthesis Across Space and Time by Niels-Hugo Blunch, Sudharshan Canagarajah and Dhushyanth Raju 0118 Social Services Delivery through Community-Based Projects by Dinah McLeod and Maurizia Tovo 0117 Earnings Inequality in Transition Economies of Central Europe Trends and Patterns During the 1990s by Jan J. Rutkowski 0116 Viewing Microinsurance as a Social Risk Management Instrument by Paul B. Siegel, Jeffrey Alwang and Sudharshan Canagarajah 0115 Vulnerability: A View from Different Disciplines by Jeffrey Alwang, Paul B. Siegel and Steen L. Jorgensen 0114 Individual Accounts as Social Insurance: A World Bank perspective by Robert Holzmann and Robert Palacios 0113 Regulating Private Pension Funds' Structure, Perfornance and Investments: Cross-country Evidence by P.S. Srinivas, Edward Whitehouse and Juan Yermo 0112 The World Bank and the Provision of Assistance to Redundant Workers: Experience with Enterprise Restructuring and Future Directions by Yi Chen 0111 Labor Markets in Transition Economies: Recent Developments and Future Challenges by Mansoora Rashid and Jan Rutkowski 0110 A Review of Social Investment Fund Operations Manuals by Juliana Weissman  Social Protection Discussion Paper Series Titles continued No. Title 0109 Risk and Vulnerability: The Forward Looking Role of Social Protection in a Globalizing World by Robert Holzmann 0108 Australia's Mandatory Retirement Saving Policy: A View from the New Millennium by Hazel Bateman and John Piggott 0107 Annuity Markets and Benefit Design in Multipillar Pension Schemes: Experience and Lessons from Four Latin American Countries by Robert Palacios and Rafael Rofman 0106 Guide for Task Teams on Procurement Procedures Used in Social Funds by Jorge A. Cavero Uriona 0105 Programmes Actifs Pour Le Marche Du Travail: Un Apercu General Des Evidences Resultant Des Evaluations by Zafiris Tzannatos and Amit Dar 0104 Kazakhstan: An Ambitious Pension Reform by Emily S. Andrews 0103 Long-term Consequences of an Innovative Redundancy-retraining Project: The Austrian Steel Foundation by Rudolf Winter-Ebmer 0102 Community Based Targeting Mechanisms for Social Safety Nets by Jonathan Conning and Michael Kevane 0101 Disability and Work in Poland by Tom Hoopengardner 0024 Do Market Wages Influence Child Labor and Child Schooling? by Jackline Wahba 0023 Including the Most Vulnerable: Social Funds and People with Disabilities by Pamela Dudzik and Dinah McLeod 0022 Promoting Good Local Governance through Social Funds and Decentralization by Andrew Parker and Rodrigo Serrano  Social Protection Discussion Paper Series Titles continued No. Title 0021 Creating Partnerships with Working Children and Youth by Per Miljeteig 0020 Contractual Savings or Stock Market Development. Which Leads? by Mario Catalan, Gregorio Impavido and Alberto R. Musalem 0019 Pension Reform and Public Information in Poland by Agnieszka Chlon 0018 Worker Reallocation During Estonia's Transition to Market: How Efficient and How Equitable? by Milan Vodopivec 0017 How Poor are the Old? A Survey of Evidence from 44 Countries by Edward Whitehouse 0016 Administrative Charges for Funded Pensions: An Intemational Comparison and Assessment by Edward Whitehouse 0015 The Pension System in Argentina: Six years after the Reform by Rafael Rofman 0014 Pension Systems in East Asia and the Pacific: Challenges and Opportunities by Robert Holzmann, Ian W. Mac Arthur and Yvonne Sin 0013 Survey of Disability Projects. The Experience of SHIA, Swedish International Aid for Solidarity and Humanity by Kaj Nordquist 0012 The Swedish Pension Reform Model: Framework and Issues by Edward Palmer 0011 Ratcheting Labor Standards: Regulation for continuous Improvement in the Global Workplace by Charles Sabel, Dara O'Rourke and Archon Fung 0010 Can Investments in Emerging Markets Help to Solve the Aging problem? by Robert Holzmann 0009 International Patterns of Pension Provision by Robert Palacios and Montserrat Pallares-Miralles  Social Protection Discussion Paper Series Titles continued No. Title 0008 Regulation of Withdrawals in Individual Account Systems by Jan Walliser 0007 Disability Issues, Trends and Recommendations for the World Bank by Robert L. Metts 0006 Social Risk Management: A New Conceptual Framework for Social Protection and Beyond by Robert Holzmann and Steen J0rgensen 0005 Active Labor Market Programs: Policy Issues for East Asia by Gordon Betcherman, Amit Dar, Amy Luinstra, and Makoto Ogawa 0004 Pension Reform, Financial Literacy and Public Information: A Case Study of the United Kingdom by Edward Whitehouse 0003 Managing Public Pension Reserves Part I: Evidence from the International Experience by Augusto Iglesias and Robert J. Palacios 0002 Extending Coverage in Multi-Pillar Pension Systems: Constraints and Hypotheses, Preliminary Evidence and Future Research Agenda by Robert Holzmann, Truman Packard and Jose Cuesta 0001 Contribution pour une Strategie de Protection Sociale au Benin by Maurizia Tovo and Regina Bendokat * The papers below (No. 9801-9818 and 9901-9934) are no longer being printed, but are available for download from our website at www.worldbank.org/sp 9934 Helping the Poor Manage Risk Better: The Role of Social Funds by Steen J0rgensen and Julie Van Domelen 9933 Coordinating Poverty Alleviation Programs with Regional and Local Governments: The Experience of the Chilean Social Fund - FOSIS by Jorge C. Barrientos 9932 Poverty and Disability: A Survey of the Literature by Ann Elwan  Social Protection Discussion Paper Series Titles continued No. Title 9931 Uncertainty About Children's Survival and Fertility: A Test Using Indian Microdata by Vincenzo Atella and Furio Camillo Rosati 9930 Beneficiary Assessment of Social Funds by Lawrence F. Salmen 9929 Improving the Regulation and Supervision of Pension Funds: Are there Lessons from the Banking Sector? by Roberto Rocha, Richard Hinz, and Joaquin Gutierrez 9928 Notional Accounts as a Pension Reform Strategy: An Evaluation By Richard Disney 9927 Parametric Reforms to Pay-As-You-Go Pension Systems by Sheetal K. Chand and Albert Jaeger 9926 An Asset-Based Approach to Social Risk Management: A Conceptual Framework by Paul Siegel and Jeffrey Alwang 9925 Migration from the Russian North During the Transition Period by Timothy Heleniak 9924 Pension Plans and Retirement Incentives by Richard Disney and Edward Whitehouse 9923 Shaping Pension Reform in Poland: Security Through Diversity by Agnieszka Chlon, Marek G6ra and Michal Rutkowski 9922 Latvian Pension Reform by Louise Fox and Edward Palmer 9921 OECD Public Pension Programmes in Crisis: An Evaluation of the Reform Options by Richard Disney 9920 A Social Protection Strategy for Togo by Regina Bendokat and Amit Dar 9919 The Pension System in Singapore by Mukul G. Asher  Social Protection Discussion Paper Series Titles continued No. Title 9918 Labor Markets and Poverty in Bulgaria by Jan J. Rutkowski 9917 Taking Stock of Pension Reforms Around the World by Anita M. Schwarz and Asli Demirguc-Kunt 9916 Child Labor and Schooling in Africa: A Comparative Study by Sudharshan Canagarajah and Helena Skyt Nielsen 9915 Evaluating the Impact of Active Labor Programs: Results of Cross Country Studies in Europe and Central Asia by David H. Fretwell, Jacob Benus, and Christopher J. O'Leary 9914 Safety Nets in Transition Economies: Toward a Reform Strategy by Emily S. Andrews and Dena Ringold 9913 Public Service Employment: A Review of Programs in Selected OECD Countries and Transition Economies by Sandra Wilson and David Fretwell 9912 The Role of NPOs in Policies to Combat Social Exclusion by Christoph Badelt 9911 Unemployment and Unemployment Protection in Three Groups of Countries by Wayne Vroman 9910 The Tax Treatment of Funded Pensions by Edward Whitehouse 9909 Russia's Social Protection Malaise: Key Reform Priorities as a Response to the Present Crisis edited by Michal Rutkowski 9908 Causalities Between Social Capital and Social Funds by Jesper Kammersgaard 9907 Collecting and Transferring Pension Contributions by Rafael Rofman and Gustavo Demarco 9906 Optimal Unemployment Insurance: A Guide to the Literature by Edi Kami  Social Protection Discussion Paper Series Titles continued No. Title 9905 The Effects of Legislative Change on Female Labour Supply: Marriage and Divorce, Child and Spousal Support, Property Division and Pension Splitting by Antony Dnes 9904 Social Protection as Social Risk Management: Conceptual Underpinnings for the Social Protection Sector Strategy Paper by Robert Holzmann and Steen Jorgensen 9903 A Bundle of Joy or an Expensive Luxury: A Comparative Analysis of the Economic Environment for Family Formation in Western Europe by Pierella Paci 9902 World Bank Lending for Labor Markets: 1991 to 1998 by Amit Dar and Zafiris Tzannatos 9901 Active Labor Market Programs: A Review of the Evidence from Evaluations by Amit Dar and Zafiris Tzannatos 9818 Child Labor and School Enrollment in Thailand in the 1990s By Zafiris Tzannatos 9817 Supervising Mandatory Funded Pension Systems: Issues and Challenges by Gustavo Demarco and Rafael Rofman 9816 Getting an Earful: A Review of Beneficiary Assessments of Social Funds by Daniel Owen and Julie Van Domelen 984-5 This paper has been revised, see Discussion Paper No. 9923 9814 Family Allowances by Suzanne Roddis and Zafiris Tzannatos 9813 Unemployment Benefits by Zafiris Tzannatos and Suzanne Roddis 9812 The Role of Choice in the Transition to a Funded Pension System by Robert Palacios and Edward Whitehouse 9811 An Alternative Technical Education System: A Case Study of Mexico by Kye Woo Lee  Social Protection Discussion Paper Series Titles continued No. Title 9810 Pension Reforn in Britain by Edward Whitehouse 9809 Financing the Transition to Multipillar by Robert Holzmann 9808 Women and Labor Market Changes in the Global Economy: Growth Helps, Inequalities Hurt and Public Policy Matters by Zafiris Tzannatos 9807 The World Bank Approach to Pension Reform by Robert Holzmann 9806 Government Guarantees on Pension Fund Returns by George Pennacchi 9805 The Hungarian Pension System in Transition by Robert Palacios and Roberto Rocha 9804 Risks in Pensions and Annuities: Efficient Designs by Salvador Valdes-Prieto 9803 Building an Environment for Pension Reform in Developing Countries by Olivia S. Mitchell 9802 Export Processing Zones: A Review in Need of Update by Takayoshi Kusago and Zafiris Tzannatos 9801 World Bank Lending for Labor Markets: 1991 to 1996 by Amit Dar and Zafiris Tzannatos  Summary Findings While the issue of child labor in developing countries has received increased attention in recent years, most of the empirical analysis has been based on one-time cross sectional samples. While this may give an idea of the incidence and determinants of child labor at one point in time, it is silent about the dynamics of child labor over time and sometimes may even be influencing policy choices against child labor adversely. This paper attempts to fill this void, analyzing the dynamics of child labor and schooling in Ghana, aiming at investigating the impact of broad economic reforms on child labor and schooling in the short, medium and long run. Starting from a premise that the simple - direct - relationship between poverty and child labor, which has often been seen as the feature of child labor, may not adequately capture the multi facetted nature of child labor, we find evidence of asymmetries in the child labor-poverty link, as well as quite complex dynamics in the evolution of child labor and schooling and their determinants over time. Most notably, child labor is found to be responsive to poverty in the short run, but not in the long run, while child schooling is unaffected by poverty in short run but responds in the medium- to long run. These results suggest that child labor acts as an economic buffer of the household in the short run, regardless of changes in the economic environment or perceptions of the latter following economic reforms, thus supporting - and refining - the poverty explanation of child labor. HUMAN DEVELOPMENT NETWORK About this series... Papers in this series are not formal publications of the World Bank. They present preliminary and unpolished results of analysis that are circulated to encourage discussion and comment; citation and the use of such a paper should take account of its provisional character. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations or to members of its Board of Executive Directors or the countries they represent. For free copies of this paper, please contact the Social Protection Advisory Service, The World Bank, 1818 H Street, N.W., Room G8-138, Washington, D.C. 20433-0001. Telephone: (202) 458-5267, Fax: (202) 614-0471, E-mail: socialprotection@worldbank.org or visit the Social Protection website at www.worldbank.org/sp.