You are on page 1of 23

This article was downloaded by: [University of Nottingham] On: 12 February 2014, At: 13:35 Publisher: Routledge Informa

Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

The Journal of Development Studies


Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/fjds20

The tradeoff between child labour and human capital formation: A Tanzanian case study
Hideo Akabayashi & George Psacharopoulos
a a b

Department of Economics , Keio University , Japan


b

Department of Economics , University of Athens , Greece Published online: 23 Nov 2007.

To cite this article: Hideo Akabayashi & George Psacharopoulos (1999) The tradeoff between child labour and human capital formation: A Tanzanian case study, The Journal of Development Studies, 35:5, 120-140, DOI: 10.1080/00220389908422594 To link to this article: http://dx.doi.org/10.1080/00220389908422594

PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the Content) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and

views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-andconditions

Downloaded by [University of Nottingham] at 13:35 12 February 2014

The Trade-off between Child Labour and Human Capital Formation: A Tanzanian Case Study
HIDEO AKABAYASHI and GEORGE PSACHAROPOULOS

Downloaded by [University of Nottingham] at 13:35 12 February 2014

We investigate the degree to which there is a trade-off between child labour and human capital formation using time-log data of children from a Tanzanian household survey. We find that a tradeoff between hours of work and study exists, and hours of work tend to be more affected by social conditions than hours of study. Hours of work are negatively correlated to reading and mathematical skills through the reduction of human capital investment activities, indicating a trade-off between child labour and human capital. The results point up the complexity of the issue and the need for detailed time allocation data. I. INTRODUCTION There has been an increasing interest on the status of child work1 in developing countries and its implications for children's human capital development and poverty alleviation. Despite this increased concern, research on child work has suffered from both conceptual and practical difficulties that are closely interrelated. One difficulty stems from the fact that child work is not a uniform phenomenon. Children in developing countries work in a variety of settings, not only as wage workers but also as domestic/informal workers. The implications of child work differ greatly depending on the context. We have seen news that in some industries child work may be exploited under the name of apprenticeship, an abuse which has been much publicised during the past few years.2 However, child work does not automatically lead to
Hideo Akabayashi and George Psacharopoulos, Department of Economics, Keio University, Japan and Department of Economics, the University of Athens, Greece, respectively. The first draft of this article was written when Psacharopoulos was Senior Advisor and Akabayashi was a consultant economist at the Human Development Department of the World Bank. The views expressed here are those of the authors and should not be attributed to the World Bank. The Journal of Development Studies, Vol.35, No.5, June 1999, pp.120-140 PUBLISHED BY FRANK CASS, LONDON

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA

121

Downloaded by [University of Nottingham] at 13:35 12 February 2014

exploitation. Moreover, some argue that work is often an important form of socialisation in developing countries, where survival is still a primary issue. For example, Rodgers and Standing have already noted in their classic survey that, 'we have to be careful not to make an automatic assumption that work by children impairs education and intellectual development ... work itself may be an important component of "education" especially in household-based production systems ...' [Rodgers and Standing, 1981:33]. A more recent survey by Grootaert and Kanbur also notes that, 'In an economic environment where survival depends on work in the informal sector, many parents conclude that taking children out of school and putting them to work is the most sensible solution for survival and the education method which offers the best prospects for the future' [Grootaert and Kanbur, 1995: 193]. Working and human capital investment are not mutually exclusive. Several studies show that many children who attend school also work on the farm or the street in developing countries [Rodgers and Standing, 1981; Patrinos and Psacharopoulos, 1995]. Child work can be harmful to the development of human capital - probably to the extent that it results in a reduction of other inputs of human capital, such as time to study. It is not clear, theoretically or empirically, to what extent child work actually leads to the reduction of human capital development, especially when the children are engaged in household production. For instance, more concerned parents may try to encourage their children to both work and study if these two activities are complementary in the acquisition of survival skills. Such possibilities have been mentioned in the literature,3 but the issue has not been explicitly investigated thus far. The reason for this oversight may have been partly practical. Most survey data in developing countries typically include only questions on the respondent's 'main' economic activity. A survey so designed forces an assumed trade-off between the main activity (for example, working) and other activities (for example, attending school). Our study is an attempt to understand the degree to which child work and human capital development are in a trade-off relationship in developing countries. To address this issue explicitly, we use the time-log data from the Human Resource Development Survey in Tanzania (HRDS) collected in 1993.4 The objective of this survey was to provide information regarding household use of social services, household investment in education and health, and incidence of public spending, among other things. Time-log data of children's activities during the school year were available only for limited districts in the Tanga region. Tanga is located 200 km north of Dar Es Salaam with a population of 1,284 thousand (5.7 per cent of the Mainland Tanzania total population) in 1988 [World Bank, 1996b]. Its population is

122

THE JOURNAL OF DEVELOPMENT STUDIES

Downloaded by [University of Nottingham] at 13:35 12 February 2014

growing at about three per cent a year and the average income level is ranked above the Tanzanian average. Tanga is mainly an agricultural area, although there are some government-owned manufacturing, such as steel and wood products, in the urban area. It also has the second largest port in the country. Because of these special circumstances the implications of this data set may not be interpreted to apply to the other regions of Tanzania. Despite such a constraint, our data set is still rich enough to investigate the issue we are concerned with. In Tanzania, school attendance is not exceptionally low compared to other developing countries. In 1993, the gross primary school enrolment rate was 71 per cent for males and 69 per cent for females, while the corresponding figure was 82 per cent and 67 per cent, respectively, among 49 low income countries [World Bank, 1996a]. Nevertheless, delayed enrolment in primary school is quite common in Tanzania.5 Figure 1 shows the age profile of primary school participation rate in 1993 for all Tanzania and for the Tanga region. Although formal primary education starts at age seven in Tanzania, the attendance rate at age eight is only 40 per cent. The attendance rate becomes about 85 per cent at the age of 13. School attendance is further delayed in the Tanga region by about one year, relative to the national average. Figure 2 shows hours of work per day by children aged 7-14 in the Tanga region, as taken from the sample.6 Most children work regardless of their school attendance. Yet, 327 out of 337 children have answered that
FIGURE 1 PRIMARY SCHOOL PARTICIPATION RATE, BY AGE AND REGION

Rate -All Tanzania -Tanga

10 11 12 13

14 15 16 17 18 19 20

Source: HRDS Tanzania 1993. The vertical axis shows the percentage of children who attend primary school for a given age.

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA


FIGURE 2 WORK HOURS BY SCHOOL ATTENDANCE AND GENDER, TANGA REGION

123

(a) In-school children Girls

Boys

c o s 2

Downloaded by [University of Nottingham] at 13:35 12 February 2014

LL

10

12

14

10

12

14

Hours of work (b) Out-of-school children Girls .4c g g

Hours of work

Boys

1
OH
0 2 4 6 10 12 14

10

12

14

Hours of work

Hours of work

Source: HRDS Tanzania 1993. The vertical axis measures the sample fraction.

their main activity is 'not working' to a separate question in the survey. This is why a 'main activity' question can be fatally misleading in studying the issue of 'child labour'. Figure 3 is a two-way plot of hours of work and hours of study at home for boys and girls who regularly attend primary school in Tanga. This figure shows that it is far from clear how much trade-off there is between work and study, an issue we pursue below. The rest of the article is organised as follows. The next section provides a very brief survey of the recent literature on child work and schooling in developing countries. It is not meant be a complete list of the expanding literature, and we cite only those studies which are closely related to the issue under investigation in this article. Section III describes the data-set and variables used, and our empirical methodology. We estimate the

124

THE JOURNAL OF DEVELOPMENT STUDIES FIGURE 3 HOURS OF WORK AND STUDY FOR CHILDREN WHO ATTEND SCHOOL

(a) Girls Hours of study

Downloaded by [University of Nottingham] at 13:35 12 February 2014

o _
2 4 6
I 10 i 12

Hours of work (b) Boys Hours of study

cP
I OO

_ 4 10

oo
12

Hours of work

Source: HRDS Tanzania 1993. Hours are average per day during a typical school week.

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA

125

determinants of school attendance, hours of work, and hours of study at home in reduced forms. We investigate to what extent a variety of social and household conditions affect the trade-off between hours of work and hours of study at home. Then we estimate the determinants of children's reading and mathematical skills as reported by the respondents (mostly heads of household), including the effects of working and human capital investment activities. Section IV presents the results of our empirical analysis, and section V gives the conclusion. Downloaded by [University of Nottingham] at 13:35 12 February 2014 II. PREVIOUS LITERATURE Our empirical model is based on a general household time allocation decision framework, initially proposed by Becker [1965]. Rosenzweig and Evenson [1977] were among the first to apply Becker's model to households' simultaneous decision of fertility, schooling, and child work in the context of developing countries. They used district level aggregate data from rural India and found that substitution effects tend to dominate income effects in the supply of child work when these two conflict, and that the wage of children has positive effects on their hours of work. Rodgers and Standing [1981] were the first to attempt to shed light systematically on child work problems in developing countries. Since then, there have been many studies on child work and schooling decisions. Levy [1985] applied Rosenzweig-Evenson's model to regional aggregate data from rural Egypt in 1974-77. His results mostly conform to those of Rosenzweig and Evenson. Patrinos and Psacharopoulos [1995] used 1990 household survey data from Paraguay and found significant effects of poverty and mother's education on children's working status. The studies cited here maintain the dichotomy of school and work, and some of them use aggregated data of those binary variables. In one study which explicitly addressed the determinants of time use among children in developing countries,7 Skoufias [1994] used household panel data from semi-arid India to estimate reduced form time-allocation equations (home, school, and leisure) of children, with selectivity corrections. Surprisingly he found, among other things, that the child's wage has a positive effect on time spent at school while female adult wages have a negative effect - results opposite to what Rosenzweig and Evenson found using village level aggregate data for school attendance. In Skoufias' analysis, wages have no significant impact on children's school attendance. Few studies have explicitly associated children's time use decisions with human capital development. Khandker, Lavy, and Filmer [1994] used the 1990-91 Morocco Living Standards Survey to evaluate the determinants of school attendance and grade achievement in a reduced form specification.

126

THE JOURNAL OF DEVELOPMENT STUDIES

Downloaded by [University of Nottingham] at 13:35 12 February 2014

Although they argued descriptively how school attendance can affect children's math and language test scores, they did not include test score equations in their regressions. Khandker [1996] used a similar method to analyse household data from Bangladesh collected in 1991-92 to estimate the determinants of school attendance, failure, and dropping out. Psacharopoulos [1997] studied the effect of children's work status on educational attainment, such as school failure and years of schooling, using 1990 data from Bolivia and 1992 data from Venezuela. He found that the fact that a child is working increases the probability of failing a grade and reduces his or her educational attainment by one to two years. Mason and Khandker [1996], using the same data set as ours, analysed the determinants of primary and secondary school attendance and enrolment delay. They used the HRDS time-log survey together with predicted district level wages from a separate survey in order to approximate the children's opportunity costs. In contrast to them, we focus on observations from a small number of districts in the Tanga region where a complete set of time-log data is available. The usage of time-log data in our study is distinct from theirs in that we explicitly treat children's time allocation to studying at home and working, as well as school attendance, as being endogenous. Furthermore, we relate child work to human capital measures in order to test for the existence of a possible trade-off.
III. DATA AND ESTIMATION METHOD

The Tanzania HRDS 1993 survey consists of the main survey and a supplementary time-log survey. The objectives of this survey were to provide information regarding household use of social services, household investment in education and health, distribution of public spending, and the households' evaluation of social services, among others. The main survey was conducted from June 1993 to January 1994, and covered 5,183 households in the Mainland Tanzania and the island of Zanzibar. Time-log data, however, are only available for children in limited clusters in the Tanga region of the Mainland Tanzania. There are 542 observations for primary and secondary school age children. The uniqueness of this data-set is that it includes information not only on the children's work and school attendance, but also on their studies at home. Interviewers asked household representatives (household heads most of the time) how many hours a child in the household spent working or studying at home on each day of a typical week. Figures 2 and 3, above, are constructed by calculating the average hours of study and work per day. School enrolment is defined, and asked categorically, by whether or not a

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA

127

child attended school during the year before the interview. We selected the children used in this sample in the following way. From the 542 cases with at least one valid observation of time-use, we selected children who were aged 7-14, unmarried, had lived with the head at least for four months of the year before the interview, and had not yet finished primary school.8 All the children were also biologically related to the household heads. We included children who had not attended school during the year before the interview in the time allocation analysis because some of those children might have been attending to school at the time of interview. Also the 'time for studying' question in HRDS includes not only school homework but also self-study and coaching outside of school. Due to the design of HRDS, we do not have measures of the children's final intellectual level or success in the labour market, which are supposed to be objective measures of human capital outcomes. Therefore, we use children's reading and mathematical skills as observed by household representatives as indicators of their human capital stock. One potential problem is that these answers can be plagued by the respondents' bias. For instance, if the respondent cannot read, he or she may overevaluate the child's reading ability. Instead of income data, HRDS includes yearly, monthly, and weekly household expenditure data, both in cash and in kind. Yearly expenditure refers to the past year, while monthly and weekly expenditures refers to a 'typical' month or week. Theoretically, expenditure is a good proxy for the permanent income level, if there is access to perfect capital markets. Provided that the respondents' recall error is small, answers to 'typical month/week' questions can be a better approximation of income than answers to 'past year' - past year expenditure can be subject to idiosyncratic shocks due to illness or natural disaster. In the following, we use an estimate of the sum of monthly and weekly expenditures with instruments such as the number of adults in the household, the value of house rent, father's age and literacy, and total acres of the household's land. We also use a number of variables which characterise households parents' primary education, number of children,9 use of electricity, per adult acres of farm land used by household, whether the household has bank accounts, log distance to the closest public school, distance to the source of drinking water,10 rural or urban community, number of schools per 1,000 population in the cluster, and the respondent's evaluation of the quality of the schools. Per adult acres of land is calculated by summing all the acres of land used during the past 12 months by household members, except the ones owned but rented out, and dividing it by the number of adult household members aged under 50. Since we do not have wage data for children working on the farm, per adult farm size is included and expected to capture

Downloaded by [University of Nottingham] at 13:35 12 February 2014

128

THE JOURNAL OF DEVELOPMENT STUDIES


TABLE 1 DEFINITION OF VARIABLES AND DESCRIPTIVE STATISTICS

Variables Dependent Primary School Hours Worked Hours of Study Reading Ability Math Ability Independent Sex Age
2

Definition

Mean

S.D.

Downloaded by [University of Nottingham] at 13:35 12 February 2014

Dummy: one if having attended primary school in the past year, zero if otherwise Average hours per day of work on the farm, looking after children, other household tasks or work for pay during a regular school week Average hours per day of reading, doing homework, getting extra coaching outside school hours during a regular school week Dummy: one if being able to read a newspaper Dummy: one if being able to do written calculations

0.608 3.348

0.489 2.921

0.891 0.613 0.619

1.061 0.488 0.486

Rural Father's Education Mother's Education Number of Children Log Expenditure Bank Land Electricity Distance to Water Log Distance to School School Quality Number of Schools

Dummy: one if male, zero if female 0.499 0.501 Age in years 10.724 2.274 Age squared 120.160 47.856 Dummy: One if living in rural areas, zero if living in urban areas 0.368 0,483 Dummy: One if father finished primary schooling 0.534 0.500 Dummy: One if mother finished primary schooling 0.472 0.500 Number of children under age 14 in the same household 4.825 2.946 Log total household expenditure per year 10.688 0.473 Dummy: One if the household has a bank/saving/ post office account 0.304 0.459 Farm size in acres per adult used in the past year by the household 2.116 3.614 Dummy: One if there is electricity in the household 0.570 0.496 Distance in miles to the source of drinking water. Zero is assigned if a household has water access at home 0.275 0.726 Log distance in miles to the closest public primary school -0.974 1.788 Dummy: Quality of schools in the region evaluated by respondents (mostly household heads) 0.810 0.393 Number of schools in a cluster per 1,000 population 0.438 0.111

the productivity of an additional child working on the household farm [Grootaert and Kanbur, 7995]. Table 1 shows descriptive statistics of the variables used in the analysis. We model a child's human capital development using the household production framework. First, a household decides simultaneously whether they should send a child to school regularly and how many hours he or she should work and study at home. The first choice is binary, while the second and the third are continuous choices censored at zero. Such a decision depends on several exogenous household and community characteristics

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA

129

TABLE 2 DETERMINANTS OF PRIMARY SCHOOL ATTENDANCE AND TIME ALLOCATION (a) Boys and girls pooled Variable School Attendance (Probit) -0.055 (-0.896) 0.765*** (5.024) -0.031*** (-4.353) -0.111 (-0.810) -0.061 (-0.788) 0.065 (0.925) 0.021 (1.450) -0.097 (-0.800) 0.242*** (2.932) -0.001 (-0.136) -0.055 (-0.608) -0.007 (-0.182) 0.025 (1.314) 0.014 (0.173) 0.793* (1.724) 0.267 -158.141 336 0.608 Hours of Work (Tobit) -0.756** (-2.430) -1.111 (-1.520) 0.044 (1.268) -0.668 (-0.966) 0.328 (0.868) -0.473 (-1.328) 0.142** (1.998) -2.464*** (-3.842) -0.187 (-0.487) 0.023 (0.589) -2.043*** (-4.509) 0.630*** (2.850) -O.210** (-2.123) 0.634 (1.543) -8.421*** (-3.589) 0.046 -767.132 335 3.348 Hours of Study (Tobit) 0.004 (0.023) 2.097*** (4.550) -0.078*** (-3.740) -0.577 (-1.633) -0.182 (-0.911) 0.577*** (3.067) -0.120*** (-3.209) 0.137 (0.427) 0.709*** (3.623) -0.023 (-1.156) 0.035 (0.145) -0.067 (-0.570) 0.101* (1.850) -0.098 (-0.446) 1.604 (1.300) 0.180 -384.902 335 0.891

Sex Age

Downloaded by [University of Nottingham] at 13:35 12 February 2014

Age2 Rural Father's Education Mother's Education Number of Children Log Expenditure Bank Accounts Land Electricity Distance to Water Log Distance to School School Quality Number of Schools Pseudo R-squared Log Likelihood N Mean Dependent Variable

Notes: * significance level at ten per cent; ** significance level at five per cent; *** significance level at one per cent; the results shown in the probit estimation are marginal effects around the mean probability; numbers in parenthesis are t-values.

130
(b) Girls Variable

THE JOURNAL OF DEVELOPMENT STUDIES TABLE 2 (cont.)

School Attendance (Probit) 0.828*** (3.545) -0.035*** (-3.139) -0.159 (-0.810) -0.167 (-1.405) 0.181* (1.694) 0.054** (2.116) 0.003 (0.018) 0.285** (2.238) -0.014 (-0.785) 0.032 (0.241) -0.037 (-0.677) 0.024 (0.827) -0.014 (-0.124) 2.019*** (2.973) 0.287 -77.562 168 0.604

Hours of Work (Tobit) -1.545 (-1.373) 0.067 (1.252) -0.921 (-0.921) 0.299 (0.534) -0.823 (-1.584) -0.053 (-0.453) -2.818*** (-2.705) -0.401 (-0.676) 0.136 (1.436) -1.949*** (-2.916) 0.755** (2.416) -0.090 (-0.620) 1.065* (1.826) -9.340*** (-2.819) 0.054 -394.796 168 3.745

Hours of Study (Tobit) 2.633*** (3.883) -0.103*** (-3.329) 0.002 (0.005) -0.207 (-0.752) 0.317 (1.289) -0.077 (-1.457) 0.543 (1.176) 0.496* (1.842) -0.139** (-2.111) 0.418 (1.303) -0.279* (-1.771) 0.048 (0.683) -0.336 (-1.144) 3.612** (2.155) 0.223 -180.750 168 0.848

Age Age2

Downloaded by [University of Nottingham] at 13:35 12 February 2014

Rural Father's Education Mother's Education Number of Children Log Expenditure Bank Accounts Land Electricity Distance to Water Log Distance to School School Quality Number of Schools Pseudo R-squared Log Likelihood N Mean Dependent Varaiable

Notes: * significance level at ten per cent; ** significance level at five per cent; *** significance level at one per cent; the results shown in the probit estimation are marginal effects around the mean probability; numbers in parenthesis are t-values.

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA


TABLE 2 (cont.) (c) Boys Variable School Attendance (Probit) 0.971*** (3.789) -0.039*** (-3.336) -0.142 (-0.664) 0.047 (0.394) 0.016 (0.146) 0.002 (0.120) -0.162 (-0.965) 0.246** (2.059) 0.005 (0.387) -0.136 (-0.977) 0.029 (0.398) 0.031 (1.048) 0.079 (0.619) 0.271 (0.376) 0.352 -69.485 168 0.613 Hours of Work (Tobit) -0.201 (-0.208) -0.002 (-0.046) -0.069 (-0.075) 0.316 (0.644) -0.385 (-0.814) 0.280*** (3.310) -2.186*** (-2.808) 0.194 (0.405) -0.010 (-0.240) -1.882*** (-3.140) 0.687** (2.239) -0.309** (-2.391) 0.099 (0.174) -9.135*** (-2.832) 0.057 -361.384 167 2.946

131

Hours of Study (Tobit) 1.767*** (2.750) -0.063** (-2.176) -1.238** (-2.246) -0.478 (-1.612) 1.095*** (3.718) -0.156*** (-3.091) -0.164 (-0.369) 1.005*** (3.745) -0.018 (-0.823) -0.314 (-0.897) 0.078 (0.439) 0.216** (2.526) 0.024 (0.067) 0.848 (0.466) 0.200 -188.901 167 0.937

Age Age^

Downloaded by [University of Nottingham] at 13:35 12 February 2014

Rural Father's Education Mother's Education Number of Children Log Expenditure Bank Accounts Land Electricity Distance to Water Log Distance to School School Quality Number of Schools Pseudo R-squared Log Likelihood N Mean Dependent Variable

Notes: * significance level at ten per cent; ** significance level at five per cent; *** significance level at one per cent; the results shown in the probit estimation are marginal effects around the mean probability; numbers in parenthesis are t-values.

132

THE JOURNAL OF DEVELOPMENT STUDIES

Downloaded by [University of Nottingham] at 13:35 12 February 2014

that reflect household members' preferences and the resource constraints they are facing. These choices are supposed to be derived from the maximisation of a single utility function. However, it is still possible that those activities are not in a pure trade-off relationship because a child spends time not only working or studying but also playing, sleeping, and so on, with distinct determinants. Then, depending on the child's activities and other direct influences from the household and the community, the child's human capital develops. In this article, we focus on the effects of hours of work, hours of study, and school attendance on a child's reading and mathematical skills. We first estimate the determinants of children's school attendance and time allocation separately. We use a standard probit or Tobit model to get consistent estimates of the parameters for family and community characteristics." Then we estimate the effect of the children's time allocation and school attendance on their reading and mathematical skills. Since those activities are endogenous, including measured hours directly in the regression of skills may generate bias in the estimates.12 Therefore, we use the predicted hours of work and study and the predicted probability of school attendance in the first stage as instruments to consistently estimate the effects of those activities on the development of cognitive skills [Lee, 1981]. IV. FINDINGS AND DISCUSSION Table 2 shows the results of a probit estimation of primary school attendance and a Tobit estimation of hours of work and study. The results shown in the probit estimation are marginal effects around the mean probability. Table 2(a) shows the results with a dummy for sex included in the explanatory variables, and Tables 2(b) and 2(c) show the results for boys and girls estimated separately. We performed the likelihood ratio test on the equality of separate sex models as a hypothesis. We get the test statistics for the school attendance probit (22.188), hours of work Tobit (21.904), and hours of study Tobit (30.502) from the differences of the log likelihood. With 14 degrees of freedom, the equality hypothesis is rejected for hours of study, but not rejected for school attendance and hours worked. Regarding primary school attendance, age is clearly the most important determinant, which is consistent with Figure 1. The mother's education has a positive impact on the school attendance of girls. Availability of school (number of schools) has a positive effect on school attendance of girls. Community characteristics other than this variable do not seem to be important in the school attendance decision. Interestingly, neither the subjective evaluation of school quality nor household expenditure has an

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA

133

Downloaded by [University of Nottingham] at 13:35 12 February 2014

impact on the school attendance decision. The number of children seems to have a positive impact on girls' school attendance. It may be because more siblings help a female child avoid household chores and find time to attend school. A significantly positive effect of bank accounts suggests the existence of credit constraints to the households' decision in sending children to school. Hours of work are determined in a slightly different way. Girls tend to work longer hours than boys, by about 0.76 hours per day. Children in wealthier households (measured by expenditure) tend to work less. More educated mothers tend not to let their children work, especially girls, although the respective coefficients are not statistically significant. Community variables play an important role in children's hours of work. Children tend to work longer if there are fewer schools available in the community, if there is no electricity available, if the source of drinking water is far from home, or if the distance to the school is shorter. Among household characteristics, size of farm land per adult does not have a significant impact on working hours, contrary to our expectation. The effect of having a bank account is not significant, although the signs of pointestimates are all negative. Age tends to have a negative effect on hours of work, probably because children start going to school as they age. A boy with many siblings tends to work longer. This is against our intuition that with more siblings the working load per child should be smaller. This result is probably biased due to the endogeneity of the number of siblings: a larger demand for household labour would increase both fertility and hours of work per child, generating a positive correlation between the two measures. Quality of school tends to increase boys' working hours, which is surprising. Determinants of hours of study tend to be similar to those of school attendance. Among others, age, mother's education, and bank accounts seem to be important determinants of hours of study at home. The positive effect of age is larger than the negative effect of age on hours worked. One notable result is that, while the number of children has a significant negative effect on hours of study at home only for boys, it has a positive effect on school attendance for girls. This shows there might be a difference in the role of siblings which determines school attendance and time allocation at home. To what extent is there a trade-off between working and studying at home? For almost all exogenous variables, the signs of the marginal effects of variables on working hours are opposite to those to hours of study. In that sense, there is clearly an overall trade-off between working and studying at home; when a change in an exogenous factor increases a child's hours of work, almost always hours of study decrease. In fact, no variable is found

134

THE JOURNAL OF DEVELOPMENT STUDIES

to increase both working and studying significantly at the same time. However, if we look at the degree and significance level of those effects, some factors do influence both hours of study and work to the same extent, but others do not. Increasing the number of schools in the community tends to decrease hours of work and increase hours of study. A shorter distance to school increases hours of work and decreases studying for boys. Many community variables that have a significant impact on hours of work, such as distance to water and electricity, do not have a significant impact on hours of study. Hours of study tend to be influenced by household characteristics such as bank accounts, mother's education, and land per adult (for boys). Table 3 reports the results of a probit estimation of reading and mathematical skills for (a) both boys and girls pooled with a dummy for sex, (b) girls only, and (c) boys only. For each sample, we first included the predicted hours of work in the set of explanatory variables (column 1). Then we added the predicted school attendance probability and the predicted hours of study (column 2). The results shown are marginal effects around the mean. Living in rural areas seems to be a disadvantage for the development of mathematical skills for girls. The mother's level of education has a positive influence on boys' mathematical skills. On the other hand, whether or not a father has finished primary school has a significant 'negative' impact on reading and mathematical skills. This may suggest that skills evaluated by the household head (mostly fathers) are biased based on their education and skill level. Household expenditure tends to have a positive impact most significantly for boys' reading ability. From column (1) of each regression, hours of work are found to be always negatively related to the development of these skills, especially for girls' reading and boys' mathematical skills. Column (2) of each regression shows that the adverse effect of children's work on the development of skills is not very much affected by the addition of the two 'study' variables, although there are some exceptions. The addition of the 'study' variables substantially lowers both the slope and the significance level of the effect that hours of work has on skill development for girls but not for boys. The effect of studying at home on skill development seems more robust for girls than for boys. Moreover, from Table 3(a), the negative impact that hours of work has seems more robust for mathematics than for reading. These results imply that the direct adverse effects of child work on skill development more severely influence boys and mathematics skills whereas the indirect effects, through the reduction of hours of study and school attendance, more severely influence girls and reading skills. The results show that there tends to be a trade-off between a child's

Downloaded by [University of Nottingham] at 13:35 12 February 2014

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA


TABLE 3 PROBIT MODEL FOR READING AND MATHEMATICAL SKILLS (a) Boys and girls pooled Variable (1) Sex -0.026 (-0.379) 0.138 (0.835) 0.001 (0.107) -0.104 (-1.031) -0.190** (_2.443) 0.001 (0.012) 0.018 (1.213) 0.163 (1.374) -0.039 (-1.412) 0.336 -148.768 335 0.613 Readin:g (2) 0.019 (0.263) -0.222 (-0.703) 0.014 (1.026) -0.090 (-0.845) -0.193** (-2.423) -0.066 (-0.770) 0.014 (0.763) 0.204 (1.638) -0.021 (-0.711) 0.587 (1.385) 0.111 (0.974) 0.346 -146.426 335 0.611 (1) -0.030 (-0.443) 0.241 (1.491) -0.004 (-0.540) -0.080 (-0.801) -0.130* (-1.689) 0.047 (0.603) 0.023 (1.576) -0.009 (-0.078) -0.057** (-2.040) 0.339 -145.159 335 0.619 Mathematics (2)

135

0.011 (0.155) -0.074 (-0.242) 0.007 (0.556) -0.066 (-0.601) -0.133* (-1.697) -0.012 (-0.137) 0.020 (1.070) 0.032 (0.253) -0.040 (-1.368) 0.519 (1.261) 0.100 (0.875) 0.347 -143.269 335 0.637

Downloaded by [University of Nottingham] at 13:35 12 February 2014

Age Age^ Rural Father's Education Mother's Education Number of Children Log Expenditure Predicted Hours of Work Predicted School Attendance Prob. Predicted Hours of Study Pseudo R-squared Log Likelihood N Mean Dependent Variable

Notes: * significance level at ten per cent; ** significance level at five per cent; *** significance level at one per cent; the results shown in the probit estimation are marginal effects around the mean probability; numbers in parenthesis are t-values.

136

THE JOURNAL OF DEVELOPMENT STUDIES


TABLE 3 (cont.)

(b) Girls Variable (1) Age Age2 -0.031 (-0.129) 0.008 (0.714) -0.269* (-1.770) -0.212* (-1.863) -0.087 (-0.790) -0.009 (-0.353) 0.019 (0.102) -0.084** (-2.179) 0.335 -74.895 167 0.607 Reading (2) -0.375 (-1.116) 0.0195 (1.294) -0.282* (-1.820) -0.209 (-1.632) -0.154 (-1.253) -0.004 (-0.142) 0.124 (0.609) 0.003 (0.050) 0.665 (1.425) 0.290* (1.740) 0.362 -71.863 167 0.597 (1) 0.366 (1.578) -0.011 (-1.015) -0.258* (-1.741) -0.168 (-1.531) -0.072 (-0.674) -0.006 (-0.237) -0.185 (-0.988) -0.058 (-1.643) Mathematics (2) 0.140 (0.434) -0.005 (-0.332) -0.261* (-1.734) -0.185 (-1.501) -0.113 (-0.960) 0.007 (0.255) -0.122 (-0.602) 0.017 (0.337) 0.437 (0.965) 0.310* (1.955) 0.326 -75.531 168 0.613

Downloaded by [University of Nottingham] at 13:35 12 February 2014

Rural Father's Education Mother's Education Number of Children Log Expenditure Predicted Hours of Work Predicted School Attendance Prob. Predicted Hours of Study Pseudo R-squared Log Likelihood N Mean Dependent Variable

0.301 -78.342 168 0.604

Notes: * significance level at ten per cent; ** significance level at five per cent; *** significance level at one per cent; the results shown in the probit estimation are marginal effects around the mean probability; numbers in parenthesis are t-values.

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA TABLE 3 (cont.)


(c) Boys Variable (1) Age Age2 0.370 (1.375) -0.009 (-0.705) 0.089 (0.597) -0.175 (-1.487) 0.078 (0.656) 0.026 (1.283) 0.339** (2.005) -0.023 (-0.553) 0.377 -69.293 168 0.619 Reading (2) -0.007 (-0.014) 0.004 (0.191) 0.172 (1.056) -0.194 (-1.579) -0.034 (-0.249) 0.044* (1.874) 0.344* (1.944) -0.032 (-0.735) 0.531 (0.889) 0.166 (1.238) 0.397 -67.118 168 0.623 (1) 0.250 (0.884) -0.002 (-0.168) 0.192 (1.222) -0.094 (-0.762) 0.222* (1.740) 0.040* (1.878) 0.217 (1.198) -0.072* (-1.664) 0.442 -59.927 167 0.635 Mathematics (2)

137

-0.278 (-0.559) 0.018 (0.828) 0.232 (1.341) -0.120 (-0.942) 0.163 (1.134) 0.048** (2.007) 0.264 (1.415) -0.084* (-1.892) 0.763 (1.270) 0.041 (0.301) 0.453 -58.682 167 0.658

Downloaded by [University of Nottingham] at 13:35 12 February 2014

Rural Father's Education Mother's Education Number of Children Log Expenditure Predicted Hours of Work Predicted School Attendance Prob. Predicted Hours of Study Pseudo R-squared Log Likelihood N Mean Dependent Variable

Notes: * significance level at ten per cent; ** significance level at five per cent; *** significance level at one per cent; the results shown in the probit estimation are marginal effects around the mean probability; numbers in parenthesis are t-values.

138

THE JOURNAL OF DEVELOPMENT STUDIES

development of basic skills and long hours of work, directly or indirectly. For example, Table 3 predicts that the reduction of a girl's work by one hour per day would increase the probability of her being able to read a newspaper by eight percentage points, due to the increased hours of study and the probability of attending school. According to Table l(b), the reduction of roughly two hours of work could be achieved, for example, by supplying electricity to the household. Thus, roughly speaking, the introduction of electricity may improve the probability of a girl's being able to read by 16 percentage points. Downloaded by [University of Nottingham] at 13:35 12 February 2014
V. SUMMARY

We investigated the degree to which child work and human capital development are in a trade-off relationship using rich time-log data from a survey conducted in the Tanga region of Tanzania in 1993. Our empirical results show that factors that tend to increase children's working hours generally tend to decrease their hours of study, indicating a trade-off between these two activities. However, we also show that household and community conditions affect a child's work and study decision to different degrees. Particularly, we show that hours of work tend to be affected by community variables, such as electricity supply and distance to water, while hours of study tend to be less affected by these factors. Hours of work are found to be negatively correlated to the development of reading and mathematical skills. This shows that there tends to exist a trade-off between a child's development of basic skills and long hours of work. We also show that increasing hours of work tend to affect adversely the development of children's reading and mathematical skills, not only directly but also indirectly through the reduction of such human capital investment activities as studying at home. By way of summary, given the complexity of the relationship between child work and human capital development in developing countries, this study shows that our knowledge can be much improved by looking at the detailed time allocation of children.
final version received October 1998

NOTES 1. We use 'child work' rather than 'child labour' throughout this article since 'labour' has a more narrow meaning than 'work'. 2. See Silvers [1996], for example. 3. See Rodgers and Standing [1981: 34] for example.

CHILD LABOUR AND HUMAN CAPITAL FORMATION: TANZANIA

139

Downloaded by [University of Nottingham] at 13:35 12 February 2014

4. See Ferreira and Griffin [7996] for a full documentation of this survey. 5. The determinants of delay of enrolment in Tanzania are studied in Mason and Khandker [1996]. 6. See Table 1 for the definitions of 'Hours of Work' and 'Hours of Study'. 7. For developed countries, Timmer, Eccles, and O'Brien [1985] investigated this issue using US data. 8. Selecting children in this way does not cause a problem for our purpose since, first, children who can finish primary school within this age range are rare in Tanzania, and second, we focus on children who are supposed to be in school. 9. Fertility is treated as exogenous here, unlike some other studies such as Rosenzweig and Evenson [1977]. 10. Log value is not used for distance to the source of drinking water because many observations have value zero. Log value is used, however, for school distance since there is no zero value and its distribution is skewed. 11. We also fitted the self-selection model for hours worked, but did not find much difference from the Tobit results. 12. For example, a child might have had a chronic illness that prevented him/her from both school attendance and sound intellectual development.

REFERENCES Becker, Gary S, 1965, 'A Theory of Allocation of Time', Economic Journal, Vol.75, No.299, pp.493-517. Ferreira, M. Luisa and Charles C. Griffin, 1996, Tanzania Human Resource Development Survey: Final Report, Population and Human Resources, Eastern Africa Department, Washington, DC: World Bank. Grootaert, Christiaan and Ravi Kanbur, 1995, 'Child Labour: An Economic Perspective', International Labour Review, Vol.134, No.2, pp. 187-203. Khandker, Shahidur R., 1996, Education Achievements and School Efficiency in Rural Bangladesh, World Bank Discussion Papers No.319, Washington, DC: World Bank. Khandker, Shahidur R., Victor Lavy, and Deon Filmer, 1994, Schooling and Cognitive Achievements of Children in Morocco: Can the Government Improve Outcomes?, World Bank Discussion Papers No.264, Washington, DC: World Bank. Lee, Lung-Fei, 1981, 'Simultaneous Equation Models with Discrete and Censored Dependent Variables', in Charles F. Manski and Daniel McFadden (eds.), Structural Analysis of Discrete Data with Econometric Applications, Cambridge, MA: MIT Press. Levy, Victor, 1985, 'Cropping Pattern, Mechanization, Child Labor, and Fertility Behavior in a Farming Economy: Rural Egypt', Economic Development and Cultural Change, Vol.33, No.4, pp.777-91. Mason, Andrew D. and Shahidur R. Khandker, 1996, 'Household Schooling Decisions in Tanzania', manuscript, Poverty and Social Policy Department, Human Capital Development, Washington, DC: World Bank. Patrinos, Harry Anthony and George Psacharopoulos, 1995, 'Educational Performance and Child Labor in Paraguay', International Journal of Educational Development, Vol.15, No.l, pp.47-60. Psacharopoulos, George, 1997, 'Child Labor versus Educational Attainment: Some Evidence from Latin America', Journal of Population Economics, Vol.10, No.4, pp.377-86. Rodgers, Gerry and Guy Standing, 1981, 'The Economic Roles of Children: Issues for Analysis', in Gerry Rodgers and Guy Standing (eds.), Child Work, Poverty and Underdevelopment, Geneva: International Labour Office. Rosenzweig, Mark R. and Robert Evenson, 1977, 'Fertility, Schooling, and the Economic Contribution of Children in Rural India: An Econometric Analysis', Econometrica, Vol.45, No.5, pp. 1065-79. Silvers, Jonathan. 1996, 'Child Labor in Pakistan', The Atlantic Monthly, Feb., pp.79-92.

140

THE JOURNAL OF DEVELOPMENT STUDIES

Skoufias, Emmanuel, 1994, 'Market Wages, Family Composition and the Time Allocation of Children in Agricultural Households', Journal of Development Studies, Vol.30, No.2, pp.335-60. Timmer, Susan G., Eccles, Jacquelynne and Kerth O'Brien, 1985, 'How Children Use Time', in F. Thomas Juster and Frank P. Stafford (eds.), Tune, Goods, and Well-Being, Institute for Social Research, University of Michigan, Ann Arbor. World Bank, 1996a, World Development Report 1996, Oxford: Oxford University Press. World Bank, 1996b, Tanzania: The Challenge of Reforms: Growth, Incomes and Welfare, Report NO.14982-TA, Washington, DC: World Bank.

Downloaded by [University of Nottingham] at 13:35 12 February 2014

You might also like