#Hild, Abor and 3chool !ttendance 4wo Oint $ecisions: Helena Skyt Nielsen

#HILD ,ABOR AND 3CHOOL !
TTENDANCE
4WO *OINT $ECISIONS
Helena Skyt Nielsen
Working Paper 98-15

October 1998
Published by
Centre for Labour Market and Social Research

Science Park Aarhus
Gustav Wieds Vej 10C, 8000 Aarhus C, Denmark
Editor: Henning Bunzel
Copyrights: Helena Skyt Nielsen
ISSN 0908-8962
#HILD ,ABOR AND 3CHOOL !TTENDANCE
4WO *OINT $ECISIONS
(ELENA 3KYT .IELSEN
Working Paper 98-15

October 1998
*%, CLASSIFICATIONS: I21, J13, J24
+EYWORDS Africa, Zambia, child labor, school attendance, community effects
!BSTRACT
In this paper, a variety of potentially explanatory indicators for child labor and school atten-
dance in Zambia is scrutinized. By analyzing the results from a bivariate probit model, new
doubt is raised with regard to the income sensitivity of the child labor choice. Different fac-
tors in the model influence the child labor and schooling choice. Based on the results, sensiti-
zation, narrowly targeted subsidies or stipends are suggested to at-risk households in selected
provinces, provided that children are sent to school.
!CKNOWLEDGEMENT
This project was financed by the World Bank and the Danish National Research Foundation.
Comments on earlier versions of this paper from Paul Glewwe, Sudharshan Canagarajah,
Martin Paldam and seminar participants in Aarhus are appreciated. Of course, any remaining
errors are my own. The World Bank does not accept responsibility for the views expressed
here, which are mine alone and should not be attributed to the World Bank or its affiliated or-
ganizations.
!DDRESSES
Department of Economics, University of Aarhus, and Centre for Labour Market and Social
Research, Science Park Aarhus, Gustav Wieds Vej 10C, DK-8000 Aarhus C, Denmark .
e-mail: HSNielsen@cls.dk
)NTRODUCTION
During the first decade following the independence in 1964, the economy of Zambia did well due to
the large copper reserves and the favorable trend in the world market copper prices. The revenue was
spent on constructing a state-owned industry. This made the country incapable of coping with the
crisis that followed the oil price crises and the end of the Vietnam War, where the price and the
demand for copper dropped. Zambia has been fighting through economic difficulties ever since.1 The
Government had to reduce public expenditures forcing social services to rely more on private
payments. Since the 1970s, the ministries of the educational sector accounted for a declining
proportion of the GDP. In 1975, 7% of the GDP was devoted to the educational sector, but by 1990
this number had fallen to 3%. In real terms the decline from 1975 to 1990 was even larger: public
expenditure on education per capita fell from US$28 to US$9. Primary school children were one
group suffering heavily from the cut backs. Public expenditures on primary education per capita
declined from US$11 per capita in 1975 to US$2 in 1990.2 Another characteristic of the public school
expenses during the decline was that salaries were seen as unavoidable expenditures. Thus, teaching
materials, equipment, repair, maintenance, improvement, and construction of additional facilities have
become more and more the responsibility of the parents and the local communities. Since 1985, the
expenditure to primary school teachers’ salaries have been relatively constant at 95% of the total
primary level spending, which means that they have risen by more than 10 percentage points since
1975.3
For poor parents it may prove to be too expensive to send children to school. As of 1993, the payment
to the school fund and the Parents’ and Teachers’ Association (PTA) was K147 (US$0.35) per term on
average for primary school.4 On top of this come uniforms, books, and possibly private tuition. On
average the total school expenses per student, including fees, private tuition, books, and uniforms
amounted to about K3,000 (US$7) per term in 1993. Even if uniforms may last long and can be
handed down to younger siblings, they account for a substantial amount of the household expenses
related to education. One uniform for primary school easily costs about US$30. However, in rural and
in poor urban areas, the requirement to wear a uniform is not enforced. More substantial school
expenses, which are not directly observable, are transportation costs and foregone earnings. When
1
For details of the decline of the Zambian economy, see Andersson and Kayizzi-Mugerwa (1993), EIU (1993), and Bratton (1994).
2
Public expenditure on education per capita computed from World Bank (1997), IMF (1996), Kelly(1992), MoE(1996b).
3
For details on education in the declining economy of Zambia see Kelly (1991, 1992) and MoE (1995, 1996, 1996a, 1996b). The
development is similar in most African and Latin-American countries, Graham-Brown (1991) reports that on average the proportion of
educational expenditure on teaching material fell from 7.6% to 4.2% in Africa and from 4.4% to 1.8% in Latin-America and the
Caribbean during the period 1975-85.
4
An exchange rate of K420=US$1 is used throughout the paper.
-1-
taken together, it is understandable if some parents find that schooling is out of the question because
of the high costs. The empirical analysis below sheds some more light on this issue.
In spite of the picture sketched above, Zambia differs from other countries in that relatively many
children attend school and relatively few work. If child labor is defined on the basis of the main
activity during a randomly selected week, 13% of the children aged 7-14 are involved in child labor,
whereas 75% attend school. These numbers are more favorable than for many other Sub-Saharan
African countries, see Canagarajah and Nielsen (1998). Even though the extent of child labor on
average is relatively moderate in Zambia, in certain regions it is considerable. This paper looks at the
factors that characterize child workers, and scrutinizes different strategies to reduce child labor.
This paper analyzes the decisions to attend school and to work. Because the school and work decisions
are closely related, they are treated as simultaneous decisions which can be analyzed in the context of
a bivariate probit model.
The paper is organized as follows: Section 2 presents the Zambian data that are used for the empirical
analysis. Section 3 describes the incidence of child labor and school attendance in terms of cross
tabulations. Section 4 consists of the results from estimation of a bivariate probit model, and section 5
contains a discussion of which policy implications can be drawn from the model. Concluding remarks
are given in section 6.
$ATA )SSUES
For the empirical analysis, Priority Survey II for Zambia is used. The data were collected by the
Zambian Central Statistical Office with financial support from the World Bank. It is a household
survey carried out in 1993; 10,000 households and approximately 55,000 individuals are included. The
data set is augmented with data from the Community Survey 1993 on education and infrastructure in
the 631 different communities in which the individuals reside.
When considering child labor, ideally children in the age group 5-14 years should be analyzed.
However, some of the information which is needed for the empirical analysis is only available for
children in the age group 7-14 years, therefore this age group only is analyzed. The upper age limit is
chosen because it defines the age where some individuals finish primary school. Because of a limited
number of secondary school places, it might be expected that schooling is less of a choice after the age
of 14, when the children may have finished primary school and are ready to enter secondary school.
Individuals with missing values for essential variables are excluded, which leaves us with 12,558
individuals. About one-half of the sample are girls and the other one-half are boys; 4,210 reside in
-2-
rural areas and 8,348 reside in urban areas.5 Means and standard deviations of the variables are shown
in Table 1.
The questionnaire asks about the activities of the individuals in the seven days before the survey.
‘Only school’ is an indicator variable given the value one when the child has not been working, and
has attended school, full-time or part-time. ‘Only work’ is equal to one for children who have been
involved in labor activities, and who did not attend school. ’Both work and school’ is equal to one if
the child is involved in both school and work activities. This includes attending school on a part-time
or a full-time basis, while working the main part of the last week. The survey does not reveal part-time
work, which is therefore not defined as child labor.
For the schooling decision the question asked is 'Are you currently attending school?' This defines
school attendance in the broadest sense possible. For the work decision, the question is: ‘During the
last seven days were you 1) working, 2) not working but looking for work, 3) not working and not
looking for work but available for work, 4) student, 5) housewife, 6) other‘. Thus according to this
definition, child workers are children that spent more time working than attending school during the
last week before the survey.
In other words, only child labor that actually interfered with schooling is defined as child labor.
Numerous children attend school and work part time, but they are not counted in this survey as child
workers. The children working only in the harvest seasons have not been caught in the Zambian data
set because the survey was collected from January to April 1993, and the harvest is in May. However,
if the months January to April are representative of the year as a whole, a good estimate on child labor
has been obtained.
5
The small number of observations from rural areas reflects stratification. The fact that the sample is stratified implies that weights are a
part of the data.
-3-
4ABLE . $ESCRIPTIVE STATISTICS
All Rural Urban Girls Boys
Variable Mean S.d. Mean S.d. Mean S.d. Mean S.d. Mean S.d.
)NDIVIDUAL CHARACTERISTICS
Female 0.50 0.42 0.49 0.49 0.51 0.47 1.00 0.00 0.00 0.00
Rural 0.59 0.42 1.00 0.00 0.00 0.00 0.58 0.42 0.60 0.42
Only school 0.70 0.39 0.60 0.48 0.86 0.33 0.70 0.39 0.71 0.39
Only work 0.08 0.23 0.14 0.33 0.01 0.08 0.09 0.24 0.08 0.23
Both work and school 0.05 0.18 0.07 0.25 0.01 0.10 0.04 0.17 0.05 0.18
Age 10.5 1.90 10.4 2.18 10.5 2.12 10.5 1.91 10.5 1.90
Non-biological relation to the household head 0.19 0.33 0.21 0.40 0.15 0.34 0.20 0.34 0.18 0.33
Age rank among household members aged 7-31 3.83 1.67 3.64 1.92 4.11 1.81 3.83 1.69 3.83 1.65
(OUSEHOLD CHARACTERISTICS
The household owns an asset (e.g. radio, bike) 0.69 0.39 0.62 0.47 0.79 0.38 0.69 0.39 0.69 0.39
The household owns land 0.05 0.18 0.02 0.14 0.09 0.26 0.05 0.19 0.05 0.18
The household is involved in agriculture 0.70 0.39 0.98 0.15 0.28 0.42 0.69 0.39 0.71 0.39
Education of the household head 5.81 3.54 4.29 3.58 8.03 3.62 5.86 3.56 5.76 3.53
Age of the household head 45.5 9.83 47.3 12.3 43.0 8.70 45.3 9.70 45.8 10.0
The household head works in the formal sector 0.27 0.37 0.06 0.23 0.57 0.46 0.27 0.38 0.26 0.37
The household head does not work 0.04 0.16 0.02 0.14 0.06 0.22 0.04 0.16 0.04 0.16
Expenditure per AE per month (K1,000) 5.50 5.40 2.82 2.57 9.41 7.48 5.68 5.67 5.31 5.11
Expenditure per AE K0-1970 0.29 0.38 0.46 0.49 0.03 0.17 0.28 0.38 0.29 0.39
#OMMUNITY CHARACTERISTICS )NFRASTRUCTURE
Fertilizer is sold all year 0.17 0.32 0.20 0.39 0.14 0.33 0.17 0.32 0.18 0.33
Credit is available 0.65 0.41 0.59 0.48 0.72 0.42 0.64 0.41 0.65 0.40
Passable road 0.85 0.30 0.79 0.40 0.93 0.23 0.85 0.31 0.85 0.30
Public transport 0.51 0.42 0.37 0.47 0.72 0.42 0.51 0.42 0.51 0.42
#OMMUNITY CHARACTERISTICS %DUCATION
Primary school in community 0.64 0.41 0.73 0.43 0.50 0.47 0.64 0.41 0.64 0.41
Secondary school in community 0.15 0.31 0.09 0.29 0.24 0.40 0.15 0.31 0.16 0.31
Distance to primary school (kilometers) 1.02 2.42 1.27 3.49 0.66 0.97 1.01 2.19 1.04 2.62
Distance to secondary school (kilometers) 22.8 27.2 35.1 35.3 4.80 7.45 23.0 27.6 22.6 26.8
Walking distance to school (hours) 0.46 0.50 0.59 0.68 0.29 0.26 0.46 0.50 0.47 0.49
School roofing is poor 0.29 0.38 0.40 0.48 0.11 0.30 0.27 0.38 0.30 0.39
School buildings are poor 0.40 0.42 0.54 0.49 0.19 0.37 0.40 0.42 0.40 0.42
School furniture is poor 0.60 0.42 0.69 0.45 0.47 0.47 0.60 0.41 0.60 0.42
Maintenance has deteriorated 0.40 0.42 0.49 0.49 0.29 0.42 0.40 0.41 0.41 0.42
Textbooks are available 0.33 0.40 0.19 0.38 0.53 0.46 0.34 0.40 0.32 0.39
Notebooks are available 0.67 0.40 0.53 0.49 0.87 0.31 0.68 0.39 0.65 0.40
More than 2 shifts per day 0.48 0.42 0.22 0.41 0.85 0.33 0.49 0.42 0.46 0.42
Fees per term (K1,000) 0.10 0.17 0.05 0.09 0.16 0.27 0.10 0.18 0.10 0.02
Fees missing 0.32 0.39 0.31 0.45 0.33 0.44 0.32 0.40 0.31 0.39
Educ. exp. per student per term (K1,000) 3.00 2.56 1.59 1.44 5.06 3.27 3.05 2.60 2.94 2.52
0ROVINCES
Copperbelt 0.18 0.33 0.04 0.18 0.40 0.46 0.19 0.33 0.18 0.33
Eastern 0.12 0.28 0.19 0.38 0.03 0.17 0.12 0.28 0.13 0.28
Luapula 0.08 0.23 0.11 0.31 0.04 0.17 0.08 0.23 0.08 0.24
Lusaka 0.13 0.29 0.03 0.17 0.28 0.42 0.13 0.29 0.13 0.28
Northern 0.13 0.28 0.18 0.38 0.05 0.20 0.12 0.28 0.13 0.29
North Western 0.06 0.19 0.08 0.27 0.02 0.12 0.06 0.19 0.06 0.20
Southern 0.11 0.26 0.13 0.33 0.08 0.26 0.11 0.27 0.11 0.26
Western 0.09 0.24 0.13 0.33 0.03 0.16 0.09 0.25 0.09 0.24
Number of observations 12558 4210 8348 6309 6249
-4-
A child who performs housework is defined as working if housework was referred to as the main
activity during the last seven days before the survey. If housework was not explicitly stated, as the
main activity during the last seven days before the survey, the survey gives no information on
performed housework. For 0.29% of the children, housework is stated as the main activity during the
last seven days before the survey, and these are mainly girls from urban areas. Because the number of
Zambian children who are reported to do housework is so small, it does not have a significant impact
on this research, and the children doing housework are defined as working.
Clearly the definition chosen has limitations. By defining only the main activity of the child as child
labor, a child who is working seasonally may be included, while a child working year-round but part-
time is excluded.
In addition to information about the main activity during the last seven days before the survey, the
questionnaire also contains a question about the main activity within the last 12 months before the
survey. Either of these two questions can be the basis for a definition of child labor. The main reason
for using the answer to the former question is that the answer is not much affected by bad memory.
Because the survey was completed over several months, the week before the survey is considered to be
representative for the entire year. If the definition of child labor is based on the question about the
main activity over the last 12 months before the survey, the estimate of the extent of child labor
decreases. Where 13% of the children are reported to have worked the last seven days, only 10% of
the children are reported to have worked most of the last 12 months. The memory effect may explain a
great deal of this difference. Also, child labor is a sensitive subject. Therefore, it is expected that the
respondent (which is most often the father or mother of the child) feels bad about revealing that the
child worked most of the last 12 months. It might be less emotionally demanding to say that the child
worked during the last seven days.
Table 1 shows that 70% of the children only attend school, the number is higher in urban areas than in
rural areas, and higher for boys than for girls. In the sample, 8% work only whereas 5% both work and
attend school. More children work in rural areas than in urban areas. In the next section a more
thorough analysis of these figures is presented.
On average, the children in the sample are 11 years old. Most of the households, in which the children
reside, own at least one asset. About one-fifth of them have a non-biological relationship to the head
of the household. This is because the most common family structure is the extended family, and
children are often placed outside the home of their biological parents during their school age.
This analysis adopts the definition of poverty in CSO (1994) which is based on the price of a food
basket for an adult male. Two poverty lines are defined on the basis of the imputed household
-5-
expenditure:6 an extreme poverty line at K5,910, and a moderate poverty line at K8,480 per adult
equivalent (AE). The extreme poverty line is defined as the price of the necessary food basket for an
adult male per month, whereas the moderate poverty line is defined as the amount needed if this food
basket account for 70% of the monthly budget. About 81% of the children grow up in households
falling below the moderate poverty line, and 69% grow up in households falling below the extreme
poverty line. These numbers conceal the fact that 91% of the rural children and ‘only’ 37% of the
urban children live in households falling below the extreme poverty line according to this definition. 7
Because the moderate poverty line defines most Zambians as poor, only the extreme poverty line is
used. Households below the extreme poverty line are divided in three groups: those with an
expenditure per AE below K1,970, those with an expenditure per AE between K1,970 and K3,940,
and those with an expenditure per AE in the range K3,940 to K5,910. Among the sampled individuals,
29% belong to households that spend less than one third of the amount which defines poverty.
The average distance to primary and secondary schools is 1 km and 23 km, respectively, and
somewhat higher in rural areas. On average, the walking distance is just above half an hour for
children from rural areas and a quarter of an hour for children from urban areas. The condition of the
schools is considered to be poor. This is especially true in rural areas, where more than one-half of the
children come from communities in which the community leader says that school buildings are in a
poor condition. Also, more than one-half come from communities in which the community leader says
that the school furniture is in poor condition, while in rural areas 49% of the children come from
communities where the maintenance of the schools has deteriorated, whereas this figure is only 29%
for urban areas. Although the condition of urban schools seems to be better, 87% of the urban children
come from communities in which the schools run more than 2 shifts per day.
Textbooks are either usually or sometimes available at the local market for one-third of the children,
whereas notebooks are either usually or sometimes available for two-thirds of the sampled children.8
On average, fees are K100 per term (US$0.24) including payments to the Parents’ and Teachers’
Association. However, for more than 30% of the children, the information on the school fees are
missing. The mean for the non-missing observations is K110/(1-0.32)=K147 (US$0.35). Because the
number of missing observations is so large, this information is included in the analysis instead of
making an ad hoc estimate on the fees in those cases. Also the community average of household
6
‘Imputed expenditure’ is used to proxy income, because expenditure is often thought to be more reliable than income. ‘Imputed
expenditure’ includes the money spent on consumption goods plus the value of the consumed home produce. The respondent can repo rt
home produce in cash or kind, and if it is reported in kind, the CSO imputes the cash value.
7
Even though, these estimates of poverty are in accordance with the findings for 1991 (World Bank, 1994), the high rates of poverty
lead one to suspect that the necessary food basket for an adult male is overstated or mismeasured. This is a general problem with
absolute definitions of poverty. Another explanation might be measurement errors in expenditure per AE.
8
This might seem surprising, but after having travelled in rural Zambia it seems perfectly plausible.
-6-
education expenses per student is applied to measure the cost of education. Averaging over
communities reduces the selectivity bias associated with household level expenditure. The household
education expenses per student is K3,000 (US$7). In the light of the poverty discussion above, this is
actually a substantial amount.
)NCIDENCE OF #HILD ,ABOR 3OME 4ABULATIONS

In this section the incidence of child labor is described. First, how the frequency of child labor differs
across age, provinces and household expenditure is investigated. Focus is then moved to the
occupation and income of the working children. Finally, a back of the envelope calculation of lost
GDP due to child labor is presented.
$ISTRIBUTION OF CHILD LABOR ACROSS AGE
A general characteristic of children’s activities in Zambia is that school participation is high compared
to other African countries, e.g. Ghana, Tanzania, and Cote d’Ivoire (see Canagarajah and Nielsen,
1998). The activity distribution by age for Zambian children is shown in Table 2.
One characteristic is that with rising age, the activity tends to shift from neither attending school nor
working to doing either or both of the two. Another characteristic is that school participation increases
to peak at age 11 due to late enrollment and high repetition rates. After that age, school dropouts cause
a steady decline, and child labor increases correspondingly. This picture prevails for both girls and
boys, although the decline in school attendance from age 11 to 14 is much stronger for girls than for
boys (11 versus 3 percentage points). This gender difference in the decline is due to the fact that girls
often marry early or become pregnant.9 In addition, it is possible that parents sacrifice the girls’
education rather than that of boys if they are stressed economically and thus discriminating in favor of
boys.
Table 2 emphasizes the point made in the previous section that school attendance is much lower and
child labor much higher in rural areas than in urban areas. The difference peaks at age 7, where school
attendance is 39% in rural areas and 67% in urban areas. However, the rural children catch up, and
from age 11 onwards the difference is stable at 12-14 percentage points. Regarding child labor the
difference is also large: in rural areas child labor increases from 16% to 27% from age 7 to 14,
whereas in urban areas it increases from 1% to 3%.
9
Jensen and Nielsen (1997) show that 6% of the 7-18 year old school leavers left school because they married or became pregnant.
Further investigation of Priority Survey II for Zambia shows that more girls than boys marry early.
-7-
4ABLE . !CTIVITY DISTRIBUTION BY AGE
Activity Age
7 8 9 10 11 12 13 14 All
!LL
Only school 48.1 60.5 73.2 74.5 78.7 77.7 77.7 72.4 70.4
Work and school 1.8 4.1 3.0 4.9 6.2 5.8 5.8 5.4 4.6
Only work 8.0 8.5 7.1 9.6 6.1 8.3 8.2 11.0 8.4
Neither 42.1 26.8 16.7 11.0 9.0 8.3 8.3 11.3 16.6
Total 100 100 100 100 100 100 100 100 100
2URAL
Only school 36.1 47.0 61.6 62.7 70.6 68.9 69.4 63.4 59.9
Work and school 2.5 6.2 4.5 7.6 8.9 9.2 9.0 8.5 7.0
Only work 13.2 13.7 11.8 15.9 9.5 13.1 13.4 18.6 13.6
Neither 48.2 33.1 22.0 13.8 11.0 8.8 8.3 9.5 19.4
Total 100 100 100 100 100 100 100 100 100
5RBAN
Only school 66.1 82.2 89.2 92.2 91.9 90.0 89.3 83.6 85.8
Work and school 0.8 0.8 0.8 0.8 1.7 1.0 1.3 1.4 1.1
Only work 0.2 0.2 0.5 0.1 0.6 1.4 1.0 1.3 0.7
Neither 32.8 16.7 9.4 6.8 5.8 7.6 8.4 13.6 12.4
Total 100 100 100 100 100 100 100 100 100
'IRLS
Only school 49.9 61.1 74.9 73.8 79.0 78.2 75.7 69.2 70.3
Work and school 1.8 4.5 1.6 5.1 6.1 5.2 5.6 4.5 4.3
Only work 8.1 7.5 9.8 9.0 5.8 8.7 9.4 11.5 8.7
Neither 40.3 26.9 13.7 12.1 9.1 7.9 9.3 14.8 16.7
Total 100 100 100 100 100 100 100 100 100
"OYS
Only school 46.1 59.9 71.7 75.1 78.4 77.1 79.7 75.4 70.6
Work and school 1.9 3.8 4.2 4.7 6.2 6.4 5.9 6.2 4.9
Only work 8.0 9.5 4.6 10.2 6.5 7.8 7.0 10.5 8.1
Neither 44.0 26.7 19.5 10.0 8.9 8.7 7.3 7.9 16.5
Total 100 100 100 100 100 100 100 100 100
'EOGRAPHICAL DISTRIBUTION OF CHILD LABOR
Figure 1 illustrates the differences in the incidence of child labor across Zambia. Much of the
difference across provinces is explained by the varying degree of urbanization that correlates
negatively with child labor as mentioned in the previous sections. The provinces of Lusaka and the
Copperbelt are strongly urbanized, and hence the use of child labor is very rare. On the contrary, the
Western, Northern, Eastern and Southern provinces are dominated by rural areas, and here the use of
child labor is widespread. The North Western province, however, appears to be an exception: It is
mainly rural but child labor seems rare. The North Western province is almost covered by jungle, and
therefore, cattle care, which is a common occupation for children, is not very common, and child labor
may not be needed as much as in other provinces. In addition, the North Western province has its
borders to Angola and Congo, which have been war zones and this may have affected the province.
-8-
The numbers behind Figure 1 are found in Table 3, which also gives a detailed activity distribution for
each province. In addition to the information in Figure 1, the table reveals that the Copperbelt registers
the highest with regard to school attendance (87%), whereas the Eastern province registers the lowest
(58%). The other provinces with much child labor have higher school attendance rates mainly because
many children are registered as both attending school and working.
The province of Lusaka has a relatively high proportion of children registered as neither attending
school nor working. This is because the Lusaka province contains the capital, Lusaka, where most of
the poor compounds and slum districts are found, in which inactive children are more common than in
other places.
&IGURE . )NCIDENCE OF CHILD LABOR IN :AMBIAN PROVINCES
0%-5%
5 % - 15 %
15 % - 25 % Luapula
25 % - 12.9
Northern
23.1
North Western Eastern

Copperbelt
1.3 1.6 20.1
Central
8.9
Western Lusaka
2.4
32.2
Southern
18.7
The largest number of child workers is found in the Northern province. However, also the Western,
Eastern and Southern provinces account for a great deal of the total amount of child labor. Together,
the four provinces count 80% of all child laborers.
4ABLE . !CTIVITY DISTRIBUTION BY PROVINCE

Province Only school Work and School Only work Neither All Totala
Central 73.3 1.4 7.6 17.7 100 157,212
Copperbelt 86.3 0.6 1.0 12.0 100 297,097
Eastern 55.7 2.7 17.4 24.2 100 200,220
Luapula 68.6 5.6 7.4 18.5 100 130,367
Lusaka 80.4 0.5 1.9 17.2 100 209,195
Northern 65.2 9.3 13.8 11.7 100 203,132
NorthWestern 74.0 0.3 1.0 24.8 100 89,813
Southern 64.5 9.9 8.8 16.7 100 175,883
Western 54.5 13.7 18.4 13.3 100 144,214
All 70.4 4.6 8.4 16.6 100 1,607,132
a. Numbers in this column are estimates of the total size of the group.
-9-
0OVERTY AND CHILD LABOR
In Table 4 poverty is cross tabulated with activity variables. As mentioned above, the definition of
absolute poverty given in the CSO (1994) is used. From the table it can be seen that in the poorest
households, few children only attend school, and more children work or do neither. In section 4, in
which the choice of school attendance versus child labor is modelled, it can be seen that this result
does not hold as strongly when other characteristics are controlled for.
4ABLE . !CTIVITY DISTRIBUTION BY LEVEL OF HOUSEHOLD INCOME PER ADULT EQUIVALENT

Only Work and Only work Neither All Totala
school school
Expenditure per AE K0-1970 53.7 6.9 17.3 22.1 100 459,034
Expenditure per AE>K5910 86.4 2.0 1.4 10.2 100 496,670
Total 70.4 4.6 8.4 16.6 100 1,607,132
/CCUPATION AND INCOMES OF WORKING CHILDREN
Table 5 illustrates the occupations of the working children. In Table 5 it is seen that child workers are
more often found in rural areas as 197,074 of 208,516 child workers come from rural areas. It also
shows that the largest part of the working children were in subsistence farming. Among urban child
workers, more than one-third is doing housework, and 11% were in elementary occupations (e.g. street
vendors). More girls than boys are occupied in housework, and more boys than girls are occupied in
elementary occupations. The table illustrates that older child workers are, to a higher extent, employed
outside of subsistence farming, whereas younger child workers are almost exclusively employed in
subsistence farming.
4ABLE /CCUPATION OF THE WORKING CHILDREN

Subsistence Market Service Elementary House- Other Missing All Total number
farming oriented occupation work occupation of working
farming childrena
Total 87.2 2.1 0.8 6.8 2.3 0.4 0.4 100 208,516
population
Rural 90.1 2.0 0.3 6.5 0.4 0.3 0.3 100 197,074
Urban 36.2 3.1 8.6 10.8 34.5 2.6 4.2 100 11,442
Female 86.8 1.5 1.1 5.9 3.0 0.8 0.9 100 104,010
Male 87.5 2.6 0.5 7.6 1.5 0.2 0.1 100 104,505
7-10 years 90.7 2.5 0.0 3.5 1.8 0.6 0.8 100 98,274
11-14 years 84.0 1.7 1.4 9.6 2.6 0.4 0.2 100 110,241
Because most of the working children are employed in subsistence farming as, for the most part,
unpaid family workers, it is difficult to estimate the income of working children. The survey includes
- 10 -
information about income in terms of both salary and imputed value of home production, and the
respondent can either report the home production in cash or in kind. If it is reported in kind, an
appropriate price is used to compute the cash equivalent. In the survey, the imputed income from
home production is usually attributed to the adult household members, meaning that the value of child
labor cannot be assessed directly from the survey.
Table 6 presents the mean predicted potential incomes from child labor based on two different
estimates of the potential labor income. The first measure, (a), is a simple mean income for all income
earning children in the district of residence. The second measure, (b), is based on an estimation of a
simple human capital model for the sample of adult males.10 The human capital equation includes the
following variables: years of education, years of education beyond the primary level, Mincerian
experience, and Mincerian experience squared. Out of sample predictions for children are made, and
because education and experience are included, the mean predicted income for children is much lower
than that of adults.
The two estimates are quite different. Although, it is believed that measure (b) is most appropriate,
both measures are tried in the empirical analysis. The simple mean estimation would overpredict the
potential income of a child, because those children who actually work for money (and are the basis for
the calculation of this mean) are the oldest, the most experienced, and the most able. The fact that the
simple mean is higher for girls than for boys, also supports the view that measure (b) is the most
appropriate. The higher mean income for girls compared to boys reflects the fact that the income-
earning girls earn more than the income-earning boys do, although more girls than boys earn no
income at all. Had the zero incomes been included in the district mean, the estimate on the potential
income would be reduced to a low and unrealistic level.
4ABLE . -EAN PREDICTED POTENTIAL INCOMES FROM CHILD LABOR +WACHAMONTH

3IMPLE MEAN
Mean income of income earning children in the district of 7,390 3,700 13,640 7,500 7,280
residence (a)
-EAN PREDICTED POTENTIAL INCOME FROM A HUMAN CAPITAL MODEL
Prediction based on a human capital model for adults (b) 1,330 1,270 1,400 1,330 1,320
The mean of the predicted potential income varies from K1,270 to K13,640 per month. As a
comparison, the average income of unskilled adult workers was K17,350 per month at the time of the
survey, and the price of a food basket for an adult male was estimated to be K5,910 per month.
10
If gender specific estimates are used, the predicted income from child labor would be extremely low for girls. Similarly, Mason and
Khandker (1998) argue that the potential income of girls should be computed using the market wage for boys.
- 11 -
7HAT ARE THE ECONOMIC GAINS AND LOSSES OF CHILD LABOR ! #OST "ENEFIT
!NALYSIS
Gains and losses from child labor are analyzed in the context of a cost-benefit analysis of primary
education. Each household decides whether a child should work and earn money now, or attend school
and thereby increase potential future earnings. Poor households with high discount rates would tend to
choose child labor, and rich households with lower discount rates would choose education. If
education pays off or if education is cheap, households would tend to choose education. Table 7
illustrates the central points:
4ABLE . #OSTS AND BENEFITS OF CHILD LABOR AND PRIMARY EDUCATION A
Choice Year 1-7 Year 8-50

No education
Earnings X per year Y per year
Complete primary school
Earnings 0 7RY per year
Private costs K3,000 per term 0
Public costs US$16-22 per yearb 0
a. X is the income from child labor, Y is the income of an unskilled adult, and R is the annual return to education.
b. Ministry of Education (1996b).
The gains from child labor are difficult to assess because the main part consists of income from child
labor. However, in Table 6 an attempt was made to evaluate the potential income from child labor
based on a human capital model. Here two different valuations of child labor are used, one is based on
measure (b) in Table 6, the other measure assumes that children earn 30% of the average income of an
unskilled adult which is K17,350.11 Another potential gain from child labor is saved costs of
education. Both the saved private costs and the saved public costs are substantial.
The most important loss from using child labor is the lost future income gain because of foregone
investment in human capital. A standard human capital model with correction for selection bias due to
the participation decision would show a private return of about 7% per year of primary education.12
For Y the mean income for adults without education (K17,350 per month) is used. The cost-benefit
analysis is performed under the assumption of a discount rate of both 5% and 20%. The former is
relevant for the government or the aid organizations, whereas the latter would be relevant for the
private Zambians.
Table 8 presents the present values of relevant income streams. Clearly, if the return to primary
education was zero, the net private benefit of child labor would be positive. Although the public costs
11
Like Tzannatos (1997).
12
See Nielsen and Westergård-Nielsen (1998).
- 12 -
are substantial, the net public benefit could be positive or negative, depending on the social gain from
being an educated society. When the return to primary education is not equal to zero, the conclusion is
less clear cut. If the discount rate is 5%, child labor would never be an optimal choice, neither for the
household nor for society. However, if the discount rate is 20%, child labor might represent an optimal
choice.
Under the assumption of the largest income from child labor (column two), and a 20% discount rate
(the upper part of the table), the private benefit from child labor is positive. Under the assumption of
an income from child labor of K1,330 per month, the private loss from using child labor would be
K52,316 (US$125) per child.
If the externalities of being an educated society are disregarded (e.g. reduction in fertility, increased
complexity of production), a conservative estimate on the social loss from child labor can be made.
The social loss should be calculated from the lower part of Table 8, because the society has a lower
discount rate than the households do. Under the assumption of the largest income from child labor, the
social loss would be K816,412 (US$1,943) per child. If they earn only K1,330 per month it would be
K1,085,479 (US$2,584). The number of working children is 208,516 of which two thirds are not
enrolled in school. If all the children who are both working and attending school accumulate human
capital and finish primary school, the loss would be about US$300 mill. If, on the other hand, all the
children who are working forego investment in education because they are not able to do both
efficiently, the loss would amount to about US$400 mill.-US$500 mill. This is the present value of the
loss from the 7 cohorts of children who attended primary school at the time of the survey. Since the
GDP in 1993 was US$3,564 mill, the annual GDP loss would, in the best case, be about 1% and, in
the worst case, 2%.13
In a country where the per capita growth rate is often negative, these numbers are huge. And because
most child workers are found in rural areas, one would expect that residents in rural areas suffer the
most from the loss in GDP. This back of the envelope cost benefit analysis serves as a motivation to
find out what determines whether children work or attend school. This is the issue of the next section.
13
Of course, the conclusion hinges on the assumptions about the discount rate, the estimated return to human capital, and the estimated
income from child labor.
- 13 -
4ABLE . 0RESENT VALUES AT THE BEGINNING OF YEAR AGE YEARS
Income from child labor per year
K1,330*12 K17,350*12*0.3
Discount rate=0.20
Income year 1-7 K57,529 K225,143
Income gain from education (R=7%) K142,286 K142,286
Private costs K32,441 K32,411
Public costs (US$16 per year) K24,223 K24,223
Discount rate=0.05
Income year 1-7 K92,350 K361,417
Income gain from education (R=7%) K1,268,790 K1,268,790
Private costs K52,077 K52,077
Public costs (US$16 per child) K38,884 K38,884
%CONOMETRIC !NALYSIS
This section first presents the econometric technique, which is a bivariate probit model with random
effects and correction for endogeneity of an explanatory variable. After that, the model estimation
results are presented.
-ETHODOLOGY
The dependent variables are two indicator variables:
1 if individual I works
Y1 = 
I
0 otherwise
1 if individual I attends school
Y2 = 
I
0 otherwise
The first indicator variable indicates whether or not an individual worked during the last seven days
before the survey, and the second indicates whether or not the individual attended school on a part
time or full-time basis at the time of the survey. In the empirical analysis, the simultaneous choice of
letting children work, attend school, or both is analyzed.
A bivariate probit model is estimated. The likelihood function for the bivariate probit looks as follows:
.
l(β 1, β 2 ) = ∑ {1
ln Φ 2 (− β 1 X1 ,− β 2 X2 , ρ ) + ln Φ 2 ( β 1X1 ,−β 2 X2 ,− ρ )
44444 42444444
I I
3 1444442444443
I I
I =1 nothing only work
+ ln Φ 2 ( β 1 X1 , β 2 X2 , ρ ) + ln Φ 2 ( − β 1 X1 , β 2 X2 ,− ρ )}
I I I I
14444 4244444
both work and school
3 14444442444444
only school
3
where X X are the explanatory variables explaining the probability of working and the probability of
I I
attending school, respectively. The coefficients to be estimated are called β and β , and the
correlation coefficient is ρ Among the explanatory variables is household expenditure per adult
equivalent. This variable might be endogenous. Rivers and Vuong (1988) suggest a two-stage
- 14 -
conditional maximum likelihood procedure to estimate a probit model with endogenous regressors,
and test for exogeneity of the presumed endogenous regressors. The first step implies an OLS
estimation of each of the endogenous regressors on a set of instrumental variables. The second step
consists of a probit analysis with the endogenous regressor and the least squares residuals from step
one among the explanatory variables. A test of whether the coefficient to the least squares residual
equals zero is a test of exogeneity. A rejection of exogeneity implies that standard errors should be
corrected as a result of using instrumental variables.
To correct for unobserved characteristics that are constant across communities, a normally distributed
random community effect is included. A one-factor loading approach is used, implying that the effect
of the random community effect on the two dependent variables is allowed to differ. This is equivalent
to including two random effects, which are restricted to be perfectly correlated. To correct for the
possible bias due to correlation between the explanatory variables and the random effects, intra-
community means of the explanatory variables are included.
2ESULTS
In Table 9 and 10 below, results from the estimation of bivariate probit models are presented. The first
part of the tables gives the estimates of the parameters that affect the work decision, whereas the last
part of the tables gives the estimates of the parameters that affect the schooling decision. Table 9
presents results from the estimation of simple bivariate probit models, whereas Table 10 contains
results from the estimation of bivariate probit models with random community effects. Including
random effects entails a risk of inconsistency due to correlation between the explanatory variables and
the random effect. Therefore, Table 10 also contains the results after correcting for this inconsistency.
Different likelihood ratio (LR) tests are performed, and they are presented at the bottom of the tables.
A LR test clearly suggests that the sample should be split by area of residence, hence this is done
before random community effects are included. Furthermore, a LR test for exogeneity of expenditure
per AE is performed. The instruments used: education, age and age squared of the household head.
Furthermore, if he or she is working, indicator variables for the occupation and industry of the
household head are used. 2 is about 0.35 in all regressions of household expenditure per AE on the

instrumental variables. A LR-test for exogeneity rejects exogeneity of expenditure per AE.
Increasing expenditure per AE decreases the probability of working in the composite sample, in the
rural sample, and in the sample of boys, whereas it increases the probability of school attendance in all
- 15 -
estimations.14 A higher level of education of the household head decreases the probability of working
in all but the urban sample, whereas it increases school attendance in all samples.
Table 9 shows that the probability of working increases with increasing age.15 This could be
interpreted as an indication of the fact that the accumulated human capital increases potential wages
and therefore the probability of working. Correspondingly, after age 11, these increased opportunity
costs stemming from increasing age decrease the probability of school attendance.
Two variables related to the age of a child are included in the analysis: age and age rank. The age
variable catches the effect of the ABSOLUTE value of the labor of a child of a given age, whereas the age
rank variable catches the effect of the RELATIVE value of the labor of a family of brothers and sisters.
The age rank of a child is high for the young children with many older siblings. The higher the age
rank, the lower the probability of working and the higher the probability of attending school, meaning
that children who have older siblings are less likely to work and more likely to attend school than
those without. The effect of the age rank probably reflects the fact that the older children are worth
more in the labor market than their younger siblings. Had a quantity–quality trade-off been relevant,
the opposite signs on this coefficient would have been expected.16
Having a non-biological relationship to the head of the household decreases the probability of
attending school for all but children in rural areas. This could indicate that rural and urban children
live in other households for different reasons.
As in the findings by Lavy (1996), school fees and quality of schooling have relatively weak effects on
the probability of school attendance while the transportation costs are more important. Actually, fees
were insignificant and therefore excluded from the equation, whereas the community average of
education expenses per student was significant. In most estimations, the walking distance to the
nearest primary school, the presence of schools in the community and the distance to the nearest
primary school in kilometers if it is situated outside the community, affect the probability of working
and the probability of attending school. Furthermore, availability of passable roads or public
transportation decreases the work probability and increases the probability of school attendance. The
effects of transportation costs are stronger in rural than in urban areas as would be expected.
14
This effect increases after correction for endogeneity.
15
Age has an increasing effect after age 7.
16
This is consistent with Montgomery et al. (1995) who report that the existence of a quantity-quality trade-off is not established for
Sub-Saharan Africa.
- 16 -
4ABLE 2ESULTS FROM ESTIMATION OF A BIVARIATE PROBIT MODEL

Variable Coeff. Std.err Coeff. Std.err Coeff. Std.err Coeff. Std.err Coeff. Std.
7ORK EQUATION
Constant -0.16 0.64 -2.07 1.31
Female 0.01 0.03 0.02 0.05 -0.07 0.09
Rural
Age -0.43 1.07 -0.31 2.18 -1.16 0.84 -0.99 0.85
Age squared 0.38 0.51 0.40 1.01 0.60 0.40
Non-biological relation to the household -0.01 0.03 -0.08 0.06 0.16 0.13 -0.02 0.05 -0.02 0.05
head
Age rank among household members aged
7-31
The household owns an asset 0.07 0.12 -0.06 0.04
The household owns land 0.00 0.08 0.08 0.16 0.04 0.21 -0.08 0.13 0.07 0.11
The household is involved in agriculture 0.21 0.20 0.24 0.16
Education of the household head 0.01 0.20
Age of the household head 0.64 0.60 -0.08 0.17
The household head works in the formal 0.12 0.18 -0.03 0.12 -0.03 0.09
sector
The household head does not work -0.31 0.27
Expenditure per adult equivalent -0.61 2.35 -1.54 1.22
Fertilizer is sold all year 0.15 0.21
Credit is available -0.01 0.03 -0.01 0.05 -0.23 0.17 0.05 0.04 -0.06 0.04
Passable road -0.28 0.28
Public transport
Primary school in community -0.33 0.22 -0.04 0.06
Secondary school in community -0.28 0.29
Distance to primary school -1.44 1.24 -0.11 0.10
Distance to secondary school 0.10 0.08 0.42 0.88 0.01 0.06
Walking distance to school 0.31 0.23
School roofing is poor
School buildings are poor 0.04 0.03 0.06 0.06 0.20 0.21 0.00 0.05 0.08 0.05
School furniture is poor
Maintenance has deteriorated 0.02 0.03 0.01 0.06 -0.09 0.25 0.07 0.04 -0.04 0.05
Textbooks are available 0.00 0.04 -0.02 0.07 0.03 0.15 0.07 0.05 -0.08 0.05
Notebooks are available
More than 2 shifts per day 0.17 0.16
Education expenses 0.00 0.01 0.01 0.01 -0.02 0.01
2ESIDUAL FROM )6 REGRESSION 0.00 0.84 -1.04 2.58 0.05 1.13 -0.31 1.32
3CHOOLING EQUATION
Constant
Female -0.05 0.04 -0.03 0.04
Rural -0.03 0.09
Age
Age squared
Non-biological relation to the household -0.02 0.04
head
Age rank among household members aged 0.13 0.10
7-31
- 17 -
The household owns an asset
The household owns land 0.11 0.07 -0.01 0.17 0.01 0.10
The household is involved in agriculture 0.08 0.06 0.16 0.16 0.11 0.08 0.02 0.08
Education of the household head
Age of the household head
The household head works in the formal 0.15 0.16 0.09 0.07
sector
The household head does not work
Expenditure per adult equivalent
Fertilizer is sold all year -0.02 0.03 0.02 0.06 -0.01 0.05 -0.03 0.05
Credit is available 0.01 0.05 -0.04 0.04
Passable road
Public transport 0.02 0.03 -0.02 0.05 0.02 0.04 0.01 0.04
Primary school in community 0.04 0.07
Secondary school in community -0.03 0.04 0.00 0.09 -0.02 0.07 0.04 0.06
Distance to primary school 0.01 0.04 0.06 0.08 -0.35 0.40 -0.10 0.07 0.07 0.06
Distance to secondary school
Walking distance to school
School roofing is poor -0.08 0.05 0.04 0.07 0.02 0.04
School buildings are poor 0.04 0.03 0.08 0.06 -0.04 0.06 0.05 0.04 0.02 0.04
School furniture is poor -0.05 0.04
Maintenance has deteriorated -0.01 0.03 -0.01 0.05 -0.05 0.05 0.02 0.04 -0.04 0.04
Textbooks are available 0.05 0.03 0.03 0.07 -0.01 0.05 0.04 0.04 0.07 0.05
Notebooks are available -0.03 0.05 -0.01 0.07 -0.01 0.04
More than 2 shifts per day 0.02 0.03 0.02 0.06 -0.03 0.06 -0.02 0.04 0.06 0.05
Education expenses 0.00 0.01
2ESIDUAL FROM )6 REGRESSION -1.58 0.99
Correlation, ρ
Log-likelihood -8932 -3822 -3189 -4499 -4368
Number of observations 12558 4210 8348 6309 6249
,IKELIHOOD RATIO TESTS
Ho: Expenditure is exogenous (df=2) 27.1 13.7 12.0 17.1 10.1
Ho: No split by gender/rural-urban (df=79) 3839.6 128.9
a. Bold figures indicate that the coefficients are significant at a 5%-level, italics indicates significance at a 10%-level, and underlined figure indicate
significance at a 15%-level.
b. Explanatory variables are scaled to be of similar size.
c. Province indicators are also included, although the parameter estimates are not reported.
- 18 -
4ABLE . 2ESULTS FROM ESTIMATION OF BIVARIATE PROBIT MODELS WITH RANDOM COMMUNITY EFFECTS A
Rural Urban
Biv. probit Mundlak Biv. probit Mundlak
with random correction with random correction
effects effects
Variable Coeff. Std.err Coeff. Std.err Coeff. Std.err Coeff. Std.err
7ORK EQUATION
Constant -0.38 1.00 -0.14 1.13 -2.16 2.36 -2.24 2.41
Female 0.09 0.07 0.09 0.07 -0.07 0.16 -0.06 0.16
Age -0.53 1.56 -0.43 1.62 -0.35 4.75 -0.41 4.71
Age squared 0.58 0.74 0.54 0.77 0.44 2.19 0.48 2.18
Non-biological relation to the household head -0.09 0.07 -0.05 0.08 0.25 0.17 0.24 0.17
Age rank among household members aged 7-31 -0.06 0.14 -0.06 0.13
The household owns an asset 0.13 0.19 0.12 0.20
The household owns land 0.08 0.23 0.19 0.25 0.00 0.24 -0.02 0.23
The household is involved in agriculture 0.04 0.28 0.10 0.28 0.20 0.19 0.24 0.22
Education of the household head 0.09 0.22 0.08 0.23
Age of the household head 0.69 0.76 1.03 0.81
The household head works in the formal sector 0.07 0.20 0.04 0.20 0.00 0.16 -0.03 0.18
The household head does not work -0.38 0.28 -0.37 0.28 -0.32 0.28 -0.38 0.31
Expenditure per adult equivalent -0.54 2.87 -0.35 3.13
Fertilizer is sold all year 0.12 0.24 0.13 0.30
Credit is available 0.08 0.16 0.05 0.16 -0.19 0.19 -0.16 0.23
Passable road -0.33 0.32 -0.17 0.37
Public transport
Primary school in community -0.20 0.26 -0.17 0.27 -0.39 0.29 -0.37 0.34
Secondary school in community -0.42 0.28 -0.35 0.29 -0.41 0.35
Distance to primary school -0.30 0.42 -0.27 0.43 -1.46 1.62 -1.61 1.99
Distance to secondary school 0.03 0.29 -0.02 0.29 0.43 0.88 0.36 1.15
Walking distance to school 0.16 0.12 0.27 0.29 0.28 0.36
School roofing is poor 0.37 0.23 0.30 0.29
School buildings are poor -0.06 0.18 -0.13 0.20 0.24 0.24 0.25 0.28
School furniture is poor -0.07 0.17 -0.12 0.18
Maintenance has deteriorated 0.05 0.17 0.10 0.17 -0.11 0.21 -0.15 0.26
Textbooks are available 0.11 0.22 0.12 0.22 -0.01 0.16 0.00 0.18
Notebooks are available 0.23 0.17 0.18 0.18 0.44 0.29
More than 2 shifts per day 0.10 0.21 0.15 0.21 0.10 0.23 0.04 0.27
Education expenses 0.01 0.04 0.01 0.04
)NTRA COMMUNITY MEANS
Non-biological relation to the household head -0.42 0.49
The household owns land -0.25 1.05
Education of the household head -0.30 0.55
The household head works in the formal sector -0.02 0.50
2ESIDUAL FROM )6 REGRESSION -1.05 2.79 -1.09 2.98
3TDDEV ON RANDOM COMMUNITY EFFECT
- 19 -
3CHOOLING EQUATION
Constant
Female -0.05 0.05 -0.05 0.05 -0.03 0.05 -0.03 0.05
Age
Age squared
Non-biological relation to the household head 0.03 0.06
Age rank among household members aged 7-31 0.20 0.13 0.17 0.13
The household owns an asset
The household owns land -0.03 0.17 -0.21 0.23
The household is involved in agriculture 0.20 0.14 0.19 0.14
Age of the household head
The household head works in the formal sector 0.15 0.15 0.16 0.16 0.03 0.05
The household head does not work
Expenditure per adult equivalent
Fertilizer is sold all year 0.00 0.06 -0.01 0.06 -0.14 0.15 -0.18 0.13
Credit is available 0.00 0.05 -0.01 0.05
Passable road 0.15 0.19 0.10 0.15
Public transport -0.01 0.05 -0.04 0.05 0.14 0.09 -0.01 0.08
Primary school in community 0.06 0.12 0.04 0.13
Secondary school in community 0.01 0.08 -0.03 0.08 0.02 0.15 -0.09 0.13
Distance to primary school 0.06 0.08 0.07 0.08 -0.24 0.33 -0.05 0.49
Distance to secondary school -0.38 0.68 -0.41 0.76
Walking distance to school -0.15 0.20 -0.15 0.18
School roofing is poor -0.08 0.05 0.02 0.15 0.04 0.15
School buildings are poor 0.09 0.06 0.08 0.06 -0.01 0.13 -0.01 0.12
School furniture is poor 0.10 0.10 0.13 0.09
Maintenance has deteriorated -0.01 0.05 -0.02 0.05 -0.07 0.13 -0.02 0.12
Textbooks are available 0.03 0.06 0.03 0.06 0.01 0.10 0.07 0.09
Notebooks are available -0.04 0.05 -0.01 0.05 -0.01 0.16 0.08 0.15
More than 2 shifts per day 0.04 0.06 0.01 0.06 -0.05 0.15 -0.07 0.13
Education expenses -0.01 0.01
Non-biological relation to the household head
The household owns land
The household head works in the formal sector
2ESIDUAL FROM )6 REGRESSION -1.58 1.23 -1.82 1.21
3TDDEV ON RANDOM COMMUNITY EFFECT
Correlation, ρ
Log-likelihood -3523 -3510 -3128 -3097
Number of observations 4210 8348
a. Bold figures indicate significance at a 5%-level, italics at a 10%-level., and underlined figures at a 15%-level.
b. All explanatory variables are scaled to be of similar size.
c. Province indicators are also included, although the parameter estimates are not reported.
d. The potential income measure and interaction effects with being female and education of the head of the household are included. For measure (b)
for rural areas, the model is also estimated without cross terms, but this does not make the variables significant.
Many of the estimated parameters to the quality variables have the opposite sign of that which would
be expected. This could, of course, reflect the fact that the variables are poor indicators of the quality
- 20 -
of schooling. Although the coefficients to the quality indicators are, in general, very badly determined,
some of them allow for interpretation. The indicator for whether school roofing is poor increases the
probability of working and decreases the probability of school attendance. For all but the urban
children, the indicator for whether school furniture is poor decrease the probability of school
attendance. In addition, the indicator for whether the school runs more than 2 shifts per day is found to
increase the probability of working in some regressions. However, some of the other quality variables
have reverse effects.
Lavy’s hypothesis, that supply constraints on secondary schooling affect the demand for primary
schooling because primary school may be a ticket to secondary school, finds some support. School
attendance decreases with an increase in the distance to the nearest secondary school, although the
effect is not always well determined for the sample from urban areas. In some estimations, the
presence of a secondary school in the community decreases the probability of working, or the distance
to the nearest secondary school increases the probability of working.
As mentioned above, the availability of public transportation and the availability of passable roads
allow for interpretation as they reflect the transportation costs of schooling. However, the other
infrastructure variables often turn out to be insignificant, or have opposite signs of those which would
be expected; availability of fertilizer increases the probability of working, whereas availability of
credit decreases the probability of school attendance.
The correlation coefficient, ρ, is significantly negative in all the estimations. This means that there is a
negative relationship between attending school and working. The interpretation would be that some
unobserved factors that increase the probability of attending school decrease the probability of
working. On the other hand, as the definition of child labor does not include part-time work, there may
be a reason to modify the importance of this result.
The inclusion of random community effects increases the likelihood values, and more so in rural areas.
This means that unobserved community factors are important determinants of the probability of
working and attending school. Unobserved community factors include omitted community variables
and measurement errors in included community variables. The coefficients to the random effect are
clearly significant with different signs, meaning that the effect of the random effect on the probability
of working and attending school is perfectly, negatively correlated.
A general drawback of random effects models is that the random effects might be correlated with
included explanatory variables resulting in inconsistent parameter estimates. To correct for this
inconsistency, the Mundlak approach is used. It implies that means of explanatory variables that vary
- 21 -
across communities should be included as explanatory variables. The means are included, the model is
reduced, and the result is seen in Table 10.
The intra-community means ensure consistency, but they can also be interpreted. Regarding the
education of the household head, Table 10 shows that the higher the education of the household head,
the higher the probability of school attendance. This effect is enforced if the household resides in a
community with a high average education of the household heads. For rural areas there is also a
negative effect of the education of the household head on the probability of working which is enforced
by a community effect. Furthermore, the probability of attending school is increased if the child lives
in a community where many households own land. For urban areas, the estimation without intra-
community means showed that a household head working in the formal sector exerted a positive effect
on school attendance. When intra-community means are included this turns out to be a community
effect: living in a community where it is common that the household heads work in the formal sector,
increases school attendance.
The results from including a measure of the potential income from child labor are not reported. In
general, the measures do not seem to catch the potential income of children, since the measures
decrease the probability of working and increase the probability of attending school. Rather they catch
a general income effect or they reflect the effect of the highest level of education obtained in the past,
either of which might be endogenous.
0OLICY 2ECOMMENDATIONS
In the previous subsection the results from estimating bivariate probit models were presented. These
results allow something to be said about the direction of the effects, but not much about their size. For
policy purposes something must be known about the magnitude of the effect of changing a policy
variable on the probability of attending school and working. Therefore, the inference of this section is
based on the marginal effects of changes in the explanatory variables for a number of standard
persons.
Because, in most cases, the gender of the child has a negligible effect on the probability of working
and attending school, all the standard persons are girls. They vary with respect to age (8 and 13 years)
and area of residence (rural or urban). The rest of the characteristics are those of the average girl from
that respective area of residence.
5NOBSERVED COMMUNITY EFFECTS
From Table 9 and 10 it was seen that including an unobserved community effect improves the
likelihood significantly, especially in rural areas. Table 11 shows the probability of working and the
- 22 -
probability of school attendance for standard persons from communities with three different sets of
unobserved characteristics: a mean community, a ‘good community’, (one standard deviation above
the mean), and a ‘bad community’ (one standard deviation below the mean).17
4ABLE 0ROBABILITY OF WORKING AND PROBABILITY OF SCHOOL ATTENDANCE IN THREE COMMUNITIES
Rural Urban
8-year-old 13-year-old 8-year-old 13-year-old
- percent -
0ROBABILITY OF WORKING
Average community (community effect=0) 3.39 7.20 0.32 1.26
Good community (community effect=std.dev.) 0.14 0.43 0.06 0.28
Bad community (community effect=-std.dev.) 25.51 38.47 1.38 4.34
0ROBABILITY OF SCHOOL ATTENDANCE
Average community (community effect=0) 57.27 81.40 88.87 95.20
Good community (community effect=std.dev.) 63.35 85.32 94.33 97.87
Bad community (community effect=-std.dev.) 51.02 76.87 80.42 90.35
Table 11 shows that the probability of working for an 8-year-old girl from a rural area is 22 percentage
points higher if she comes from a bad community than if she comes from an average community. The
similar number for a 13-year-old is 31 percentage points. For urban areas the numbers are smaller in
magnitude. The probability of school attendance is less affected by unobserved community
characteristics; for 8-year-olds from urban areas where the effect is largest, the probability of school
attendance declines by 8 percentage points if the individual comes from a bad community.
In urban areas, the probability of working is zero unless the individual comes from a community with
bad unobserved characteristics. Nothing else matters. In rural areas the unobserved community
characteristics are important, but also other variables have an effect.
)NCOME EFFECT
Table 12 and 13 present selected marginal effects for standard persons from rural and urban areas,
respectively. The first thing to notice is that the marginal effect of an income change is surprisingly
small, even though the effect has increased after correction for endogeneity of income. Therefore, a
striking conclusion from the analysis is that poverty and low income are not the main explanations for
child labor in Zambia. Since poverty is mentioned in the theoretical literature as a main hypothesis in
explaining child labor (e.g. Bonnet, 1993; Grootaert and Kanbur, 1995; Fallon and Tzannatos, 1998),
one would expect a huge effect of income on the probability of working. However, that hypothesis is
17
It might seem strange that the probability of working and the probability of school attendance do not correspond directly to the
probabilities presented in Table 2. However, this is because the model is non-linear, which means that the probabilities for the mean
person are not equal to the mean probability in the sample. This is why coming from a bad community affects the probabilities more
(measured in percentage points) than coming from a good community does.
- 23 -
not confirmed here. Canagarajah and Coulombe (1998) find a similar conclusion for Ghana, and field
studies in India show the same result, see Sinha (1996) and M.V.Foundation (1996). The latter states
that “ … the Foundation has realized that poverty as a factor behind non-enrollment and child labor
comes third in importance. The tradition of the family and access to school are the first and second
most important factors, respectively.”
4ABLE 3ELECTED MARGINAL EFFECTS FOR STANDARD PERSONS FROM RURAL AREAS
Average community Bad community
7ORK EQUATION std.person marg. effect marg. effect marg. effect marg. effect
The household owns an asset 1 -1.86
Education of the household head 4.3 -0.26 -0.47
Expenditure per AE (K1,000) 2.9 -1.30
Passable road 1 -4.05
Public transport 0 -1.65 -3.16 -8.25
School roofing is poor 0 4.31
Secondary school in community 0 -2.15 -4.19 -11.43 -14.62
Education of the household head 4.3 -0.54 -0.98
0ROBABILITY OF WORKING 3.39 7.20 25.51 38.47
3CHOOLING EQUATION
The household owns an asset 1
Education of the household head 4.3
Expenditure per AE (K1,000) 2.9
Passable road 1
Public transport 0 -1.57 -1.09 -1.60 -1.24
Primary school in community 1
Secondary school in community 0 -1.05 -0.73 -1.07 -0.82
Distance to primary school (kilometers) 1.2 0.26 0.18 0.26 0.20
Distance to secondary school (kilometers) 35.9
Walking distance to school (hrs) 0.6
School roofing is poor 0 -3.07 -2.15 -3.10 -2.43
School furniture is poor 1
Education expenses (K1,000) 1.6
0ROBABILITY OF SCHOOL ATTENDANCE 57.27 81.40 51.02 76.87
Note. Marginal effects are reported for variables of special interest. Bold figures indicate significance at a 5%-level, italics indicate significance at a
10%-level, whereas underlined figures indicate significance at a 15%-level.
A second finding from this analysis is that the marginal effects of income changes differ with age, and
they are larger for children from communities with unfavorable unobserved characteristics than for
their peers from communities with more favorable unobserved characteristics. A third finding is that
the income effects are higher in rural than in urban areas.
- 24 -
For rural children the income elasticity of child labor is below one. Rural children work more
frequently than urban children as they account for 197,074 out of 208,516 working children (see Table
5). Of the working children, 80% come from the four provinces (Western, Northern, Southern,
Eastern) with the highest incidence of child labor (see Table 3). Most child laborers are in the upper
end of the analyzed age-interval (see Table 2). Furthermore, Table 11 shows that the probability of
working increases by 31 percentage points if the child comes from a community with bad unobserved
characteristics, and hence it is expected that most child workers come from this type of community.
Therefore, the income effect which is calculated for a 13-year-old girl from a ‘bad’ community in the
rural part of the Northern province is probably close to being representative for the household of a
potential child worker. This marginal effect of a rise in income of K1,000, i.e. a 34% increase, is -6.61
percentage points, see Table 12.18 Therefore, even for this group of children who are most exposed to
the risk of working, the probability of working is rather insensitive towards changes in household
expenditure per capita. Roughly, a countrywide real growth of 2% through 15 years or 1% through 30
years would decrease child labor by one third.
4ABLE 3ELECTED MARGINAL EFFECTS FOR STANDARD PERSONS FROM URBAN AREAS
Average community Bad community

3CHOOLING EQUATION std.person marg. effect marg. effect marg. effect marg. effect
Non-biological relation to the household 0
head
Age rank 3 and 5
The household owns an asset 1
The household owns land 0
Expenditure per AE (K1,000) 9.6
Education expenses (K1,000) 5.1
0ROBABILITY OF SCHOOL ATTENDANCE 88.87 95.20 80.42 90.35
Note. Marginal effects are reported for variables of special interest. Bold figures indicate that the effects are significant at a 5%-level, italics indicate
significance at a 10%-level, whereas underlined figures indicate significance at a 15%-level.
The indicator variable for whether or not the household owns an asset also to some extent reflects
poverty. Owning an asset decreases the probability of working by 7-8 percentage points and increases
the probability of attending school by 10 percentage points for the children who are most exposed to
the risk of working. If a household is not allowed to borrow on the financial capital market, and if they
18
E.g. for a 13-year-old from a ‘bad’ community, this corresponds to an income elasticity of (6.61/38.47)/(1/2.9))=0.5.
- 25 -
own no assets, their only possibility of covering a sudden drop in income is to borrow on the human
capital market, i.e. let the children work instead of attending school.19
The education of the household head is often suggested as an indicator that can be used to identify
vulnerable households. It is found that one extra year of education of the child’s own household head
would decrease child labor in the ‘bad’ communities by about one percentage point and increase the
probability of school attendance by about two percentage points in rural areas. An increase in the
average education of household heads in the community has an effect of similar magnitude. In urban
areas, the effect of the education of the child’s own household head is a bit smaller, although the effect
of living in a well-educated community is of similar size.
#OST EFFECTS
The effect of reducing the education expenses by K1,000 would be an increase in school attendance by
about 3 percentage points in rural areas and an increase in school attendance by less than one
percentage point in urban areas.20 Hence, reducing school costs is also a relatively inefficient way of
reducing child labor and increasing school attendance.
In rural areas, reducing the transportation costs might be an successful way of reducing child labor and
increasing school attendance. Having an all year passable road in the community reduces child labor in
the ‘bad’ communities by more than 10 percentage points, whereas it increases school attendance by
7-9 percentage points. Similarly, the presence of a primary school in the community increases school
attendance by 7-10 percentage points. However, the effect of reducing supply constraints of secondary
education, as suggested by Lavy (1996), would have a negligible effect: decreasing the distance to
secondary schools in rural areas by 10 kilometers would increase school attendance by half a
percentage point.
3UBSIDY
The design of a subsidy to reduce child labor is important. As the analysis has shown, some groups are
more frequently using child labor than others, and also the school attendance of some groups of
children is more sensitive to changes in costs than that of others. The effectiveness of a subsidy would
obviously improve if the subsidy is targeted to the groups that use child labor the most, and those who
are most sensitive to changes in income and costs. In practice, one way to target a subsidy is to give
19
See discussion by Parsons and Goldin (1989).
20
This corresponds to elasticities of less than 0.1.
- 26 -
subsidies to households in the geographical areas where the mentioned characteristics of high child
labor and low school attendance are common.
Subsidies should be targeted to areas where school attendance is low and child labor prevalent.
Furthermore, the subsidy should not be paid to children who fail to attend school, which could be
avoided if the subsidy was paid on a daily or a weekly basis. Another possibility is subsidies like the
food-for-school programs, where the children get food at school if they are present. It is often seen that
mothers care more about the welfare of the children than fathers, and therefore, one could argue that
the money should be paid to the mothers and not the fathers.21
The results presented in Tables 9 and 10, give an indication of who should be the target of the subsidy.
It was found that including random community effects improved the model immensely. Coming from
a community with unobserved characteristics which are worse than the mean minus one standard
deviation increases the probability of working by more than 30 percentage points in some cases.
Furthermore, child labor, in particular, but also the school attendance of these children are generally
more sensitive to changes in income than that of others. Therefore, identifying the above-mentioned
communities, would be a good investment.
A subsidy could be given either in the form of an educational subsidy or an income subsidy. As the
fees in Zambia are negligible, the costs of an educational subsidy which refunds the fees for the
working children would be small in comparison to the cost of other policies. However, the previous
section showed that fees have no impact on whether or not a child attends school. Therefore, it is
plausible that this subsidy would not be successful in reducing child labor and increasing school
attendance. In addition to payment of fees comes textbooks, notebooks, and the expensive uniform,
which makes about K3,000 per term (see Table 1). On top of that comes direct costs in terms of
transportation costs, and indirect costs in terms of foregone earnings. It is likely that the parents would
need compensation for all the costs of schooling before they would decide to send the children to
school, and that would increase the cost of the subsidy substantially. The previous sub-section showed
a significant, though small, impact of decreasing the total education expenses and a larger effect of
decreasing transportation costs on school attendance.
An income subsidy could be targeted to the poor, to rural areas, or especially to the Western, Northern,
Southern, and Eastern provinces. In rural areas, there are almost one million children of primary
school age, and most of these live in households below the extreme poverty line. Any significant
income subsidy to such a large group would cost a fortune. In the Western province, in which the
incidence of child labor is the largest, there is 144,214 children of primary school age, one-third of
21
Regarding the design of a subsidy to reduce child labor, see the discussion by Tzannatos (1997).
- 27 -
which are working. In the four mentioned provinces, there are 723,449 children of which 23% works.
An income subsidy to this group, or more realistically to the poor, or to residents in rural areas of these
provinces would be slightly more viable and more effective.
It would be even more effective to identify communities with characteristics that induce child labor or
reduce school attendance. The analysis identifies some ‘fixed community effects’ that indicate who
should be the recipients of such a subsidy. School attendance is low for children residing in
communities where the household heads have a low level of education. In addition, in rural areas,
school attendance is low for children who live in communities where it is unusual to own land or
unusual to have children who are not closely biologically related to the head of the household.
In order to reduce child labor it would be more effective if the communities with ‘bad’ unobserved
characteristics could be identified. According to the analysis above, targeting those communities
would increase the efficiency of subsidization.
The incidence of child labor is not very sensitive to income. In the Eastern province, where the
sensitivity is relatively large, the effect of an increase in expenditure per AE of K1,000 would be a
reduction of the probability of child labor by about two percentage points, and about six percentage
points if the community have ‘bad’ unobserved characteristics. In urban areas and in the North-
Western and the Copperbelt provinces, the effect would be smaller.
If it was possible to give the income subsidy under the condition that the child stops working and
attends school, the subsidy would work as a compensation for lost labor income, and the effect might
be larger. However, the proposed measures of potential labor income were not successful in giving an
indication about the size of this effect.
7HAT IS LIKELY TO BE THE MOST SUCCESSFUL POLICY IN :AMBIA
If the reason for child labor is economic externalities such as failure of the market for education, or a
too low weight of child welfare in the household welfare function, the optimal policy would be to
internalize the externalities, and not to ban child labor. In either of the two cases, the problem is that
the private benefits fall below the private costs of education, whereas the social gains exceed the social
costs, and thus the household would make an inoptimal choice from society’s point of view. In these
two cases, child labor represents economic rationality, and it would be inoptimal to ban it, a better
policy would be to change parameters that would change incentives.
However, an enforced ban of child labor would be a viable policy if the reasons for child labor are
traditions or attitudes. It might be that tribal traditions and attitudes towards education play a role as to
the extent of child labor in some areas of Zambia. Actually, that could be the explanation why the
- 28 -
unobserved community effects are important. A less excessive policy would be sensitization, which
might change traditions and attitudes.
In the long run, the annual GDP loss caused by child labor may be substantial. Especially, the rural
areas would suffer because child workers mainly come from rural areas. An educational subsidy in
terms of a reduction of school fees or payments to PTAs would be relatively cheap, but it would not
reduce child labor significantly. Introducing an income subsidy, on the other hand, would be expensive
even if it has a narrow target. However, if it is possible to condition the subsidy on no work and school
attendance, the effect of a moderate subsidy would be significant for some groups of children, since
the subsidy would play the role of reducing the potential labor income or the general education
expenses. In practice ’food-for-school-programs’ or bonuses could implement this after completing a
grade.
Reducing transportation cost in rural areas would have a significant impact. Availability of public
transport and/or a passable road in the community decreases child labor and increases school
attendance. Similarly, the presence of a primary school and the walking distance measured in hours is
important.
Some effect of changing school quality was found. However, the coefficients to the quality variables
had different signs and were often not significantly different from zero. Only poor condition of school
roofing and poor condition of school furniture decreases school attendance significantly in rural areas.
Quality variables cover a whole community, and they express the opinion about quality from the
community leader only, therefore, there might be omitted variables or measurement error. Among
other things, the unobserved community effects catch the effect of unobserved school quality if it is
constant across communities. And because the unobserved community effects were found to be
important, it might imply that unobserved school quality is important, but that cannot be said for
certain.
In the case of Zambia, it is found that the poor, children from rural areas, and from the Western,
Northern, Southern, and Eastern provinces, are more likely to work than others. Therefore, targeted
income subsidies, and an effort to reduce transportation costs or enhance quality are expected to be
more effective for these groups than for others.
#ONCLUSION
A cost-benefit analysis predicts a total GDP loss of 1-2% from using child labor instead of sending the
children to school. This is rather substantial in a country where the annual growth is occasionally
negative. The most worrisome fact about this GDP loss is that it takes place in rural areas where most
- 29 -
of the poor live. This observation is a good motivation for scrutinizing the reasons for child labor and
lack of school attendance.
In this paper, the characteristics of child workers and the determinants of the incidence of child labor
in Zambia have been analyzed. As expected, more child workers were found among the poor than
among the better off, and more child workers among the older children than among the young
children. Furthermore, more child labor was found in rural areas than in urban areas, which is a
general characteristic of child labor in Africa. In Zambia, child workers are concentrated in the
Western, Northern, Eastern, and Southern provinces, and the incidence of child labor is the highest in
the Western province.
In contrast to what is said in the theoretical literature, this empirical analysis for the case of Zambia
shows that child labor is not particularly responsive to changes in income. The effect of an income
change is significantly positive for most groups of children, and it is somewhat larger for children
from communities with ‘bad’ unobserved characteristics than it is for others. Surprisingly, child labor
and schooling are only a moderately sensitive to changes in education expenses and quality of
schooling, if the physical condition of the schools, availability of books, and the number of shifts can
be taken as a valid indicator of that.
The analysis suggests certain policy recommendations. An income subsidy is a useful device to reduce
the incidence of child labor. However, even in the case of narrow targeting, an income subsidy is
rather expensive, and it should be substantial to have any effect at all. A viable alternative is to target
the subsidy to the rural parts of the four mentioned provinces, and insert a condition that the children
in the households receiving the subsidy must attend school and are not allowed to work at all.
Changing economic incentives through poverty reduction, introduction of income subsidies or

enhancing school quality might have an effect, but based on the present analysis, other policies are
expected to be more effective. An operational approach would be to identify and visit the communities
where child labor is common, and investigate what community characteristics affect their behavior.
This field work would uncover the unobserved community factors that the analysis finds so important
for the use of child labor. It would reveal whether sensitization is needed to change traditions or
attitudes, or whether school quality, credit availability or infrastructure should be improved upon; a
question, which is beyond the scope of the present analysis.
- 30 -
2EFERENCES
Andersson, P.-Å., and S. Kayizzi-Mugerwa, 1993. “External Shocks and the Search for diversification
in Zambia.” in (eds.) M. Blomström and M. Lundahl %CONOMIC #RISIS IN !FRICA 0ERSPECTIVES ON
0OLICY 2ESPONSES Routledge: London.
Bonnet, M., 1993. “Child Labor in Africa.” )NTERNATIONAL ,ABOR 2EVIEW 132: 371-389.
Bratton, M., 1994. “Economic Crisis and Political Realignment in Zambia.” in (ed.) J.A. Widner
%CONOMIC #HANGE AND 0OLITICAL ,IBERALIZATION IN 3UB 3AHARAN !FRICA Johns Hopkins UP:
Baltimore.
Canagarajah, S., and Nielsen, H. S., eds., 1998. #HILD ,ABOR IN !FRICA. Forthcoming. Department for
Human Development and Social Protection. The World Bank: Washington D.C.
Canagarajah, S. and Coulombe, H., 1998. “Child Labor and School Attendance in Ghana.” in (eds.) S.
Canagarajah and H. S. Nielsen: #HILD ,ABOR IN !FRICA. Forthcoming. Department for Human
Development and Social Protection. The World Bank: Washington D.C.
Central Statistical Office, 1994. 3OCIAL $IMENSIONS OF !DJUSTMENT 0RIORITY 3URVEY )) 4ABULATION
2EPORT CSO, Lusaka, Zambia.
Economist Intelligence Unit , 1993. #OUNTRY 0ROFILE :AMBIA.
Fallon, P. and Tzannatos, Z., 1998. #HILD ,ABOR )SSUES AND $IRECTIONS FOR THE 7ORLD "ANK. HDNSP,
The World Bank.
Graham-Brown, S., 1991. %DUCATION IN THE $EVELOPING 7ORLD #ONFLICT AND #RISIS World University
Service. Longman: UK.
Grootaert, C. and Kanbur, R., 1995. “Child Labor: An Economic Perspective.” )NTERNATIONAL ,ABOUR
2EVIEW 134: 187-203.
International Monetary Fund, 1996. )NTERNATIONAL &INANCIAL 3TATISTICS 9EARBOOK.
Jensen, P. and Nielsen, H.S. , 1997. “Child Labour or School Attendance ? Evidence from Zambia.”
*OURNAL OF 0OPULATION %CONOMICS (10): 407-24.
Kelly, M. J., 1991. “Education in a Declining Economy: The Case of Zambia 1975-85.” %CONOMIC
$EVELOPMENT )NSTITUTE $EVELOPMENT 0OLICY #ASES !NALYTICAL #ASE 3TUDIES no. 8 The World Bank,
Washington D.C.
Kelly, M. J., ed., 1992. “National Conference on Education for All: Issues, Challenges and the Way
Ahead for Zambia.“ #ONFERENCE 0APERS Vol. II, National task Force on Education for All,
Ministry of Education.
Lavy, V., 1996. “School Supply Constraints and Children’s Educational Outcomes in Rural Ghana.”
*OURNAL OF $EVELOPMENT %CONOMICS 51: 291-314.
Mason, A. and Khandker, S. R., 1998. “Children’s Work, Opportunity Cost, and Schooling in
Tanzania.” 7ORKING 0APER The World Bank.
Ministry of Education, 1995. %DUCATION 3TATISTICS "ULLETIN Government of the Republic of
Zambia.
Ministry of Education, 1996. )NVESTING IN /UR 0EOPLE )NTEGRATED %DUCATION 3ECTOR )NVESTMENT
0ROGRAMME Government of the Republic of Zambia.
Ministry of Education, 1996a. %DUCATING /UR &UTURE .ATIONAL 0OLICY ON %DUCATION, Government of
the Republic of Zambia.
Ministry of Education, 1996b. Education Expenditure Review, $RAFT.
- 31 -
Montgomery, M., Kouamé, A., and Oliver, P., 1995. "The Trade-off between Number of Children and
Child Schooling: Evidence from Côte d'Ivoire and Ghana", ,IVING 3TANDARDS -EASUREMENT 3URVEY
, The World Bank: Washington D.C.
M.V.Foundation, 1996. “The M.V.Foundation’s Efforts to Mainstream the Child Workers in
Shankarpally Mandal.” 4HE !DMINISTRATOR 41: 125-135.
Nielsen, H. S. and Westergård-Nielsen, N., 1998. “Returns to Schooling in LDCs: New Evidence from
Zambia.” #,3 70 98-11 Centre for Labour Market and Social Research, Aarhus, Denmark.
Parsons, D. O. and Goldin, C., 1989. “Parental Altruism and Self-Interest: Child Labor among Late
Nineteenth-Century American Families.” %CONOMIC )NQUIRY 27: 637-59.
Rivers, D. and Vuong, Q. H., 1988. “Limited Information Estimators and Exogeneity Tests for
Simultaneous Probit Models.” *OURNAL OF %CONOMETRICS 39: 347-366.
Sinha, S., 1996. “Child Labour and Education Policy in India.” 4HE !DMINISTRATOR 41: 17-29.
Tzannatos, Z., 1997. “Child Labor and School Enrollment in Thailand in the 1990s.” 5NPUBLISHED
PAPER World Bank.
World Bank, 1994. :AMBIA 0OVERTY !SSESSMENT Vol. I: Main Report, Report No. 12985-ZA. Human
Resources Division, Southern Africa Department, Africa Regional Office.
World Bank, 1997. 7ORLD $EVELOPMENT )NDICATORS . CD-ROM. World Bank: Washington D.C.
- 32 -
Working
Paper
95-01 Christian Belzil: Contiguous Duration Dependence and Nonstationarity in Job

Search
95-02 Christian Belzil: Unemployment Insurance and Unemployment Over Time: An

Analysis with Event History Data.
95-03 Christian Belzil: Unemployment Duration Stigma and Reemployment Earnings.
95-04 Christian Belzil: Relative Efficiencies and Comparative Advantages in Job

Search.
Published in *OURNAL OF ,ABOR %CONOMICS, 14, no. 1, pp. 154-173.
95-05 Niels Henning Bjørn: Causes and Consequences of Persistent Unemployment.
95-06 Nicholas M. Kiefer and Mark F.J. Steel: Bayesian Analysis of the Prototypal
Search Model.
Published in *OURNAL OF "USINESS AND %CONOMIC 3TATISTICS, 16, no. 2, pp.
178-186.
95-07 Nicholas M. Kiefer, Ranjini Natarajan and Charles E. McCulloch: Maximum

Likelihood for the Multinomial Probit Model.
95-08 Christian Belzil and Philip Hergel: Fertility and the Human Capital Loss of
Non-Participation
95-09 Christian Belzil, William A. Sims and Philip Hergel: Endogeneity, Self-Selecti-
vity and the Sensitivity of Female Earnings to Non-Participation.
95-10 Paul Bingley, Niels Henning Bjørn and Niels Westergård-Nielsen: Wage Mobi-
lity in Denmark 1980-1990.
95-11 Audra J. Bowlus, Nicholas M. Kiefer and George R Neumann: Estimation of

Equilibrium Wage Distributions with Heterogeneity.
Published in *OURNAL OF !PPLIED %CONOMETRICS, no. 10, pp. 119-131.
95-12 Anders Björklund and Tor Eriksson: Unemployment and Mental Health: Evi-
dence from Research in the Nordic Countries
Published in 3CANDINAVIAN *OURNAL OF 3OCIAL 7ELFARE, 7, pp. 219-235.
Gustav Wieds Vej 10C Phone: +45 8942 2350 Email: cls@cls.dk
DK-8000 Århus C Fax: +45 8942 2365 WWW: http://www.cls.dk

95-13 Melvyn G. Coles and John G. Treble: Here Today, Gone Tomorrow: Calculat-
ing the Price of Worker Reliability.
95-14 Christian Belzil: Employment Reallocation, the Return to Human Capital and
the Allocation of Workers Between Expanding and Declining Firms.
95-15 John T. Addison and Jean-Luc Grosso: Job Security Provisions and Employ-
ment: Revised Estimates.
95-16 John T. Addison and McKinley L. Blackburn: A Puzzling Aspect of the Effect
of Advance Notice on Unemployment.
95-17 Peder J. Pedersen and Nina Smith: The Welfare State and the Labour Market.
95-18 Mette Lausten: Inter-Industry Wage Differentials in Denmark ?
96-01 Mark Yuying An: Log-concave Probability Distributions: Theory and Statistical
Testing.
Published in: *OURNAL OF %CONOMIC 4HEORY 80, p. 350-3369, 1998.
96-02 Audra Bowlus, Nicholas M. Kiefer and George R. Neumann: Fitting Equilibri-
um Search Models to Labour Market Data.
96-03 Karsten Albæk, Mahmood Arai, Rita Asplund, Erling Barth and Erik Strøyer
Madsen: Employer Size-Wage Effects in the Nordic Countries.
96-04 Bent J. Christensen and Nicholas M. Kiefer: Inference in Non-Linear Panels

with Partially Missing Observations: The Case of the Equilibrium Search
Model.
Published in *OURNAL OF %CONOMETRICS, 1997, no. 79, pp. 201-219.
96-05 Michèle Naur and Nina Smith: Cohort Effects on the Gender Wage Gap in
Denmark.
96-06 Elizabeth J. Cunningham: The Relationship between Recruiting and Screening

within the Employer Search Framework
96-07 Tim Barmby and Nina Smith: Household Labour Supply in Britain and
Denmark: Some Interpretations Using a Model of Pareto Optimal Behaviour.
96-08 Michael Rosholm: Unemployment Duration over the Business Cycle.
96-09 Mark Yuying An and Ming Liu: Structural Analysis of Labor Market Trans-
itions Using Indirect Inference

96-10 Paul Bingley and Niels Westergård-Nielsen: Worker and Plant Wages: Esti-
mates from a Multi-Level Model
96-11 Paul Bingley and Gauthier Lanot: Danish Private Sector Wage Policies and
Male Retirement Decisions
96-12 George R. Neumann and Gauthier Lanot: Measuring Productivity Differences

in Equilibrium Search Models.
96-13 Tor Eriksson: Executive Compensation and Tournament Theory: Empirical

Tests on Danish Data.
96-14 Peter Jensen and Helena Skyt Nielsen: Child Labour or School Attendance ?
Evidence from Zambia.
Published in *OURNAL OF 0OPULATION %CONOMICS, 10, pp. 407-424.
96-15 Ebbe Krogh Graversen: Male and Female Labour Supply in Denmark.
96-16 Tor Eriksson and Markus Jäntti: The Distribution of Earnings in Finland 1971-
1990.
Published in %UROPEAN %CONOMIC 2EVIEW, 1997, no. 41, pp. 1763-1779.
96-17 Ebbe Krogh Graversen: Measuring Labour Supply Responses to Tax Changes
by Use of Exogenous Tax Reforms.
97-01 Report 1993 - 1996.
97-02 Paul Bingley and Ian Walker: Labour Supply with In-Work and In-Kind Trans-
fers.
97-03 Paul Bingley and Ian Walker: Household Unemployment and the Labour Sup-
ply of Married Women.
97-04 Christian Belzil: Job Creation and Destruction, Worker Reallocation and
Wages.
97-05 Christian Belzil: The Dynamics of Female Time Allocation upon a First Birth
97-06 Christian Belzil and Jörgen Hansen: Estimating the Returns to Education from
a Non-Stationary Dynamic Programming Model
97-07 Niels Westergård-Nielsen and Anders Rue Rasmussen: Apprenticeship Train-

ing in Denmark - the impacts of subsidies.

97-08 H. Bunzel, B.J. Christensen, P. Jensen, N.M. Kiefer, L. Korsholm, L. Muus,
G.R. Neumann, M. Rosholm: Specification and Estimation of Equilibrium
Search Models.
97-09 Ebbe Krogh Graversen: Work disincentive effects of taxes among Danish
married men and women
97-10 Jukka Vittaniemi: Top Executive Compensation and Company Performance in

Finland.
97-11 Peder J. Pedersen and Nina Smith: Trends in the Danish Income Distribution,
1976-90.
97-12 Ronald L. Oaxaca and Michael R. Ransom: Identification in Detailed Wage

Decompositions
97-13 Bent J. Christensen and Nicholas M. Kiefer: Panel Data, Local Cuts and
Orthogeodesic Models
97-14 Michael Rosholm: The risk of marginalization in the labour market:

Application of a three state dependent competing risks duration model.
97-15 Helena Skyt Nielsen and Michael Rosholm: The Incidence of Unemployment:
Identifying Quits and Layoffs
97-16 Tor Eriksson: Long-Term Earnings Mobility of Low-Paid Workers
97-17 Lars Korsholm: The Semiparametric Normal Variance-Mean Mixture Model
98-01 Helena Skyt Nielsen: Two Notes on Discrimination and Decomposition
98-02 Esben Agerbo, Tor Eriksson, Preben Bo Mortensen and Niels Westergård-Niel-
sen: Unemployment and mental disorders - an empirical analysis
98-03 Birthe Larsen: Minimum Wages, Technological Progress and Loss of Skill
98-04 Kevin T. Reilly and Tony S. Wirjanto: Does More Mean Less ? The
Male/Female Wage Gap and the Proportion of Females at the Establishment
Level
98-05 Helena Skyt Nielsen: Low Demand for Primary Education: Traditions or
Economic Incentives ?
98-06 Ebbe Krogh Graversen and Nina Smith: Labour supply, overtime work and
taxation in Denmark

98-07 Christian Bontemps, Jean-Marc Robin, and Gerard J. van den Berg: Equilibri-
um Search with Continuous Productivity Dispersion: Theory and Non-Parame-
tric Estimation.
98-08 Mark Y. An, Bent J. Christensen, and Nicholas M. Kiefer: Approximate

Distributions in Essentially Linear Models.
98-09 Morten Bennedsen: Political Ownership.
98-10 Helena Skyt Nielsen and Niels Westergård-Nielsen: Returns to Schooling in

LDCs: New Evidence from Zambia.
98-11 Lars Korsholm: Likelihood Ratio Test in the Correlated Gamma-Frailty Model.
98-12 Mark Y. An: Statistical Inference of a Bivariate Proportional Hazard Model

with Grouped Data.
98-13 Lars Korsholm: An Equilibrium Search Model with Human Capital

Accumulation
98-14 Dale T. Mortensen: Equilibrium Unemployment with Wage Posting: Burdett-

Mortensen Meet Pissarides
98-15 Helena Skyt Nielsen: Child Labor and School Attendance: Two Joint Deci-
sions.

)33.
#%.42% &/2 ,!"/52 -!2+%4 !.$ 3/#)!, 2%3%!2#(

3CIENCE 0ARK !ARHUS 'USTAV 7IEDS 6EJ # !ARHUS # $ENMARK
0HONE &AX %MAIL CLS CLSDK 777 HTTPWWWCLSDK
#,3 IS FUNDED BY 4HE $ANISH .ATIONAL 2ESEARCH &OUNDATION

#Hild, Abor and 3chool !ttendance 4wo Oint $ecisions: Helena Skyt Nielsen

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

#Hild, Abor and 3chool !ttendance 4wo Oint $ecisions: Helena Skyt Nielsen

Uploaded by

Copyright:

Available Formats

#HILD ,ABOR AND 3CHOOL !

Helena Skyt Nielsen

Working Paper 98-15

Centre for Labour Market and Social Research

Editor: Henning Bunzel

Copyrights: Helena Skyt Nielsen

(ELENA 3KYT .IELSEN

Working Paper 98-15

*%, CLASSIFICATIONS: I21, J13, J24

+EYWORDS Africa, Zambia, child labor, school attendance, community effects

 )NCIDENCE OF #HILD ,ABOR 3OME 4ABULATIONS

 $ISTRIBUTION OF CHILD LABOR ACROSS AGE

 'EOGRAPHICAL DISTRIBUTION OF CHILD LABOR

&IGURE . )NCIDENCE OF CHILD LABOR IN :AMBIAN PROVINCES

North Western Eastern

4ABLE . !CTIVITY DISTRIBUTION BY PROVINCE

4ABLE . !CTIVITY DISTRIBUTION BY LEVEL OF HOUSEHOLD INCOME PER ADULT EQUIVALENT

 /CCUPATION AND INCOMES OF WORKING CHILDREN

4ABLE  /CCUPATION OF THE WORKING CHILDREN

4ABLE . -EAN PREDICTED POTENTIAL INCOMES FROM CHILD LABOR +WACHAMONTH 

4ABLE . #OSTS AND BENEFITS OF CHILD LABOR AND PRIMARY EDUCATION A

Choice Year 1-7 Year 8-50

The dependent variables are two indicator variables:

All Rural Urban Girls Boys

 5NOBSERVED COMMUNITY EFFECTS

 )NCOME EFFECT

Average community Bad community

 #OST EFFECTS

 7HAT IS LIKELY TO BE THE MOST SUCCESSFUL POLICY IN :AMBIA 

Changing economic incentives through poverty reduction, introduction of income subsidies or

95-01 Christian Belzil: Contiguous Duration Dependence and Nonstationarity in Job

95-02 Christian Belzil: Unemployment Insurance and Unemployment Over Time: An

95-03 Christian Belzil: Unemployment Duration Stigma and Reemployment Earnings.

95-04 Christian Belzil: Relative Efficiencies and Comparative Advantages in Job

95-05 Niels Henning Bjørn: Causes and Consequences of Persistent Unemployment.

95-07 Nicholas M. Kiefer, Ranjini Natarajan and Charles E. McCulloch: Maximum

95-11 Audra J. Bowlus, Nicholas M. Kiefer and George R Neumann: Estimation of

DK-8000 Århus C Fax: +45 8942 2365 WWW: http://www.cls.dk

95-18 Mette Lausten: Inter-Industry Wage Differentials in Denmark ?

96-04 Bent J. Christensen and Nicholas M. Kiefer: Inference in Non-Linear Panels

96-06 Elizabeth J. Cunningham: The Relationship between Recruiting and Screening

96-08 Michael Rosholm: Unemployment Duration over the Business Cycle.

DK-8000 Århus C Fax: +45 8942 2365 WWW: http://www.cls.dk

96-12 George R. Neumann and Gauthier Lanot: Measuring Productivity Differences

96-13 Tor Eriksson: Executive Compensation and Tournament Theory: Empirical

97-01 Report 1993 - 1996.

97-07 Niels Westergård-Nielsen and Anders Rue Rasmussen: Apprenticeship Train-

DK-8000 Århus C Fax: +45 8942 2365 WWW: http://www.cls.dk

97-10 Jukka Vittaniemi: Top Executive Compensation and Company Performance in

97-12 Ronald L. Oaxaca and Michael R. Ransom: Identification in Detailed Wage

97-14 Michael Rosholm: The risk of marginalization in the labour market:

97-16 Tor Eriksson: Long-Term Earnings Mobility of Low-Paid Workers

97-17 Lars Korsholm: The Semiparametric Normal Variance-Mean Mixture Model

98-01 Helena Skyt Nielsen: Two Notes on Discrimination and Decomposition

DK-8000 Århus C Fax: +45 8942 2365 WWW: http://www.cls.dk

98-08 Mark Y. An, Bent J. Christensen, and Nicholas M. Kiefer: Approximate

98-09 Morten Bennedsen: Political Ownership.

98-10 Helena Skyt Nielsen and Niels Westergård-Nielsen: Returns to Schooling in

98-12 Mark Y. An: Statistical Inference of a Bivariate Proportional Hazard Model

98-13 Lars Korsholm: An Equilibrium Search Model with Human Capital

98-14 Dale T. Mortensen: Equilibrium Unemployment with Wage Posting: Burdett-

DK-8000 Århus C Fax: +45 8942 2365 WWW: http://www.cls.dk

#%.42% &/2 ,!"/52 -!2+%4 !.$ 3/#)!, 2%3%!2#(

+EYWORDS Africa, Zambia, child labor, school attendance, community effects

)NCIDENCE OF #HILD ,ABOR 3OME 4ABULATIONS

$ISTRIBUTION OF CHILD LABOR ACROSS AGE

'EOGRAPHICAL DISTRIBUTION OF CHILD LABOR

&IGURE . )NCIDENCE OF CHILD LABOR IN :AMBIAN PROVINCES

4ABLE . !CTIVITY DISTRIBUTION BY PROVINCE

4ABLE . !CTIVITY DISTRIBUTION BY LEVEL OF HOUSEHOLD INCOME PER ADULT EQUIVALENT

/CCUPATION AND INCOMES OF WORKING CHILDREN

4ABLE /CCUPATION OF THE WORKING CHILDREN

4ABLE . -EAN PREDICTED POTENTIAL INCOMES FROM CHILD LABOR +WACHAMONTH

4ABLE . #OSTS AND BENEFITS OF CHILD LABOR AND PRIMARY EDUCATION A

5NOBSERVED COMMUNITY EFFECTS

)NCOME EFFECT

#OST EFFECTS

7HAT IS LIKELY TO BE THE MOST SUCCESSFUL POLICY IN :AMBIA