UNICEF Staff Working Papers

Division of Policy and Planning Series

Child Labour, Education

and the
Principle of Non-
Discrimination

Elizabeth D. Gibbons
Friedrich Huebler
and
Edilberto Loaiza

UNICEF
New York, NY
Child Labour, Education and the Principle of Non-Discrimination
© United Nations Children's Fund (UNICEF), New York, 2003

UNICEF
3 UN Plaza, NY, NY 10017
November, 2005

This is a working document. It has been prepared to facilitate the exchange of knowledge and to
stimulate discussion.

The text has not been edited to official publication standards and UNICEF accepts no
responsibility for errors.

The designations in this publication do not imply an opinion on legal status of any country or
territory, or of its authorities, or the delimitation of frontiers.
Table of Contents

Executive Summary……………………………………………………………………………..i

Resumen Ejecutivo………………………………………………………………………..…….ii

Résumé…………………………………………………………………………………………..iii

Part 1: The Human Rights Framework for Analysing Child Labour and Education……..1

1.1: International conventions against child labour: standards and reality………..1

1.2: Implications of a human rights approach to development……………………...2
1.3: Education as a preventive strategy against child labour………………………..3

Part 2: Overcoming the Discrimination Against Girls in Analyses of Child Labour………4

2.1: Existing data sources for assessing child labour…………………………………4

2.2: Creating an equalizing indicator between girls and boys: The role of household
chores in the measurement of child labour……………………………………………5

Part 3: Does Participation in Child Labour Significantly Constrain School Attendance and
Achievement?...............................................................................................................................7

3.1: Data coverage and analytical model………………………………………………7

3.2: School attendance and child labour: Descriptive Statistics……………………...8
3.3: Regression analysis: Determinants of school attendance……………………….12
3.4: The role of gender in determining school attendance…………………………...14
3.5: Child labour, repetition and dropout…………………………………………….16

Part 4: Conclusions and Policy Recommendations…………………………………………...20

4.1: The human rights principle of non-discrimination applied to

statistical analyses………………………………………………………………………20
4.2: Limitations of survey data and child labour indicators…………………………21
4.3: Some reasons why children in Sub-Saharan Africa are not attending school…22
4.4 Policy recommendations……………………………………………………………24

Appendix: Regression Results…………………………………………………………………26

Executive Summary
UNICEF estimates that out of the 115 million children out of school, 62 million of them are
girls. Many of these children work but traditional indicators of child labour often underestimate
the amount of girls’ work because they ignore household chores. The human rights principle of
non-discrimination requires that all work by children – whether of a domestic nature or not – be
considered equally in the analysis of child labour. This paper presents estimates of child labour
in Sub-Saharan Africa that include household chores and thus reveal the discrimination against
girls. The authors also investigate to what extent participation in child labour leads to lower
school attendance and increased repetition and dropout rates, and whether child labour affects
girls and boys differently.

The data in the study was collected in MICS and DHS household surveys from 18 African
countries. 60 percent of children aged 7 to 14 years in the sample are attending school and 38
percent are engaged in child labour. 20 percent of all children are combining school attendance
and child labour. A regression analysis shows that household wealth and education of the mother
are the most important determinants of school attendance. Children from wealthier households
and children of mothers with a formal education are more likely to attend school. In the majority
of the countries in the study, boys, urban residents, and children not engaged in labour also have
an increased probability of school attendance.

i
Resumen Ejecutivo

UNICEF calcula que unos 115 millones de niños no asisten a la escuela, y de ellos, 62 millones
son niñas. Muchos de estos niños y niñas trabajan, pero los indicadores tradicionales del trabajo
infantil por lo general no tienen suficientemente en cuenta el volumen del trabajo de las niñas
porque dejan de lado las labores domésticas. El principio de no discriminación inspirado en los
derechos humanos requiere que en los análisis del trabajo infantil se tenga en cuenta todo trabajo
realizado por niños, se trate o no de labores de índole doméstica. Los cálculos estimados sobre el
trabajo infantil en África subsahariana que se ofrecen en este estudio incluyen esas tareas
domésticas y, por lo tanto, revelan la discriminación de las niñas. Los autores investigan también
en qué medida el trabajo infantil determina tasas más bajas de asistencia a la escuela y aumenta
las tasas de repetición de grados y de deserción escolar, así como si el trabajo infantil afecta de
manera diferente a las niñas y los niños.

Los datos contenidos en el estudio fueron recabados por medio de Encuestas agrupadas de
indicadores múltiples y de Encuestas de Demografía y Salud (DHS) que se llevaron a cabo en 18
países africanos. Un 60% de los niños de 7 a 14 años de edad de la muestra asisten a clases, y el
38% participan en el trabajo infantil. Un 20% de todos los niños asiste a la escuela y trabaja. Un
análisis de regresión demuestra que los principales factores determinantes de la asistencia a clase
de los niños y niñas son el nivel de ingresos de sus respectivas familias y el grado de educación
de sus madres. Los niños y niñas provenientes de hogares con mayores recursos y los hijos de
mujeres con un mayor nivel de instrucción siempre muestran una mayor propensión a ir a la
escuela. En la mayoría de los países considerados en el estudio, los niños varones, los niños que
viven en zonas urbanas y los que no trabajan también tienen mayores probabilidades de asistir a
la escuela.

ii
Résumé
L’UNICEF estime que 115 millions d’enfants ne vont pas à l’école, dont 62 millions de filles.
Un bon nombre de ces enfants travaillent, mais les indicateurs habituels du travail des enfants
sous-estiment souvent le nombre de filles qui travaillent, ne tenant pas compte des travaux
ménagers. Le principe de non-discrimination, qui est un droit de l’homme, exige que tous les
travaux effectués par des enfants – qu’ils soient domestiques ou non – soient pris en
considération sur un pied d’égalité, lorsqu’on analyse le travail des enfants. La présente étude
donnent des estimations du travail des enfants en Afrique subsaharienne qui comprennent le
travail ménager et qui révèlent donc la discrimination à l’encontre des filles. Les auteurs
recherchent aussi dans quelle mesure le travail des enfants fait baisser le taux de fréquentation
scolaire, augmente celui de redoublement et celui d’abandon de l’école. Les auteurs se
demandent en outre si le travail des enfants affecte différemment les garçons et les filles.

Les données rassemblées dans cette étude proviennent d’enquêtes en grappes à indicateurs
multiples et d’enquêtes démographiques et sanitaires effectuées dans 18 pays africains. Dans
l’échantillon, composé d’enfants âgés de 7 à 14 ans, 60 pour cent de ces enfants vont à l’école et
38 pour cent sont obligés de travailler, alors que 20 pour cent de la totalité des enfants vont à
l’école tout en travaillant. Une analyse économétrique indique que le niveau économique du
ménage et l’éducation de la mère constituent les facteurs déterminants de la fréquentation
scolaire. Les enfants issus d’un ménage aisé et ceux dont la mère a bénéficié d’un enseignement
scolaire ont toujours plus de chances d’aller à l’école ; c’est également le cas, dans la majorité
des pays étudiés, des garçons, des habitants des villes et des enfants qui ne sont pas obligés de
travailler..

iii
Child Labour, Education and the Principle of Non-Discrimination
Elizabeth D. Gibbons, Friedrich Huebler, and Edilberto Loaiza1

1 The Human Rights Framework for Analysing Child Labour and Education
1.1 International conventions against child labour: standards and reality

As early as 1921, when the International Labour Organisation (ILO) passed the
first Minimum Age Convention, the world has attempted to protect children’s right to an
education and to prevent any child labour which would prejudice their school
attendance.2 The ILO’s Minimum Age Convention 138 of 19733 set the standard for the
minimum age for admission to employment as 15 years, or in special cases where
economic and educational facilities are insufficiently developed, 14 years; light work not
harmful to the child or prejudicial to his or her attendance at school is permissible after
age 12. Since 1990, with the entry into force of the Convention on the Rights of the
Child,4 the child’s right to be protected from “any work that is likely to be hazardous or
to interfere with the child’s education” (Article 32) and his or her right, on an equal, non-
discriminatory basis to “primary education compulsory and available free to all” (Article
28) have gained the status of internationally recognised norms, while imposing an
obligation on the 192 states parties to the Convention to realise these rights for the
children under their jurisdiction.5 In 2000, children were provided further protection
through the entry into force of ILO Convention 182,6 which was ratified by 150 countries
as of May 2004.7 Convention 182 prohibits the worst forms of child labour, defined as
all forms of slavery and similar practices; child prostitution and pornography; illicit
activities (in particular the production and trafficking of drugs); and work that is likely to
harm the health, safety or morals of children.
However, as is well known, many governments have thus far failed to realise
these rights for their children. For 2002, the United Nations’ Children’s Fund (UNICEF)

1
Invaluable inputs to this paper were received from Anna-Karin Irvine, Meredith Slopen and Radhika Gore
of the Global Policy Section, Division of Policy and Planning, UNICEF. The views expressed in this paper
are those of the authors and do not necessarily represent the views of UNICEF. This paper was originally
prepared for a conference organized by the Center for Human Rights and Global Justice, New York
University School of Law, and appeared in the volume: Human Rights and Development: Towards Mutual
Reinforcement, edited by P. Alston and M. Robinson, Oxford University Press, 2005, based on the
conference.

2
“Children under the age of fourteen years may not be employed or work in any public or private
agricultural undertaking, or in any branch thereof, save outside the hours fixed for school attendance. If
they are employed outside the hours of school attendance, the employment shall not be such as to prejudice
their attendance at school.” Article 1 in: ILO. 1921. C10: Minimum age (agriculture) convention.
http://www.ilo.org/ilolex/cgi-lex/convde.pl?C010 (accessed 17 May 2004).
3
June 26, 1973, 1015 U.N.T.S. 297, entered into force June 19, 1976 (Convention 138).
4
G.A. res. 44/25, annex, 44 U.N. GAOR Supp. (No. 49) at 167, U.N. Doc. A/44/49 (1989), entered into
force Sept. 2, 1990.
5
UNICEF. 2000. First call for children: World declaration and plan of action from the World Summit for
Children, Convention on the rights of the child. New York: UNICEF.
6
38 I.L.M. 1207 (1999), entered into force Nov. 19, 2000 (Convention 182).
7
Database of International Labour Standards. http://www.ilo.org/ilolex/cgi-lex/ratifce.pl?C182 (accessed
17 May 2004).

1
estimated that 121 million children were out of school, 65 million of them girls.8 In
2002, the ILO estimated that, worldwide, 211 million children aged 5 to 14 years were
economically active, 111 million of them in hazardous work.9 In fact, since this figure
only counts children working in economic activities and excludes those working in
household chores (which if excessive, can also affect school attendance), the number of
working children in the world is likely to greatly exceed this estimate. For these millions
of children, their rights to education, to a childhood protected from work detrimental to
their development, and to human dignity, are all being violated. This is a scandal for the
21st century, a harvest of ignorance and lost potential which mortgages the future of these
children and their countries.

1.2 Implications of a human rights approach to development

Since 1996, when UNICEF adopted a Mission Statement whereby the

Organisation in all its work is “guided by the Convention on the Rights of the Child and
strives to establish children’s rights as enduring ethical principles,”10 it has struggled to
understand and apply a human rights approach to development in its programmes of
cooperation. Learning is continuous, but to date we have seen, amongst other effects of
the rights approach, that UNICEF is driven, beyond the utilitarian principle of the
greatest good for the greatest number of children, to give attention to those children and
vulnerable members of society living at the margins of the mainstream, and to push for
services to reach the “last 10 percent” of the un-reached. As a corollary, applying the
principles of universality and non-discrimination pushes UNICEF to direct the state’s
attention and resources to marginalised children and their families. This includes
children engaged in child labour. Likewise, application of human rights principles has
led UNICEF into actions which identify, advocate for, and support communities of ethnic
and racial groups suffering from discrimination, and to seek gender equity in all its
actions. This means that the gender dimension of child labour has to be visible, as a
prelude to being understood and acted upon.
Applying the principle of indivisibility of human rights has led to programmes
which incorporate the inter-relatedness of causes and are increasingly inter-sectoral in
content, addressing the rights of the whole child. This has particular relevance for
programmes simultaneously seeking to increase school attendance and decrease child
labour, acting through a complex nexus of interacting factors. Thus, the rights approach
to child labour calls for a profound analysis of the causes contributing to child labour, and
a multi-sectoral response to the problem: provision of accessible, affordable and quality
education; interventions aimed at increasing household income of poor families; reform
and implementation of laws on minimum age of employment, truancy, teachers’
minimum qualification and mandatory teaching hours; birth registration (without which it
is impossible to establish whether a child is old enough to attend school or to work); and
civic education aimed at all levels of society to promote and respect the fulfilment of
child rights, and ensure that law enforcement effectively suppresses the demand for child
labour.

8
UNICEF, The State of the World's Children 2004 (New York: UNICEF, 2004) 7.
9
ILO, Every Child Counts: New Global Estimates on Child Labour (Geneva: ILO, 2002) 20.
10
UNICEF, The Mission of UNICEF, UN Doc.E/ICEF/1996/AB/L.2.

2
 Overall, applying human rights principles derived from the Convention on the
Rights of the Child has resulted in a shift in the mix of strategies under-girding
UNICEF’s development work:

a) Increased support to capacity building of the state, its policies and its institutions, so
as to enable it to better meet its obligations to the citizenry.
b) Considerable widening of partnerships, well beyond the state, into civil society
organisations at all levels, and greater clarity on the importance of community
capacity building and citizen empowerment.
c) A deeper causal analysis for non-realisation of rights, leading to a better
understanding of the interrelationship of causes impeding children’s growth and
development, and to programmes which address structural causes of inequity. Prior
to adoption of the human rights approach, structural causes of children’s problems —
including poverty, which is often at the root of child labour — were taken as a given
and not subject to change through programme action.
d) A much higher investment by UNICEF programmes in advocacy with (for example)
parliaments to change discriminatory laws and to increase budgets for social
development, with international financial institutions and with power brokers in
general.
e) A considerable decrease in direct support to service delivery, (except in situations of
humanitarian emergency) as this is the state’s duty, to which UNICEF contributes
indirectly through capacity building of its institutions, through empowering
communities to demand the quality services to which they have a right, and to know
how they can hold state agents accountable for poor services.

1.3 Education as a preventive strategy against child labour

Stimulated by the Oslo International Conference on Child Labour in 1997,

UNICEF, with support of seven partners,11 developed the Global Child Labour
Programme, whose most important sub-programme was “Education as a Preventive
Strategy against Child Labour.” Implemented in 30 countries between 1999 and 2002,
the programme used a multi-sectoral, child rights approach to implementing four
components: provision of quality, relevant and affordable education; improvement in
family economies; raising of awareness in and respect for children’s rights; and the
enforcement of child labour laws. All four components were to be implemented at
policy, institutional, school and community level.
In 2003, the programme was evaluated.12 The experience showed that it is
relevant to use education as a main entry point for the targeted children (i.e., those
engaged in child labour), but there was only partial effectiveness in interventions. The
evaluation concluded that education can only be an adequate alternative to child labour if
it is accessible, affordable, of good quality, non-discriminatory, safe, and linked with

11
These funding partners were Finland, Luxembourg, the Netherlands, Norway, Sweden, the ILO, and the
World Bank.
12
UNICEF, Education as a preventive strategy against child labour: Evaluation of the cornerstone
programme of UNICEF's global child labour programme, Evaluation Working Paper (New York:
UNICEF, 2003).

3
other programmes in an integrated way. This finding reinforces the multi-sectoral
approach to development problems, as derived from the human rights principle of
indivisibility. However, the complexity of the determinants and the inter-relationship
between child labour and education were such that predictive factors, favouring the
abandonment of child labour and enrolment in school, were difficult to identify. The
present study attempts to fill this gap by analysing the constraining effect of child labour
on school attendance and achievement.
Overall, the programme failed to systematically collect and synthesize
quantitative data which could have helped to further explain the inconclusive results.
However, the programme confirmed that there is not invariably an inverse relationship
between school attendance and child labour, as many other factors determine whether a
child will or will not attend school. Furthermore, the programme “focused on working
children, children who have never been to school, and children at risk of dropping out of
school to join the workforce. ... Except for an intended link to girls’ education, there was
no particular attention to girls at work.”13 To that degree the programme failed to address
the gender dimensions of child labour, perhaps because the tools for assessing the extent
of household chores did not exist, and because there are, to date, no internationally
accepted definitions of child labour that include the tasks disproportionately carried out
by girls, as well as the corresponding indicators to measure and report the existing
empirical evidence.

2. Overcoming the Discrimination Against Girls in Analyses of Child Labour

2.1 Existing data sources for assessing child labour

The vast majority of studies on child labour limit themselves to analysing time
spent in economic activity (whether inside or outside the home); they do not take into
account time spent on household chores. Since girls are almost always more likely to be
occupied in household chores than are boys, this way of analysing the extent of child
labour, and its impact on schooling, is significantly biased against girls. Non-
discrimination is a key human rights principle; it is important to develop an “equalizing”
indicator of child labour, so that its impact on girls access to education, and school
attainment, is made visible in similar terms as the effect of child labour on boys.
UNICEF has been able to collect data on the time children spend in household
chores through its Multiple Indicator Cluster Survey (MICS). The MICS is a household
survey developed by UNICEF to fill data gaps in areas critical to the survival of children.
The methodology was developed in collaboration with the World Health Organization
(WHO), UNESCO, the United Nations Statistics Division, MEASURE (USAID), the
London School of Hygiene and Tropical Medicine, and the United States Center for
Disease Control and Prevention (CDC). Specifically, MICS was developed to obtain data
on key indicators for assessing progress towards the goals of the World Summit for
Children for the year 2000. The end-decade MICS (MICS2) collected data for 63 of the
75 Indicators for Monitoring Progress at End-Decade.14 MICS2 drew heavily on

13
Ibid 49.
14
The World Summit for Children adopted 27 goals with 75 related indicators. The goals and indicators are
listed in Appendix 1 to: UNICEF, End-decade assessment: Indicators for assessing progress globally,
UNICEF Executive Directive CF/EXD/1999-03, New York, 23 April 1999.

4
experiences with the mid-decade MICS, which was conducted around 1995, and the
subsequent mid-decade MICS evaluation.
By 2001, 65 developing countries had carried out MICS2 studies. The studies
were conducted between 1999 and 2001 by government agencies (mostly Statistical
Offices and Ministries of Health) with the technical and financial support of UNICEF.
The MICS2 model questionnaire includes 19 core modules and 4 optional modules to
obtain information for households, household members, women 15 to 49 years of age,
and children under five years of age. Since the main objective of MICS2 was to help
countries fill data gaps, not all modules were necessarily included in a country’s
questionnaire. With the data from the surveys, the respective governments completed
country reports that documented the progress toward the end-decade goals defined at the
World Summit for Children. The results are presently being used by national
governments to define priorities regarding women and children for the period 2000-
2005.15 UNICEF used the results to prepare the report We the Children that the UN
Secretary-General presented at the Special Session for Children in May 2002,16 the
accompanying statistical review,17 and the outcome document, A World Fit for
Children.18 The survey data also helped define UNICEF’s medium-term strategic plan
for the period 2000-2005.19
The MICS surveys collected information on the number of hours children aged 5
to 14 years spent working for others (paid or unpaid); working for the family, whether on
a farm or in a business (paid or unpaid); and on the time spent in household chores such
as cleaning, fetching water, laundry, or child care. However, due to limitations of the
methodology, it was not possible to collect data on the kind of work children engaged in.
Thus, MICS data cannot be used to analyse the worst forms of child labour (bonded
labour, prostitution, drug trafficking, etc.).
A further source of data are the Demographic and Health Surveys (DHS) that are
conducted by Macro International with funding from the U.S. Agency for International
Development (USAID). As a result of coordination between USAID and UNICEF, some
countries implementing DHS surveys decided to include the child labour module from
the MICS.20

2.2 Creating an equalizing indicator between girls and boys: The role of household
chores in the measurement of child labour

Because MICS and DHS provide data on household chores it is possible to apply
the principle of non-discrimination by making the extent of girls’ engagement in child
labour visible. A study by Friedrich Huebler and Edilberto Loaiza of UNICEF’s
Strategic Information Section compared the rates of child labour with and without

15
MICS documentation and results can be obtained at a dedicated UNICEF Web site, www.childinfo.org.
16
UNICEF, We the Children: Meeting the promises of the World Summit for Children (New York:
UNICEF, 2002).
17
UNICEF, Progress since the World Summit for Children: A statistical review (New York: UNICEF,
2001).
18
United Nations, A world fit for children, UN Doc. A/S-27/19/Rev. 1. (New York: UN, 2002).
19
UNICEF, Medium-term strategic plan for the period 2002-2005, UN Doc. E/ICEF/2001/13, November 7
(New York: UNICEF, 2001).
20
Documentation and survey data can be obtained at the DHS Web site, www.measuredhs.com.

5
household chores in 25 countries from Sub-Saharan Africa.21 To ensure that the extent of
girls’ labour is not underestimated, UNICEF includes household chores in excess of four
hours per day in calculations of child labour. Household chores of less than four hours
per day are, for the sake of argument, not considered harmful to the child’s development,
and hence are not counted as child labour. Of course, depending on the chore, one could
question this assertion; four hours a day carrying heavy buckets of water could certainly
be both physically detrimental to the child, and prevent him or her from attending
school.22 Even though with a minimum measure of four hours household chores per day,
the extent of child labour may be underestimated, without some arbitrary cut-off, almost
all children could be considered to be engaged in child labour.
The inclusion of at least 28 hours per week of household-chores in UNICEF’s
definition of child labour creates, however imperfectly, an “equalizing indicator” between
boys and girls, and expands the usual ILO definition of child labour as follows:

Indicator Definition
1. Child labour (with Ages 5-11: at least (a) one hour of economic activity or (b) 28 hours of
household chores) household chores per week.
Ages 12-14: at least (a) 14 hours of economic activity or (b) 28 hours of
household chores per week.
2. Child labour (without Ages 5-11: at least one hour of economic activity per week.
household chores) Ages 12-14: at least 14 hours of economic activity per week.

The study of 25 African countries by Huebler and Loaiza indeed shows that once
household chores are included, the disparity in child labour rates for boys and girls
narrows by more than half, as illustrated by table 1.
Table 1: Child labour in Sub-Saharan Africa, 25 countries, children 5-14 years
Without Household Chores With Household Chores
Total Child Labour 25.3% 30.8%
Girls 23.6% 30.2%
Boys 27.0% 31.5%
Difference boys-girls 3.4% 1.5%
Source: Friedrich Huebler and Edilberto Loaiza, Child Labour and school attendance in Sub-Saharan
Africa: Empirical evidence from UNICEF’s Multiple Indicator Cluster Surveys, Working Paper (New
York, UNICEF, 2002).

When household chores are taken into account, the estimate of child labour among girls
is increased by more than one quarter, from 23.6 percent to 30.2 percent. In contrast, the
estimate of child labour among boys is increased by one sixth, from 27.0 percent to 31.5
percent. The gap between boys’ and girls’ child labour falls from 3.4 percent, to 1.5
percent. Thus, despite its imperfections, the UNICEF indicator of child labour, which
includes household chores, contributes to a better visibility and understanding of gender

21
Friedrich Huebler and Edilberto Loaiza, Child Labour and school attendance in Sub-Saharan Africa:
Empirical evidence from UNICEF’s Multiple Indicator Cluster Surveys, Working Paper (New York,
UNICEF, 2002).
22
Researchers at the joint ILO-UNICEF-World Bank project “Understanding Children’s Work,” based at
UNICEF’s Innocenti Research Centre in Florence, are currently engaged in studies to determine at what
threshold the number of hours of household chores becomes detrimental to school attendance and
achievement.

6
disparities in child labour. As such, its use allows policy makers to better apply the
human rights principle of non-discrimination.

3. Does Participation in Child Labour Significantly Constrain School Attendance and

Achievement?
3.1 Data coverage and analytical model

The present study, stimulated by the ambiguous results of the UNICEF

programme “Education as a preventive strategy against child labour,” seeks, through
analysis of household survey data, to understand quantitatively (a) the extent to which
child labour is, in and of itself (i.e., independent of other factors) preventing school
attendance and contributing to repetition and dropout rates, and (b) the extent to which
other factors may influence the realisation of children’s right to education. The present
study covers the 18 countries listed in table 2, which is a reduced sample from that in the
study by Huebler and Loaiza.23 For 14 countries, data from MICS surveys was used and
for the remaining four countries, DHS data was used.

Table 2: Data sources and country population

Country Survey Year Population in
2000 (1,000)
Burundi MICS 2000 6,267
Central African Rep. MICS 2000 3,715
Comoros MICS 2000 705
Congo (DRC) MICS 2000 48,571
Côte d'Ivoire MICS 2000 15,827
Gambia MICS 2000 1,312
Guinea-Bissau MICS 2000 1,367
Kenya MICS 2000 30,549
Lesotho MICS 2000 1,785
Malawi DHS 2000 11,370
Mali DHS 2001 11,904
Niger MICS 2000 10,742
Senegal MICS 2000 9,393
Sierra Leone MICS 2000 4,415
Somalia MICS 1999 8,720
Swaziland MICS 2000 1,044
Tanzania DHS 1999 34,837
Uganda DHS EdData 2001 23,487
Population figures from: United Nations Population Division, World Population Prospects: The 2002
revision (New York: United Nations, 2003).

23
In some of the 25 Sub-Saharan African countries in the earlier study, surveys were conducted during a
period of school vacation, which means that no concurrent data on child work and school attendance was
available. In these countries it is thus not possible to evaluate the trade-off between school attendance and
child labour.

7
We use the same definition of child labour as the earlier study, whereby household chores
of 28 hours or more per week are included, in order to introduce an equalizing indicator
and make girls’ work as visible as boys’ work. Our working hypothesis is that being
engaged in child labour invariably constrains school attendance. To test this, we want to
investigate the extent to which child labour, together with other socioeconomic factors
such as gender and wealth, is a significant factor constraining school attendance and
educational attainment among children 7 to 14 years of age.24
For the purpose of this study we define a child to be in school if he or she was
attending either primary or secondary education at the moment of the survey. Children in
preschool or in non-standard schools are counted as not attending school. A child is
considered to have dropped out from school if he or she was attending school the year
before the survey and is not currently attending. Similarly, a child is considered to be a
repeater if, at the moment of the survey, the grade of school attended is the same as
during the year before the survey. Repetition and dropout rates are taken as proxies for
educational achievement, although it is recognised that such measures cannot fully reflect
school performance. Since the surveys represent a point in time, they cannot account for
the cumulative effect of child labour on attainment over time.
A regression analysis of the data seeks to show the relative weight of six factors
in influencing school attendance, grade repetition and dropping out. These factors are:
age (7-10 or 11-14 years), gender, area of residence (rural or urban), household wealth
(indicated by wealth quintile), mother’s or caretaker’s education (for more than 9 out of
10 children the caretaker is the mother), and child labour. This is a very simple model
that cannot take into account any variable related to access to school (in terms of the non-
existence of a school as a reason for non-attendance), to the quality of education
(availability of books, materials, or trained teachers, which is known to have a strong
influence on parents’ preference for work over school), or to the intensity of child labour
(beyond the hourly threshold established by the definition above). As previously noted,
due to data collection limitations, the extent of worst forms of child labour cannot be
analysed. However, the model does have the merit of producing quantitative data,
comparable over a large number of countries. By showing the relative weight of each of
these factors, analysis of the data should present some considerations for policies aimed
at increasing school attendance and eliminating child labour.

3.2 School attendance and child labour: Descriptive statistics

Table 3 and figure 1 summarise the school attendance and child labour rates from
the 18 countries in this study.
Table 3: School attendance and child labour in Sub-Saharan Africa (%), children 7-14 years
Category Attending Child School School CL, no No
school labour only and CL school school,
no CL
Burundi 47.3 28.2 36.0 11.3 17.0 35.7
CAR 49.8 60.8 21.6 28.2 32.6 17.6
Comoros 35.9 29.0 25.7 10.2 18.8 45.3

24
Although child labour data for children aged 5 to 14 years is available, we limit the analysis to children 7
years and older because 7 years is the minimum age by which children in all countries are supposed to
attend primary school, according to national legislation.

8
Congo (DRC) 61.7 31.8 48.2 19.3 12.4 20.1
Côte d'Ivoire 61.7 39.0 41.9 19.7 19.3 19.1
Gambia 52.7 21.6 43.5 9.2 12.5 34.8
Guinea-Bissau 43.3 54.7 25.8 17.5 37.1 19.5
Kenya 81.3 28.8 59.6 21.7 7.1 11.6
Lesotho 75.3 20.0 61.4 14.0 6.0 18.7
Malawi 83.4 20.9 66.1 17.3 3.5 13.1
Mali 38.7 40.8 27.7 11.0 29.7 31.6
Niger 30.2 71.8 12.2 17.9 53.8 16.0
Senegal 47.0 39.5 33.2 13.8 25.7 27.3
Sierra Leone 43.2 58.5 19.3 23.8 34.6 22.2
Somalia 13.4 41.9 8.9 4.5 37.4 49.2
Swaziland 77.8 10.1 70.1 7.8 2.3 19.9
Tanzania 51.1 42.4 31.6 19.5 22.9 26.0
Uganda 87.8 43.4 49.2 38.6 4.8 7.4
Male 62.4 37.9 42.8 21.1 16.8 19.3
Female 58.3 38.2 40.1 19.2 19.0 21.7
Total 60.3 38.0 41.4 20.1 17.9 20.5
Notes: Averages are weighted by country population. – School attendance rate for Congo (DRC) is for
children 7-14 years, remaining columns for Congo show estimates for children 10-14 years because no
child labour data was available for ages below 10 years.

Figure 1: Activity status, children 7-14 years

Boys Girls

19.3% 21.7%

42.8% 40.1%

16.8%
19.0%

21.1% 19.2%

Total

20.5%

41.4%

17.9% School only

School and child labour
20.1% Child labour only
No school, no child labour

In the sample overall, 60 percent of children 7 to 14 years of age are attending school. It
is also evident that for many children, labour is part of their daily lives: 38% of all
children are labourers. Among these children, slightly more than half (or 20 percent) also
attend school while another 18 percent are only engaged in labour and have their right to
education denied. On the other hand, among the non-labouring children, two thirds attend
school. In spite of significant overlap between school attendance and child labour, 41

9
percent of the children attend school only and the remaining 21 percent are neither in
school nor working.25
Girls attend school less than boys (58 percent compared to 62 percent) but thanks
to the inclusion of household chores in the analysis, we are able to see that the share of
child labourers among girls is the same as among boys—about 38 percent. Because of
their lower attendance rate, girls are more likely to be engaged in child labour only (19
percent compared to 17 percent for boys), or to neither attend school nor do child labour
(22 percent compared to 19 percent for boys).
Figure 2 plots the child labour and school attendance rates against each other. We
observe that in countries with a high proportion of child labourers, school attendance
tends to be low. At the extreme ends of the distribution are Swaziland, with a school
attendance rate of 78 percent and a child labour rate of 10 percent, and Niger, with a
school attendance rate of 30 percent and a child labour rate of 72 percent. Uganda and
Somalia appear to be two outliers. In Uganda, 43 percent of all children are labourers
and yet 88 percent of all children are in school. Somalia has the lowest school attendance
rate of all countries, with 13 percent, but with a share of 42 percent there are almost as
many child labourers as in Uganda.

Figure 2: Child labour and school attendance in Sub-Saharan Africa, children 7-14 years

100

Malawi Uganda
Swaziland Kenya
80
Lesotho
School attendance (%)

Cote d'Ivoire
Congo (DRC)
60
Country average
Gambia Tanzania CAR
Burundi Senegal
Sierra Leone
40 Mali Guinea-Bissau
Comoros
Niger

20
Somalia

0
0 10 20 30 40 50 60 70 80
Child labour (%)

The evidence from table 3 and figures 1 and 2 indicates that child labour is a constraint
for school attendance but many other factors can have an effect on the schooling decision
for a child. A multivariate analysis is necessary to examine the relative weight of some
of the other determinants of school attendance. As a first step, table 4 disaggregates the

25
The 21% not in school and not in child labour includes 2% doing light work (1 to 13 hours of
economic activities per week among children 11-14), 13% doing 1 to 27 hours of household
chores per week, and 6% that are idle (no school and no work at all).

10
estimates for the activities of children in the 18 countries in our sample by various
background characteristics.
Table 4: School attendance and child labour by background characteristics (%), 18
countries in Sub-Saharan Africa, children 7-14 years
Category Attending Child School School CL, no No
school labour only and CL school school,
no CL
7 years 41.8 29.6 30.0 11.0 18.6 40.4
8 years 53.9 36.6 36.4 17.3 19.3 27.0
9 years 60.4 43.2 36.9 23.7 19.5 19.8
10 years 63.8 47.6 36.3 27.5 20.1 16.1
11 years 68.2 50.2 36.6 31.6 18.5 13.2
12 years 67.4 32.2 50.9 16.4 15.8 16.8
13 years 66.5 34.1 49.4 17.0 17.1 16.5
14 years 65.2 35.8 48.0 17.2 18.7 16.1
7-10 years 54.7 39.9 35.6 21.2 18.7 24.5
11-14 years 66.7 37.7 46.5 20.3 17.5 15.8
Female 58.3 38.2 40.1 19.2 19.0 21.7
Male 62.4 37.9 42.8 21.1 16.8 19.3
Rural 55.5 42.7 35.6 21.4 21.3 21.7
Urban 73.3 24.6 58.6 15.3 9.3 16.7
Poorest quintile 48.8 44.7 31.2 19.4 25.3 24.1
Second quintile 52.1 43.9 33.0 20.3 23.7 23.0
Third quintile 57.1 39.5 38.0 20.6 19.0 22.4
Fourth quintile 65.8 36.4 44.9 22.3 14.2 18.7
Richest quintile 78.4 23.8 61.9 16.7 7.1 14.2
Caretaker no ed. 51.4 40.6 34.4 18.4 22.1 25.1
Caretaker ed. 72.7 33.1 51.4 22.6 10.5 15.5
Total 60.3 38.0 41.4 20.1 17.9 20.5
Note: Averages are weighted by country population.

A number of important observations can be made from table 4. The school attendance
rate of girls is 4 percentage points below that of boys. School attendance increases with
age, reaching a peak at 11 years and falls again. If the child lives in an urban area, his or
her chance of attending school is three in four, compared to roughly one in two, if he or
she lives in a rural area.
As one might expect, a child in the richest 20 percent of the population is much
more likely to be in school and not working than one in the poorest 20 percent; children
from the richest households are approximately twice as likely to be in school only, and
the gap between the two groups of children is 31 percentage points. Further, a child from
the richest quintile is almost half as likely to be working as a child from the poorest
quintile. Children from the poorest quintile have a fifty-fifty chance of not being in
school at all, but at the same time it is important to emphasize that even among the richest
20 percent of the population one fifth of all children do not attend school. This could be
due to lack of access to schools, even for the richest, poor quality of education (leaving
parents to believe school is a waste of time for their children), or simply that in poor
countries only the top 5 percent of families have the wherewithal to fulfil their children’s
right to an education.
The other very striking observation is the role of mother’s education as a
determinant of school attendance, with 73 percent of children of educated mothers

11
attending, compared with only 51 percent of children of uneducated mothers.26 This
shows, in dramatic and incontrovertible terms, the inter-generational payoff from an
investment in girls’ education. The potential development gain from that inter-
generational pay-off also reinforces the need to understand and address the obstacles
preventing girls from attending school today, and hence for non-discriminatory indicators
capable of revealing those obstacles.

3.3 Regression analysis: Determinants of school attendance

To assess the statistical significance of the findings in the preceding section, a

multivariate analysis was implemented for all 18 countries. In the model, the probability
of school attendance is assumed to be a function of six factors: the child’s age and
gender, the place of residence (urban or rural), the wealth of the household, the mother’s
or caretaker’s education, and the child labour status. For each country, the following set
of binary variables is created:

Variable Description
Dependent variable:
School attendance 1 if the child is attending school, 0 if not
Independent variables:
Age 11-14 years 1 if the child is 11-14 years old, 0 if 7-10 years old
Male 1 if the child is male, 0 if female
Urban 1 if the household is located in an urban area, 0 if in a rural area
Second quintile 1 if the household is in the second poorest wealth quintile, 0 if not
Third quintile 1 if the household is in the third wealth quintile, 0 if not
Fourth quintile 1 if the household is in the second richest wealth quintile, 0 if not
Richest quintile 1 if the household is in the richest wealth quintile, 0 if not
Caretaker educated 1 if the mother or caretaker has a formal education, 0 if not
Child labour 1 if the child is a labourer, 0 if not

For households in the poorest wealth quintile, all four household wealth variables are
equal to 0. We use a probit regression to test the model, with school attendance as the
dependent variable.
The results of the regression are presented in table A1 in the appendix. For each
country, the table shows the net attendance ratio (NAR), that is the share of all children
aged 7-14 years who are in school, as well as the number of observations in the sample.27
For example, the Burundi sample consisted of 5,166 children between 7 and 14 years,
with a net school attendance ratio of 47.3 percent.
Instead of regression coefficients, the marginal effects are listed. The marginal
effect is the change in the dependent variable for a one-unit change in the independent
variable — from girl to boy (0 to 1) or rural to urban (again 0 to 1), for example. In
Burundi, the marginal effect of the variable “Male” on school attendance is 0.092; this
means that compared to girls, boys have a probability of attending school that is 9.2
26
A mother or caretaker is considered to be educated if she or he has attended any formal education at
primary level or higher. Those who attended only preschool or non-standard schools are considered to have
no formal education.
27
In a few cases (Congo, Somalia, Swaziland) the school attendance rates differ from those in table 3. This
is because the regressions were run with a reduced number of observations, due to missing values in some
of the variables.

12
percentage points higher. The marginal effect of the variable “Child labour” is -0.104 in
Burundi, meaning that the probability of attending school is 10.4 percentage points lower
for child labourers compared to non-labourers.
Statistical significance is indicated with asterisks; if a marginal effect has no
asterisk the effect of the respective variable on school attendance is statistically
insignificant. In the case of Burundi, belonging to a household in the second wealth
quintile has no statistically significant effect on school attendance; this means that there
is no difference in school attendance between these children and those from the poorest
quintile (the reference category for the household wealth variables).
To ease interpretation, the regression results are summarized in table 5. The table
counts the countries where each of the independent variables has (a) a positive and
statistically significant effect on school attendance, (b) a negative and statistically
significant effect, or (c) no statistically significant effect. The last two columns list the
mean values of the positive and negative marginal effects, weighted by country
population.
Table 5: Summary of multivariate analysis in table A1: determinants of school attendance, children
7-14 years (18 countries, average attendance rate 60.3%)
Variable Positive and Negative and Insignificant Average marginal effect
significant* significant* Positive and Negative and
significant* significant*
Age 11-14 years 12 2 4 0.185 -0.030
Male 10 2 6 0.120 -0.026
Urban 11 3 4 0.079 -0.064
Second quintile 8 1 9 0.058 -0.053
Third quintile 16 . 2 0.075 .
Fourth quintile 17 . 1 0.131 .
Richest quintile 18 . . 0.210 .
Caretaker educated 18 . . 0.170 .
Child labour 1 10 7 0.017 -0.092
* Minimum 5% level of significance. – Averages are weighted by country population.

The most striking finding of this analysis is that belonging to the richest
household quintile and having a formally educated mother or caretaker have a positive
and statistically significant effect on school attendance in all 18 countries. These effects
are statistically significant regardless of other variables in the model (age, sex, place of
residence, and child labour). Moreover, their marginal effect is substantial. A wealthy
child’s chances of attending school are increased on average by 21 percentage points
(compared to a child from the poorest 20 percent of the population) and the chances of a
child with an educated mother by 17 points. Compared to the bottom 20 percent of the
population in terms of household wealth, children from the third and fourth wealth
quintiles also have a significantly increased probability of attending school.
Almost as striking is that child labour, as a constraining determinant of school
attendance, is significant in more than half of all countries (10 of 18). Among these
countries, the strongest negative effect of child labour can be observed in Mali and
Senegal, where the likelihood of attending school is reduced by 14 percentage points for
child labourers. The average among the countries in which child labour has a negative
and significant effect on school attendance is -9 points, ranging from -14 points to -4
points in Lesotho. In other words, in and of itself, being a labourer would on average

13
decrease a child’s likelihood of attending school by 9 percentage points, compared to a
child who is not working.
In one country, Uganda, the opposite is observed; here child labour has a positive
effect on school attendance (+2 points). In the remaining 7 countries, no relationship
between child labour and school attendance can be observed. Although child labour
should be eliminated because of its violation of children’s basic rights, the findings for
these 7 countries indicate that doing so, even if possible, would not significantly increase
school attendance nationwide.28 Thus other strategies are required for children to realize
their right to education. In the Central African Republic, for example, where child labour
is not significantly associated with school attendance, the findings point to the
importance of poverty (being in the richest quintile of the population produces a 30 point
difference in the levels of school attendance, compared to the poorest quintile), mother’s
education (20 points difference), gender and place of residence (13 points difference in
both cases).
For 11 of the 18 countries we also found that children living in urban areas tend to
have greater levels of school attendance. However, in 3 other countries (Côte d’Ivoire,
Kenya, and Uganda) this relationship works in the opposite direction, meaning that
residents of rural areas tend to have greater levels of school attendance. In the remaining
4 countries the relationship between area of residence and school attendance is
statistically insignificant. The average marginal effect for the 11 countries in our sample
in which living in an urban area has a positive and significant effect on school attendance,
compared with rural residents, is 8 percentage points. Conversely, for the 3 countries in
which the opposite is true, there is an average marginal effect of 6 percentage points in
favour of rural children. Policy interventions that target urban or rural populations need
to be informed by the findings presented here, and tailored to the specific situation.

3.4 The role of gender in determining school attendance

It is frequently argued that, in general, girls tend to attend school less than boys.
For the 18 countries in our sample, boys have an attendance rate that is 4 percentage
points above that of girls, as shown in table 6. The gender disparity is greatest in the
Central African Republic, Côte d’Ivoire, Mali, and Niger, with about 12 percentage
points difference between boys’ and girls’ school attendance ratios.29 In a few countries,
girls are more likely to be in school than boys, including Lesotho (9 points difference)
and Tanzania (3 points difference).

28
This conclusion does not preclude that child labour may have an effect on school attendance at a sub-
national level. Interventions targeted at certain provinces or parts of the population may help increase
school attendance.
29
These countries continue a widespread practice of early marriage, which may contribute to girls’ lower
school attendance.

14
Table 6: School attendance in Sub-Saharan Africa by gender (%), children 7-14 years
Country Total Male Female Male-
Female
Burundi 47.3 51.2 43.8 7.4
CAR 49.8 55.7 44.0 11.7
Comoros 35.9 35.5 36.3 -0.8
Congo (DRC) 61.7 66.0 57.6 8.4
Côte d'Ivoire 61.7 67.6 55.3 12.3
Gambia 52.7 56.5 49.2 7.3
Guinea-Bissau 43.3 46.7 39.9 6.8
Kenya 81.3 80.7 81.8 -1.1
Lesotho 75.3 71.0 79.6 -8.6
Malawi 83.4 82.6 84.2 -1.6
Mali 38.7 45.0 32.7 12.3
Niger 30.2 36.2 24.2 12.0
Senegal 47.0 51.1 42.9 8.2
Sierra Leone 43.2 45.9 40.4 5.5
Somalia 13.4 14.2 12.7 1.5
Swaziland 77.8 77.7 78.0 -0.3
Tanzania 51.1 49.6 52.6 -3.0
Uganda 87.8 87.8 87.7 0.1
Total 60.3 62.4 58.3 4.1
Note: Averages are weighted by country population.

The regression analysis confirms that, controlling for all other factors, girls are
less likely to attend school than boys in 10 of the 18 countries. In 2 of the other 8
countries (Lesotho and Malawi), boys are at a disadvantage compared to girls, and in the
remaining 6 countries being a boy or a girl does not have a significant effect on the
likelihood of school attendance (Comoros, Kenya, Somalia, Swaziland, Tanzania, and
Uganda).
However, when we examine the regressions to select only countries where there
were significant gender disparities in school attendance, we find a much stronger
correlation between child labour and school attendance than in the sample as a whole.
While overall, in 10 out of 18 countries child labour has a strong negative effect on
school attendance, 7 of these 10 countries (Burundi, Congo, Côte d’Ivoire, Gambia, Mali,
Niger, and Senegal) are among the 10 countries that also have school attendance rates
favouring boys (in the Central African Republic and Guinea-Bissau the impact of child
labour on school attendance is not significant, and in Lesotho girls are more likely to be
in school). Without further analysis at the country level it is not possible to say with
certainty why child labour tends to affect school attendance more in countries with
significant gender disparities. We can say, however, that child labour and gender are
important determinants of school attendance, as shown in Table 5, although their net
effects may be in the opposite direction of the conventional wisdom.
Delving further into the question of gender, we divided the samples for the 18
countries and ran separate regressions for boys and girls. The results for the individual
countries are shown in tables A2 and A3 in the appendix, and summaries are presented in
tables 7 and 8. In the 9 countries in which child labour has a negative and statistically
significant effect on school attendance, a girl engaged in child labour is on average 11
percentage points less likely to attend school (table 7), while a boy in the same

15
circumstances would be close to 8 percentage points less likely to attend school (table 8),
compared to their non-working peers. Moreover, while in three countries (Central
African Republic, Comoros, and Uganda) being a boy child labourer has a positive and
significant impact on his school attendance (possibly because it gives him both needed
financial resources and time-management skills), this is never the case for girls. In other
words, the impact of child labour on girls’ school attendance, when statistically
significant, is always negative and on average more severe than for boys. This gender-
differentiated effect is an important finding, one that may have been impossible to
observe without the inclusion of household chores in our definition of child labour. The
separate regressions also confirm a result from the summary regression shown in table 5:
for both boys and girls, having an educated mother or being in the richest 20 percent of
the population is not only positively associated with school attendance, it also has the
greatest net effects on attendance, compared to other factors like area of residence.

Table 7: Summary of multivariate analysis in table A2: determinants of school attendance, girls 7-14
years (18 countries, average attendance rate 58.3%)
Variable Positive and Negative and Insignificant Average marginal effect
significant* significant* Positive and Negative and
significant* significant*
Age 11-14 years 10 4 4 0.175 -0.053
Urban 9 4 5 0.148 -0.089
Second quintile 7 1 10 0.068 -0.071
Third quintile 15 . 3 0.078 .
Fourth quintile 17 . 1 0.141 .
Richest quintile 18 . . 0.220 .
Caretaker educated 18 . . 0.178 .
Child labour . 9 9 . -0.108
* Minimum 5% level of significance. – Averages are weighted by country population.

Table 8: Summary of multivariate analysis in table A3: determinants of school attendance: boys 7-14
years (18 countries, average attendance rate 62.4%)
Variable Positive and Negative and Insignificant Average marginal effect
significant* significant* Positive and Negative and
significant* significant*
Age 11-14 years 12 . 6 0.203 -0.055
Urban 9 1 8 0.113 .
Second quintile 6 . 12 0.069 .
Third quintile 14 . 4 0.089 .
Fourth quintile 17 . 1 0.120 .
Richest quintile 18 . . 0.199 .
Caretaker educated 18 . . 0.161 .
Child labour 3 9 6 0.026 -0.076
* Minimum 5% level of significance. – Averages are weighted by country population.

3.5 Child labour, repetition, and dropout

The analysis in the previous section showed that in our sample of 18 countries in
Sub-Saharan Africa child labour had no effect on school attendance in 7 countries, in one
country the effect was positive, and in the remaining 10 countries the effect on attendance
was negative. In addition, we found that the effect of child labour on girls’ school
attendance was substantially greater than the effect on boys’ attendance. In this section

16
we examine the effect of child labour on academic performance. Could it be that child
labourers who manage to attend school are too tired to learn? If this is so, then one
would expect that children repeating a grade or dropping out of school would be
disproportionately those engaged in child labour. As a recent study from Ghana and a
review of similar studies by the ILO have shown, work has a detrimental effect on
learning achievements in the key areas of language and mathematics.30
To test the hypothesis that children engaged in labour would have higher
repetition and dropout rates, the survey data was further analysed. Table 9 lists the
attendance rates among children aged 7 to 14 years in the current and previous year, the
share of repeaters, and the dropout rate. On average, 13 percent of the children in school
are repeaters, with a range of 1 percent in Tanzania to 36 percent in Comoros. Of the
children who were in school during the year preceding the survey, 5 percent had dropped
out by the time the surveys were conducted. The dropout rate ranges from 1 percent in
Mali to 28 percent in Comoros.

Table 9: Levels of school attendance, repetition and dropout in Sub-Saharan Africa (%), children 7-
14 years
Category Attending Attended Repeaters Dropouts
school school last
year
Burundi 47.3 40.0 23.8 4.3
CAR 49.8 48.2 33.9 5.6
Comoros 35.9 41.4 36.4 27.5
Congo (DRC) 61.7 57.1 20.9 10.2
Côte d’Ivoire 61.7 56.8 13.3 1.9
Gambia 52.7 45.1 3.6 1.6
Guinea-Bissau 43.3 34.5 16.5 4.9
Kenya 81.3 70.9 9.4 2.1
Lesotho 75.3 62.3 10.0 3.7
Malawi 83.4 75.6 29.9 2.2
Mali 38.7 34.6 10.4 1.3
Niger 30.2 25.5 8.8 3.5
Senegal 47.0 46.7 14.5 14.1
Sierra Leone 43.2 30.1 11.1 8.4
Somalia 13.4 10.4 3.3 13.1
Swaziland 77.8 68.1 13.2 6.4
Tanzania 51.1 39.8 0.9 1.9
Uganda 87.8 75.8 10.9 1.7
Total 60.3 52.9 13.0 5.1
Note: Averages are weighted by country population.

In Table 10, the data is disaggregated by various background characteristics. Young

children, boys, and rural children are more likely to repeat a grade. The dropout rate falls
from age 7 to age 11 and then rises again. Girls and rural children are in greater danger
of dropping out than boys and urban children. However, with the average dropout rate at
5 percent, there is on average less than one point difference between girls and boys,
between urban and rural children, and between children aged 7 to 10 and 11 to 14 years.

30
C. Heady, What is the effect of child labour on learning achievement? Evidence from Ghana, Innocenti
Working Paper No. 79, (Florence, Italy: UNICEF 2000). P.Orazem and V. Gunnarsson, Child labour,
school attendance and academic performance: A review, ILO/IPEC working paper (Geneva: ILO, 2003)

17
Both the repetition and dropout rate fall with increasing household wealth. Children with
a formally educated mother or caretaker are less likely to repeat a grade or drop out of
school. Finally, child labour is associated with higher repetition and dropout rates.

Table 10: School attendance, repetition and dropout for 18 countries in Sub-Saharan Africa
(%), children 7-14 years
Category Attending Attended Repeater Dropout
school school last
year
7 years 41.8 21.7 23.4 6.8
8 years 53.9 38.5 16.7 4.8
9 years 60.4 50.3 14.8 4.5
10 years 63.8 57.9 12.9 4.5
11 years 68.2 63.7 11.9 4.2
12 years 67.4 65.8 10.2 4.8
13 years 66.5 67.0 11.0 5.7
14 years 65.2 66.8 11.6 6.3
7-10 years 54.7 41.6 15.7 4.7
11-14 years 66.7 65.8 11.1 5.3
Female 58.3 51.2 12.8 5.3
Male 62.4 54.6 13.2 4.8
Rural 55.5 47.8 14.5 5.4
Urban 73.3 67.0 9.3 4.5
Poorest quintile 48.8 41.0 15.6 6.4
Second quintile 52.1 44.5 15.0 6.4
Third quintile 57.1 49.9 14.1 5.8
Fourth quintile 65.8 58.6 12.9 4.8
Richest quintile 78.4 71.4 9.3 3.2
Caretaker has no formal ed. 51.4 44.7 14.1 6.6
Caretaker has formal ed. 72.7 64.1 12.5 3.8
No child labour 65.3 58.3 11.8 4.3
Child labour 55.9 50.7 13.2 6.7
Total 60.3 52.9 13.0 5.1
Note: Averages are weighted by country population.

To test whether these findings are statistically significant, we run probit regressions with
repetition and dropout as the dependent variables. The set of explanatory variables is the
same as in the regressions for school attendance.
Variable Description
Repetition 1 if the child is repeating a grade, 0 if not
Dropout 1 if the child dropped out of school, 0 if not

Complete regression results for each country are shown in tables A4 and A5 in the
appendix. Table 11 summarises the results for grade repetition from table A4.31 The most
important determinant of repetition seems to be age: in 11 of 17 countries children 11 to
14 years old are less likely to repeat a grade than children 7 to 10 years old; the average
marginal effect is -7 percentage points. This is not surprising, given that young children
(most of whom will not have benefited from preschool education) have to adapt to the

31
The summary table for grade repetition excludes Somalia because of unreliable regression results for this
country.

18
rhythm and demands of school. The data may also show the existence of a selective
process which tends to promote students with higher abilities. However, in 3 countries
the net effect of age is reversed: in Mali, Niger, and Senegal older children are more
likely to repeat a grade.
The other explanatory variables are insignificant in the majority of countries.
Increasing household wealth reduces the likelihood of repetition, but this effect is mostly
limited to children from the richest quintile, and here only in 8 countries. Having an
educated caretaker reduces the repetition rate in 6 countries, on average by 3 percentage
points. Children from urban areas are less likely to repeat a grade in 5 countries and the
average marginal effect is relatively strong with -6 percentage points.
Table 11: Summary of multivariate analysis in table A4: determinants of grade repetition, children
7-14 years (17 countries, average repetition rate 13.0%)
Variable Positive and Negative and Insignificant Average marginal effect
significant* significant* Positive and Negative and
significant* significant*
Age 11-14 years 3 11 3 0.047 -0.072
Male 4 1 12 0.017 -0.020
Urban 1 5 11 0.075 -0.064
Second quintile . 2 15 . -0.046
Third quintile 1 1 15 0.100 -0.044
Fourth quintile . 4 13 . -0.045
Richest quintile . 8 9 . -0.051
Caretaker educated 1 6 10 0.053 -0.031
Child labour 4 3 10 0.028 -0.023
* Minimum 5% level of significance. – Averages are weighted by country population.
Note: Somalia is excluded from this table because of unreliable regression results; marginal effects near 1 for the third, fourth, and
richest wealth quintile would strongly skew the mean values.

The evidence for child labour is mixed. As table 11 shows, child labour has a
significant net effect on grade repetition in 7 of 17 countries. However, in three of these
countries (Malawi, Senegal, and Uganda), children who worked were on average 2
percentage points less likely to repeat a grade than the average child. In the countries
where child labourers are more likely to repeat a grade (Comoros, Guinea-Bissau, Mali,
and Swaziland), the average marginal effect is 3 percentage points.
We can conclude that in most countries the likelihood of repeating a grade is not
significantly associated with changes in the independent variables included in the model
used in the present study. It would also seem that, contrary to our prior expectations,
being a child labourer does not automatically affect school performance, or at least not so
dramatically that the child cannot pass his or her end-of-year exams. At the same time,
we also need to recognize that household wealth and education of the primary caretaker
have significant net effects on grade repetition in more than one third of the countries in
our sample; for school attendance, as will be recalled, these factors had a significant
effect in all countries studied.
The regression results for school dropout, presented in table A5 in the appendix,
are similarly mixed. In the majority of the 18 countries, the determinants included in our
model have no effect on the probability of dropping out. On the other hand, we find that
children with educated caretakers are less likely to drop out in 8 countries, with an
average marginal effect of -3 percentage points. In 7 of 18 countries, children from the
richest household quintile are less likely to drop out; compared to children from the

19
poorest quintile their dropout rate is 5 percentage points lower. The effect of the area of
residence on the dropout rate is ambiguous. The marginal effect is statistically significant
in only 7 countries, and among those we observe a positive effect in 4 countries and a
negative effect in 3 countries. In 5 countries, younger children are more likely to drop
out than older children, and in one country, they are less likely to drop out.
The results for child labour as a determinant of dropping out are clearer than they
were for repeating a grade. Child labour increases the probability of dropping out of
school in 8 countries, with an average marginal effect of 3 percentage points. In one
country, the Central African Republic, being a child labourer lowers the probability of
dropping out, but only by 1 percentage point.
Table 12: Summary of multivariate analysis in table A5: determinants of dropping out of school,
children 7-14 years (18 countries, average dropout rate 5.1%)
Variable Positive and Negative and Insignificant Average marginal effect
significant* significant* Positive and Negative and
significant* significant*
Age 11-14 years 5 1 12 0.013 -0.031
Male 1 1 16 0.050 -0.031
Urban 4 3 11 0.037 -0.052
Second quintile 1 2 15 0.062 -0.019
Third quintile . 3 15 . -0.015
Fourth quintile 1 6 11 0.042 -0.033
Richest quintile . 7 11 . -0.054
Caretaker educated . 8 10 . -0.034
Child labour 8 1 9 0.029 -0.010
* Minimum 5% level of significance. – Averages are weighted by country population.

In summary, it is fair to say that in one third to half of all countries, we observed
significant net effects on dropping out of school that can be associated, though not
uniformly or predictably, with age, place of residence, household wealth, mother’s
education, or child labour. Once again we see that any policy aimed at school dropouts
or repeaters must be designed for each country individually.

4. Conclusions and Policy Recommendations

4.1 The human rights principle of non-discrimination applied to statistical analyses

The gender-differentiated impact of child labour revealed by this study confirms the
importance of applying the human rights principle of non-discrimination to analyses of
the nexus between school attendance and child labour. A universally recognized and
applied gender-equalizing indicator of child labour, such as that used here on an
experimental basis, must take household chores into account. The human rights approach
to analysis of issues in human development aims to reveal, as a prelude to appropriate
policy responses, who’s rights are not being fulfilled and why. This demands a more
refined approach to statistical analysis, requiring investigators to:

a) Use household survey data: Survey data can provide more accurate information than
administrative records. Survey data can be disaggregated and analyzed at various
levels, which helps reveal discrimination. Surveys are a good (perhaps the best) tool
to collect information on child labour; however they are unlikely to capture hidden

20
 child labour (this includes the worst forms of child labour: prostitution, bonded
labour, drug trafficking).

b) Use appropriate indicators: Indicators developed with the idea of revealing

discrimination provide a more accurate picture. Deciding what indicators to use
requires an understanding of prevailing cultural, ethnic and gender patterns which
may produce discrimination. A child labour indicator that considers household
chores shows that girls work as much or more than boys. This indicator also reveals
that girls fall into stereotypical roles assigned to women by traditional society (cook,
caretaker of children, etc.).

c) Propose new data collection methods: Current indicators based on available data may
hide existing discrimination. Equality in school attendance rates, for example, does
not mean that there is no discrimination: teacher attitudes may mean that girls are not
given as much opportunity to participate in class as boys, reducing their learning
achievement. Our analysis showed that child labourers are more likely to drop out,
but we do not know if child labourers who stay in school perform worse or the same
as their non-working peers. Thus, we need data on school achievement but existing
data is insufficient. The Program for International Student Assessment (PISA), which
was begun in 2000, provides data on reading, mathematical and scientific literacy of
15-year-olds in school but these surveys are mostly implemented in OECD countries
where child labour is less widespread than in Africa and other parts of the world.32

However, even when all these measures to reveal discrimination are considered,
we are often left with a result which will only show statistical averages within a larger
population. A weak link between child labour and education at the country level may
hide stronger relationships between the two variables in certain areas of the country or
among minority populations, which cannot always be identified because the survey data
may not be representative at a lower level. Thus, even in countries where child labour is
not a statistically significant constraint to school attendance, it still may prevent large
numbers of children from enjoying their right to education. More refined analysis will
always be required to determine who those children at the margins actually are, and
whether they are statistically invisible due to discrimination, poverty or other factors.
The human rights principle of universality means that ultimately, all children must have
their rights fulfilled. Policy makers cannot be satisfied with averages which hide
disparities.

4.2 Limitations of survey data and child labour indicator

In addition to inherent difficulties of applying the human rights principle of non-

discrimination to statistical analyses in general, the weak results with regard to child
labour for some countries in this study may at least partly be due to the way child labour
is measured. The data from the MICS and DHS surveys does not tell us exactly what
kind of work children do, only whether they work for a family farm or business, for
someone who is not a member of the household, or whether they are engaged in domestic
32
OECD, Literacy skills for the world of tomorrow: Further results from PISA 2000 (Paris: OECD, 2003).

21
work. Tending cows for one’s parents, perhaps accompanied by friends, and working
long hours on a cacao plantation far away from home both carry the same weight in our
analysis.33 Yet, these two activities are likely to have a different impact on a child’s
physical and mental well-being. If we find that child labour has no statistically
significant effect on school attendance and achievement in a country, then this is perhaps
due to the fact that the children engage in relatively light forms of labour.34 If such work
does not interfere with children’s education it would be less cause for concern from a
human-rights perspective, even though there could be other implications, such as the
child’s reduced enjoyment of the right to recreation, which is integral to his or her human
development.

4.3 Some reasons why children in Sub-Saharan Africa are not attending school

The main conclusion to be drawn from this study of 18 countries in Sub-Saharan Africa
is that the strongest determinants of school attendance are, overwhelmingly and without
exception, household wealth and mother’s education. This finding clearly indicates that
in order to increase the levels of school attendance, policies should focus on poverty
reduction, for example through income transfers to poor families. This would allow
parents to forgo the contribution of their children to household income, were the latter
attending school instead of working. The effect of household wealth is clearly seen in
figure 3, which plots child labour and school attendance rates for the poorest and richest
20 percent of the population in 18 countries. Children from the poorest households are
more likely to work and less likely to be in school than children from the richest
households.

33
Both activities are child labour, as long as they meet the minimum threshold of one hour per week for
children aged 5 to 11 years, and 14 hours per week for children aged 12 to 14 years.
34
The authors thank Craig Scott of York University for bringing this point to our attention.

22
Figure 3: Child labour and school attendance in Sub-Saharan Africa, by household wealth, children
7-14 years

100
Richest 20%
Poorest 20%

80 Richest average
School attendance (%)

60

Poorest average

40

20

0
0 10 20 30 40 50 60 70 80
Child labour (%)

Note: Each country is represented by two points, one for the richest and one for the poorest
household quintile.

Similarly, the findings related to mother’s education indicate that educating the mothers
and caretakers of the future will lead to increased levels of school attendance. This
finding should give renewed vigour to the international community’s commitment to
reach gender parity in primary school by 2005.35
The second conclusion to be drawn from the present study is that child labour is
one of the major factors constraining school attendance for the majority of the countries
analysed. For 10 of 18 countries, being engaged in child labour adds on average 9
percentage points to the probability that a child will not attend school. It is important to
mention also that the probability of not attending school is disproportionately skewed
against girls. For girls, the marginal effect of child labour on school attendance (-11
percentage points) is, in relative terms, 40 percent higher than the marginal effect of child
labour on boys’ school attendance (-7 percentage points). Considerably more research is
required to understand why this should be. One hypothesis could be that the countries
with high gender disparities are also those with the least developed economic and social
infrastructure. The absence of such infrastructure, such as rural electrification,
community water systems, preventive health services and preschools, all increase the
time which girls’ must spend on household chores, thus limiting time available for
attending school.

35
Target 4 of the Millennium Development Goals, adopted at the UN Millennium Summit in 2000, aims to
“eliminate gender disparity in primary and secondary education, preferably by 2005 and in all levels of
education no later than 2015” (UNDP, Human Development Report 2003, New York: Oxford University
Press, 2003).

23
 However, it is also important to mention that the net effects of child labour on
school attendance were not statistically significant in 7 countries. Our original
hypothesis, that child labour invariably constrains school attendance, has thus been
proved false. This study shows that there are other factors besides time spent working
that prevent children from realizing their right to education; these factors need to be
identified on a case by case basis to ensure that policy interventions are appropriate.
Such factors can include the quality of education, the physical access to schools, the
child’s nutritional and health status as well as cultural barriers. More research is
especially required to determine the causes of apparently anomalous results. For
example, of particular concern is the finding that around one in five children from the
richest 20 percent of the population are not attending school, even though household
wealth is the strongest determinant of school attendance.
With more than half of all child labourers also attending school, we expected that
due to less time available for studying, they would be more prone to repeat grades or drop
out of school, compared to their non-working peers. With regard to repetition, this
hypothesis is true for 4 countries, but the opposite is observed in 3 other countries. For
the remaining 10 countries, the marginal effects are not significant, which would also
appear to indicate that time spent in child labour does not affect school performance
enough to keep the child from being promoted to the next grade. One must also keep in
mind that with the MICS and DHS data it is not possible to show if this result would hold
over time because the surveys only show the relationship between labour and repetition at
one moment in time.
Finally, for 8 countries child labour increases the probability of dropping out of
school by 3 percentage points on average. On the other hand, in one country, the Central
African Republic, children involved in child labour tend to have dropout rates that are on
average 1 percentage point lower than among non-labouring children. In the remaining 9
countries, child labour was not significantly associated with the probability of dropping
out of school. Thus, the conventional wisdom that child labour significantly increases the
probability of dropping out of school has been proved true for only half of the countries
in the sample. This confirms the need for much more sophisticated causal analyses when
designing policy responses to the problem of retention in school. Attention must be paid,
for example, to the structure of the local economy (which could be a “pull” factor on
children’s participation in the labour market), the legal framework (including the age
until which schooling is compulsory), the minimum age of employment, and the
minimum age of marriage.36

4.4 Policy recommendations

Based on the results presented in this study, several recommendations for policy
can be made.
a) The importance of using a human rights lens when analyzing the data has been
confirmed. The human rights approach to statistical analysis is a useful tool for getting at
the causes of violations of each child’s right to education, revealing discriminations on

36
These ages can be inconsistent. Niger, for example, the country with the lowest rate of school attendance
in the study (with the exception of Somalia), has a school-leaving age of 16, a minimum employment age
of 14 and no legally-established minimum age of marriage.

24
the basis of gender and poverty. Better indicators, more refined survey methodologies
and deeper analyses are needed to identify the scope of the child labour problem and who
is affected, and its relationship with school attendance and achievement. A universally
accepted “gender equalizing” indicator of child labour must be created.
b) The present study, by demonstrating that the relationship between child labour
and education is not a simple or predictable one, has also confirmed that a holistic, multi-
sectoral development approach, consistent with the principle of indivisibility of rights, is
needed to address both issues. Only through a multi-sectoral response will all children be
enabled to enjoy the rights laid down in the Convention on the Rights of the Child.
c) Poverty reduction strategies must be vigorously pursued as an integral response
to poor school attendance, given that in all 18 countries in the study, poverty was a
constraining factor on school attendance for both boys and girls.
d) The mother’s education as a determinant for assuring a child’s school
attendance reinforces and gives renewed impetus to the Millennium Development Goals
for gender equity in schools by 2005, and universal primary education for girls and boys
by 2015. Countries must invest in girls’ education over the long-term.
As with most of the research in social sciences, the analysis in the present study
leaves unanswered many questions about the determinants of school access and
achievement. The reality is perhaps as complex and diverse as the analysed countries.
Overall, the study establishes that there is an intricate nexus of factors that influence
whether a child attends school or not, and child labour is but one of them. Even though
most parents want education for their children, some may perceive school as irrelevant,
since they see that children end up doing the same work as their peers who have dropped
out of school and are seen as getting a head start in the labour market. Other factors
affecting school attendance and achievement can include teachers’ and administrators’
attitudes, biased and irrelevant curricula, discriminatory and abusive treatment of children
such as corporal punishment and sexual abuse, or deprived conditions (poor
infrastructure, no books, poorly prepared teachers). In reality, the lack of respect for
children’s rights within the school, along with the poor quality of education itself, can
drive children out of the system and into child labour. To what extent this may be so is a
topic for further research.

25
 Appendix: Regression results
Table A1: Multivariate analysis (probit regression): marginal effects of the determinants of school attendance in 18 sub-Saharan countries, children 7-14 years

Burundi CAR Comoros Congo (DRC) Côte d’Ivoire Gambia Guinea-Bissau Kenya Lesotho
NAR 47.3 NAR 49.8 NAR 35.9 NAR 67.5 NAR 61.7 NAR 52.7 NAR 43.3 NAR 81.3 NAR 75.3
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years 0.136** 0.050** 0.084** 0.015 -0.034** 0.089** 0.148** 0.112** 0.139**
(9.56) (7.38) (6.50) (1.10) (3.89) (6.93) (11.02) (15.40) (13.26)
Male 0.092** 0.134** -0.006 0.114** 0.141** 0.091** 0.109** -0.011 -0.092**
(6.43) (20.04) (0.44) (10.46) (15.98) (7.22) (8.57) (1.49) (8.81)
Urban 0.102** 0.134** 0.108** 0.034* -0.029* 0.075** 0.295** -0.091** -0.031
(3.23) (15.78) (6.39) (2.10) (2.53) (4.74) (16.58) (5.59) (1.88)
Second wealth quintile 0.022 0.128** 0.019 -0.053** 0.091** 0.087** 0.019 0.055** 0.089**
(0.91) (12.02) (0.95) (3.14) (6.95) (4.93) (0.90) (5.76) (6.28)
Third wealth quintile 0.057* 0.146** 0.073** 0.016 0.106** 0.115** 0.051* 0.063** 0.141**
(2.35) (13.74) (3.64) (0.95) (8.10) (6.17) (2.43) (6.48) (10.15)
Fourth wealth quintile 0.139** 0.241** 0.083** 0.113** 0.177** 0.223** 0.162** 0.105** 0.190**
(6.20) (21.10) (4.07) (6.73) (12.02) (11.35) (7.32) (10.24) (13.55)
Richest wealth quintile 0.204** 0.304** 0.095** 0.212** 0.243** 0.310** 0.252** 0.087** 0.234**
(8.50) (25.44) (4.37) (10.47) (15.26) (13.50) (9.41) (5.75) (15.85)
Caretaker has formal ed. 0.165** 0.199** 0.097** 0.211** 0.218** 0.126** 0.218** 0.163** 0.146**
(9.86) (28.17) (6.18) (17.97) (21.52) (6.04) (11.28) (18.55) (8.21)
Child labour -0.104** 0.002 0.013 -0.068** -0.105** -0.053** -0.025 -0.062** -0.040**
(6.55) (0.34) (0.92) (5.78) (10.98) (3.39) (1.77) (7.63) (3.09)
Observations 5166 24780 5858 7671 13055 6697 7448 11206 6827
Probit regressions: * significant at 5%; ** significant at 1%

Malawi Mali Niger Senegal Sierra Leone Somalia Swaziland Tanzania Uganda
NAR 83.4 NAR 38.7 NAR 30.2 NAR 47.0 NAR 43.2 NAR 13.7 NAR 78.1 NAR 51.1 NAR 87.8
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years 0.053** -0.002 -0.015 -0.023* -0.006 0.054** 0.027* 0.430** 0.083**
(8.71) (0.25) (1.13) (2.55) (0.34) (5.59) (2.46) (25.83) (12.00)
Male -0.016** 0.148** 0.147** 0.096** 0.073** 0.013 -0.001 -0.026 0.001
(2.70) (17.95) (11.63) (10.89) (5.02) (1.33) (0.12) (1.56) (0.14)
Urban 0.036** 0.119** 0.208** 0.076** 0.121** -0.019 -0.022 0.038 -0.053**
(3.10) (9.37) (9.27) (5.92) (6.31) (1.55) (1.03) (1.45) (3.56)
Second wealth quintile 0.020* 0.027* 0.030 0.010 0.023 0.001 0.091** 0.012 0.005
(2.31) (2.06) (1.38) (0.72) (0.99) (0.05) (6.46) (0.47) (0.46)
Third wealth quintile 0.033** 0.055** 0.030 0.070** 0.105** 0.081** 0.109** 0.102** 0.043**
(3.91) (4.15) (1.48) (5.00) (4.50) (4.06) (7.21) (3.86) (4.60)
Fourth wealth quintile 0.080** 0.144** 0.029 0.167** 0.172** 0.151** 0.108** 0.175** 0.070**
(9.66) (10.67) (1.36) (10.22) (7.33) (6.40) (6.51) (6.76) (7.58)
Richest wealth quintile 0.108** 0.332** 0.228** 0.247** 0.286** 0.250** 0.089** 0.332** 0.085**
(11.40) (19.60) (9.36) (13.28) (10.54) (9.44) (4.23) (11.02) (8.06)
Caretaker has formal ed. 0.123** 0.240** 0.240** 0.268** 0.249** 0.095** 0.120** 0.127** 0.060**
(19.47) (16.72) (10.25) (21.44) (12.68) (8.00) (9.48) (7.24) (8.71)
Child labour -0.003 -0.143** -0.102** -0.135** 0.010 -0.016 0.001 -0.117** 0.017*
(0.37) (16.69) (7.05) (14.83) (0.60) (1.56) (0.05) (6.64) (2.51)
Observations 13886 15567 5787 14309 5119 4599 5434 4179 8627
Probit regressions: * significant at 5%; ** significant at 1%

26
Note: Data for Congo (DRC) is for children 10-14 years of age.

27
Table A2: Multivariate analysis (probit regression): marginal effects of the determinants of school attendance in 18 sub-Saharan countries, girls 7-14 years

Burundi CAR Comoros Congo (DRC) Côte d’Ivoire Gambia Guinea-Bissau Kenya Lesotho
NAR 43.8 NAR 44.0 NAR 36.3 NAR 62.4 NAR 55.3 NAR 49.2 NAR 39.9 NAR 81.8 NAR 79.6
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years 0.114** 0.003 0.069** 0.004 -0.070** 0.065** 0.118** 0.102** 0.139**
(5.81) (0.32) (3.72) (0.22) (5.25) (3.68) (6.30) (10.10) (10.12)
Urban 0.128** 0.174** 0.075** 0.046 -0.026 0.078** 0.284** -0.119** -0.077**
(2.87) (14.55) (3.08) (1.94) (1.53) (3.56) (11.55) (5.18) (3.49)
Second wealth quintile -0.019 0.092** 0.021 -0.071** 0.111** 0.072** 0.053 0.059** 0.062**
(0.57) (5.85) (0.72) (2.80) (5.38) (2.85) (1.65) (4.55) (3.44)
Third wealth quintile 0.037 0.141** 0.086** -0.032 0.103** 0.119** 0.079* 0.074** 0.109**
(1.08) (9.14) (2.97) (1.28) (5.08) (4.57) (2.56) (5.54) (6.13)
Fourth wealth quintile 0.118** 0.235** 0.109** 0.126** 0.193** 0.225** 0.211** 0.102** 0.163**
(3.79) (13.88) (3.70) (4.98) (8.49) (8.03) (6.44) (7.26) (9.13)
Richest wealth quintile 0.190** 0.311** 0.105** 0.239** 0.243** 0.302** 0.330** 0.104** 0.196**
(5.65) (17.94) (3.38) (7.79) (9.88) (9.33) (8.66) (5.26) (10.28)
Caretaker has formal ed. 0.166** 0.215** 0.092** 0.247** 0.235** 0.163** 0.217** 0.141** 0.130**
(7.36) (21.48) (4.06) (14.29) (15.90) (5.72) (8.14) (11.52) (5.64)
Child labour -0.132** -0.019 -0.015 -0.065** -0.140** -0.068** -0.016 -0.095** -0.007
(5.95) (1.86) (0.72) (3.82) (10.12) (3.14) (0.78) (8.22) (0.41)
Observations 2706 12378 2842 3890 6248 3526 3719 5600 3419
* significant at 5%; ** significant at 1%

Malawi Mali Niger Senegal Sierra Leone Somalia Swaziland Tanzania Uganda
NAR 84.2 NAR 32.7 NAR 24.2 NAR 43.0 NAR 40.4 NAR 12.9 NAR 78.2 NAR 52.6 NAR 87.7
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years 0.038** -0.025* -0.049** -0.063** -0.025 0.029* 0.023 0.418** 0.062**
(4.57) (2.19) (2.96) (5.04) (1.11) (2.15) (1.46) (17.56) (6.46)
Urban -0.010 0.082** 0.291** 0.083** 0.108** -0.038* -0.031 0.024 -0.070**
(0.65) (4.81) (9.78) (4.68) (4.04) (2.18) (1.06) (0.67) (3.33)
Second wealth quintile 0.025* 0.019 0.021 0.021 0.000 0.003 0.083** 0.049 0.017
(2.22) (1.07) (0.74) (1.04) (0.01) (0.09) (4.08) (1.32) (1.24)
Third wealth quintile 0.036** 0.040* 0.074** 0.111** 0.120** 0.089** 0.105** 0.041 0.069**
(3.11) (2.23) (2.63) (5.50) (3.53) (3.14) (4.85) (1.11) (5.45)
Fourth wealth quintile 0.092** 0.138** 0.035 0.208** 0.206** 0.183** 0.117** 0.180** 0.087**
(8.14) (7.44) (1.25) (8.90) (6.06) (5.23) (4.92) (4.93) (6.99)
Richest wealth quintile 0.103** 0.346** 0.178** 0.285** 0.330** 0.243** 0.068* 0.334** 0.094**
(8.02) (14.79) (5.54) (10.81) (8.46) (6.39) (2.29) (7.97) (6.60)
Caretaker has formal ed. 0.128** 0.275** 0.228** 0.291** 0.222** 0.074** 0.090** 0.125** 0.067**
(14.69) (13.97) (7.38) (16.33) (7.93) (4.41) (5.04) (5.16) (6.98)
Child labour -0.007 -0.120** -0.119** -0.132** 0.036 0.000 0.011 -0.147** 0.008
(0.68) (10.47) (6.62) (10.29) (1.60) (0.03) (0.42) (5.87) (0.86)
Observations 7162 7938 2920 7187 2523 2211 2666 2087 4390
* significant at 5%; ** significant at 1%
Note: Data for Congo (DRC) is for girls 10-14 years of age.

28
Table A3: Multivariate analysis (probit regression): marginal effects of the determinants of school attendance in 18 sub-Saharan countries, boys 7-14 years

Burundi CAR Comoros Congo (DRC) Côte d’Ivoire Gambia Guinea-Bissau Kenya Lesotho
NAR 51.2 NAR 55.7 NAR 35.5 NAR 72.8 NAR 67.6 NAR 56.4 NAR 46.7 NAR 80.7 NAR 71.0
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years 0.161** 0.099** 0.101** 0.023 -0.001 0.113** 0.172** 0.124** 0.138**
(7.83) (10.42) (5.51) (1.25) (0.06) (6.07) (9.10) (11.83) (8.77)
Urban 0.073 0.090** 0.140** 0.020 -0.025 0.072** 0.301** -0.055* 0.018
(1.63) (7.61) (5.97) (0.91) (1.64) (3.13) (11.85) (2.41) (0.74)
Second wealth quintile 0.063 0.158** 0.017 -0.036 0.074** 0.100** -0.005 0.050** 0.119**
(1.86) (11.24) (0.60) (1.66) (4.61) (4.08) (0.17) (3.64) (5.38)
Third wealth quintile 0.075* 0.151** 0.060* 0.058** 0.107** 0.107** 0.032 0.053** 0.175**
(2.18) (10.55) (2.16) (2.72) (6.51) (4.07) (1.14) (3.76) (8.15)
Fourth wealth quintile 0.158** 0.243** 0.059* 0.098** 0.156** 0.217** 0.124** 0.109** 0.219**
(4.98) (16.15) (2.09) (4.48) (8.42) (7.94) (4.11) (7.36) (10.05)
Richest wealth quintile 0.216** 0.291** 0.086** 0.180** 0.238** 0.313** 0.175** 0.065** 0.275**
(6.36) (18.02) (2.85) (6.84) (11.80) (9.61) (4.57) (2.85) (12.10)
Caretaker has formal ed. 0.164** 0.176** 0.103** 0.174** 0.198** 0.086** 0.218** 0.185** 0.164**
(6.64) (18.08) (4.73) (11.11) (14.60) (2.83) (7.83) (14.74) (6.09)
Child labour -0.073** 0.024* 0.042* -0.071** -0.066** -0.036 -0.036 -0.028* -0.068**
(3.23) (2.43) (2.07) (4.39) (5.09) (1.62) (1.78) (2.38) (3.54)
Observations 2460 12402 3016 3781 6807 3171 3729 5606 3408
* significant at 5%; ** significant at 1%

Malawi Mali Niger Senegal Sierra Leone Somalia Swaziland Tanzania Uganda
NAR 82.6 NAR 45.0 NAR 36.2 NAR 51.1 NAR 45.9 NAR 14.4 NAR 78.0 NAR 49.6 NAR 87.9
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years 0.071** 0.020 0.021 0.016 0.012 0.076** 0.031* 0.448** 0.104**
(8.00) (1.61) (1.08) (1.30) (0.53) (5.58) (1.98) (19.04) (10.48)
Urban 0.093** 0.152** 0.114** 0.065** 0.133** 0.001 -0.011 0.050 -0.034
(5.57) (8.25) (3.42) (3.56) (4.90) (0.03) (0.34) (1.30) (1.58)
Second wealth quintile 0.013 0.035 0.043 0.002 0.047 -0.001 0.098** -0.024 -0.007
(1.08) (1.87) (1.31) (0.12) (1.47) (0.03) (5.05) (0.66) (0.48)
Third wealth quintile 0.030* 0.068** -0.005 0.034 0.097** 0.066* 0.113** 0.175** 0.015
(2.41) (3.62) (0.16) (1.77) (3.02) (2.38) (5.38) (4.62) (1.11)
Fourth wealth quintile 0.067** 0.145** 0.023 0.133** 0.141** 0.115** 0.100** 0.168** 0.051**
(5.58) (7.61) (0.74) (5.83) (4.34) (3.65) (4.30) (4.56) (3.71)
Richest wealth quintile 0.115** 0.315** 0.271** 0.214** 0.245** 0.249** 0.114** 0.334** 0.076**
(8.30) (13.10) (7.64) (8.21) (6.49) (6.77) (3.87) (7.64) (4.93)
Caretaker has formal ed. 0.116** 0.199** 0.241** 0.243** 0.273** 0.115** 0.148** 0.129** 0.052**
(12.71) (9.67) (7.06) (13.93) (9.93) (6.90) (8.31) (5.05) (5.35)
Child labour 0.004 -0.159** -0.079** -0.139** -0.017 -0.028 -0.007 -0.086** 0.026**
(0.40) (12.75) (3.51) (10.71) (0.76) (1.91) (0.27) (3.46) (2.69)
Observations 6724 7629 2867 7122 2596 2388 2768 2092 4237
* significant at 5%; ** significant at 1%
Note: Data for Congo (DRC) is for boys 10-14 years of age.

29
Table A4: Multivariate analysis (probit regression): marginal effects of the determinants of grade repetition in 18 sub-Saharan countries, children 7-14 years

Burundi CAR Comoros Congo (DRC) Côte d’Ivoire Gambia Guinea-Bissau Kenya Lesotho
Repeaters 23.8 Repeaters 33.9 Repeaters 36.4 Repeaters 17.4 Repeaters 13.3 Repeaters 3.6 Repeaters 16.5 Repeaters 9.4 Repeaters 10.0
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years -0.093** 0.002 -0.066** -0.064** 0.013 -0.020** -0.072** -0.040** -0.092**
(4.67) (0.22) (2.86) (4.51) (1.60) (2.94) (4.72) (5.84) (9.19)
Male 0.001 -0.014 0.007 0.005 0.003 0.017* 0.005 0.012 0.025**
(0.07) (1.56) (0.29) (0.48) (0.36) (2.56) (0.31) (1.89) (2.71)
Urban -0.041 -0.013 0.022 -0.079** -0.005 0.002 -0.035 -0.049** -0.040**
(1.18) (1.17) (0.77) (5.28) (0.45) (0.28) (1.54) (3.89) (3.40)
Second wealth quintile -0.056 0.028 -0.023 0.001 -0.008 0.018 0.021 -0.012 -0.011
(1.62) (1.60) (0.58) (0.07) (0.54) (1.76) (0.56) (1.27) (0.68)
Third wealth quintile -0.010 0.005 -0.008 -0.025 -0.013 -0.007 0.100** -0.003 -0.008
(0.29) (0.32) (0.20) (1.39) (0.94) (0.72) (2.66) (0.28) (0.49)
Fourth wealth quintile -0.037 -0.022 0.061 -0.007 -0.013 -0.008 0.041 -0.006 0.006
(1.21) (1.26) (1.65) (0.39) (0.87) (0.78) (1.16) (0.65) (0.40)
Richest wealth quintile -0.030 -0.076** -0.023 -0.037 -0.045** -0.010 0.072 -0.031* -0.009
(0.92) (4.26) (0.60) (1.76) (2.92) (0.88) (1.93) (2.00) (0.57)
Caretaker has formal ed. -0.023 0.053** -0.079** -0.009 0.001 -0.018 0.009 -0.028** -0.065**
(1.10) (5.71) (3.06) (0.74) (0.15) (1.85) (0.50) (3.46) (3.49)
Child labour -0.022 0.011 0.123** 0.021 0.002 -0.001 0.042* 0.007 0.008
(0.99) (1.22) (4.70) (1.69) (0.19) (0.09) (2.47) (0.99) (0.67)
Observations 1976 11560 1814 4827 7013 2877 2595 7688 4082
Probit regressions: * significant at 5%; ** significant at 1%

Malawi Mali Niger Senegal Sierra Leone Somalia Swaziland Tanzania Uganda
Repeaters 29.9 Repeaters 10.4 Repeaters 8.8 Repeaters 14.5 Repeaters 11.1 Repeaters 3.1 Repeaters 13.3 Repeaters 0.9 Repeaters 10.9
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years -0.164** 0.036** 0.051** 0.055** -0.097** -0.004 -0.033** -0.005 -0.078**
(18.06) (4.30) (4.01) (5.83) (5.81) (0.51) (2.89) (1.56) (9.97)
Male 0.005 -0.020* -0.016 -0.007 -0.022 0.000 0.027* 0.002 0.016*
(0.59) (2.36) (1.22) (0.78) (1.50) (0.04) (2.38) (0.68) (2.15)
Urban -0.085** 0.019 0.024 0.025 -0.033 0.009 0.075** -0.005 -0.042**
(5.76) (1.58) (1.32) (1.94) (1.85) (1.06) (3.22) (1.73) (3.10)
Second wealth quintile 0.026 -0.006 -0.047* -0.027 0.055 0.983 -0.032* 0.006 0.001
(1.76) (0.38) (2.30) (1.54) (1.44) (.) (2.04) (0.95) (0.07)
Third wealth quintile -0.026 0.014 -0.025 -0.010 0.005 0.948** -0.044** 0.003 -0.005
(1.83) (0.86) (1.24) (0.62) (0.15) (8.22) (2.71) (0.54) (0.43)
Fourth wealth quintile -0.052** 0.000 -0.054** -0.034 0.041 0.838** -0.047** -0.006 -0.038**
(3.71) (0.01) (2.68) (1.92) (1.36) (8.02) (2.72) (1.12) (3.34)
Richest wealth quintile -0.120** -0.016 -0.062** -0.050** -0.015 0.628** -0.065** 0.008 -0.038**
(8.12) (0.93) (2.68) (2.60) (0.49) (7.67) (3.11) (1.33) (3.00)
Caretaker has formal ed. 0.003 -0.026* 0.023 -0.036** -0.033* -0.011 0.003 0.005 0.011
(0.37) (2.45) (1.38) (3.38) (2.08) (1.50) (0.19) (1.76) (1.52)
Child labour -0.022* 0.019* -0.000 -0.028** 0.026 -0.010 0.051** 0.001 -0.021**
(2.07) (1.97) (0.02) (2.73) (1.60) (1.34) (2.61) (0.49) (2.77)
Observations 10505 5250 2007 5600 1488 388 3467 1809 6489
Probit regressions: * significant at 5%; ** significant at 1%
Notes: Data for Congo (DRC) is for children 10-14 years of age. – The results for Somalia are unreliable, possibly due to a small sample size. Somalia is therefore excluded from the summary in table 11.

30
Table A5: Multivariate analysis (probit regression): marginal effects of the determinants of dropping out in 18 sub-Saharan countries, children 7-14 years

Burundi CAR Comoros Congo (DRC) Côte d’Ivoire Gambia Guinea-Bissau Kenya Lesotho
Dropouts 4.3 Dropouts 5.6 Dropouts 27.5 Dropouts 10.0 Dropouts 1.9 Dropouts 1.6 Dropouts 4.9 Dropouts 2.1 Dropouts 3.7
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years 0.002 0.005 0.000 0.003 0.003 0.001 -0.007 0.006* 0.018**
(0.23) (1.26) (0.01) (0.25) (1.08) (0.21) (0.87) (2.05) (3.22)
Male 0.000 0.005 0.050** -0.031** 0.001 -0.005 -0.007 0.002 0.007
(0.05) (1.16) (2.84) (3.86) (0.46) (1.16) (0.90) (0.54) (1.28)
Urban 0.013 -0.011* 0.034 0.048** -0.001 0.003 -0.045** 0.015* -0.002
(0.69) (2.19) (1.57) (4.24) (0.21) (0.55) (3.34) (2.13) (0.22)
Second wealth quintile -0.005 -0.015* 0.062* -0.012 -0.003 0.008 0.039 0.001 -0.008
(0.35) (2.25) (2.04) (0.93) (0.68) (1.05) (1.82) (0.33) (1.02)
Third wealth quintile -0.014 -0.009 0.014 -0.018 -0.003 -0.005 0.011 -0.014** -0.016*
(1.11) (1.31) (0.49) (1.52) (0.59) (0.81) (0.57) (3.57) (2.18)
Fourth wealth quintile -0.017 -0.029** -0.011 -0.054** -0.004 0.003 0.021 -0.010* -0.020**
(1.50) (4.32) (0.38) (4.68) (0.73) (0.45) (1.13) (2.46) (2.68)
Richest wealth quintile -0.025* -0.024** 0.034 -0.095** -0.010 -0.012 0.023 -0.009 -0.032**
(2.05) (3.30) (1.12) (7.31) (1.93) (1.76) (1.15) (1.53) (3.99)
Caretaker has formal ed. -0.015 -0.014** -0.055** -0.040** -0.002 0.014 -0.014 -0.005 -0.016
(1.66) (3.16) (2.74) (4.33) (0.52) (1.93) (1.47) (1.41) (1.64)
Child labour 0.053** -0.010* 0.016 0.043** 0.005 0.000 -0.015 0.023** 0.020**
(5.00) (2.24) (0.78) (4.76) (1.46) (0.01) (1.62) (5.91) (2.72)
Observations 2065 12262 2560 5346 7186 2929 2740 7840 4250
Probit regressions: * significant at 5%; ** significant at 1%

Malawi Mali Niger Senegal Sierra Leone Somalia Swaziland Tanzania Uganda
Dropouts 2.2 Dropouts 1.3 Dropouts 3.5 Dropouts 14.1 Dropouts 8.4 Dropouts 13.0 Dropouts 6.4 Dropouts 1.9 Dropouts 1.7
Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect Marg. effect
(z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.) (z stat.)
Age 11-14 years 0.014** 0.006* 0.041** -0.031** -0.009 0.015 0.009 0.012 0.005
(5.75) (2.11) (5.33) (3.93) (0.64) (0.51) (1.14) (1.94) (1.71)
Male 0.004 -0.003 -0.002 0.014 -0.011 -0.005 -0.003 -0.008 -0.001
(1.89) (1.23) (0.31) (1.81) (0.85) (0.17) (0.33) (1.63) (0.38)
Urban -0.007 -0.001 0.004 -0.069** -0.008 0.110** -0.004 0.007 0.015*
(1.77) (0.19) (0.42) (6.18) (0.50) (2.98) (0.29) (0.96) (2.19)
Second wealth quintile -0.004 -0.002 0.006 0.002 -0.031 0.186 -0.031** 0.005 0.002
(1.51) (0.40) (0.38) (0.13) (1.14) (1.35) (3.10) (0.59) (0.42)
Third wealth quintile -0.002 0.006 -0.003 -0.006 -0.046 0.191 -0.027* -0.011 0.003
(0.52) (1.18) (0.26) (0.45) (1.90) (1.60) (2.57) (1.45) (0.69)
Fourth wealth quintile -0.010** -0.004 0.042* -0.013 0.020 0.094 -0.036** 0.002 -0.006
(3.33) (0.86) (2.35) (0.90) (0.78) (0.89) (3.16) (0.21) (1.32)
Richest wealth quintile -0.014** -0.007 0.010 -0.041* -0.012 0.015 0.009 -0.003 -0.010*
(4.13) (1.35) (0.73) (2.50) (0.47) (0.15) (0.59) (0.28) (2.08)
Caretaker has formal ed. -0.016** -0.007 -0.018* -0.100** -0.040** -0.007 -0.005 -0.008 -0.015**
(6.64) (1.82) (2.25) (9.81) (2.82) (0.23) (0.59) (1.53) (4.72)
Child labour 0.008** 0.001 0.014* 0.041** -0.007 -0.000 0.006 0.018** 0.005
(2.97) (0.28) (2.01) (4.75) (0.47) (0.00) (0.46) (3.12) (1.81)
Observations 10731 5320 2084 6512 1653 494 3718 1844 6610
Probit regressions: * significant at 5%; ** significant at 1%
Note: Data for Congo (DRC) is for children 10-14 years of age.

31