WpS 190o5'
POLICY RESEARCH WORKING PAPER 1905
Child Labor in Cote d' Ivoire Most children in
perform some kind of vork
In rural areas, more tean fohjr
Incidence and Determinants of five children work, with
only a third combining work
Christiaan Grootaert with schooling.
The World Bank
Social Development Department
March 1998
LO iCE RFSEARCr7 WORKING PAI'ER 1905
Summary findings
Child labor in C6te d'lvoire increased in the 1980s * The education- and eimployviernt sratuQ or the parents
because of a severe economic crisis. Two out of three (low parental education IS a good targeting variable for
urban children aged 7 to 17 work; half of them also interventions).
attend school. In rural areas, more than four out of five * The .ivailabilit,; of wi:himn-houschold eirnploynment
children work, but only a third of them manage to opportunities.
combine work with schooling. * The household's poverty status.
Full-time work is less prevalent, but not negligible. The household's location (calling for geographical
Roughly 7 percent of urban children work full time (an targeting).
average 46 hours a week). More than a third of rural With improved martvo-cononihc gtovto, it is hoped,
children work full time (an average of 35 hours a week), child labor will decline -hut a significant d(crline could
with the highest incidence in the Savannah region. take several generanions. Nleanwhile, it ii ortant to:
The incidence of such full-time viork rises with age but Use a graduai approach toward rh elimination of
is by no means limited to older children. The average age child wvork by "iun'g initial intervent',,ns at racilitating
of the full-time child worker in Cote d'Ivoire is 12.7. combined wvork andn schooling.
These children have received an average 1.2 years of e Support the devcvlopment of hone erl errises as
schooling. That child is also more likely to be ill or par; of poverry alley iarn prngram.s, but ombine it with
injured and is less likely to receive medical attention than incentives for school attendance.
other children. Make school ho-nrs and vacation periods flexible
Urban children in the interior cities are far more likely (accommodating harvest times) in rural arcas. l hiis would
to work and their working hours are much longer. also improve children s health.
Among rural children, those in the Savannah region Improve rural school attendance bhs ng a school
(where educational infrastructure lags far behind the rest in the village rather thall I to 5 kiionmerers aura!V.
of the country) are most likely to work. * Improve edcLcational investrmeurct im tlic 8Svannah.
Five factors affect a household's decision to supply
child labor:
The age and gender of the child (girls are more
likely to work, especially when the head of household is
a worran).
This paper is a product of the Social Development Department. The study was funded by the Bank's Research Support
Budget under the research project "Child Labor: What Role for Demand-Side Interventions" (Rl'O 680-64). Copies of the
paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Gra-cie Ochieng,
room MC5-158, telephone 202-473-1123, fax 202-522-3247, Internet address gochicng( worldhank.org March 1998.
(75 pages)
The Policy Research Working Paper Series disseminates the findings of wtork It1 Progress ,'o ,ck exckl',>ca' Zg,) olt
development issues. An obrective of the series is to get the findings Out quickly, cZ Zp i ..S less ,'i'o' , I ' e
| papers carry the namnes of the authors and should be cited accordingly. Tie fiisi , ss ir7ter�r's, a.i rwh:a e i . ih;
paper are entirely those of the authors. They do not necessarily represent the vieic thf)ri Pink. cS iX'' !), s ,
countries they represent.
Produced by the Policy Research DisscimaatrionV enell.r
Social Development Department
Environmentally and Socially
Sustainable Development Network
The World Bank
Child Labor in C8te d'Ivoire:
Incidence and Determinants
Christiaan Grootaert*
* Senior Economist, Social Development Department, The World Bank.
The author would like to thank the National Institute of Statistics of C8te
d'Ivoire for making the data set available for this study. Thanks also are due to
Gi-Taik Oh, who managed the data base and programmed all the computer
calculations, to Kimberly Cartwright and Harry Patrinos-colleagues in the
comparative study "Child Labor: What Role for Demand-side Interventions"
(RPO 680-64) - for many helpful discussions on the topic of child labor, and to
Ravi Kanbur, Roger Key, Felicia Knaul and John Mcintire who provided
valuable comments on an earlier draft of this paper.
Child Labor in C6te d'Ivoire:
Incidence and Determinants
TABLE OF CONTENTS
Pae
1. Introduction .......................................... . 1
2. Trends in Child labor in C6te d'Ivoire in the 1980s ............................. 6
3. Child Labor and Schooling in 1988 ................................... 15
4. Multivariate Analysis .................. ........................ 27
A Model for the Determinants of Child Labor ................................... 31
Results for Urban Areas ........................................... 40
Results for Rural Areas .......................................... 52
5. Conclusions ...................................... .... 63
Appendix: Multinomial Logit Results.. . . ... 68
Referencesf.... 73
1
1. Introduction
Child labor is prevalent in the developing world but the estimates vary
widely. The ILO estimated that in 1990 there were about 78 million
economically active children under the age of 15 (Ashagrie, 1993). UNICEF
(1991) estimated that there were 80 million children aged 10-14 who undertook
work so long or onerous that it interfered with their normal development.
Recently, the ILO (1996) has increased its estimates to 120 million working
children in the ages 5-14 who are fully at work. If part-time work is included,
the total number of working children approaches 250 million. Labor force
participation rates for children 5-14 vary greatly from country to country,
ranging from close to zero in most developed countries to an average of 20% in
Latin America and 40% in Africa.
Most empirical work on the incidence and determinants of child labor
covers a sub-national area, often one or a few villages, at best a province or
region. (Reviews of the child labor literature can be found in ILO (1986) and
Grootaert and Kanbur (1995)). The dearth of direct data on child labor has led
many researchers to focus on the determinants of school attendance, even
though it is recognized that school attendance is not the "inverse" of child labor.
Nevertheless, much of this literature views schooling as the most important
2
means of drawing children away from the labor market (Siddiqi and Patrinos,
1995).
The range of usable policy variables extends however well beyond
education. In their review of these variables, Grootaert and Kanbur (1995)
discuss, the role of fertility behavior, the household's risk management, and
government policies with respect to social expenditure and population control as
variables which affect the supply of child labor. On the demand side, the
structure of the labor market and the prevailing production technology are the
two main determinants of child labor. To these economic variables must be
added the legislative framework (nationally and internationally), which usually
involves a ban on child labor that is rarely enforced effectively, and social factors
such as advocacy, awareness raising and community-based efforts to help child
workers and street children. As a final factor, war and civic strife often draw
children into militia.
Each one of these variables offers several policy angles and, as discussed
by Grootaert and Kanbur (1995), conventional welfare economics provides a
useful framework to analyze child labor issues. The starting point is the
household decision making process which must allocate children's time between
labor and non-labor activities, taking into account the private returns to each.
Each household will allocate the time of its children to wherever the perceived
private return is highest, untfl the marginal return is equalized across all uses of
3
child time.' The crucial question is whether, at that point, equality is achieved
with the marginal social return. When the private return of child labor exceeds
the social retum, there is arguably "too much" child labor and interventions are
called for. These can occur in the labor market itself, in the market for
education, or elsewhere, depending upon where the market failure occurs.
The key element to come out of the welfare economic analysis is that there
is not a simple, or even a dominant, way of approaching the elimination of child
labor. A single intervention has the potential of making the working child and
its household worse off, if the intervention is not where the market failure
occurs. One example is a ban on child labor imposed when child labor occurs as
a result of a failure in the education market. This situation can lead to a further
reduction of the child's already limited opportunity set since after the ban
(assuming it is enforced) the child can neither work nor attend school. Indeed,
the ban does not address the failure in the education market. Hence, an array of
policy instruments is likely to be required, addressing different aspects of
market failures, and taking both efficiency and distributional considerations into
account.
l It is to be noted here that the private returns in question are those to the household, which
can differ from the returns to the dcild itself. As Grootaert and Kanbur (1995) explain, the
household's utility function can be dominated by the head of household and the welfare of
the cbild may have low weights in the decision malkng process. These weights are a function
of the nature of the intra-household bargaining process.
4
Empirically, the challenge is to estimate a model of the child labor
decision which captures the household's behavior with respect to labor market
participation, education, fertility, risk management and other relevant factors.
The paper below presents one such approach, relying on a reduced-form model
which portrays the child labor decision as a three-stage sequential process. An
alternative view and model of the child labor decision as a simultaneous process
is also presented in an appendix. The case study is for C6te d'Ivoire in 1988, in
an economic setting of severe recession and, as a result, rising child labor.
One of the main difficulties in furthering the empirical analysis of the
determinants of child labor is the dearth of national household surveys that
include questions on labor market participation addressed at adults and children
in the household. Most labor force surveys use a minimum age cut-off of 14 or
15 years, so that, by definition, most official labor force statistics will exclude
child labor. This age cut-off is a matter of national practice, and not the result of
international guidelines. The latter do indicate that the measurement of the
economically active population must use a minimum age limit, but no particular
value is specified. The guidelines mention that countries where a large
proportion of the labor force works in agriculture should use a lower age limit
than highly industrialized countries (Hussmanns et al, 1990).
Because of this, multi-purpose household surveys are often the best
source of data on child labor. Such surveys include a wide variety of questions
5
on the socioeconomic conditions of the household, and employment questions
are often asked with a lower age cut-off. The C8te d'Ivoire data set used in this
study is a multipurpose household survey with national coverage, which
recorded labor force participation for all household members aged 7 years and
above.
The results of the case study confirm the validity of a multi-angled policy
approach towards the elimination of child labor. In particular, the case is made
for a gradual policy approach, whereby initially the combination of child labor
and schooling is made more attractive, relative to only work. This presents a
more realistic approach for poor households who are likely to select work
options for their children, and avoids interventions which can make the child
worse off.
2. Trends in Child Labor in C6te d'Ivoire in the 1980s
The investigation of the incidence and determinants of child labor in C6te
d'Ivoire in this paper is based on the 1988 C6te d'Ivoire Living Standards Survey
(CILSS). The CILSS was canvassed annually between 1985 and 1988 over a
representative sample of 1600 households. The survey collected detailed
information on employment, income, expenditure, assets, basic needs and other
socioeconomic characteristics of households and their members. Over the four
years, coverage and methodology of the survey were held constant so that
6
results are comparable over time. The survey is described in more detail in
Grootaert (1986, 1993).
The years 1985-88 are of particular importance in the recent economic
history of C6te d'Ivoire. Throughout the eighties the country experienced an
economic recession. The downturn is attributed to the collapse of the world
prices of coffee and cocoa-the country's two main export crops-in the late
seventies, and to unsustainable macro-economic policies (Demery, 1994). Of the
decade, 1988 was one of the worst years. Between 1987 and 1988, GDP per
capita fell by 5% in real terms, but private consumption fell by almost 17%, and
the poverty rate rose from 35% to 46% (Grootaert, 1995). At the same time, the
labor market underwent drastic changes. As a result of the recession,
employment in the formal sector (including the public sector) shrunk by 14%.
Many of the workers who were laid off as well as the vast majority of labor
market entrants, had to find jobs in the informal sector. Between 1980 and 1990,
employment in the informal sector more than doubled, and unemployment
nearly tripled. The informal sector was characterized by underemployment, low
productivity work, and low earnings -on average, one-fifth of earnings in the
formal economy. The incidence of poverty among informal sector workers was
hence high and rose rapidly during the 1980s (Grootaert, 1996).
7
Table 1. Labor force participation rates
Abidian Other Cities I Rural Areas C8te d'Ivoire
1985
Very poor 29.5 47.1 67.9 63.7
Mid-poor 28.1 40.5 64.7 59.6
Non-poor 33.8 40.0 64.9 50.3
AI 1 33.6 .. 40.6 65.2 1_ 53.4
1988
Very poor _ 43.8 69.0 66.4
Mid-poor 21.9 43.3 66.9 58.0
Non-poor 33.5 37.9 64.2 49.7
All 32.0 40.0 66.1 54.4
Table 1 shows the labor force participation rates in 1985 and 1988, by
poverty status. Labor force participation is defined here as any form of
economic activity for wages or own account, whether paid or not. Labor force
participants thus consist of employees, employers, self-employed, and unpaid
family workers. Two conclusions stand out from Table 1. First and foremost,
labor force participation is inversely related to welfare level: the very poor have
the highest rate of participation. This immediately dispenses with the
hypothesis that the poor are poor because they do not participate in the labor
force. Second, participation is much higher in rural than urban areas. This is
consistent with the previous finding, since poverty in Cote d'Ivoire is higher in
rural areas. (The sole exception seems to be Abidjan, where the participation of
the non-poor is higher.)
Between 1985 and 1988, labor force participation did not change much.
Only the very poor showed a slight increase, which suggests that the increase in
8
extreme poverty during the sharp economic downtum in 1987-88 was not due to
falling employment among the poorest
The disappearance of high-wage and stable jobs in the 1980s led many
households to implement risk-reducing strategies, by diversifying income
sources across jobs and across household members. This is reflected in the labor
force participation rates by age and gender (Table 2). At the national -level, the
participation rate of adult men -the key income earners for the household -was
79% in 1985. Again, this rate is inversely related to poverty status, and reaches
90% for the very poor. This inverse relation is in fact observed for all
age/gender groups, but is especially pronounced for children and adolescents.
Among non-poor households, 14% of children and 36% of adolescents worked,
but for very poor households the corresponding percentages are 31% and 73%.
By and large this relationship holds regionally as well, although the difference
between poor and non-poor is greater in urban than in rural areas. (In Abidjan,
the relation cannot be established because there are too few observations of poor
households.)
9
Table 2. I I
AjIis1 I I.pn*1. A Anlkqj1 .....JE!eriy
_____________ (7-14 years) ,J (15-18 years) j (19-59 years) j (1949 years)
1985
Abidjan
Very poor 0. 0.0* I 69.3* 44.4" I 0.0*
Mid-poor I 0. 0.0" j 33.0 55.8 36.4"
Non-poor 2. 11.5 65.8 41.0 52.3
All 11.3 64.6 41.4 51.7
Other Cities
Very poor 10.5 I 30.3 j 88.1 I 66.5 76.9
Mid-poor 9.4 26.0 79.4 58.6 56.6
Non-poor I 152.8157.9
All 5.6 22.8 75.3 54.8 583
Rural Areas
Very poor 36.1 86.6 90.8 I 83.7 72.7
Mid-poor I 31.7 73.8 86.6 81.3 74.9
Non-poor 126.1169.3190.1180.7 72.4
All IiI89 8L3{ 73.0
C6te d'Ivoire
Very poor 30.6 72.6 89.7 80.4 73.1
Mid-poor 26.8 61.5 82.0 76.9 72.6
Non-poor 14.0 35.7 77.6 62.5 67.3
All 18.5 44.0 79.2 67.0 69.0
1988
Abidjan
Mid-poor I 0. 10.0154.8130.6 42.9
Non-poor 0. 7.3 67.4 38.3 45.6
All 10.81 7.7 166A 1 37.41451
Other Cities
Very poor 24.0 53.8 80.0 J 57.5 40.0
Mid-poor 7.5 32.8 82.1 62.9 66.2
Non-poor 3.4 18.3 79.4 51.8 59.3
All 6.3 23.7 j 80.2 56.2 60.3
Rural Areas
Very poor 46.3 82.7 96.2 85.0 I 66.4
Mid-poor I 29.8 81.2 93.7 87.9
Non-poor J 19.4 J 69.9 J 92.4 78.8 78.3
AU 129.4 1 76.5j 93.4 83.2176.4
C8te d'Ivolre
Very poor 43.9 79.6 95.2 81A 64.3
Mid-poor 21.9 60.3 87.5 77.2 74.8
Non-poor 10.2 35.0 80.4 60.4 72.7
AU 19.3 47.7 83.5 68.3 72.3
'WTI -
10
By 1988, under conditions of severe economic recession, participation
rates of children in very poor households had risen from 31% to 44%. In the
households of the mid-poor on the other hand, children's participation had
fallen. The same pattern existed for adolescents. The participation of male
adults increased between 1985 and 1988 for all groups, but more so for the mid-
poor and very poor.
It would appear that households of all income levels responded to the
recession by increasing the participation of male adults. Very poor household
however also increased the participation of secondary earners, i.e. children and
adolescents. (Changes in the participation of women were minor.) There were
differences between urban and rural areas. The sharpest increase in
participation of very poor children and adolescents occurred in cities other than
Abidjan, which is an area where participation of adult members declined. This
suggests that any drop in labor force participation of adults in very poor
households is compensated by an increase in the participation of younger
household members.
Participation rates only reflect whether household members are
economically active or not, and do not measure the extent of labor supply. The
latter is measured in Table 3, which shows actual hours of labor supplied per
11
Table 3. Average labor supply of children and adolescents
1985
Hours Per Year % of Total Household Labor Supply
Children Adolescents Children Adolescents
(7-14 years) (15-18 years) (7-14 years) (15-18 years)
Abidjan
Very poor _ _ _
Mid-poor _ _ _
Non-poor 2,181* 2,469 1.8* 4.7
All 2,181* _ 2,469 - -L8- 4.5
C)ther Cities
Very poor 1,171* 2,059* 5.2* 9.2*
Mid-poor 1,273* 1,600* 6.5* 9.4*
Non-poor 1,040 1,459 1.7 6.7
All 1,131 __1,5136 . 2-7 1 7.4
Rural Areas
Very poor 1,275 1,709 13.6 12.3
Mid-poor 956 1,335 10.6 11.1
No-poor 848 1,293 7.2 8.7
All L 962 1,375 9.2 ____ 10.0
MCte d'Ivoire
Very poor 1,268 1,743 12.2 11.8
Mid-poor 977 1,361 9.8 10.6
Non-poor 920 1,444 4.6 7.2
All poor 1,001 1,464 6.6 8.5
1988
.I = = = g =__ Hours Per Year | % of Total Household Labor Supply
Abidjan
Very poor _ _ _
Nid-poor 2 1,8249* 4.6*
Non-poor 2,947* 2,4699*0.* 2.4*
All L 2,947* 2,352 ______ 0.9* 2.5
other Cities
Very Poor 1,245* 2,171* 15.5* 15.7*
Mid-poor 1,473 2,101 4.7 8.2
Non-poor 1,874 2,263 2.5 7.3
All 1,538 _ 2,196 4.0 l__ 8
Rural Areas
Very poor 1,742 1,467 26.2 10.5
Mid-poor 1,475 1,657 14.0 9.4
Non-poor 1,558 1,581 8.0 8.4
All 1,593 1,578 13.8 l 9.2
CMte d'Ivoire
Very poor 1,713 l 1,518 25.5 10.8
Md-poor 1,475 11,728 11.6 9.0
Non-poor 1,619 1,754 5.2 6.8
All 1,598 1,692 10.2 8.1
Based on fewer than 10 observations.
12
year by children and adolescents, and the percentage of total household labor
supply that this represents.
In 1985, children who participated in the labor force worked an average of
1,001 hours per year, and adolescents worked 1,464 hours. These are very high
figures. To put them in perspective, the average economically active male adult
in C6te d'Ivoire worked 1,876 hours and the average economically active female
adult worked 1,424 hours in 1985. Hence, adolescents put in more hours than
adult women. Again, hours supplied are systematically higher in poor than
non-poor households. For those children who are economically active, hours
supplied are higher in urban than in rural areas. This contrasts with the
participation rates, which are higher in rural areas (Table 2). In other words,
fewer children and adolescents work in urban areas in C6te d'Ivoire than in
rural areas, but those who do, work longer hours.
In 1988, the labor supply of children had increased by more than 50%, to
1,598 hours. Noteworthy is that this increase also took place in non-poor
households, indicating how wide-spread the impact of the economic recession
was.
The share in total household labor supply represented by children and
adolescents is significant in 1985 it was 15.1%, and this rose to 18.3% in 1988
13
(Table 3). In very poor households, however, the figures are much higher:
24.0% in 1985, and 36.3% in 1988.
Two conclusions emerge so far. First, labor supply in very poor
households is higher than among other households, indicating that the quantity
of labor supply is not a cause of their poverty. The key factor is hence low
hourly earnings. Second, between 1985 and 1988, very poor households had to
rely to an increasing extent on the work of children to compensate for falling
incomes.
Children's contributions to household well-being are not limited to hours
of labor supply. Many children also undertake home care activities, such as
cleaning, cooking, child care, etc. that frees other household members to work
for pay. Table 4 shows that this requires, on average, another 12 hours per week
of children, and 14-15 hours of adolescents. These figures did not change much
between 1985 and 1988 (this is also the case for adult household members).
Perhaps one could have expected a decline, due to crowding out from increased
labor supply, but this did not happen. The result, obviously, was reduced
leisure time.
14
Table 4. Average time (hours per week) spent in home-
care activities by economically active children and
adolescents
1985
Children Adolescents
(7-14 years) (15-18 years)
Very poor 12.8 17.6
Mid-poor 12.3 15.1
Non-poor 11.8 14.9
Abidjan 8.2 7.6
Other Cities 12.7 12.2
Rural Areas 12.2 16.6
All 12.2 15.4
1988 =
Very poor 11.0 12.7
Mid-poor 11.9 14.6
Non-poor 13.8 13.3
Abidjan 21.0 14.1
Other Cities 10.5 10.7
Rural Areas 12.2 14.1
All 12.1 13.6
3. Child Labor and Schooling in 1988
The previous section highlighted the importance of child labor for Ivorian
households in absorbing the shock of falling incomes during the recession of the
1980s. As 1988 is the last year for which the detailed data of the CILSS exist, we
investigate the determinants of child labor in more detail for that year. In this
section, we do so by means of tabulations, focusing on the interplay between
work and schooling. The next section will consist of a multivariate analysis.
15
Table 5 shows that the 1,600 households in the 1988 CIISS consisted of
9,860 people, of which 5,310 (i.e. 54%) were children or adolescents. This high
percentage is the result, of course, of the high population growth rate (3%) in the
country. Of these children, 3,897 (73%) were children of the head of household,
while the others were children of other members of the household or of non-
members. This reflects the fact that extended households as well as the practice
of child fostering are common in Cote d'Ivoire.
For our purposes, we have used an age cut-off of 7 years, at which point
all children should legally be in school. This gives an effective sample for
analysis of 2,828 children.2
Table 5. The 1988 CILSS S___
Urban Rural All
Households 624 976 1,600
Individuals 3,820 6,040 9,860
Children 0-17 years 2,093 3,217 5,310
Children of heads of household 0-17 years 1,538 2,359 3,897
Children 7-17 years 1,177 1,650 2,828
Children of heads of household 7-17 years 795 1,232 2,028
2 The analysis of the CILSS data requires the use of sampling weights to reflect varying
sampling probabilities. All results in this paper use these weights. The construction and use
of these weights is explained in Demery and Grootaert (1993).
16
Each of these children and their households face the choice of allocating
his/her time among five activities:
* going to school,
* working in the labor market outside the home,
* working in an enterprise or farm belonging to the household,
* helping with home care tasks, and
* leisure.
In the CILSS, there is direct information on the first four activities. Since the
personal development needs of the child are best served by school attendance, it
behooves to look first at the extent to which this time allocation is not chosen by
the child, or more likely, for the child by the parents. By age 7, almost 50% of
children in C6te d'Ivoire are not enrolled in school yet (Table 6 and Figure 1).
The figure decreases to 32% at age 9, then rises steadily to the 40% range at ages
12-14. As of age 15, there is a sudden jump to above 60%. This corresponds to
the end of the primary schooling cycle, at which point many children end their
school careers. This calls for distinguishing in the analysis (as we have done so
far), children in the 7-14 age range from adolescents in the 15-17 age range.
17
Table 6. Non-School Enrollment (%) by Location, Gender and Age
Age Urban Rural Boys Girls All
(years) (%) l%) (%) (%) (%)
7 29.1 57.6 40.2 55.1 47.3
8 19.4 52.8 28.5 49.5 38.6
9 19.8 41.4 22.4 44.0 32.4
10 18.3 47.4 27.0 46.8 36.1
11 21.3 43.3 28.6 42.1 34.9
12 29.9 50.4 31.0 55.8 42.4
13 32.5 54.9 38.0 54.9 46.2
14 24.6 55.8 32.1 52.4 41.4
15 44.2 73.4 55.3 65.7 60.1
16 46.2 86.7 59.7 71.4 66.0
17 56.1 96.3 69.2 80.5 74.1
All 29.2 55.4 36.1 53.9 44.5
Figure IA. Non-School Enrollment by Age and Location
100
90
80
11
5U -~.,Lan
40'
0'
7 8 9 10 11 12 13 14 15 18 17
Age of Child
18
Figure lB. Non-School Enrollment by Age and Gender
90 M
2333B2' 33:23 .8 .3....2.. . . ; -j j ...... ,-- . .................. ;- ,-, - $<,,"."
70
160 -
5 0 ........ Boys
z 0.... Girls
30-
70 8 9 10 1 1 13 16 1
Age of Child
It is also clear that non-enrollment is muc ihe nrural than in urban
areas, at all ages. There is a distinct gender dimension: at all ages, girls' school
enrollment is lower than boys. The result of these location and gender
differences is an important difference in educational achievement at age 17:
35.5% of urban children and 63.0% of rural children have less than complete
primary education; 41.3% of girls and 51.3% of boys have completed primary
education (Table 7). At age 17, the average urban child has received 6 years of
education, against only 3.1 years for the average rural child. The average 17-
year old boy has received 5.4 years, and the average 17-year old girl has received
3.8 years.
19
Table 7. Educational Achievement at Age 17
Urban Rural AR
Male Female AR Male Female AU Male Female All
Level of education completed
None 23.9 47.5 35.5 61.8 65.0 63.0 42.9 54.2 47.8
Primary 64.6 45.2 55.0 38.2 35.0 37.0 51.3 41.3 46.9
Lower secondary 11.6 7.3 9.5 - - - 5.8 4.5 5.2
All 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0
Average years of education 7.2 4.7 6.0 3.6 2.3 3.1 5.4 3.8 4.7
20
As we pointed out earlier, children can devote their time to five activities
and most children in C6te d'Ivoire combine several of these, especially work and
school. For purposes of this analysis, and considering the limitations imposed
by the CILSS sample size, we have classified children in 4 categories:
(1) children attending school, and not reporting any work ("school
only")
(2) children attending school and reporting work ("school and work")
(3) children not attending school and reporting work ("work only")
(4) children not attending school and reporting no work or home care
activities only ("home care")
The fourth category deserves some clarification. In the CILSS sample,
12% of children report no school attendance, no work inside or outside the
home, and participation in home care activities. Another 10% report no school,
no work, and no home care activities either. In the context of C6te d'Ivoire, it
would be most unusual for children in the age group 7-17 to not attend school
and to make no contribution at all to the household. We must consider the
possibility of reporting errors for those cases. It is most likely that those children
forgot to report home care activities, and we have therefore grouped them
21
together with children reporting no school, no work and home care activities.
The fourth category is thus a "residual" category, for whom we have somewhat
less certainty about the nature of children's activities than for the other three
categories.
Table 8 shows the distribution of children across the four categories:
* Only 25% of children in C6te d'Ivoire attend school as their only
activity. This represents 34% of urban children and 14% of rural
children, 35% of boys and 14% of girls.
* Another 30% of children combine schooling with work inside or
outside the home.3 The figure is higher in urban areas and for girls.
* More than one in five children works as their primary activity. This
situation is predominant in rural areas, where it pertains to 34% of
children (against only 6.5% in urban areas). The frequency of the
work-only situation rises sharply with the child's age.
3 This is a very high figure and important feature of the child labor situation in Cote d'Ivoire.
In leighboring Ghana, 19% of children combined work and school (Canagarajah and
Coulombe, 1997).
22
* Slightly more than 20% of children report only home care activities,
but the figure exceeds 30% for girls.
Table 8. School and Work: Mutually Exclusive Categories of
Child Activities, by Location
Urban Rural All
__ __ __ _ __ __ __ __ __(%) (%)(%
School only 34.3 18.8 25.3
All School and work 36.4 25.8 30.2
Children Work only 6.5 34.4 22.8
Home care 22.8 20.9 21.7
All 100.0 100.0 100.0
School only 48.4 27.2 35.4
School and work 32.1 26.2 28.5
Boys Work only 6.0 30.9 21.2
Home care 13.5 15.8 14.9
All 100.0 100.0 100.0
School only 20.7 8.4 13.9
School and work 40.6 25.3 32.2
Girls Work only 6.9 38.9 24.5
Home care 31.8 27.3 29.4
All 100.0 100.0 100.0
School only 39.3 21.3 28.5
School and work 36.6 28.4 31.7
Ages 7-14 Work only 3.7 27.9 18.3
Home care 20.3 22.4 21.5
All 100.0 100.0 100.0
School only 16.0 4.9 10.5
School and work 35.5 11.7 23.6
Ages 15-17 Work only 16.4 70.4 43.4
Home care 32.0 13.0 22.5
All 100.0 100.0 100.0
We documented in the previous section the strong link between child
labor and poverty, and the fact that the poor increased the supply of child labor
23
the most in the 1985-88 period, in response to the economic recession. Table 9
explores this relation further for the four categories of child work and schooling.
In the table, households have been ranked by income per capita (excluding the
income from child labor) and grouped in quintiles. We excluded income from
child labor in order to display the household situation prior to the child labor
decision.
Table 9. School and Work: Mutually Exclusive Categories of
Child Activities, by Income Quintiles
Quintiles of Per Capita Household Income
1 2 3 4 5 All
(%) (%) (%) (%) (%) (%
School only 20.6 21.7 27.4 24.7 38.1 25.3
School and work 23.0 25.5 31.5 38.5 38.2 30.2
Country Work only 30.9 27.9 21.3 17.1 8.9 22.8
Home care 25.5 24.9 19.8 19.8 14.8 21.7
All 100.0 100.0 100.0 100.0 100.0 100.0
School only 33.8 28.6 37.7 26.5 43.3 34.3
School and work 29.7 34.2 33.8 43.7 36.9 36.4
Urban Work only 5.1 9.9 7.6 6.0 4.5 6.5
Home care 31.5 27.2 20.9 23.8 15.3 22.8
All 100.0 100.0 100.0 100.0 100.0 100.0
School only 16.4 18.8 19.6 22.7 20.0 18.8
School and work 20.9 21.8 29.7 32.8 42.7 25.8
Rural Work only 39.0 35.5 31.9 29.3 24.0 34.4
Home care 23.6 23.9 18.8 15.3 13.3 20.9
All 100.0 100.0 100.0 100.0 100.0 100.0
Over most of the income range, the incidence of the "school-only"
situation shows little relation with income level. Only in households in the
highest income quintile is there a clearly higher presence of children who go to
24
school only, and this result is at least partly attributable to the fact that most of
these households live in the urban areas, where the supply of education is better.
The school-work combination displays a more pronounced positive correlation
with income, especially in rural areas. Conversely, children who only work are
found mostly in the two lowest quintiles, and to a very large extent in rural
areas.
The final task we wish to undertake in this section is to portray better the
full-time child worker in C6te d'Ivoire, defined as are the children who do not
attend school and report work outside the home or on a household enterprise or
farm as their sole activity (i.e. category 3, "work only").
The full-time child worker is on average 12.7 years old, and has a very
low average education of only 1.2 years. This category is split evenly among
boys and girls, but is found much more frequently in the poorest 40% of
households. Almost 90% of these child workers live in rural areas. Of those,
60% live in the Savannah, which is C6te d'Ivoire's poorest region. This is clearly
a critical observation for policy interventions. Savannah is the zone where Cote
d'Ivoire's main cash crops (cocoa and coffee) cannot be grown. Farmers are
predominantly subsistence farmers and only cotton can provide cash income. It
is also the zone that lags the most in education facilities and enrollment We will
return to this in the next section, when we undertake the multivariate analysis.
25
Table 10. A Portrait of the Full-Time Child Worker in C=te d'lvoire
Full-Time
Child Worker All Children
(Category 3) 7-17
Average age 12.7 11.2
Average years of education 1.2 2.5
% Girls 50.7 50.4
% Boys 49.3 49.6
% In Poorest 40% of Households 62.1 48.0
% In Urban Areas 11.8 41.6
% In Rural Areas 88.2 58.4
Average working hours per week:
* Boys 38.9 --
* Girls 34.2
* Urban 45.8
* Rural 34.9
Average hours per week spent on home
care tasks:
* Boys 9.5 7.5
* Girls 19.9 15.7
* Urban 14.2 12.4
* Rural 16.1 12.4
The children who work only, do so for an average of 34 hours (girls) to 39
hours (boys) per week, i.e. their work is truly full-time. As we observed earlier,
in urban areas the work hours are much higher than in rural areas (46 and 35
hours, respectively). In addition, the full-time child workers spend many hours
doing home care tasks, for an average of 9.5 hours per week (boys) and 20 hours
26
(girls). This is significantly more than non-working or part-time working
children.
4. Multivariate Analysis.
As we mentioned in the introduction, the literature on child labor has
identified several critical supply and demand factors. In the analysis below we
focus on supply factors at the household level, i.e. those characteristics of the
child and the household which can exercise an influence over the household's
decision to allocate children's time away from schooling and towards work. We
also include measures of the cost of schooling and proxies for demand factors.
Characteristics of the child. The tabular presentation in the previous
sections, as well as virtually all empirical work on child labor, has indicated that
the age and gender of the child are important determinants of the probability of
work. The magnitude and direction of these effects are however country-
specific, and determined by cultural factors, labor market opportunities, and
wage patterns.
Parents' characteristics. There is ample empirical evidence that education
and employment status of the parents affect the child labor decision (ILO, 1992;
Grootaert and Kanbur, 1995; Patrinos and Psacharopoulos, 1995). The usual
assumption is that the father's education and employment affects boys the most,
27
and mother's education and employment affects girls the most In the model, we
include the number of years of education of each parent, and an interaction
variable with the gender of the child. The nature of parents' employment also
matters-if the parents have no or irregular employment, it creates the need for
additional income sources to be provided by children. Due to sample size
limitations, the employment aspect in the model below is captured only by a
categorical variable indicating whether the parent is employed as wage earner or
self-employed (i.e. excluding unpaid family workers). This variable is also
interacted with the gender of the child.4
Household characteristics. Several demographic and economic features of
the household as a unit affect the supply of child labor.' On the demographic
side, household size and composition are of foremost importance. Ceteris
paribus, the more children there are in the household, the more likely it is that
one of them will work. The literature has clearly established that larger
household size reduces children's educational participation and reduces parental
investment in schooling (Lloyd, 1994). A larger household size decreases income
per capita and increases the dependency ratio, and both factors increase the
likelihood that a child will need to generate income (in cash or in kind) to
maintain the household's level of living. However, each child does not have the
4 If the parent is not a member of the household, we selected the education and employment
characteristics of the oldest male or female person in the household.
5 Cultural household characteristics could also be relevant in the child labor decision, e.g.
religion. This type of information is not available in the CILSS.
28
same probability to be called upon to work; it depends on the child's age and
gender, but also on the age and gender of the siblings present in the household
(Lloyd, 1993; Jomo, 1992; De Graff et al, 1993; Patrinos and Psacharopoulos,
1997). In the model below, we enter variables that capture the numbers of
sblings, by gender and age group.
We also include in the model the stage in the life cycle as captured by the
age of the head of household. The gender of the head of household is also
relevant because female-headed households usually have higher dependency
ratios, although this can be offset by an income effect (their smaller size implies
higher income per capita).
On the economic side, the key variables are the ownership by the
household of income generating assets. In the model, we have included two
such assets which we consider exogenous in the short term, namely, the
ownership of a farm or a non-farm household enterprise.
In spite of the strong observed correlation between poverty (or income in
general) and child labor, it would not be appropriate to include household
income as a variable in the model, because this variable is endogenous. We have
indeed already included the main household endowments of human and
29
physical capital that determine income.6 We have however included a
categorical variable to indicate whether the household fell in the lowest income
quintile. This is not intended as an income variable, rather it captures the special
constraints faced by the poorest segments of the population in terms of access to
credit and insurance. This lack of access prevents a poor household from relying
on outside markets to reduce income risk and is a major reason why child labor
is predominant among poor households (Grootaert and Kanbur, 1995).
Cost of schooling. Since schooling is the main competing time use for
children, it stands to reason that the cost of schooling would be an important
determinant of the likelihood of child work (Siddiqi and Patrinos, 1995). The
CILSS contains information on household expenditures for education, but these
cannot be included as explanatory variables in the model because they are
endogenous to the child labor decision (by definition, expenses on education are
incurred only for children for whom the decision was made to enroll them in
school). We have hence averaged, for each cluster in the survey, household
expenditures on education per enrolled child. This average can be considered
an independent measure of the cost of schooling in that cluster, and this variable
has been included as a regressor in the model. We have also included the
distance to the school (also averaged by cluster) as a partial measure for the
opportunity cost of school attendance. Unfortunately, a direct measure of
6 The model we estimate below is a reduced form equation. In a structural model, it would be
appropriate to have a separate equation to determine household income as a function of
30
foregone earnings could not be calculated because there are too few cases in the
CILSS for which income from child labor is reported.
Demand factors. As a household survey, the CILSS does not furnish data
on employment opportunities for children. Likewise, the data on wages for
children cannot be used here because the number of cases of reported wages are
too few to use a cluster- or region-specific averaging procedure (as we did for
the cost of schooling) to produce a useable exogenous measure of wages. Hence,
the model below does not include any direct demand variables. As a (weak)
proxy, we have included in the model dummy variables for the region of
residence of the household.
A Model for the Determinants of Child Labor
Several formal models of the household economy that explicitly take into
account the economic contributions of children have been discussed in the
literature (Levy, 1985; Rivera-Batiz, 1985; Sharif, 1994). Much of this work is
based on Rosenzweig and Evenson (1977). The setting is the standard
constrained utility maximization model of the household. A consumption vector
is maximized, subject to the resource endowment of the household and the
market determined returns to these assets. A structural formulation of this
assets deemed exogenous. The child labor equation can then include an instrumental
variable for income, e.g. its value predicted by the first equation.
31
household economy includes equations explaining the supply of labor of
different household members, including children.
For this paper our objective is more modest We wish to estimate a
reduced-form model of the determinants of child labor. As we explained earlier,
we lack demand variables in the data set and our focus is therefore on the
supply side. Of course, the "conventional" policy approach to child labor has
been focused on the demand side, mainly by trying to affect the behavior of
owners of firms to reduce their demand for child labor, e.g. by legislation
prohibiting child labor, by foreign boycotts of the products manufactured with
child labor, or by increasing society's awareness of child labor and stigmatizing
entrepreneurs who use child labor. As Grootaert and Kanbur (1995) have
argued, the range of policy variables needs to be enhanced, in part by providing
proper incentives to the households who provide the child labor. This calls for a
look at the supply side, and this is the focus of this analysis. The reduced-form
model estimated below contains the most relevant supply variables.
There are several ways to model econometrically the supply of child labor
depending upon the view one holds about the decision making process within
the household. The key aspect of this process is whether the decision maker in
the household considers all options open to the child simultaneously, or whether
preferred options (especially schooling) are considered first, followed by a
32
hierarchical decision making process.7 As far as we know, the literature does not
contain any evidence on this, and at any rate it is likely that the process differs
across households. A simultaneous decision making process would call for a
multinomial choice model, whereby the choices are schooling, work for wages,
work in home enterprise, work on farm, no work, or variations thereof. A
hierarchical decision making process can be modeled with a sequential choice
model, whereby the first step models the choice between the preferred option,
say, school attendance, against all other options combined. The second step
models the second best choice against the remaining options, conditional upon
not having opted for the first best choice. This process continues until the
choices are exhausted.
There are advantages and disadvantages to each approach. The appeal of
the multinomial choice approach is that only one equation needs to be estimated,
which by construction, will yield a consistent set of probabilities showing the
effect of a change in each explanatory variable on the probability to select each
option. There are, however, several drawbacks. The most important is that the
multinomial logit model requires the assumption of independence of irrelevant
altematives (IIA) (Maddala, 1983). This assumption states that the odds ratios
derived from the model remain the same, irrespective of the number of choices
7 As Grootaert and Kanbur (1995) discuss, the sole decision maker can be the head of
household, or there can be an intra-household bargaining process, e.g. between the father and
the mother-child nexus. This is not immediately relevant for the model formulation in this
paper, because each type of decision making process can consider the child's options
simultaneously or sequentially.
33
offered. In practice, the HA assumption is inappropriate in many applications.
In the case of child labor, it requires that, e.g., the choices between wage work
and work at a home enterprise are seen by the decision maker as independent
from other options, and not affected by whether or not a schooling option is
available. Obviously, this is a very unlikely situation. If non-independent
choices are included in the multinomial logit model, the model will overestimate
the selection probability for those options. An attractive alternative is the
multinomial probit model, in which the residuals have a multivariate normal
distribution, and which is not subject to the IIA assumption. The problem here
is that, for computational reasons, the model can only handle a small number of
alternatives (in practice, at most four).
The multinomial probit and logit models also share the requirement that
the relevant set of explanatory variables is the same for all choices. In the case of
the child labor options, this is to some degree defensible, but not entirely. E.g.
the cost of schooling is clearly a relevant variable in the schooling-work choice,
but not for the choice among work options. Likewise, ownership of a farm may
matter for the choice between work for wages and work at a home enterprise,
but not for the other options.
The sequential model approach solves many of these difficulties. The IIA
assumption is not required, since each alternative is introduced one at the time,
and the vector of explanatory variables, if needed, can be adjusted for each set of
34
alternatives. Furthermore, the use of a set of binomial choice equations makes it
convenient to extend the model estimation to include a labor supply equation
(with hours supplied as the dependent variable). This equation is censored and
needs to be corrected for possible selection bias, which can readily be done with
Heckman's well-known two-step procedure (whereby the first step is the binary
choice equation). The drawbacks of the sequential model are that multiple
equations need to be estimated and, more importantly, that the probabilities
derived from the model are conditional upon previous choices. This means that
estimation results will depend upon the ordering of options. The sequential
approach is thus most indicated for applications where a clear ordering of
options is possible.
On balance, in the case of the child labor choices, we think that the
benefits of the sequential approach outweigh the drawbacks. This is particularly
so because we would argue that it is possible to determine the "proper"
hierarchy of choices, namely: (1) schooling, (2) wage work, (3) home enterprise
work, (4) no work. The criteria underlying this ranking are, first, the welfare of
the child, and, second, the income contribution to the household. We expect
little dispute with the proposition that schooling is the preferred option from the
point of view of the child's welfare. If that option is not chosen, wage labor on
35
average will yield more income to the household than labor in a home
enterprise.'
The discussion below will hence analyze the supply of child labor as a
sequential decision making process, using three binary probit models. The
appendix to this paper presents, for comparative purposes, the results of a
multinomial logit model.9
The hierarchy of the four choices outlined above needs some modification
in the case of C6te d'Ivoire, for two reasons. First, fewer than 2% of children
work for wages. There are hence too few cases in the sample to permit model
estimation with wage work as a separate choice. Second, almost one-third of
children in Cote d'Ivoire combine work and school (Table 8). This calls for
considering this combination as a separate choice category. This leads to the
following four choices, and choice probabilities, to be estimated for each child:
P1 = probability to go to school and not to work.
P2 = probability to go to school and to work.
8 The use of an income crterion must be evaluated within specific social and cultural settings.
E.g., in some countries, work at home would be preferred to wage work for young women
because of religious considerations. In the case of C6te d'Ivoire, our assessment is that
income is a valid criterion.
9 Either one of these models represents an improvement over the most common approach in
the empirical literature, which is to use a single binary probit or logit model for the work or
school choice (see, e.g., Jensen and Nielsen, 1997; Patrinos and Psacharopoulos, 1995, 1997;
Mason and Khandker, 1997). Camagarajah and Coulombe (1997) use a bivariate probit model
allowing for interdependency between the work and school choice.
36
P3 = probability not to go to school and to work.
P4 = probability not to go to school and not to work.
In the sequential probit model, these probabilities are determined as
follows:
P1 = F(b'1 X)
P2 = [1 - F(b'1 X)] F(b'2 X)
P3 = [1 - F(b'1 X)] [1 - F(b'2 X)] F(b'3 X)
P4 = [1 - F(b'i X)] [1 - F(b'2 X)] [1 - F(b'3 X)]
where F represents the standard normal distribution function, and bi, b2, and b3
are vectors of the model parameters. The vector X contains the explanatory
variables. Parameters b1 are estimated over the entire sample. Parameters b2 are
estimated over the sample of children excluding those who go to school only.
Parameters b3 are estimated over the sample of children who do not go to school.
The pyramid in Figure 2 summarizes this process, and shows the sample sizes
involved.
37
Figure 2: Samples for Sequential Probit Estimation
Urban Rural
/Children
P3 n 344 notin n = 914
P4 school
(categories 3 &4)
All children except
P2 n= 773 in school only n= 1340
(categories 24)
P1 n 1177 All children n = 1650
(categories 1-4)
38
Table 11: List of Variables
Child Characteristics
AGE - age of child
AGESQ - age of child squared
FEMALE - gender (female = 1)
Parent Characteristics
EDUCFA - years of education of father
EDUCFA X FEMALE - years of education of father X gender of child
EDUCMO - years of education of mother
EDUCMO X FEMALE - years of education of mother X gender of child
EMPFA - father employed
EMPFA X FEMALE - father employed X gender of child
EMPMO - mother employed
EMPMO X FEMALE -motheremployed X gender of child
Household Characteristics
HEADAGE - age of head
HEADAGESQ - age of head squared
HEADFEMALE - gender of head (female = 1)
#BOYS 0-5 - # of other boys in household 0-5 years
#BOYS 6-9 - # of other boys in household 6-9 years
#BOYS 10-15 - # of other boys in household 10-15 years
#BOYS 16-17 - # of other boys in household 16-17 years
#GIRLS 0-5 - # of other girls 0-5 years
#GIRLS 6-9 - # of other girls 6-9 years
#GIRLS 10-15 - # of other girls 10-15 years
#GIRLS 16-17 - # of other girls 16-17 years
FARM - household owns farm
BUSINESS - household owns non-farm enterprise
POOR - household in poorest quintile
Cost of Schooling
COST - cluster average of household education expenditure
per pupil ('000 CFAF)
- school less than 1 km away (omitted)
DISTANCE 1-5 - school 1-5 km away
DISTANCE 5+ - school >5 km away
Location (urban)
- Abidjan (omitted)
OTHERCITIES - other cities
Location (rural)
- East Forest (omitted)
WFOREST - West Forest
SAVANNAH - Savannah
39
Results for Urban Areas
Table 12 shows the sequential probit results for urban areas, for all
children ages 7-17.'� The first two columns of the table contain the probit
coefficients and their standard error (an asterisk indicates that the coefficient is
significantly different from zero at the 90% confidence level). The third column
shows the partial derivatives of the estimates, computed at the means of the
explanatory variables. They show the change in probability, expressed in
percentage points, due to a one-unit increase at the mean of a given explanatory
variable, while holding all other variables constant at the mean.
The first stage results show the determinants of the probability to go to
school and not to work. The first striking finding is that this probability is not
influenced by the child's age. This is surprising given the U-shaped pattern of
labor force participation which we observed in Figure 1, but obviously these
differences are not statistically significant and/or are explained away by the
other factors in the equation. Girls, however, have a 30 percentage points lower
probability of going to school and not working than boys, ceteris paribus.
10 We attempted to estimate the model separately for children in the age groups 7-14 and 15-17,
in view of the higher labor force participation rates for the latter group. The small sample
size however created difficulties and not all steps could be estimated successfully. The
results we did obtain did not suggest any major differences between the two age groups, in
terms of the key determinants of child labor. We mention the few noteworthy differences in
the text.
40
The characteristics of the household have an important influence. Among
the parent characteristics, the father's education and the mother's employment
have the greatest impact, and in both cases they contribute to increasing a child's
probability of going to school and not working. The interaction variables with
the child's gender are not significant One interesting finding of the regression
estimated for younger children only (7-14 years) is that for girls the effect of the
mother's education is twice as strong as in the regression for the whole sample.
Stage in the life cycle also matters: the older the head of the household,
the more likely it is that a child will be attending school and not working -the
peak of the function occurs at age 53. The gender of the head of the household is
insignificant. If the household owns a non-farm business, the child has a
10 percentage points lower probability of going to school and not working. The
presence of other siblings has a fairly small effect the presence of brothers or
sisters in the 10-15 age group matters most, but only increases the probability of
going to school and not working by 3-4 percentage points.
Since age, education, employment and assets are the main determinants
of income, our results suggest that income is a key determinant of child labor.
Over and above these effects though, the dummy variable for lowest income
quintile suggests that the constraints faced by the poorest further decrease the
probability of attending school and not working by 8.6 percentage points.
41
Lastly, none of the cost-of-schooling variables were significant We think
that this result reflects the weakness of the available cost measures.
The second estimation stage eliminates from the sample the children who
go to school and do not work. The probability to be determined is that of
combining schooling and work. Unlike in the first stage, the child's age matters
a lot the probability of both working and going to school increases between the
ages 7 and 11 and declines thereafter. Girls are less likely than boys to combine
school and work and more likely to drop out of school.
Parents' education also matters more at this stage: each additional year of
education of the father reduces the probability that a child will drop out of
school and work by 1.8 percentage points, and each year of education of the
mother does so by 3.5 percentage points. This effect is not specific to the gender
of the child.
As in the previous stage, there is a pronounced life cycle effect the older
the head (up to age 57), the more likely children will attempt to combine school
and work rather than drop out Also as before, the gender of the head has no
additional influence on this outcome. The role played by siblings is different at
this stage: the presence of brothers at the ages 6-9 and 16-17 increases the odds
of being able to combine school and work; sisters in the 11-15 age group have a
similar effect
42
The presence of a non-farm household enterprise reduces the probability
that a child can combine work with school. In Cote d'Ivoire the ownership of
such enterprises, which for the most part are in the informal sector, is associated
with lower income and higher poverty (Grootaert, 1996). In contrast, wage
employment is associated with higher incomes. Poverty status has an additional
effect of increasing the likelihood of selecting non-schooling options. This effect
shows up stronger when the regression is limited to younger children.
Lastly, the cost-of-schooling variables are again not significant. On the
demand side, there is a location effect all other things being the same, children
in cities other than Abidjan are 10 percentage points more likely to combine
school and work.
The third stage of the estimation looks only at the children who are not in
school and determines the probability that they will work for wages or in
household enterprises as opposed to doing only home care tasks or no work at
all. The pattern of determinants is entirely different at this stage. The age of the
child is one of the most powerful factors: the older the children, the more likely
that they will work for wages or in a household's enterprise-each year
increases this probability by 9 percentage points. Girls have a higher probability
of being engaged only in home care tasks or not working.
43
Interestingly, the only parental characteristic that has a significant effect at
this stage is mother's employment, which increases the odds that girls will work
This is perhaps a surprising result, given that it is sometimes argued that
mothers and daughters are substitutes: if the mother works, the daughters need
to take over the care of the home. This does not appear to be the case in C8te
d'Ivoire. However, since most women's work in urban C8te d'Ivoire is in
household enterprises, the meaning of this result is that mothers involve their
daughters in this enterprise-and, likewise, they share the home care
responsibilities.
Life cycle, gender of the head, and the presence of siblings have no
statistically significant effects at this stage (except for sisters in the 5-9 age
range). Poverty status also has no effect on the work choice at this stage. In
contrast, the presence of a household farm or non-farm enterprise has a strong
positive influence on the likelihood to work.
The children who work and do not go to school can rightfully be labeled
"full-time workers" since their mean working hours are 44 hours per week. In
order to see whether the actual supply of hours is a function of the characteristics
of the child and the parent, we estimated a labor supply equation, suitably
corrected for selection bias using the two-step Heckman method. We imposed
two (somewhat arbitrary) identifying restrictions on the equations by deleting
from the first step (the probit choice equation) the education characteristics of the
44
parents and from the second step (the hours-supplied equation) the head-of-
household characteristics. The estimated coefficient of the hours equation are
reported in the last column of the third-stage results in Table 12.
The strongest determinants of supplied hours of child labor are the age of
the child and location. Hours rise sharply after age 12. Working children in
other cities work an average of 20 hours per week more than working children
in Abidjan.
Considering the other variables, mother's education tends to reduce the
labor supply of boys but increase that of girls. This suggests again that in C6te
d'Ivoire the labor supply of mothers and daughters is complementary rather
than being substitutes for one another. While children in urban households who
own a farm are more likely to work, the negative coefficient on the farm variable
indicates that they work on average fewer hours. Children from the poorest
households also work less on average. This finding is different from the tabular
results presented earlier, which showed that children from poor households
worked more hours. The multivariate result in Table 12 is of course a partial
result, after controlling for all other relevant variables, and suggests that the
poorest households face constraints that affect negatively their ability to supply
labor. The observed higher labor supply results from above average presence in
poor households of factors which tend to increase child labor supply-the most
important one being location, since the poverty rate in other cities is much
45
higher than in Abidjan. Lastly, as an econometric point, we note that the hours
equation is not subject to selection bias, since the coefficient of "lambda" (the
inverse Mills-ratio) is not significantly different from zero.
Summary. In urban areas in Cote d'Ivoire, the decision to supply child
labor is influenced significantly by the age and gender of the child, and by the
characteristics of the parents and the household in general. A very pronounced
gender gap exists at all three decision stages: girls are less likely to only attend
school or to combine work and school, and they are more likely to undertake
home care activities or not work. The continued promotion of girls' schooling
through appropriate incentives must thus remain a priority in C6te d'Ivoire.
Every additional year of age above 11, greatly increases the odds that the child
works. Parents' own education, the presence of a non-farm business in the
household, and the constraints from being among the poorest households are the
most important variables in determining the child work/schooling outcome in
Cote d'Ivoire.
Parents' characteristics, especially education, matter the most at the first
two decision stages relating to schooling options. Parents with no or low
education are more likely to choose work options for their children. This effect
is particularly strong for younger children. This underlines the
transgenerational aspect of lack of schooling and child labor. The effect is also
accentuated with younger parents. While parental education is in itself not a
46
policy variable, low parental education could be used as a targeting variable for
interventions.
The results also underline the importance of a gradual policy approach
towards the elimination of child labor. More than one in three urban children in
C6te d'Ivoire combine work and school. It would be a big step forward if
children who currently only work or are engaged in home care tasks could be
induced to combine this with school attendance. Flexibility in school hours is an
important policy variable in this context This would have benefits for the
children beyond education, and also improve their health status. Children who
work report a much higher rate of illness and injury and a lower rate of
consultation with a health care professional than children who combine work
and schooling.
The employment situation of the parents and the sources of income of the
household are a double-edged knife as far as child labor is concemed. An
employed mother will contribute to household income, thus reducing the need
for child labor and leading to much higher probabilities that the child will go to
school. However, in Cote d'Ivoire, the bulk of urban female employment is in
household enterprises, and the presence of these (all other things being the
same) increases the odds of child labor. The results of the third stage estimation
moreover show that mothers and daughters are not substitutes in employment,
but complement each others' work, both in the household enterprise and in
47
home care. Ownership of a household enterprise is a positive correlate of
poverty in C6te d'Ivoire, and among the poorest households child labor is more
likely. Care will thus have to be exercised that poverty alleviation policies
which include the provision of credit and other forms of support to household
enterprises do not have the inadvertent effect of increasing child labor.
The solution to this dilemma is the joint provision of support measures to
increase household income of the poor and incentives towards school
attendance. As an interim measure, facilitating the work/school combination
(e.g. with flexible school hours) may well be needed. Unfortunately, due to data
limitations, our results are weak in suggesting the nature of schooling incentives.
Neither the cost nor distance variables yielded significant coefficients. Still, one
should not conclude that cost of schooling is not a suitable policy variable. More
analysis with better cost data is needed. What we can say though, is that
targeting towards girls, towards children above age 11 (when drop-out
probabilities begin to increase) and towards children in the poorest households
and with the youngest parents is called for.
48
Table 12:. Sequential Probit Results - Urban Areas
First Stage: P1 = Probability of going to school and not working
Probit Standard Probability
Coefficent Error Derivative
l _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ (% po in ts)
Intect_ -0.5812 0.9413
Child Characteristics
AGE -0.1566 0.1224 -5.45
AGESQ -0.0010 0.0053 -0.03
FEMALE -0.8563* 0.2131 -29.80*
Parent Characteristics
EDUCFA 0.0424* 0.0152 1.48*
EDUCFA X FEMALE -0.0065 0.0224 -0.23
EDUCMO -0.0232 0.0178 -0.81
EDUCMO X FEMALE 0.0313 0.0250 1.09
EMPFA -0.1140 0.1579 -3.97
EMPFA X FEMALE 0.0612 0.1999 2.13
EMPMO 0.3055* 0.1324 10.63*
EMPMO X FEMALE -0.2677 0.1744 -9.32
Household Characteristics
HEADAGE 0.0837* 0.0266 2.91*
HEADAGESQ -0.0008* 0.0003 -0.03*
HEADFEMALE 0.1214 0.1767 4.22
#BOYS 0-5 -0.0631 0.0460 -2.20
#BOYS 6-9 0.0544 0.0609 1.89
#BOYS 10-15 0.1049* 0.0650 3.65*
#BOYS 16-17 0.1058 0.1370 3.68
#GIRLS 0-5 -0.0090 0.0458 -0.31
#GIRLS 6-9 0.0598 0.0602 2.08
#GIRLS 10-15 0.1262* 0.0652 4.39*
#GIRLS 16-17 0.1213 0.1385 4.22
FARM -0.2151 0.1412 -7.48
BUSINESS -0.2885* 0.1154 -10.04*
POOR -0.2480* 0.1123 -8.63*
Cost of Schooling
COST 0.0031 0.0024 0.11
DISTANCE 1-5 0.0939 0.1005 3.27
DISTANCE 5+ 0.2547 0.1630 8.86
Location
OTHERCITIES 0.1116 0.1160 3.88
Log. Likelihood -603.9
Restricted Log. Likelihood -795.6
Chi-Squared 383.4*
% Correct Predictions 72.5
49
Table 12: Sequential Probit Results -Urban Areas
Second Stage: P2 = Probability of combining work and school
Probit Standard Probability
Coefficient Error Derivative
Inten --t -4.700* 1.1033 - - - - - - - -
Child Characteristics
AGE 0.5357* 0.1368 21.05*
AGESQ -0.0251* 0.0056 -O.99*
FEMALE -0.4174* 0.2461 -16.40*
Parent Characteristics
EDUCFA 0.0468* 0.0210 1.84*
EDUCFA X FEMALE -0.0247 0.0252 -0.97
EDUCMO 0.0901* 0.0285 3.54*
EDUCMO X FEMALE -0.0534 0.0342 -2.10
EMPFA 0.0143 0.2112 0.56
EMPFA X FEMALE 0.3461 0.2381 13.60
EMPMO 0.1838 0.1861 7.22
EMPMO X FEMALE 0.0302 0.2116 1.18
Household Characteristics
HEADAGE 0.0696* 0.0309 2.73*
HEADAGESQ -0.0006* 0.0003 -0.02*
HEADFEMALE 0.1037 0.1947 4.07
#BOYS 0-5 -0.0217 0.0504 -0.85
#BOYS 6-9 0.1171* 0.0717 4.60*
#BOYS 10-15 -0.0217 0.0803 -0.85
#BOYS 16-17 0.3546* 0.1835 13.93*
#GIRLS 0-5 0.0122 0.0499 0.48
#GIRLS 6-9 -0.0004 0.0681 -0.02
#GIRLS 10-15 0.2296* 0.0825 9.02*
#GIRLS 16-17 0.0719 0.1800 2.82
FARM -0.1936 0.1547 -7.60
BUSINESS -0.3213* 0.1298 -12.62*
POOR -0.3184* 0.1248 -12.51*
�__ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ - - - - - - - - - - -- - - - - - - - - - -- - - - - - - - - -
Cost of Schooling
COST 0.0027 0.0032 0.11
DISTANCE 1-5 0.0809 0.1153 3.18
DISTANCE 5+ 0.2373 0.1867 9.32
Location
OTHERClTIES 0.2686* 0.1375 10.55*
Log. Likelihood -457.8
Restricted Log. Likelihood -494.3
Chi-Squared 73.1*
% Conect Predictions 66.3
50
Table 12: Sequential Probit Results - Urban Areas
Third Stage: P3 = Probability of only woriing
Probit Standard Probability Weekly Hours
Coefficient Error Derivative Worked (OLS)
_0o_2utS) Coefficientl
InteneEt 4.6326* 2.2369 _ 115.6*
Child Characteristics
AGE 0.5139* 0.3006 9.33* -11.23*
AGESQ -0.0127 0.0118 -0.23 0.46*
FEMALE -0.8580* 0.4909 -15.58* -5.97
Parent Characteristics
EDUCFA 0.0172 0.0429 0.31 0.14
EDUCFA X FEMALE -0.0707 0.0525 -1.28 -1.00
EDUCMO -0.1394 0.1033 -2.53 -6.96*
EDUCMO X FEMALE 0.0582 0.1147 1.06 8.32*
EMPFA -0.1283 0.4114 -2.33 2.40
EMPFA X FEMALE 0.2419 0.4570 4.39 4.00
EMPMO -0.3016 0.3592 -5.48 3.96
EMPMO X FEMALE 1.0052* 0.4280 18.25* -3.26
Household Characteristics
HEADAGE -0.0169 0.0534 -0.31 _
HEADAGESQ -0.0001 0.0005 -0.00 _
HEADFEMALE 0.2242 0.3583 4.07 -
#BOYS 0-5 -0.1546 0.1057 -2.81 1.85
#BOYS 6-9 -0.1147 0.1455 -2.08 1.41
#BOYS 10-15 0.0916 0.1539 1.66 0.07
#BOYS 16-17 -0.5927 0.5748 -10.76 1.24
#GIRLS 0-5 -0.0890 0.0859 -1.61 -2.38
#GIRLS 6-9 0.3486* 0.1382 6.33* -2.91
#GIRLS 10-15 0.1842 0.1594 3.34 -0.42
#GIRLS 16-17 -0.2081 0.4008 -3.78 -12.99
FARM 1.6093* 0.2907 29.22* -22.51*
BUSINESS 0.5075* 0.2653 9.22* -1.18
POOR 0.0927 0.2271 1.68 -13.43*
Cost of Schooling
COST
DISTANCE 1-5
DISTANCE 5+
Location
OTHERCITIES 0.2917 0.2691 5.30 20.08*
Lambda -1.76
Log. Ukelihood -116.3
Restricted Log. Likelihood -202.8
Chi-Squared 172.9*
% Correct Predictions 84.5
R-Squared - _ - 0.49
51
Results for Rural Areas
The first stage estimation results for rural areas (Table 13) suggest that, as
was the case in urban areas, the age of the child is not a significant determinant
of the probability of only going to school. Gender is, however, a powerful
determinant: girls in rural areas are 15 percentage points less likely to only be in
school than boys, after controlling for other relevant variables (in urban areas,
this differential was 30 percentage points).
Parents' education matters more in rural areas than in urban areas.
However, while the father's education increases the probability that girls attend
school and do not work, the mother's education decreases it Parents'
employment status has no further significant effect on this.
The role of the characteristics of the head of household are the reverse of
what we found in urban areas. In rural areas, there is no life cycle effect on the
supply of child labor, but a female head of household significantly decreases the
odds of a child going to school and not working. In urban areas, we found no
gender effect, but a strong age effect
52
The presence of siblings seems to matter less than in urban areas.
Likewise, the ownership of a non-farm business or the household's poverty
status has no an independent effect on the child labor decision.
Among the cost-of-schooling variables, only the dummy variable
indicating a distance in excess of 5 km has a significant coefficient. Its positive
sign, however, is wrong from a theoretical perspective: one would expect
distance to be a hindrance to school attendance. It is likely that we are
estimating a reverse causality, whereby attending a far-away school makes it
difficult to work at the same time.
While we found no strong differences between Abidjan and the other
cities in Cote d'Ivoire, in rural areas there is a pronounced regional effect All
other things equal, children in West Forest are 4 percentage points less likely to
go to school and not to work than in the reference region of East Forest, and
those in Savannah are 19 percentage points less likely to do so.
In the second estimation stage, the determinants of the probability to
combine work and school display an overall pattern similar to what was
observed for urban areas. The probability of combining work and school rises
with the child's age until age 11, after which point it becomes more likely that
the child drops out of school.
53
Parents' education again exerts a powerful influence on this outcome.
The more educated the parents, the more likely a child will combine education
with work-but this effect is markedly lower for girls. Older heads of
household (up to age 56) are also more likely to decide in favor of the work-
school combination.
As far as siblings is concerned, the key age group appears to be 10-15
years. Having brothers or sisters in that age group greatly reduces the odds of
school drop-out and increases that of maintaining the work-school combination.
The large negative coefficient for "number of boys 16-17' is out of line with all
others; given that only 3% of the children in the sample have these siblings, we
suspect that this result is unduly influenced by a few (unusual) observations in
the sample.
As we found in urban areas, the second decision stage is the one where
household assets and poverty status matter the most The presence of a non-
farm business decreases the probability of a combined work-school outcome by
9 percentage points,"1 and being among the poorest 20% of households further
lowers it by 27 percentage points.
In the rural equations, there is no variable for the ownership of a farm because almost all
rural households own a farm.
54
In rural areas, distance to the school also matters. If the school is 1-5 kms.
away, rather than being in the village, it reduces the probability that the child
can combine work and school by 18 percentage points.
Location effects are again very pronounced. In West Forest, a child is
14 percentage points more likely to be able to combine work and school relative
to East Forest, but in Savannah this outcome is 11 percentage points less likely.
This result probably reflects the poor educational infrastructure in Savannah.
The third and final stage models the choice between work for wages (rare
in rural areas) or in the household farm or enterprise versus undertaking home
care tasks only or not working. This outcome is quite strongly related to the age
of the child, with younger children being more likely to be assigned home care
tasks or not working. In urban areas we found that girls are much more likely to
receive home care assignments or not to work, but in rural areas this gender
effect is absent
Again, as we observed in urban areas, in the third decision stage, parents'
education ceases to be a significant determinant, but employment status remains
significant. Although not all coefficients are significantly different from zero, the
results suggest that an employed father increased the odds that a son will also be
employed and a daughter be assigned to home care, while an employed mother
has the reverse effect. This finding again undercuts the hypothesis that mothers
55
and daughters are substitutes for one another when it comes to home care, and
rather suggests that mother's employment leads to a situation whereby both
work and home care duties are shared.
The presence of siblings has little impact at the third decision stage, but
the presence of a non-farm enterprise does. Strangely enough, the direction of
the effect is opposite from that in urban areas. In cities, the presence of a home
enterprise increases the odds of a child's work in this enterprise, but in rural
areas it decreases these odds. There are two reasons for this result First, almost
all rural households have a farm, and this makes a far greater claim on child
labor. (The farm variable, of course, is not in the rural equation because there is
no variation across households). Second, home enterprises in rural areas are
mostly a subordinate activity and can more easily be combined with home care.
Lastly, the strong regional diversification continues to manifest itself, with
children in Savannah being much more likely to work on the farm or the home
enterprise.
As we did for urban areas, we also estimated an hours worked equation
suitably corrected for selection bias. The average hours worked by children who
work on the farm or in a household enterprise is 34.8 hours per week.
56
The results from this equation (Table 12, third stage, column 4) are not
very illuminating. The equation has a fairly poor fit (R2= 0.16) and the main
finding is that children in Savannah work on average 9 hours longer than
elsewhere. This is in addition to their already higher probability to work. The
selection variable lambda has a large positive coefficient, indicating that any
unobserved variables which make selection into child work more likely also
contribute to increasing work hours above average. This result underlines the
double disadvantage faced by children in Savannah and the need to make
intervention in this region a top policy priority.
Summary. The results from the rural sequential probit model identify
several key characteristics of the household which affect the child labor decision,
but the overriding finding (and major difference with the results for urban areas)
is the strong location effect. All other factors being the same, children in the
Savannah have a far lower probability to go to school or to combine work and
school than children elsewhere. This reflects the thin educational infrastructure
in Savannah-a disadvantage that has been present for several generations as
reflected e.g. in literacy rates in that region which are less than one-third the
national average (Grootaert, 1993). The prerequisite for any successful child
labor policy in rural C8te d'Ivoire is therefore to reduce the gap in education
investment between Savannah and the rest of the country.
57
Girls in rural Cote d'Ivoire are less likely to be given options involving
schooling than boys, but the gender gap is less than in urban areas (primarily
because more children work overall in rural areas). Unlike in urban areas, a
female head of household increases the chances that a child will work. Parents'
education is an even more critical variable in rural than in urban areas, because
it is a more rare attribute.
Poverty status of the household matters the most in the decision between
work-only and the work-school combination. This underlines the usefulness of
the gradual policy approach towards child labor whereby initially interventions
aim to make possible the combination of work and schooling, rather than to
eliminate immediately all child work. In the short run, having no children work
is not a viable strategy for many poor households. In the rural setting, flexibility
of school hours and vacation periods that coincide with harvest times are two
potentially effective measures to allow children to stay in school while helping
on the household farm.
The rural results identify the importance of having a school in the village
as opposed to 1-5 kms. away. The multinomial logit results, discussed in the
appendix, suggest that cost of schooling also matters in rural areas (at a hefty
rate of a one percentage point reduction in the probability to work for every $3
reduction in schooling cost). However, the same caveat regarding data quality
58
applies as for urban areas, and better cost data would have to confirm this
finding before it could be used to support concrete interventions.
Lastly, the rural results suggest that measures need be targeted to
children above age 11, at which point the probability to drop out of school
begins to rise. To this needs to be added of course the targeting towards girls
and towards all children in the Savannah region.
59
Table 13: Sequential Probit Results-Rural Areas
First Stage: P1 = Probability of going to school and not woiring
Probit Standard Probability
Coefficient Error Derivative
_____________ - _________0.0471 _0. 91 2
Child Characteristics
AGE 0.0130 0.1296 0.26
AGESQ -0.0075 0.0058 -0.15
FEMALE -0.7652* 0.2483 -15.17*
Parent Characteristics
EDUCFA 0.0279 0.0192 0.55
EDUCFA X FEMALE 0.0608* 0.0305 1.21*
EDUCMO 0.0609* 0.0331 1.21*
EDUCMO X FEMALE -0.2256* 0.0720 4.47*
EMPFA -0.1217 0.1485 -2.41
EMPFA X FEMALE -0.0504 0.2512 -1.00
EMPMO -0.0685 0.1109 -1.36
EMPMO X FEMALE -0.1198 0.1884 -2.38
Household Characteristics
HEADAGE 0.0081 0.0236 0.16
HEADAGESQ 0.0000 0.0002 0.00
HEADFEMALE Q0.5347* 0.2216 -10.60*
#BOYS 0-5 0.0188 0.0422 0.37
#BOYS 6-9 0.0588 0.0604 1.17
#BOYS 10-15 0.1141* 0.0652 2.26*
#BOYS 16-17 -0.0301 0.2682 -0.60
#GIRLS 0-5 0.0043 0.0404 0.08
#GIRLS 6-9 0.0857 0.0675 1.70
#GIRLS 10-15 0.1117 0.0755 2.21
#GIRLS 16-17 -0.0850 0.2261 -1.69
BUSINESS 0.2304 0.1496 4.57
POOR -0.1452 0.1064 -2.88
Cost of Schooling
COST -0.0086 0.0073 -0.17
DISTANCE 1-5 -0.1851 0.1203 -3.67
DISTANCE 5+ 0.2888* 0.1161 5.73*
�~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Location
WFOREST -0.2015* 0.1028 4.00*
SAVANNAH -0.9786* 0.1229 -19.40*
Log. Likelihood -622.2
Restricted Log. Likelihood -743.5
Chi-Squared 242.5*
% Correct Predictions 82.1
60
Table 13: Sequential Probit Results - Rural Areas
Second Stage: P2 = Probability of combining wori and school
Probit Standard Probability
Coefficient Error Derivative
Intet_ _- -__ 8.7698* _ 1.0688 ____=______
Child Characteristics
AGE 1.1550* 0.1346 37.22*
AGESQ -0.0534* 0.0059 -1.72*
FEMALE -0.2139 0.2256 -6.89
___ ___ ___ ___ ___ ___ ___------ -- - -- - - -- -- -- - _-_-_ _-__-_-____ -__ --_
Parent Characteristics
EDUCFA 0.2061* 0.0290 6.64*
EDUCFA X FEMALE -0.1221* 0.0353 -3.93*
EDUCMO 0.0986* 0.0567 3.18*
EDUCMO X FEMALE -0.0300 0.0707 -0.97
EMPFA -0.1525 0.1750 -4.91
EMPFA X FEMALE 0.0343 0.2253 1.10
EMPMO -0.1269 0.1306 4.09
EMPMO X FEMALE 0.0214 0.1728 0.69
Household Characteristics
HEADAGE 0.1118* 0.0284 3.60*
HEADAGESQ -0.0010* 0.0003 -0.03*
HEADFEMALE -0.1096 0.2105 -3.53
#BOYS 0-5 -0.0023 0.0442 -0.07
#BOYS 6-9 -0.0318 0.0618 -1.02
#BOYS 10-15 0.1854* 0.0670 5.97*
#BOYS 16-17 -0.6486* 0.3041 -20.90*
#GIRLS 0-5 0.0244 0.0407 0.78
#GIRLS 6-9 0.0131 0.0696 0.42
#GIRLS 10-15 0.1958* 0.0838 6.31*
#GIRLS 16-17 0.0679 0.1804 2.19
BUSINESS -0.2827* 0.1607 -9.11*
POOR -0.8245* 0.1095 -26.57*
Cost of Schooling
COST -0.0066 0.0079 -0.21
DISTANCE 1-5 -0.5504* 0.1265 -17.74*
DISTANCE 5+ - 0.1124 _ 0.1240_ _ ____-3.62 ----
Location
WFOREST 0.4454* 0.1103 14.35*
SAVANNAH -0.3374* 0.1169 -10.87*
Log. likelihood -603.8
Restricted Log. Likelihood -796.1
Chi-Squared 384.7*
% Correct Predictions 76.8
61
Table 13: Sequential Probit Results - Rural Areas
Third Sta e: P3 = Probability of only woiin_
Probit Standard Probability Weekly Hours
Coeffident Error Derivative Worked (OLS
_---_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ - .._-_ _ _ _ _ _ _ _ _ %ffipoints Coeffident_
Intercet -5.7076* 1.1132 - _ =__-5.41
Child Characteristics
AGE 0.8039* 0.1423 29.46* 3.44
AGESQ -0.0244f 0.0061 -0.89* -0.10
FEMALE -0.0813 0.2557 -2.98 4.89
Parent Characteristics
EDUCFA -0.0713 0.0555 -2.61 1.27*
EDUCFA X FEMALE 0.0232 0.0632 0.85 -0.75
EDUCMO 0.0513 0.0905 1.88 0.53
EDUCMO X FEMALE -0.0784 0.1088 -2.87 -1.07
EMPFA 0.5257* 0.2104 19.26* 2.97
EMPFA X FEMALE -0.3631 0.2587 -13.30 1.14
EMPMO 0.1557 0.1586 5.71 0.92
EMPMO X FEMALE 0.3606* 0.2051 13.21* 0.46
Household Characteristics
HEADAGE -0.0146 0.0306 -0.53 _
HEADAGESQ 0.0001 0.0003 0.01 _
HEADFEMALE -0.0498 0.2446 -1.82 -
#BOYS 0-5 0.0631 0.0514 2.31 1.36*
#BOYS 6-9 0.3327* 0.0833 11.82* 1.11
#BOYS 10-15 -0.0051 0.0878 -0.19 -1.22
#BOYS 16-17 -0.0037 0.2736 -0.13 -1.22
#GIRLS 0-5 -0.0505 0.0510 -1.85 0.48
#GIRLS 6-9 0.0740 0.0849 2.71 0.02
#GIRLS 10-15 -0.1393 0.1031 -5.11 2.31*
#GIRLS 16-17 -0.1625 0.2178 -5.95 1.83
BUSINESS -0.7152* 0.1820 -26.21* 0.88
POOR -0.0715 0.1157 -2.62 -1.96
Cost of Schooling
COST
DISTANCE 1-5
DISTANCE 5+ - ____=_____ ____=_____ 7
Location
WFOREST -0.3973* 0.1531 -14.56* 0.49
SAVANNAH 0.7885* 0.1268 28.89* 9.34*
Lambda 9.92*
Log. Likelihood 409.4
Restricted Log. iUkelihood -628.8
Chi-Squared 438.8*
% Correct Predictions 80.2
R-Sauared __________ ___________ _ 0.16
62
5. Conclusions
Most children in C6te d'Ivoire perform some form of work: work for
wages, work on the farm or the household enterprise, or home care tasks. In
urban areas, two out of every three children in the age group 7-17 years work
and about half of them combine this with school attendance. In rural areas more
than four out of every five children work, but only about a third of them manage
to combine this with schooling. Full-time child work, which can be expected to
have a major negative impact on the child's personal development, is less
prevalent but not negligible. In urban areas, 7% of children work full-time, for
an average of 46 hours per week. In rural areas, more than one-third of children
work full-time for an average of 35 hours per week, with the highest incidence in
the Savannah region. While the incidence of full-time child work rises with age,
it is by no means limited to older children: the average age of the full-time child
worker in C6te d'Ivoire is 12.7 years. The damage to the development of these
children is made clear by the fact that they have received on average only 1.2
years of education, have a higher incidence of illness and injury, and are less
likely to receive medical attention.
The figures cited pertain to 1988 and reflect a gradual increase of child
labor over the decade of the 1980s which was characterized by a severe economic
crisis in Cote d'Ivoire. Our results suggest that during this crisis, reduced labor
63
force participation of adults in poor households was compensated by an increase
in the participation of younger household members. The hope is therefore that
as the macroeconomic performance of the economy improves, child labor will
decline. Like for so many economic and social problems, a sound
macroeconomic environment which makes possible sustainable economic
growth is crucial to the long-run decline of child labor. However, while such
growth is a prerequisite for the elimination of child labor in C6te d'Ivoire, it
should clearly not be relied on as sole instrument to address the problem. The
experience of the currently developed nations during their industrial revolution
suggests that it could well take several generations for economic growth to
reduce child labor significantly.
In order to identify policy variables, we examined the determinants of
child labor using a sequential probit model. (An alternative multinomial logit
model is presented in the appendix). Our results identify five key factors which
affect the household's decision to supply child labor: the age and the gender of
the child, the education and employment status of the parents, the availability of
within-household employment opportunities, the household's poverty status
and its geographic location. Due to data limitations, our results are ambivalent
about the role of schooling costs and distance to school.
At each stage of the household decision making process, a pronounced
gender gap is observed, especially in urban areas: girls are less likely to attend
64
school exclusively, they are less likely to combine work and school relative to
working only, and they are more likely to undertake home care tasks. In rural
areas, a female head of household further increases the odds that a child will
have to work. The continued promotion of girls' schooling through appropriate
incentives must thus remain a priority in Cote d'Ivoire. Efforts to increase school
attendance of children (girls and boys) need to pay special attention to children
who have reached age 11, because from that age on the probability to work
increases rapidly, i.e. well before children finish elementary school (which in
Cote d'Ivoire occurs at age 14 on average).
Parent's characteristics, especially education, matter the most at the
decision stages involving schooling options. Parents with no or low education
are more likely to choose work options for their children. This effect is most
pronounced in rural areas and for younger children, and underlines the
transgenerational aspect of lack of schooling and child labor. While parental
education in itself is not a short-term policy variable, low parental education can
be used as a targeting variable for interventions.
The presence of household enterprises as an in-house source of
employment for children is a double-edged sword. On the one hand, the direct
effect is to increase greatly the odds of a child working, but the increased income
of the enterprise reduces the odds of child labor. Since in C6te d'Ivoire,
ownership of a household enterprise is a positive correlate of poverty, the direct
65
effect is likely to outweigh the income effect. Furthermore, our results indicate
that if the mother is the entrepreneur running the household enterprise, the
chances that daughters get drawn into the enterprise as well are high. There is
thus a danger that poverty alleviation policies which include the provision of
credit and other forms of support to household enterprises may, initially at least,
have the inadvertent effect of increasing child labor. The solution is the joint
provision of support measures to increase household income of the poor and
incentives towards school attendance.
The role played by household enterprises, as well as the finding that the
poverty status of the household matters the most in the decision between the
work-only option and the work-school combination, underline the usefulness of
a gradual policy approach towards child labor. Initially, interventions should
aim to make possible the combination of work and schooling, rather than to
eliminate immediately all child work. Flexibility of school hours and vacation
periods in rural areas which coincide with harvest times are two potentially
effective measures to facilitate the work-school combination. Our data suggest
that this would also improve children's health status.
Measures to make schooling less costly and more accessible are likely to
help as well but our results are ambivalent due to the weakness of the data on
costs of schooling. For rural areas, the results indicate that school attendance can
be improved by having a school in the village rather than at a distance of
66
1-5 kms away. The multinomial logit results show that a reduction in the cost of
schooling by about $3 would lead to a one percentage point increase in the
probability of school attendance, but the sequential probit model does not
confirm this finding. Further analysis with better cost data is needed.
Lastly, our findings show the need for and the strong potential of
geographic targeting. In urban areas, children in the interior cities of Cote
d'Ivoire have a much higher probability of working and their working hours are
much longer. In rural areas, children in the Savannah region are much more
likely to work than elsewhere, after controlling for all relevant household
characteristics. The educational infrastructure in the Savannah lags far behind
the rest of the country, as it has done for generations, and the reduction of the
gap in educational investment between the Savannah and the rest of the country
is an important prerequisite for a successful child labor policy in C6te d'Ivoire.
67
Appendix: Multinomial Logit Results
As discussed in the text, the multinomial logit model provides an
alternative estimation method to the sequential probit We have argued that this
model is less appropriate because of the Independence of Irrelevant Alternatives
(HA) assumption, which is not likely to hold in the case of the child labor
decision.
There is no reason to expect that the sequential probit model and the
multinomial logit model would yield similar results. This is because the HA
assumption is not imposed on the sequential probit model, and because the
sequential probit model yields probabilities conditional upon the outcome of the
previous choice whilst the multinomial logit model yields unconditional and
simultaneously determined probabilities.
The multinomial logit results are shown in Table Al. Only derivatives
calculated at the mean of the independent variables are shown. They are
marked by an asterisk (*) if they are significantly different from zero at the 90%
confidence level. Those probabilities are constrained to sum to zero for each
variable, across the four choices.
The statistical fit of the multinomial logit model is good but its predictive
ability, at 50-60% correct predictions, is inferior to the probit models, which
68
predicted correctly in the 70-80% range. The urban model severely
underpredicts the work-only choice and overestimates the work/school
combination. The errors in the rural model are fairly evenly spread across the
four choices.
In spite of the different assumptions underlying the two models, the
results in Table Al confirm many of the major findings which we highlighted in
the main text In urban areas, both models pick up the bias for girls against
schooling and towards home care. Likewise, both models confirm the role of
parents' education in deciding the options involving schooling and the greater
importance of the mother's employment status relative to the father's
employment status. Similarly, both models confirm the role of non-farm
household enterprises, and of the household's poverty status. Where the models
differ is in the role of siblings and of location. The multinomial logit model
shows many fewer significant coefficients for the sibling variables than the
sequential probit. The multinomial logit model shows the strongest location
effect in the home care/work choice, while the probit model puts this in the
decision involving the work-school combination.
For rural areas, the main conclusions from the sequential probit
estimation are also confirmed by the multinomial logit results. For example,
both models portray the growing difficulty for older children to combine work
and school and the increased likelihood as they get older to drop out of school
69
and work only. Both models highlight the severely disadvantaged position of
children in the Savannah region.
There are two noteworthy differences between the models. First, the
multinomial logit model shows a higher importance of the cost-of-schooling
variables for rural areas. It is the only model which suggests that an increase in
cost of schooling increases the probability to opt for work (at a rate of about one
percentage point for every 1000 CFAF-about $3). Second, the probit model
shows a gender gap only for the schooling decision, while the multinomial logit
model shows this also for the work versus home care choice. This can be
explained by the fact that in the sequential probit case the probability is
conditional upon a non-schooling choice for girls, whilst in the multinomial logit
case all options are considered simultaneously.
70
Table Al: Multinomial Logit Results
(probability derivatives at the mean)
URBAN AREAS
Schooling Work and Work Home Care
Only School Only or No Work
Child Characterstics
AGE -6.89 18.01* 0.55 -11.68*
AGESQ 0.02 -0.63* 0.00 0.61*
FEMALE -29.40* 9.07 0.04 20.25*
Parent Characteristics
EDUCFA 1.66* 0.28 -0.11 -1.83*
EDUCFA X FEMALE -0.37 -0.56 -0.03 0.97
EDUCMO -0.40 2.35* -0.57 -1.38
EDUCMO X FEMALE 0.69 -1.74* 0.31 0.74
EMPFA 4.13 2.48 0.19 1.45
EMPFA X FEMALE 1.27 7.67 -0.58 -8.36
EMPMO 12.21* -2.40 -0.77 -9.03*
EMPMO X FEMALE -10.98* 7.48 1.42 2.08
Household Characternstics_________ ___________
HEADAGE 3.01* -0.72 -0.25* -2.02*
HEADAGESQ -0.03* 0.01 0.00 0.02*
HEADFEMALE 4.39 -0.58 0.34 -4.15
#BOYS 0-5 -2.31 0.41 -0.18 2.08
#BOYS 6-9 1.61 1.50 -0.48 -2.62
#BOYS 10-15 3.76 -2.86 0.09 -0.99
#BOYS 16-17 4.44 3.67 -1.66 -6.44
#GIRLS 0-5 -0.35 -0.24 -0.20 0.79
#GIRLS 6-9 2.66 -0.80 0.38 -2.24
#GIRLS 10-15 5.55* 3.14 -0.32 -8.37*
#GIRLS 16-17 4.74 -1.33 -0.18 -3.22
FARM -5.44 6.27 3.17* -4.00
BUSINESS -10.84* 1.21 1.59* 8.04*
POOR -7.97* -2.33 0.78 9.53*
Cost of Schooling
COST 0.13 -0.00 -0.00 -0.00
DISTANCE 1-5 2.80 -0.55 -0.86 -1.39
DISTANCE 5+ 9.54 -3.36 -1.31 -4.87
Location
OTHERCITIES 3.78 3.80 0.10 -7.68*
Predicted Probability (%) 33.0 44.9 1.5 20.6
Actual Frejuenec (%) 34.1 36.5 6.8 22.6
Log. Likelihood -1180.2
Restricted Log. Likelihood -1470.5
Chi-Squared 580.6*
% Correct Predictions 51.5
71
Table Al: Multinomial Logit Results
(probability derivatives at the mean)
RURAL AREAS
Schooling Work and Work Home Care
Only School Only or No Work
Child Characteristics
AGE -1.31 29A9* 0.66 -28.85*
AGESQ -0.12 -1.35* 0.34* 1.23*
FEMALE -14.95* -0.77 3A2 12.30*
Parent Characteristics
EDUCFA 1.96* 5.49* -6.17* -1.29
EDUCFA X FEMALE 0.35 -4.15* 2.76 1.05
EDUCMO 1.96* 1.35 0.09 -3A0
EDUCMO X FEMALE -5A9* 2.50 -1.31 4.30*
EMPFA -2.72 -2.68 11.84* -6.44
EMPFA X FEMALE -3.00 1.93 -1.70 2.77
EMPMO -1.74 -2.85 3.54 1.05
EMPMO X FEMALE -4.05 3.23 12A4* -11.62*
Household Characteristics
HEADAGE 0.21 2.64* -1.91* -0.95
HEADAGESQ 0.00 -0.03* 0.02* 0.01
HEADFEMALE -12.59* 3A6 5.19 3.93
#BOYS 0-5 0.71 -0.23 0.99 -1A6
#BOYS 6-9 0.80 -0.35 6.52* -6.98*
#BOYS 10-15 3.12* 4.37* -5.09* -2A1
#BOYS 16-17 -2.53 -16.54* 14.37* 4.70
#GIRLS 0-5 -0.02 0.73 -1.82 1.11
#GIRLS 6-9 2.16 0.72 0.98 -3.85*
#GIRLS 10-15 3.98* 5.38* -9.19* -0.17
#GIRLS 16-17 -2.35 1.53 -1.27 2.09
BUSINESS 4.61 -7.68 -12.30* 15.37*
POOR -4.14 -26.47* 16.80* 13.82*
Cost of Schooling
COST -0.28 -0.35 1.00* -0.37*
DISTANCE 1-5 4.01 -16.34* 11.76* 8.60*
DISTANCE 5+ 6.57* -2.53 -9.54* 5.50
Location
WFOREST -2.66 17.13* -15.53* 1.06
SAVANNAH -23.03* -3.54 35.65* -9.07*
Predicted Probability (%) 15A 29.3 31.7 23.6
Actual Fr2quen y(%� 18.8 25.8 34.4 20.9
Log. Likelihood -1635.1
Restricted Log. Likelihood -2242.0
Chi-Squared 1213.7*
% Correct Predictions 57.6 6.1
eRokXCTS'-: in percentage points; * indicates significantly different from
zero at the 90% confidence level.
72
References
Ashagrie, K. (1993). "Statistics on Child Labor: A Brief Report." Bulletin of
Labour Statistics, No. 3.
Canagarajah, S. and H. Coulombe. (1997). "Child Labor and Schooling in Ghana."
Mimeo, Africa Technical Department, Washington, DC: The World Bank.
DeGraff, D. S., R. E. Bilsborrow, and A. N. Herrin (1993). "The Iinplications of
High Fertility for Children's Time Use in the Philippines." Chapter in
Lloyd (ed.), 1993.
Demery, L. (1994). "C6te d'Ivoire: Fettered Adjustment." Chapter 3 in
I. Hussain and R. Faruqee (eds.), Adjustment in Africa- Lessons from
Country Case Studies. Washington, DC: The World Bank.
Demery, L. and C. Grootaert (1993). "Correcting for Sampling Bias in the
Measurement of Welfare and Poverty in the C6te d'Ivoire Living
Standards Survey." World Bank Economic Review, Vol. 7, No. 3.
Grootaert, C. (1996). Analyzing Poverty and Policy Reform - The Experience of C6te
d'Ivoire. Brookfield, Vermont Avebury.
Grootaert, C. (1995). "Structural Change and Poverty in Africa: A
Decomposition Analysis for C6te d'Ivoire." Journal of Development
Economics, Vol. 47.
Grootaert, C. (1993). "The Evolution of Welfare and Poverty Under Structural
Change and Economic Recession in C6te d'Ivoire, 1985-88", Policy
Research Working Paper, No. 1078, The World Bank: Washington, DC.
Grootaert, C. (1986). "Measuring and Analyzing Levels of Living in Developing
Countries: An Annotated Questionnaire", Living Standards
Measurement Study Working Paper, No. 24, The World Bank:
Washington, DC.
Grootaert, C. and R. Kanbur (1995). "Child Labor: A Review", Policy Research
Working Paper No. 1454, Washington, DC: The World Bank.
Hussmanns, R., F. Mehran and V. Verma (1990). Surveys of Economically Active
Population, Employment, Unemployment and Underemployment: An ILO
Manual on Concepts and Methods. Geneva: ILO.
ILO (1996). "Child Labour: Targeting the Intolerable", Report VI, Part 1,
International Labour Conference, 85th Session. Geneva.
73
ILO (1992). World Labour Report 1992. Geneva.
ILO (1986). Annotated Bibliography on Child Labor. Geneva: ILO.
Jensen, P. and H. S. Nielsen. (1997). "Child Labor or School Attendance? Evidence
from Zambia", Journal ofPopulation Economics, Vol. 1 0.
Jomo, K. S. (ed.) (1992). Child Labor in Malaysia. Kuala Lumpur: Varlin Press.
Levy, V. (1985). "Cropping Pattern, Mechanization, Child Labor and Fertility
Behavior in a Farming Economy: Rural Egypt" Economic Development
and Cultural Change, Vol. 33.
Uoyd, C. B. (1994). "Investing in the Next Generation: The Implications of High
Fertility at the Level of the Family." New York Population Council,
Research Division Working Paper No. 63.
Lloyd, C. B. (ed.) (1993). "Fertility, Family Size and Structure-Consequences
for Families and Children." Proceedings of a Population Council
Seminar, New York, 9-10 June, 1992. New York: The Population CounciL
Maddala, G. (1983). Limited-Dependent and Qualitative Variables in Econometrics.
Cambridge: Cambridge University Press.
Mason, A. and S. Khandker. (1997). Mimeo, Poverty, Gender and Public Sector
Management Department, Washington, DC: The World Bank.
Patrinos, H. and G. Psacharopoulos. (1994). "Educational Performance and Child
Labor in Paraguay." International Journal for Educational Development,
Vol. 15(1).
Patrinos, H., and G. Psacharopoulos (1997). "Family Size, Schooling and Child
Labor in Peru: An Empirical Analysis", Journal of Population Economics,
forthcoming.
Rivera-Batiz, F. L. (1985). "Child Labor Patterns and Legislation in Relation to
Fertility." (mimeo) Department of Economics, Indiana University.
Bloomington.
Rosenzweig, M. R., and R. Evenson (1977). "Fertility, Schooling, and the
Economic Contribution of Children in Rural India: An Economic
Analysis." Econometrica 45(5).
Sharif, M. (1994). "Child Participation, Nature of Work, and Fertility Demand:
A Theoretical Analysis." Indian Economic Journal, 40(4).
74
Siddiqi, F. and H. Patrinos (1995). " Child Labor: Issues, Causes and
Interventions." Education and Social Policy Department Discussion
Paper No. 53, Washington, DC: The World Bank.
UNICEF (1991). T7he State of the World's Children 1991. Oxford: Oxford
University Press.
75
Policy Research Working Paper Series
Contact
Title Author Date for paper
WPS1883 Intersectoral Resource Ailocation and Fumihide Takeuchi February 1998 K. Labrie
Its Impact on Economic Development Takehiko Hagino 31001
in the Philippines
WPS1884 Fiscal Aspects of Evolving David E. Wildasin February 1998 C. Bernardo
Federations: Issues for Policy and 31148
Research
WPS1885 Aid, Taxation, and Development: Christopher S. Adam February 1998 K. Labrie
Analytical Perspectives on Aid Stephen A. O'Connell 31001
Effectiveness in Sub-Saharan Africa
WPSI 886 Country Funds and Asymmetric Jeffrey A. Frankel February 1998 R. Martin
Information Sergio L. Schmukler 39065
WPS1887 The Structure of Derivatives George Tsetsekos February 1998 P. Kokila
Exchanges: Lessons from Developed Panos Varangis 3'3716
and Emerging Markets
WPS1888 What Do Doctors Want? Developing Kenneth M. Chomitz March 1998 T. Charvet
Incentives for Doctors to Serve in Gunawan Setiadi 8,7431
Indonesia's Rural and Remote Areas Azrul Azwar
Nusye Ismail
Widiyarti
WPSI 889 Development Strategy Reconsidered: Toru Yanagihara March 1998 K. Labrie
Mexico, 1960-94 Yoshiaki Hisamatsu 31001
WPS1 890 MNarket Development in the United Andrej Juris March 1998 S. Vivas
Kingdom's Natural Gas Industry 82809
WPS189i The Housing Market in the Russian Alla K. Guzanova March 1998 S. Graig
Federation: Privatization and Its 33150
Implications for Market Development
WPS1892 The Role of Non-Bank Financial Dimitri ViKtas March 1998 P. Sintim-Aboagye
Intermediaries (with Particular 38526
Reference to Egypt)
WPS1893 Regulatory Controversies of Private Dimitri Vittas March 1998 P. Sintim-Aboagye
Pension Funds 313526
WPS1894 Applying a Simple Measure of Good Jeff Huther March 1998 S., Valle
Governance to the Debate on Fiscal 84493
Decentralization
WPS1895 The Emergence of Markets in the Andrej Juris March 1998 S. Vivas
Natural Gas Industry 8'2809
WPS1896 Congestion Pricing and Network Thomas-Olivier Nasser March 1998 S. Vivas
Expansion 82809
Policy Research Working Paper Series
Contact
Title Author Data for paper
WPS1897 Development of Natural Gas and AndreJ Juris March 1998 S. Vivas
Pipeline Capacity Markets in the 82809
United States
WPS1898 Does Membership in a Regional Faezeh Foroutan March 1998 L. Tabada
Preferential Trade Arrangement Make 36896
a Country More or Less Protectionist?
WPS1899 Determinants of Emerging Market Hong G. Mir March 1998 E. Oh
Bond Spread: Do Economic 33410
Fundamentals Matter?
WPS1900 Determinants of Commercial Asli Demirgug-Kunt March 1998 P. Sintim-Aboagye
Bank interest Margins and Harry Huizinga 37656
Profitability: Some International
Evidence
WPS1901 Reaching Poor Areas in a Federal Martin Rava'lion March 1998 P. Sader
System 33902
WPS1902 When Economic Reform is Faster Martin Rava"lion Miar c 1998 P. Sader
than Statistical Reform: Measuring Shachua Chen 33902
and Explaining Inequality in Rural
China
WPS1903 Taxing Capital Income in Hungary Jean-Jacques Dethier March 1998 J. Smith
and the European Union Christoph John 87215
WPS1904 Ecuador's Rural Nonfarm Sector Peter Lanjouw March 1998 P. Lanjouw
as a Route Out of Poverty 34529