Child labor and schooling in Ghana

_A_WPs r rM POLICY RESEARCH WORKING PAPER 1844 Child Labor and Schooling Toimprove humri capio- Ghana and reduce the incidence ot in Ghana child labor in Ghana, tne country's school systems Sudbarsban Canagarajab should reduce families' Harsha Coulomb a schoofing costs, adapt to tu : !Harold Coulombe constraints on schooling in rural areas (whera most children mujst vvo-k at ieas, part-time), and provid&-- be!Per education lrmore reievari-i to the needs of the labor market). If these thin :q --e done, more families may decide that schooting is a viable option as opposed to "chifd labor- for tneir children The World Bank Human Development Technical Family Africa Region November 1997  I POLICY RESEARCH WORKING PAPER 1844 Summary findings Child labor is a widespread, growing problem in the school. Of all children between 7 and 14, about 90 developing world. About 250 million of the world's percent helped with household chores. children work, nearly half of them full-time. Child labor Boys and girls tend to do different types of work. Girls (regular participation in the labor force to earn a living do more household chores while boys work in the labor or supplement household income) prevents children force. from participating in school. The data do not convincingly show, as most literature One constraint on Ghana's economic growth has been claims, that poverty is the main cause of child labor. But inadequate human capital development. According to poverty is significantly correlated with the decision to 1992 data for Ghana, one girl in three and one boy in send children to school, and there is a significant four does not attend school. The figures are worse in negative relationship between going to school and rural areas. working. Increased demand for schooling is the most Canagarajah and Coulombe studied the dynamics of effective way to reduce child labor and ensure that how households decided whether to send children 7 Ghana's human capital is stabilized. through 14 to school or to work, using household survey The high cost of schooling and the poor quality and data for 1987-92. They do not address the issue of street irrelevance of education has also pushed many children kids, which does not imply that they are less important into work. than the others. And family characteristics play a big role in the child's Unlike child labor in Asia, most child labor in Africa, decision to work or go to school. The father's education especially Ghana, is unpaid work in family agricultural has a significant negative effect on child labor; the effect enterprises. Of the 28 percent of children engaged in is stronger on girls than on boys. So adult literacy could child labor, more than two-thirds were also going to indirectly reduce the amount of child labor. This paper - a product of the Human Development Technical Family, Africa Region - is a background paper for World Bank Economic and Sector Work on Ghana: Labor Markets and Poverty. Copies of this paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Betty Casely-Hayford, room J8-272, telephone 202-473-4672, fax 202-473-8065, Internet address bcasely-hayford@worldbank.org. November 1997. (38 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center  CHILD LABOR AND SCHOOLING IN GHANA Sudharshan Canagarajah Harold Coulombe This paper is one of a series of background papers undertaken as part of a World Bank Economic and Sector Work (ESW) on Ghana: Labor Markets and Poverty. We acknowledge funding from Dutch and Canadian Trust funds.   TABLE OF CONTENTS Pa Abstract 1 1. Introduction 2-4 2. Data 4-6 3. Child Labor and Schooling: Tabulation Results 6-12 4. The Econometric Model 12-14 5. Econometric Results 14-28 6. Conclusions and Recommendations 28-29 REFERENCES 30-32 Annex 1: Definition of Variables used in Probit 33 Annex 2: Descriptive Statistics of Variables used in Probit 34 Annex 3: Ancillary Estimations 35-37   1. Introduction Child labor is a widespread and growing phenomena in the developing world. ILO (I 996a) estimates put the prevalence of child labor as 250 million in the World, out of which 61 percent is in Asia, 32 percent in Africa and 7 percent in Latin America. The same source also indicates that 120 million children are full time workers and 80 percent of them are between 10-14 years of age. In terns of child labor force participation rates Africa ranks highest with 33 percent in East Africa, 24 percent in West Africa and 22 percent in middle Africa, followed by East Asia and South Asia with 20 and 14 percent respectively (see Figure 1 below). The above information indicates the intensity of child labor and the necessity to address it, in order to eliminate its adverse effects on human capital development and the future growth potential of developing countries. Figure 1: Child Labor Force Participation Rates in the developing countries Child Labor Force Participation Rates ., 35 30 * ~25 20 .2~~~~ 15 22. 10 a. 5 0 __ U) - ZO~~~~~~~~~~~( Regions 2  The literature distinguishes child labor and child work, where the latter is the more unhannful and probably healthy kind, and includes helping household in various chores and household activity. These activities may take place after school hours or during holidays more intensively and are probably inevitable in rural areas. ILO's Minimum Age convention authorizes the employment of children above 12 or 13 years in certain type of light work under certain conditions (ILO, 1995). On the other hand, Child Labor is defined as the participation of school- aged children on a regular basis in the labor force in order to earn a living for themselves or to supplement household income. Child labor, therefore, prevents school participation and also possibly exposes them to health hazards. Empirical studies reveal that children contribute as high as one third of household income at times and their income source can not be treated as insignificant by poor families (Patrinos and Psacharopoulos, 1995). One of the major constraints in Ghana's growth challenge has been the lack of human capital development. The enrollments rates have not been picking up fast and the future trend on human capital does not look optimistic. The non-school attendance rates in Ghana are very high with wide gender disparities. 1992 GLSS data indicate that one in every three girls and one in every 4 boys does not attend school. The rural non-schooling is higher, with 37 percent for girls and 28 percent for boys. Ghana 2000 in its strategy for accelerated growth in Ghana argued for massive investment in primary education as a way of building the necessary human capital for sustainable growth (World Bank, 1993). In this context, it is important to understand the dynamics of household decision making of whether to send children to school and/or work, to benefit from investments in education. If not, colossal public investments in education are not likely to get children into class rooms. It has been noted that inconsistency between minimum age for employment and schooling in most countries makes the implementation of these laws complicated (ILO, 1996). This seems to be the case for Ghana as well. Ghana's labor Decree (1967) prohibits employment of children under the age of 15, although the law permits undefined "light" work by children. The Education Act (1961) states that education is free and compulsory, although it does not define until what 3  age the child should be in school. This indicates the problems of addressing child labor through legislation alone. This paper tries to investigate the child labor phenomena in Ghana in conjunction with school participation trends. In addition to citing examples from literature, this paper uses three rounds of the Ghana Living Standards Survey and analyses the issue of child labor at the household level where it takes place. The study does not focus on child labor away from home, i.e. street kids and prostitution. Also the definition used for child labor force participation used in this paper excludes household chores such as fetching wood, fetching water, cooking, cleaning and child care and similar activities undertaken by a boy or girl child in the household. However, household chores are accounted for separately. This paper addresses an aspect of labor markets which has not been discussed in Ghana in any detail in the past literature. The next section describes the data sources, while the following section gives a description of tabulations on child labor and school participation trends in Ghana based on the data available. Section 4 presents the econometric model used in this paper to analyze the joint probability and trade-off of child labor and schooling in Ghana. Section 5 discusses the results of the econometric model and where relevant showing evidence of similar findings from other studies. The final section concludes with some policy lessons for eliminating child labor and ensuring higher participation in schooling which is essential for Ghana's growth challenge. 2. Data It has been noted that there is very limited information on child labor in the developing countries mainly because none of the employment and labor surveys capture child labor (Grootaert and Kanbur, 1995). However, in Ghana we do have information.' The two main sources of data in our analysis are Ghana Living Standards Survey (GLSS) 1987/88, 1988/89, and 1991/92 and See Canagarajah and Thomas (1997) and Coulombe and McKay (1995) for a detailed description of the GLSS data. 4  ILO Child Labor Survey (1996b). The latter was collected in a small sample of children who did not attend school in Accra and two rural areas. Since the sample is non-random and also not nationwide it is not wise to draw nationwide conclusions or policy recommendation. This data set does not enable us to analyze simultaneously the decision of schooling and child labor. The GLSS data sets which are mainly collected to understand poverty and welfare levels also contain information on all types of household behavior including child participation in the labor market. One interesting aspect of the GLSS data sets is that they have information on children's activity in the last seven days, especially whether they went to school, worked in the labor market or worked at home in household chores. Thus the infonnation enables us to divide the child activity into four group - work only, school only, work and school and none. The information is available for all individuals age 7 and above. The sample sizes and their categorizations are given in Table 1 below. As we can see the sample although covers on average more than 3000 households per round and more than 15000 individuals, it has only around 3000-5000 children per survey round. Since the questions are asked about schooling and work in the last seven days we use only those. who were not on school holidays in order to minimize selection bias in our child labor sample. This gives us a final child sample of 2876 for GLSS1, 3011 for GLSS2, and 3859 for GLSS3. In each of the periods more than 60 percent of the children in the sample come from rural areas. Table 1: Sample Sizes of GLSS GLSS 1 GLSS 2 GLSS 3 (1987/88) (1988/89) (1991/92) Number of households 3172 3434 4523 Number of individuals 15 227 15 369 20 403 Number aged 7-14 3357 3421 4717 Number not constraint by 2876 3011 3859 school holidays Numberinrural areas 1838 2056 2601 Source: GLSSI-3. As we have already noted the GLSS data captures majority of child labor age group and can be treated as a reliable basis for child labor analysis. The fact that the data set was selected to analyze household welfare does not bias the sample and makes the data sets more interesting. The wide set of information on household welfare also enables us to test the hypothesis whether poverty is the main determinant of child labor among other things. The schooling information on 5  children enables us to jointly exploit their linkage to understand their trade-off for a child. It is also worth noting since most children are not working due to personal convictions, their analysis in a household framework is necessary for meaningful policy analysis. Also the claim that education system is not responsive and relevant for labor markets necessitates us to analyze the two choices simultaneously, rather than separately as many studies in the past have treated. 3. Child Labor and Schooling: Tabulation Results It is estimated, based on GLSS 1992 survey, that around 28 percent of children between the ages 7-14 years were involved in child labor in Ghana. This nationally amounts to around 800,000 children in child labor. However, over the three rounds child labor rates changed from 30.5 in 1987 to 22.4 in 1988 and 28 in 1992, which corresponds to the trend in the agricultural income between 1987-92. In 1992 out of the total number of children who were working 66 percent were also going to school and 90 percent were involved in household chores. 20 percent of boys and 17 percent of girls were observed to do both - working and going to school. The main difference was in those who did nothing; 14 percent of boys and 22 percent of girls did nothing. Male labor force participation for 7-14 year age group is 33.4 compared to 27.6 for girls, although if domestic chores were to be included the participation rates will change to 88 for girls and 75 for boys. These trends are similar to what has been observed in other developing countries where data is available (ILO, 1996). On the other hand school participation rates have evolved over time with 58.6 percent in 1987, 68.0 in 1988 and 72.7 in 1992. The girls' school participation increased from 53 to 68 percent, while that of boys increased from 64 to 76 percent between 1987-92. Urban schooling participation rates for the 7-14 year age group has increased from 68 to 83 percent while rural rates increased from 53 to 67. All this indicate the positive trend in school participation rates, despite the existence of child labor. However, these figures do not give any comfort as more than one quarter of children in the school age population are not attending school. 6  In terms of total labor force participation (LFP), children constitute 12.1 percent of the labor force. Out of the total labor force in rural areas 14.4 percent and 4.4 percent in urban areas are child workers. However, in terms of total number of labor hours children in the 7-14 year age group contribute 5.3 percent with those above 65 years contributing 5.5 percent. Of the total male LFP 14 percent is from those below 15 years, while the corresponding figure for females is 10 percent. All these data indicate the magnitude of child labor in Ghana. If household chores were to be included, as noted in the literature (ILO, 1996), girls will easily outnumber the boys in LFP. A child begins to work as early as five years in rural Ghana, although the current data source only gives labor participation information for those above 7 years of age. The average age of child labor for a boy is almost twelve while for a girl it is around 11, indicating that girls start working early. Girls also work more hours than boys and this difference is more pronounced if We take hours spent on household chores. Table 2 presents a typical profile of a boy and girl child worker in Ghana. As can be seen from the table more than 90 percent of child labor is in rural areas. It is also clear that these children work as many hours as adults. More than 5 percent of total labor hours nationally is contributed by children, signifying the importance of child labor in the national economy. Table 2: Typical Profile of a Child Worker in.Ghana 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 Average hours in labor market per week 13.5 15.1 Average hours in household chores per week 13.3 17.1 Proportion of child workers in labor force 14 10 Proportion in total work hours 15.8 24.7 Child labor force participation rate 29.3 26.7 Participation in Trading (percentage) 1.3 6.4 Participation in Farming (percentage) 96.3 88.5 School Participation (percentage) 76.7 68.3 Contribution to total hours of participation nationally 5.4 5.3 7  Source: GLSS 3 One of the claims in the literature that child labor increases with high levels of welfare is not convincingly proved in our analysis. Tables 3 below shows that that there is no clear direction in this relationship whether we analyze by regional patterns of poverty or welfare quintiles of households. Poverty incidence and depth in rural Savannah and rural forest is highest but child labor in rural forest is not high. Incidence and depth of poverty in rural coastal areas are lower than rural forest, yet the incidence of child labor is not lower than rural forest. From all this, it is clear that school participation is highly correlated with household welfare, indicating that households are willing to send their children to school as long as they have enough resources to do so. Child labor probably exist as long as the threat of poverty lingers in the household, pushing households who are above the poverty line also to send children to work. In poor households2 7.3 percent and in non-poor households 8.6 percent of the children were working, while the corresponding figures for schooling were 54.2 and 56.7 percent. As we will see later from our econometric analysis household welfare is indeed weakly related to the incidence of child labor and strongly related to school participation trends. Table 3: Poverty and Child labor in Ghana 1992 HEADCOUNT POVERTY-GAP CHILD LABOR INDICATORS RATIO RATTO REGION Index Contribution Index Contribution Wor School Work & None k Only School Onl y Accra 23.0 6.0 5.6 5.7 3.1 86.3 0.8 9.8 Other Urban 27.7 22.0 7.1 22.0 3.5 75.7 6.0 14.7 Rural Coastal 28.6 12.9 6.8 11.2 12.1 48.6 24.2 15.1 Rural Forest 33.0 31.1 8.3 30.4 6.9 47.5 36,9 8.7 Rural Savannah 38.3 28.1 10.5 30.3 18.0 33.2 13.2 35.6 Source: GLSS3 2 Based on the poverty line and number of poor people as defined in Ghana Statistical Service (1995). 8  Since survey data was not available to analyze child labor and schooling simultaneously most past studies have assumed child labor and schooling as mutually exclusive categories. However, since GLSS data provides information on both we find that almost 19 percent of the children were both working and schooling. Although there is no doubt that this would have had an impact on their educational attainment, it clearly indicates that it is indeed possible in many cases Table 4: Joint Labour Force and School Participation Rate (last 7 days), by gender, age, ecological zones, expenditure quintiles, socio-economic group and religion - 1991/92 Work Only School Only Work & School None All Gender Male 9.2 56.6 20.1 14.1 100.0 Female 9.4 51.0 17.3 22.3 100.0 Age 7 4.7 56.2 7.4 31.7 100.0 8 6.7 59.1 11.4 22.8 100.0 9 6.1 57.5 17.0 19.4 100.0 10 8.8 55.2 20.2 15.9 100.0 11 8.2 56.3 23.6 11.9 100.0 12 11.5 51.3 22.6 14.6 100.0 13 14.0 46.1 28.3 11.6 100.0 14 16.4 47.1 24.3 12.2 100.0 EJcpenditure Quintile Lowest 13.1 46.4 15.5 24.9 100.0 Second 6.8 54.1 21.7 17.3 100.0 Third 10.5 53.8 18.6 17.1 100.0 Fourth - 8.7 55.2 19.2 17.0 100.0 Highest 5.7 64.6 19.1 10.6 100.0 Socio-Economic Group Public 2.8 71.1 13.5 12.5 100.0 Wage-priv-formal 1.3 75.5 13.9 9.3 100.0 Wage-priv-informal 15.2 52.5 18.2 14.1 100.0 Self-agro-export 9.3 45.2 36.3 9.3 100.0 Self-agro-crop 15.3 35.4 24.7 24.7 100.0 Self-bus 3.4 74.2 9.0 13.3 100.0 Non-working 2.2 68.9 0.0 28.9 100.0 Religion Muslim 12.4 49.7 12.8 25.1 100.0 Catholic 6.1 59.9 21.7 12.2 100.0 Protestant 4.8 62.1 25.6 7.5 100.0 Other Christian 5.5 66.0 19.5 9.1 100.0 Animist 16.6 32.6 16.2 34.6 100.0 All 9.3 53.9 18.8 18.1 100.0 Source: Authors' calculations from the GLSS 3. Table 5: Occupation Distribution, by region and gender, (1991/92) Urban Rural Male Female All 9  Farming 59.2 96.3 96.3 88.5 92.7 Trade 22.3 1.7 1.4 6.3 3.7 Processing 11.7 0.7 0.5 3.3 1.8 Other 6.8 1.3 1.8 1.8 1.8 All 100.0 100.0 100.0 100.0 100.0 #of obs. 103 954 567 490 1057 Source: GLSS 3. to have these activities coexisting. One interesting observation from Table 3 is that work and school combination is a predominantly a rural phenomenon and very marginal in Accra and other urban areas. Table 4 presents the child labor and schooling participation pattern of children in Ghana in 1992 by various disaggregations. It is useful to note that with increasing age child labor plays an increasing role in communities. In terms of welfare quintiles the child labor pattern is not conclusive, but schooling shows a steady increase with higher levels of welfare. When considering socio-economic group of parents it was noted that the children of private informal sector wage earners and food crop producing farmers had the highest incidence of child labor. Religion plays an important role in explaining child labor and schooling patterns of children. Children from Christian households are more enrolled in school followed by Muslims and animists, while the child labor pattern in relation to religion is the reverse of the schooling trend. Majority of children are unpaid family workers, involved in family farm and enterprises (see Table 5). It is worth noting that more than 90 percent of the children were involved in household level agricultural activities. This is more the norm than exception in all Sub-Saharan African (SSA) countries; in South Asian countries child labor is predominantly in the manufacturing sector. In SSA only 3 percent of the children are wage workers, the majority of whom are in urban areas and boys (Ashagrie, 1993; ILO, 1996). Wage differentials between children and adults have been discussed extensively in the literature. It is clear that on average, in Ghana, children earn one sixth of what adults earn. The minimum wage is 12,000 cedis, while only a meager 10 percent of child workers receive any where near 10  that amount. With such low wages it is no surprise why employers prefer children to do most of the work where possible. In family enterprises the ease and flexibility of household child labor makes it attractive to employ children in a variety of tasks. The table below shows the sectoral pattern of child labor and its predominance in farming activities, although there are more girls involved in trading and processing compared to boys. The Participatory Poverty Assessment (Nortan et al, 1995) found that parents did not want to send their children to school due to inferior quality of teaching and teacher absenteeism. It was also noted that some teachers wanted the children to work in their farms in return for classes for them. This practice has disgusted many parents with Ghana's schooling system and has pushed them into involving their children in their own farms instead of teachers' farms. The high opportunity cost of sending children to school has also been stated as a reason for not sending them to school by many rural households. Table 6: Schooling expenditure per student, public school only, by urban/rural, level and items, 1991/92 (in Cedis, based on students currently enrolled) Urban Rural Total Mean Median Mean Median Mean Median Primary 1-2 Fees 3091 1450 762 550 1465 650 Uniforms 2280 2000 1508 1500 1741 1500 Books 866 400 337 200 497 200 Total 11082 7650 4838 3350 6681 4050 Primary 3-6 Fees 2505 1500 957 800 1484 850 Uniforms 2581 2425 1870 1800 2111 2000 Books 1480 900 752 500 999 600 Total 13360 9450 5968 4350 8408 5390 Sources: Authors' calculations from the GLSS 3. Note: The horizontal totals refer to the above three items as well as expenditure on parent/teacher associations, transportation, food and other expenses in cash or in-kind. Regardless of the rhetoric that education is free, many parents have had to pay some amount for tuition and other direct costs in terms of uniform and books. This together with recent efforts of 11  cost-recovery schemes have pushed parents in pulling their children out of school and sending them to work. As the table below very clearly shows education is not free. This has pushed more and more parents to stop their children from school as they just can not afford it. For instance, in 1992 per capita costs for publicly provided primary education has been in the range of 7,300 cedis which accounts for more than 15 percent of households mean per capita expenditure, thus indicating the burden of school expenses on poor households. Also past studies (Demery et al, 1995) have found that the public subsidies benefit the urban non-poor more than the rural poor. This emphasizes the point that the poor do not benefit from government resources towards their human capital investments. The ever changing nature of labor markets and low returns to education have made education less attractive for many parents. This has especially been the case in rural areas, where formal education makes very little difference given limited formal sector opportunities and most skills are acquired by the "learning by doing" principle. Child labor is perceived as a process of socialization in many countries and it is believed that working rather than education enables a child to get acquainted with the skills required for being employable (Grootaert and Kanbur, 1995). 4. The Econometric Model Our model tries to understand the factors that influence the probability of child's school attendance and working behavior in a reduced form model, focusing on a mixture of demand and supply side variables. The particular choice of the estimation method has been influenced by the decision making process, and available data. We do not want to assume that schooling and work decisions of children are independent, which could be treated in a multi-nomial logit model. We also do not want to assume any sequential process in the decision making process as we believe it is not necessarily a sequential choice. Hence we treat schooling and working possibilities as two interdependent choices. 12  There are no studies yet which have used the dichotomic model for labor and education jointly due to unavailability of data. With a view to exploiting the rich information on joint participation in schooling and working of children in Ghana (GLSS), we use a bivariate probit model to test the likelihood of children working and going to school; given varied individual and household characteristics. Bivariate probit models allow for the existence of possible correlated disturbances between two probit equations. It also allows us to test whether this joint estimation makes significant difference as opposed to estimating univariate probits for each decision. In the Bivariate probit, let the latent variable y represent the decision of working and y2 represent the decision of schooling. Therefore the general specification for a two-equation model would be Y = , y, = I if y, > O, O otherwise Y; = l2 + E2, Y2 = 1 if y; > O, O otherwise, E[E] = E[s23 = 0, Var[E] = Var[S2] = 1, COV[S I,2] = P. and the likelihood function to maximise is 13;X, j3X2 L=n f f| 2 (Zi, z2; p)dz2dzI where 4 2' the Bivariate normal density function, is 02(zI,z2;p) =[27r(l -p2)12 ]-' exp[-1/2(I- p2)-'(z2 +z2 -2pz z2)] and, p - coefficient of correlation between the two equations. XI and X2 - row vectors of exogenous variables which determnine respectively, working and schooling propensities. f,r and 2 - associated parameter column vectors. The coefficients need to be adjusted to be marginal effects, unlike standard linear regression a(t)(P'x) models. In this probit model E[y] = 'D(P'x), then the marginal effects are a (X). 13  These marginal effects would obviously vary with the values of x. It is worth noting that all the coefficients P would have the same scale factor 4(P'x) applied. Except for dichotomous variables these marginal effects would be correct for infinitesimal changes in explanatory variables. In case of dichotomous variables it is better to estimate the equation with and without the variable of interest. For instance, the marginal effect for the dummy variable i,(5j), would be defined as 6, = (D(P-jx i + pi) - d(3P,JXj) where the subscript - i represent all the variables but the ih , and Y,_ are their sample means. 5. Econometric Results In the bivariate probit model there are two dependent variables. The first dependent variable is defined I if the child went to school in the last seven days, and 0 if otherwise. The second dependent variable is defined as 1 if the child is economically active in the labour market the last 7 days and 0 if otherwise. Annex Table 1 presents the definition and Annex Table 2 presents some descriptive statistics of the explanatory variables used in the analysis. The child's age and gender which revealed differences in child labour and schooling participation was included as child specific variables in the regressions. Since we also noted earlier that household characteristics are important, we included some parental and household wide characteristic variables. Parent specific variables are the education of the father and mother taken separately and variables accounting for their presence in the household. For household characteristics variables other than general household welfare, we included information on siblings, the household's main socio-economic category, religious background, and asset ownership. We also included regional dummies to take care of the demand patterns of labour markets, schooling distance and expenditure as supply variables. We felt these variables are bound to have an impact on the pattern and intensity of child labour and school attendance. We use two different estimations for each sub-sample estimation of the model. The second is similar to the first except that it includes school supply variables, to test the relevance of schooling supply in the household decision to send children to school or to work. Apart from estimating the 14  model at the national level, we also estimated regional, gender, and age-group sub-samples to test the robustness of estimates. 15  Table 7: Determinants of Labour Force Participation and School Participation, Ghana, Children Aged 7-14 Model I Model 2 Labour Force School Participation Labour Force School Participation Participation Participation Independent Marginal t-ratio Marginal t-ratio Marginal t-ratio Marginal I-ratio Variable Effect Effect Effect Effect Constant -8.3280 -2.944 -2.1243 -8.127 -8.0076 -2.837 -2.3329 -8.305 Agey 0.1551 4.430 0.1883 5.584 0.1539 4.372 0.1931 5.486 Agey2 -0.0052 -3.134 -0.0089 -5.518 -0.0051 -3.086 -0.0091 -5.440 Male 0.0146 0.986 0.1136 7.620 . 0.0131 0.886 0.1131 7.220 Relson -0.0469 -1.406 0.0006 0.018 -0.0463 -1.392 0.0106 0.316 Motherln 0.0396 1.598 -0.0111 -0.445 0.0395 1.588 -0.0179 -0.678 Fatherln -0.0877 -3.014 0.0915 3.367 -0.0877 -3.018 0.0964 3.438 Medl -0.0114 -0.449 0.0977 3.437 -0.0149 -0.587 0.0858 2.880 Med2 -0.0194 -0.846 0.1528 5.714 -0.0216 -0.943 0.1450 5.251 Med3 0.0002 0.003 0.0627 1.050 -0.0006 -0.009 0.0537 0.894 Fedl -0.0174 -0.622 0.1158 3.721 -0.0239 -0.842 0.1170 3.537 Fed2 -0.0228 -1.185 0.1160 5.598 -0.0291 -1.504 0.1065 4.962 Fed3 -0.1119 -3.303 0.1661 4.772 -0.1186 -3.488 0.1556 4.362 Lnpcwell 1.1839 2.516 0.0863 5.569 1.1066 2.354 0.0843 5.112 Lnpcwell2 -0.0494 -2.521 - - -0.0464 -2.373 - - ChildO6 -0.0009 -0.132 -0.0089 -1.301 -0.0009 -0.129 -0.0082 -1.156 Bro7I4 0.0080 0.772 0.0154 1.506 0.0079 0.752 0.0135 1.276 Sis714 -0.0129 -1.130 0.0315 2.609 -0.0137 -1.195 0.0361 2.842 MaleIS59 0.0195 2.553 -0.0151 -1.955 0.0198 2.594 -0.0141 -1.747 Feml559 0.0033 0.420 0.0070 0.848 0.0017 0.213 0.0046 0.529 91d60 0.0161 1.188 -0.0261 -1.955 0.0136 0.995 -0.0335 -2.390 Selfagro 0.1104 4.968 -0.0304 -1.319 0.1165 5.165 -0.0266 -1.078 Selfbus -0.0903 -3.328 0.0467 1.833 -0.0882 -3.229 0.0486 1.841 Mheadeco 0.0229 0.904 -0.0579 -2.503 0.0289 1.135 -0.0447 -1.885 Muslim 0.0162 0.684 0.0742 3.279 0.0072 0.301 0.0511 2.166 Catho 0.0003 0.010 0.1505 5.687 -0.0062 -0.254 0.1373 4.969 Protes 0.0430 1.749 0.1957 7.569 0.0348 1.409 0.1815 6.781 Ochris -0.0207 -0.937 0.1315 5.610 -0.0286 -1.283 0.1211 4.992 Landsize -0.00004 -0.351 0.00004 0.300 -0.00003 -0.231 0.00007 0.511 Animal 0.0016 0.653 -0.0116 -6.726 0.0021 0.847 -0.0111 -6.130 Accra -0.2646 -5.325 0.1374 3.223 -0.2984 -5.666 0.0798 1.646 Town -0.1668 -6.575 0.1084 4.267 -0.1829 -6.691 0.0750 2.543 Rcoastal 0.1098 4.239 0.0704 2.752 0.0891 3.237 0.0214 0.760 Rforest 0.1320 6.409 0.2026 8.832 0.1218 5.822 0.1829 7.439 Tschexp - - - - 0.0260 1.946 0.0345 2.490 Distance - -0.0007 -2.722 -0.0012 -5.277 Smiss2 - - - - 0.1416 1.184 -0.5117 -3.416 p -0.1252 -3.540 -0.1527 -4.165 InL -3513.7 -3432.0 InL(=0) -4500.5 -4500.5 Sample Size 3811 3811 16  The results we obtained from the econometric model are in line with past research done on child labour and schooling determinants as independent choices, except for a few differences in the intensities of these effects. In terms of gender we find that there is no significant difference between boys and girls in their likelihood to work. This is mainly because our definition of work did not include household chores where majority of the girls are active. In terrns of labour force participation this finding however may be at conflict with earlier findings, Psacharopoulos and Arrigada (1989) and Patrinos and Psacharopoulos (1995), indicated that males were more likely to be involved in the labour market. When the definition was expanded to include household chores it clearly showed that girls are more likely to participate in the more broadly defined labour market activities than boys. The gender discrimination in schooling comes out very clearly in the schooling equation where the male dummy had a higher probability of school attendance compared to girls. In literature it is often claimed that the main determinant of child labour is poverty (Grootaert and Kanbur, 1995). Hence we included a welfare index which was household per capita expenditure deflated by time and spatial price index. Although a negative relationship was expected in the labour force equation we found an inverted U shape relationship, which peaked at 152,000 cedis and is slightly below the median expenditure (figure 2). This relationship was observed to be very strong in rural areas. It may be due to the prevalence of slack season labour demand patterns in regions where the poor live, or the presence of constraints in terms of other inputs and availability of credit which distort this postulated relationship. However, the significant low effect of welfare on the probability of labour force participation has also been found by Levinson (1991). This casts doubts on the traditional, simplistic view that poverty pushes children into the labour market. However, in terms of school participation the effect of welfare of the household is rather strong and positive. The difference in school participation between the lowest and top deciles is around 12 percentage points. The above relationship is strong everywhere except in the presence of livestock in the household when the relationship between schooling and work is not different between poor and non-poor households, partly because livestock is a time-intensive activity. Our estimations also show that fathers with very high levels of education are likely to have a negative effect on the likelihood of working, while mothers' education seems to influence only schooling participation. The latter may be at odds with other empirical studies where the income 17  variables used might not have been as good as ours and thus the parent's education variable would have captured more of the permanent income effect. In general parents education has a strong positive effect on schooling participation than working. The presence of the father at home is likely to positively effect the likelihood of going to school as opposed to work, which is similar to Tienda (1979). As figure 2 below shows the probability of going to school and of working based on age coefficients shows that there is steep increase in labour force participation in rural areas. It is also possible that this result is due to inadequacy of schooling system. The probability of going to school increases with age until 11 years and then starts declining. This is consistent with the high prevalence of delayed school attendance in Ghana (Glewwe and Jacoby, 1993). It has been argued in the past that the age, presence and gender of siblings has a strong effect on schooling and working patterns of members of households (Chernichovsky, 1985). We included a series of variables to capture this effect - namely number of siblings in 0-6 and 7-14 age groups and their gender, number of female and male adults. The only significant variable turns out to be the presence of adult males in the household, whereby each additional male decreases the probability of working by approximately 2 percent. On the other hand in the schooling equation, there is a positive marginal effect on school participation, if there are other female siblings or elderly people in the household. This is because if other members are able to take care of household chores, then school aged children are liberated from household chores, which are likely to prevent them from going to school. The literature also indicates that in large households parents in general can not afford to send all children to school and hence some children attend school at the expense of others (Lloyd and Gage-Brandon, 1994). In terms of employment activity of parents we find that if parents are involved in agricultural self-employment then children in such households are 12 percent more likely to work than children from other type of households. On the other hand children from non-farming self- employment households are less likely to work. The headship of household was found to affect schooling more than labor force participation. It was observed that children from female headed 18  household are 4 percent more likely to go to school rather than male headed households, which is consistent with past research which indicates that female headed households were more rational in intra-household resource allocation pattern and investing on essential items (Haddad et al, 1996). Figure 2: The probability of going to school and working by age and welfare levels. ne pd tyfgigtDsdrd axdf Ni,Aty of goig to sood and f mIginhma, w n ina mm (yW lfram ekids) (!tras) ~~~ ~~ ~~~~~90 ------ 0 0... o