the determinants of child labor and schooling in the ... · the determinants of child labor and...

The Determinants of Child Labor and Schooling in the Philippines

Lindsay Rickey [email protected]

Thesis Adviser: Professor Seema Jayachandran

Department of Economics Stanford University

May 11, 2009

Abstract

The previous literature suggests that the determinants of child labor are largely country specific, indicating that any policies aimed at reducing child labor must look carefully at the causes of child labor in context. My thesis adds to the empirical work on child labor by investigating what household and community characteristics are most common among working children in the Philippines, using data collected by the International Labour Organization. I use a multinomial logit model with child activity as the dependent variable, where the three possible outcomes are work only, work and study, and study only. I find that poverty has a strong negative impact on the probability a child works full time or part time (relative to study only), especially in rural areas, as do the years of the household head’s education and having electricity and access to drinking water. Having a close biological relation to the head has a significantly positive effect on the probability of studying only for all groups of children, and especially for urban girls. The results also indicate that government programs like welfare and community organizations do little to reduce child labor, probably due to the lack of awareness among the majority of the populace. Keywords: multinomial logit, Filipino, poverty, income, household head, education, biological relation, household composition, community infrastructure, female headship, welfare

Acknowledgements I would like to thank my adviser, Professor Seema Jayachandran, for her invaluable advice and comments. I would also like to thank Professor Nicholas Bloom for his instruction in the research process, Professor Geoffrey Rothwell for his guidance in the Honors Program, and Professor Mark Tendall for his support during Summer Research College. Finally, I would like to thank the International Labour Organization for the detailed data sets they provided.

Lindsay Rickey - 2 - 5/11/2009

Table of Contents

1. Introduction ............................................................................................................................... 3

2. Theoretical Model ..................................................................................................................... 5

3. Review of Empirical Findings ................................................................................................ 13

3.1 Poverty................................................................................................................................. 13

3.1.1 Poverty’s Effect on Child Labor ................................................................................... 13

3.1.2 Poverty’s Effect on Child Schooling ............................................................................. 15

3.2 Parental Education.............................................................................................................. 17

3.3 Land Size ............................................................................................................................. 21

3.4 Household Size .................................................................................................................... 22

3.5 Household Composition ...................................................................................................... 23

3.6 Gender of Household Head ................................................................................................. 26

3.7 Marital Status of Household Head ...................................................................................... 27

3.8 Age of Household Head ....................................................................................................... 28

3.9 Relation to Head .................................................................................................................. 29

3.10 Child Gender ..................................................................................................................... 30

3.11 Child Age ........................................................................................................................... 32

3.12 Region Effects .................................................................................................................... 34

3.13 Community Infrastructure Effects ..................................................................................... 35

4. Methodology ............................................................................................................................ 37

5. Results ...................................................................................................................................... 44

6. Conclusions .............................................................................................................................. 52

7. Appendix .................................................................................................................................. 54

8. Reference List .......................................................................................................................... 75


1. Introduction

Over two hundred years have passed since “The Factory Health and Morals Act,” the first

piece of legislation that restricted the number of hours a child works, was first passed in Britain

in 1802. Since then, many international organizations have been created to help eliminate child

labor, most notably the International Labour Organization (ILO), the United Nations Children’s

Fund (UNICEF), and the inter-agency Understanding Children’s Work (UCW) Project. Yet,

according to the 2004 ILO estimate, 218 million children are engaged in child labor in

developing countries, of whom 126 million were in hazardous conditions.1

This is not to say all child work must be eliminated. Some economists argue that some

light, non-hazardous work can benefit the child since it provides labor market experience and

sometimes much-needed income for poverty-stricken families. The potential benefit for the

child depends largely on the type of child labor, whether it is voluntary, the number of hours a

week they work, and the extent to which work interferes with schooling. Despite these potential

benefits, there are some forms of child labor that are considered unconditionally harmful to the

child: prostitution, forced labor, military, drug trafficking, and other “hazardous” work, defined

as “work which, by its nature or the circumstances in which it is carried out, is likely to harm the

health, safety or morals of children” (ILO-IPEC, p. 16). The ILO estimates that as of 2002, an

estimated 178.9 million children are employed in the worst forms of child labor. Not only do

these forms of child labor violate the fundamental rights of the child, they also inhibit economic

development through their adverse effects on the long term development of human capital.

1 Here, the term “child labor” refers to work that is considered unacceptable by the International Labor Organization, which includes most work (apart from chores and working for family) done by children under 14, hazardous work done by children under 18, and the worst forms of child labor, defined above.


The problem of reducing child labor depends on what the actual determinants of child

labor are. If poverty is the main determinant, an outright ban on child labor might only result in

the children who must work to survive being involved in more dangerous work; if children are

not recognized by the law as workers, child workers cannot be protected by the law. Similarly, a

developed country’s decision to boycott goods from developing countries with child labor might

only worsen the well-being of children in those countries by lowering their living standards and

forcing them to work longer hours in potentially more hazardous conditions.

The determinants of child labor are highly debated, especially the effect of poverty on

parents’ decisions to send their children to work. The literature suggests that these determinants

are largely country specific, indicating that any policies aimed at reducing child labor must look

carefully at the causes of child labor in context. My paper aims to add to the empirical work on

child labor by investigating what household and community characteristics are most common

among working children in the Philippines in the year 2001. The results in my paper should

provide insights into the relationship between child labor, poverty, education, and other

household and community characteristics in the Philippines.


2. Theoretical Model

The model for this paper is based on Cigno, Rosati, and Tzannatos (2001), who develop a

simple model for household decision making. It assumes a household with one parent (the head

of the household) who is altruistic, i.e. that his life-time utility depends on his own life-time

consumption, on the consumption of his pre-school and school-age children, and on the amount

of human capital with which each of these children will enter adult life. Since the amount that a

person is able to earn, as an adult, is positively related to the person’s health (dependent on past

consumption) and personal skills (dependent on human capital), saying that the head cares about

his children’s current consumption and future human capital is equivalent to saying that he cares

about his children’s lifetime welfare.

Human capital is affected by innate ability and education. The cost of education entails

time commitments (both school attendance and study outside school hours), and the costs of

other educational inputs (books, tuition and writing material, and travel to school). Though the

child’s time and educational inputs are clearly complementary, the model assumes that there is

some substitutability of child’s time and educational inputs in the formation of human capital,

i.e. that an increase in educational inputs with no corresponding increase or perhaps even a

decrease in time spent can still lead to an increase in human capital. For example, if a child goes

to a better and more expensive school, the child might be able to spend less time studying but

still achieve the same amount or a higher amount of human capital.

The model assumes constant returns to scale in the production of a child’s human capital

with regards to both time and educational inputs. Also, it assumes that the marginal cost of

human capital is constant and equal to the prices of educational inputs plus the opportunity cost

of the child’s time, up to the point where the child’s time is fully employed in education. After


Figure 1

this point, the cost increases with human capital, as more and more has to be spent for

educational inputs in conjunction with a fixed amount of time.

The relationship between the marginal cost of human capital q and the stock of human

capital h may be interpreted as a supply curve. Figure 1 shows how this curve is affected by

changes in the prices of educational inputs, or in the opportunity cost of time spent in education.

The broken line through point I represents the supply of human capital for a particular

configuration of prices of educational inputs, and opportunity cost of time. If the opportunity

cost of time rises, the supply curve shifts to S’: the horizontal segment of the curve shifts

upwards, and the amount of human capital produced by full-time education increases. This is

because the head of the household will “spend” less on child’s time (i.e. have the child spend

fewer hours on education) and spend more on educational inputs; this would shift the frontier,

i.e. the amount of human capital that can be produced by full education, outwards. Next, take the

curve through point H as the initial situation, and consider the effect of a rise in the prices of

q

I

L

H

h

S S’

S’’

Figure 1


Figure 2

educational inputs. The new curve is S’’—the horizontal segment again shifts upwards, but the

human capital associated with full-time education decreases since the household head spends less

on educational inputs.

The household head decides how to allocate the time of his school-age children, and how

much to spend for each of them, so as to maximize his own utility (which takes into account that

of the children), subject to the family budget constraint. The possible solutions are illustrated in

Figure 2, where c stands for household consumption.

Notice that in Figure 2, the y-axis is consumption rather than the cost of human capital.

A higher cost of human capital (denoted by q on the y-axis of Figure 1) leads to lower

consumption per unit of human capital (denoted by c in Figure 2). The broken line through

points I and L is the production frontier. The x-coordinate of Point I is the amount of human

capital that the child would have in the absence of education (“innate ability”). To the right of

point L, the child’s time is fully occupied in education. The slope of the production frontier,

equal to the marginal cost of human capital, is constant to the left of point L, increasing to the

c

h

L

I T


right of it. The steeper slope is due to the assumption stated earlier that once a child is fully

employed in education, the costs to accumulating additional human capital become higher. The

choice set is bounded by the vertical line through point I to the left (the household head cannot

sell off their children’s innate ability), and by the production frontier upwards. The slope of the

indifference curve through point I is the price of human capital above which the household head

is not willing to bear any cost for his child’s education. The slope of the indifference curve

through point L is the price of human capital below which the head wants the child to study full

time.

The first type of solution is at point I, where the marginal cost of human capital is higher

than the maximum the head is willing to pay. If that is the type of solution, the child is made to

work full time. The second type of solution is at any point between I and L (e.g., at point T),

where the marginal cost of human capital is equal to its marginal rate of substitution for

consumption. If that is the case, the child works and studies at the same time. The third type of

solution is either at or to the right of point L, where the marginal cost of human capital is lower

than the minimum below which the head wants the child to study full time. If that is the case, the

child does not work at all. If the head sends his children to school at all, he also buys educational

inputs.

An increase in family income raises current consumption and the future stock of human

capital for every child, but it also raises the maximum that the head will be willing to pay for an

extra unit of human capital, and the minimum below which he wants the child to study full time

by shifting the budget curve outward. Since the marginal cost of human capital is not affected,

the probability that a child will work full time will then fall, while the probability that the child

studies full time will rise. The probability that the child works and studies may go either way.


c Figure 3 illustrates the effects of an increase in the price of an educational input, or in

the opportunity cost of time in education, holding full household income constant. Take the

broken line through point L as the frontier before the change, with the point A the optimal choice

before the change. By raising the marginal cost of human capital, an increase in either of those

variables makes the frontier steeper everywhere. Unless the child is already a full-time worker

(i.e., the initial solution happens to be at point I), this will lead to a rise in household

consumption since it leads to a child working more and contributing more income to the

household.

Other effects will depend on whether the increase was in the opportunity cost of time or

in the price of an educational input. If the former, the new frontier will be like the one through

points I and H (remember that point H is the point at which the child studies full time and that as

the opportunity cost of time increases, the amount of human capital achieved through full

education also increases). You can see from the example in the graph that this results in the

price of human capital below which parents want their children to study full time (equal to the

c

h

I

L

H

F

A B C

Figure 3


slope of the indifference curve at the point of full education) decreasing, and the price above

which they want children to work full time (equal to the slope of the indifference curve at point

I) staying the same. Therefore, the effect of a rise in the child wage rate (or, if the child works

for his parents, in the domestic productivity of child labor) is to raise the probability of full-time

work (as the increased marginal cost is more likely to be above the maximum price the head will

pay for education), and to lower the probability of full-time study (as the increased marginal cost

is more likely to be above the maximum price at which the head will enroll their child in school

full time); the effect on the probability of part-time work is ambiguous. The effect on the

demand for educational inputs (other than the child’s own time) is also ambiguous because, on

the one hand, the demand for human capital falls, but on the other, each unit of human capital is

produced with more educational inputs and less time.

If the price of educational inputs rises, the new frontier will be the one through points I

and F. The price of human capital below which children study full time may rise or fall (in the

example in the graph, it rises); that above which they work full time is again unaffected.

Therefore, if the price of, for example, books or travel to school goes up, the probability of full-

time work increases (as marginal cost of human capital increases), but we cannot say whether the

child is more likely to study full or part time. The reason for this is that the income and

substitution effects are opposed to each other—the income effect would cause children to work

more since the household is poorer due to the price increase, but the substitution effect would

lead children to spend more time in school rather than spend money on educational inputs. If the

income effect is stronger, a child will be more likely to part-time study, and if the substitution

effect is stronger, a child will be more likely to study only.


We can see that the shape of the indifference curve greatly influences the head’s decision

to send a child to work, school, or both. The indifference curve, in turn, is determined by the

head’s preferences for human capital for the child and household consumption. Someone who

values education and the accumulation of human capital a great deal will have a steeper

indifference curve than one who does not value human capital. Comparing Figure 2 and Figure

4, we see that the household head with a steeper indifference curve will have a higher maximum

price of human capital and a higher minimum below which he allows the child to study full time,

which makes sense since he is willing to give up more units of household consumption for

human capital. For the particular household head in Figure 4, human capital accumulation is

important enough to devote all of the child’s time to it.

Figure 5 shows three cases where the child will never study and work at the same time,

i.e. the indifferent curves are straight lines. In these cases the maximum price at which the head

will pay for any schooling and the minimum price below which the child goes to school only are

the same, since the slope of the indifference curve is the same everywhere. The curve that

crosses point E represents an extreme valuation of human capital above consumption, while the

curve that crosses point I represents an extreme valuation of consumption above human capital.

Figure 4 Figure 5

c

h

c

h

I I

L L

E


The curve crossing Point L represents a high, though not extreme, valuation of human capital,

leading to the child being fully employed in education.

Many factors not explicitly modeled here could affect the value placed on human capital

versus consumption. The education level of the household head, the gender and age of the

household head, the total number of adults in the household, the total number and ages of other

children in the household, and the perceived quality of the schools could all affect how the head

views education as compared with household consumption. The head’s preferences for a child’s

education might also be dependent on the characteristics of that particular child, for example,

whether the child is their biological child.


3. Review of Empirical Findings

This section reviews the findings of many of the empirical studies that use micro-level data to

analyze the possible determinants of child labor and child schooling. I have divided the review

into separate subsections for the main potential factors I have found in my research on child

labor.

3.1 Poverty

3.1.1 Poverty’s Effect on Child Labor

Many theoretical models of child labor are based on what Basu and Van (1998) called the

Luxury Axiom, i.e. that a family will send a child to work only if the family’s non-child-labor

income drops below some threshold. However, despite the seemingly obvious link between

poverty and child labor, the evidence for a significant income effect is mixed. An insignificant

income effect is reported in Coulombe (1998) in Côte d’Ivoire, Sasaki and Temesgen (1999) in

Peru, Patrinos and Psacharopoulos (1997) in Peru, Ilahi (2001) for rural boys in Peru, Ray (2000)

in Pakistan, and Ersado (2005) for urban children in Nepal, Peru, and Zimbabwe. In her survey

of field studies of child labor in India, Bhatty (1998) concludes that there is no clear association

between poverty and child labor. In a review of empirical studies of Côte d’Ivoire, Ghana and

Zambia, Canagarajah and Nielsen (2001) also conclude that there is not much evidence in favor

of the view that poverty is a significant cause of child labor.

A positive coefficient on income is obtained in Cartwright (1999) for household

farm/enterprise work in rural Colombia and in Patrinos and Psacharopoulos (1995) in Paraguay.

Bhalotra and Heady (2003) provide a theoretical justification for the positive coefficient. They

explain that owning land has both wealth and substitution effects on a household’s supply of

child labor. The wealth effect suggests that large landholdings generate higher income, making it


easier for households to forgo the income that child labor brings. However, because of

geographical inaccessibility to workers, or lack of information flows, large landholders may find

it cheaper to hire family members rather than other labor. Bhalotra and Heady test this model by

looking at households in Ghana that run their own farms and find that richer households in

developing countries tend to own more land, and households tend to employ family members

(including children) on this land. Consequently, richer households, on average, make greater use

of child labor than poorer households. This theory is also supported by Dumas (2007), who

looks at rural households in Burkina Faso and finds that child labor seems to be due to the

absence of labor market rather than to household subsistence needs.

Negative income effects are found in Cartwright for wage work in rural Colombia (1999),

in Cigno and Rosati in rural India (2000), and in Ilahi for rural girls in Peru (1999). Amin,

Quayes, and Rives (2004) also find a negative income effect for both urban and rural boys and

girls in Bangladesh, as do Rosati and Tzannatos (2000) in Vietnam, Liu (1998) for wage work in

Vietnam, Ray (2000) in Peru, Bhalotra and Heady (2000) for rural farm work for boys in

Pakistan and girls in Ghana, and Ersado (2005) for rural children in Nepal, Peru, and Zimbabwe.

Edmonds (2005) finds that income growth in Vietnam can account for a large part of the

reduction in child labor observed there during the 1990s. Carvalho (2000) examines the

introduction of an old-age pension in Brazil and finds that it resulted in a reduction in child labor

amongst children living with grandparents, with the impact of a grandmother’s pension on her

granddaughters’ labor being especially large. Edmonds (2006) looks at how cash transfers affect

child labor in South Africa and documents large declines in total hours worked when black South

African families become eligible for social pension income. Schady and Araujo (2006) study

cash transfers in Ecuador and find that the transfers have a large negative impact on work, about


17 percentage points. Considering these studies have less methodological problems than cross-

sectional data (Edmonds (2001) is based on two years of data and the rest are natural

experiments), Bhalotra and Tzannatos (2003) postulate that the variance of the income effect in

different studies might come from methodological issues rather than actual country variations.

However, Bourguignon, Ferreira, and Leite (2002) and Cardoso and Souza (2004) find that, in

Brazil, conditional income transfers requiring a child to go to school had no significant impact on

the incidence of child labor.

3.1.2 Poverty’s Effect on Child Schooling

Although child labor and schooling are not mutually exclusive, and are, in fact, often

done together, it is interesting to consider whether the effect of poverty on schooling is any

stronger than the effect (or lack thereof) of poverty on child labor. Behrman and Knowles

(1999) survey estimates of income elasticities for a range of indicators of educational enrollment

and attainment for the US and a number of developing countries. The median elasticity of 0.07 is

small, though somewhat larger estimates of about 0.20 are observed for lower income regions.

In their own analysis of five indicators of schooling in Vietnam in 1996, Behrman and Knowles

find higher income elasticities than the previous literature. According to Bhalotra and Tzannatos

(2003), this is at least partly on account of Behrman and Knowles’ more careful attention to the

choice of indicators and the specification of the equation.

A significantly positive effect of income on child schooling is found in Ray (2000) for

Pakistan, Cigno, Rosati, and Tzannatos (2001) in rural India, Ilahi (2001) for girls in Peru,

Canagarajah and Coulombe (1999) in Ghana, Coulombe (1998) in Côte d’Ivoire, and Jensen and

Nielsen (1997) in Zambia. In Côte d’Ivoire, Grootaert (1999) shows that for poor households, in


both urban and rural areas, there is a higher probability for selecting non-schooling options than

richer households. Rosati and Tzannatos (2006) find that the effect of income on schooling in

Vietnam is non-linear and that the significantly positive effect of income on the probability that a

child will only go to school decreases with the level of income. Edmonds (2006) finds that cash

transfers in the form of pensions lead to large increases in child schooling in South Africa.

Schady and Araujo (2006) study cash transfers in Ecuador and find that the transfers have a

large, positive impact on school enrollment, about 10 percentage points, perhaps partly because

some households believed that there was a school enrollment requirement attached to the

transfers (though not monitored or enforced). Similarly, Bourguignon, Ferreira, and Leite (2002)

and Cardoso and Souza (2004) found in Brazil that conditional income transfers requiring a child

to go to school increased the likelihood of schooling.

Ray (2000) finds no significant income effect on child schooling in Peru, nor does Ilahi

(2001) for boys in Peru or Ersado (2005) for urban children in Nepal, Peru, and Zimbabwe.

Patrinos and Psacharopoulos (1995) did not find monthly family income to be a significant

determinant of years of school attainment in Paraguay, but did find a positive association with

school enrollment. A negative effect of income on child schooling is found by Patrinos and

Psacharopoulos (1997) in Peru. Overall, though, it seems that income has a bigger effect on

schooling than on child labor. In fact, higher income can lead to more schooling even in regions

where higher income leads to more child labor. For example, Bhalotra and Heady (2003) find

that in Ghana and Pakistan income has a significantly positive effect on schooling attendance

even though larger farm size leads to richer households employing more child labor.


3.2 Parental Education

There is consistent evidence that the mother’s education has a negative effect on child

labor, and the size of this effect is often greater than that of the father’s education. Using data

combined for boys and girls in rural and urban areas in Ghana, Canagarajah and Coulombe

(1999) find that the father’s secondary level education has a negative effect on child work

participation while the mother’s education has no effect. Using the same data, Bhalotra and

Heady (2000) find a negative effect for the mother’s middle or secondary level education for

rural boys but no effect for the father’s education. Bhalotra and Heady also find that on family

farms in rural Pakistan, the mother’s middle or secondary level education has a negative effect

for boys and girls (larger in the case of girls) and the father’s secondary education has a negative

effect that is restricted to girls.

Cigno and Rosati (2000) find that in rural India the children of mothers with less than

primary education are significantly more likely to be in full-time work as compared with full-

time study, and having a mother who completed middle school reduces the probability of

combining work and school as compared with full-time study, while the father’s education has

no significant effect. Ravallion and Wodon (1999) find negative effects of the mother’s and

father’s education level on child labor in Bangladesh. In Vietnam, years of father’s education

have no effect on child labor but mother’s education has a negative impact on the probability of

work (full-time and part-time) as well as on the probability of being neither in work nor in school

(Rosati and Tzannatos 2000). However, Liu (1998) finds insignificant effects for both mother’s

and father’s years of schooling on child labor in Vietnam, whether market or home based.

Emerson and Souza (2008) find that in Brazil, both father’s and mother’s education have a

negative effect on child labor and a positive effect on schooling for both boys and girls. Kruger


(2007) finds these same effects in Brazil for parents’ education.

Amin, Quayes, and Rives (2004) examine child labor in Bangladesh, dividing the

children by gender, region, and age groups: younger (aged 5-11) and older (aged 12-14). They

find that the education of the male head significantly lowers the probability of working for all

rural boys (older and younger), all urban girls, urban older boys, and rural older girls, with no

effect on urban younger boys and rural younger girls. The education of the female head has a

significantly negative effect on child labor for boys in all four groups, and for all rural girls, but

no effect for urban girls (older and younger). Bhalotra and Heady (2003) find that the father’s

education has a negative effect on the probability that rural girls in Pakistan work, but no effect

on work for rural boys in Pakistan or for rural boys and girls in Ghana. The father’s education

does, however, have a positive effect on school attendance for rural boys and girls in both

Pakistan and Ghana. The mother’s education has a significantly negative impact on child labor

for rural boys in Ghana, and rural boys and girls in Pakistan, and a significantly positive effect

on schooling for rural girls in Pakistan, and rural boys and girls in Ghana. It has no effect on

labor for rural girls in Ghana and no effect on schooling for rural boys in Pakistan.

Patrinos and Psacharopoulos (1997) find that the probability of combining work and

school as compared with the probability of full-time study is reduced by the years of the father’s

education in Peru and by years of mother’s education in Paraguay (Patrinos and Psacharopoulos

1995). Using the same Peruvian data, Sasaki and Temesgen (1999) find that the probability of

combining work and school as compared with full-time study is reduced by the father’s college

education and the mother’s secondary and college level education. While the first study uses a

binomial logit, the second uses a multinomial logit, allowing for two further outcomes: work

only and neither work nor school. The mother’s education has no effect on these outcomes


relative to study only, while the father’s secondary education has a negative effect on the

probability of work only relative to study only. Controlling for household-specific effects using

a random effects probit on the Peruvian data for two years (1994 and 1997), Ilahi (1999) finds a

negative effect of the education of the oldest prime-age female on the probability of children

working in an income-generating activity in Peru, and this effect is similar for rural and urban

areas. While the results of the three studies for Peru do not contradict one another, they do show

the importance of which sub-sample is being discussed and whether the education effects are

allowed to be nonlinear (Bhalotra and Tzannatos 2003).

In Colombia, multinomial logit estimates reported in Cartwright (1999) indicate that

there is a negative effect of the father’s years of schooling on the probability of full-time child

work whether this is for wages, on the family farm or enterprise, or for full-time home care (with

full-time study as the reference category). The mother’s years of schooling have no effect on

child labor producing marketable goods but have a positive effect on the probability that a child

is at home in full-time care. This result is consistent with the view that educated women are

more likely to work and so their children may have to substitute for them at home at the expense

of going to school (Bhalotra and Tzannatos 2003). These effects are similar in rural and urban

areas, though the effects of the mother’s education are stronger in urban areas. In their analysis

of child labor in urban Bolivia, Cartwright and Patrinos (1997) find a negative effect of the

mother’s years of education on the probability of children working in wage-based labor as

opposed to being in school. Unlike in the case of Colombia, there is no effect of mother’s

education on child time in home-care. They do not include the father’s education in the model.

Using a multinomial logit, Ersado (2005) finds that the years of the mother’s education

have a significantly positive effect on schooling for rural and urban children in Nepal and


Zimbabwe and urban children in Peru, and a significantly negative effect on child labor for rural

children in Nepal and rural and urban children in Zimbabwe. However, he finds no significant

effect of the mother’s education on schooling for rural children in Peru or on child labor for

urban children in Peru, and a significantly positive effect on child labor for rural children in Peru.

With regard to the father’s education, Ersado found no effect on schooling for urban children in

Zimbabwe and no effect on child labor for all children in Zimbabwe and rural children in Peru

and Nepal, a significantly positive effect on schooling for rural children in Zimbabwe and all

children in Peru and Nepal, and a significantly negative effect on child labor for urban children

in Peru and Nepal.

Rosati and Tzannatos (2006) also use a multinomial logit model to examine child labor in

two different years in Vietnam (1993 and 1998). They find that years of the father’s education

have a significantly negative effect on the probability a child only works as opposed to only

studies for both years and on the probability a child works and studies in the 1998 survey (again,

with “only studies” as the reference group). The father’s education has no effect on the

probability a child works and studies in the 1993 survey. The mother’s education has a negative

impact on the children only working for both years but no effect on the children working and

studying (again, for both years).

Grootaert (1999) finds that in Côte d’Ivoire, the probability of full-time study as opposed

to full-time work is positively influenced by years of the father’s education in urban areas and by

years of the mother’s education in rural areas; in each case, the education of the other parent has

no effect. For rural and urban regions, both the father’s and mother’s education raise the

probability of a child combining school with work as opposed to working full time. Analysis of

the same Côte d’Ivoire data by Coulombe (1998) using a bivariate probit shows no effect of the


father’s schooling on either work or school participation. The mother’s education has no effect

on work although it does increase school participation. Nielsen (1998) also finds no effect of the

father’s schooling on child work or school decisions in bivariate probit estimates of data for

Zambia, but she does not investigate mother’s schooling. Using a fixed-effect logit model,

Tunali finds no parental education effects in Turkey, but the probit analysis by Dayioglu (2006)

shows that the mother’s and father’s education levels have a strong negative correlation with

child labor in Turkey.

3.3 Land Size

The vast majority of working children live in rural areas and work on farms,

predominantly family-run farms (Bhalotra and Tzannatos 2003). As mentioned in the previous

section on poverty’s effect on labor and schooling (3.1), the lack of a labor market can lead to

children being used for labor on a large farm, despite the increased wealth that owning land

brings the household (Bhalotra and Heady 2003). In order to separate the income effect (which

should lead to decreases in child labor as land size increases) and the substitution effect (which

would lead to increases in child labor as land size increases), some measure of income must also

be included in the model. After controlling for income, the theoretical model would assume that

land size or the mere ownership of land would be positively correlated with child labor, since the

amount of land raises the opportunity cost of children’s time.

Canagarajah and Coulombe (1999) find no effect of farm size on child work participation

rates in Ghana. Distinguishing boys and girls and restricting the sample to rural farming

households, Bhalotra and Heady (2003) find a positive effect of farm size on girls’ work in rural

Pakistan and Ghana, though no effect for boys. They also find a negative effect on school


participation for rural girls in Pakistan, though no effect for girls in Ghana or for boys. Cigno

and Rosati (2000) find a positive effect of land size on child labor in rural India, combining data

on girls and boys. Rosati and Tzannatos (2006) find that in Vietnam, the size of cultivable land

owned by the household raises the probability that children will combine work with school and

the probability of full-time work as opposed to studying full time. They also find that relative to

study, unsurprisingly, owning land reduces the probability that a child is idle.

3.4 Household Size

Since size and composition are clearly correlated, the relation between household size

and child work will depend upon whether household composition is held constant. In empirical

results, there is a tendency to find a positive association of household size and child work.

However, this finding cannot be regarded as robust since the studies differ in whether or not land

size and household composition are held constant (Bhalotra and Tzannatos 2003). Bhalotra and

Heady (2003), controlling for these factors, find negative effects of household size on child’s

labor participation for boys in rural Pakistan and girls in rural Ghana with no effect for girls in

rural Pakistan or boys in rural Ghana. They also find positive effects of household size on child

school participation for boys and girls in rural Pakistan and girls in rural Ghana, with no effect

for boys in rural Ghana.

Cigno, Rosati and Tzannatos (2001) also find a negative effect of household size for

participation in work in rural India. Ilahi (1999) finds a negative effect on child labor for boys

and no effect for girls in Peru. Also using data from Peru, Patrinos and Psacharopoulos (1997)

find a negative effect of the number of siblings (not household size) on the probability of

combining work and school relative to the probability of simply attending school if the number


of children not in school is held constant (insignificant if this control variable is not included). In

Vietnam, Rosati and Tzannatos (2006), after controlling for total number of children, find a

significant negative effect of household size on the probability of being in work and on the

probability of combining work and school, relative to the probability of simply being in school

for both 1993 and 1998 surveys. There is no effect for the children who report being in neither

work nor school.

Positive estimated effect of household size on child work are found in Patrinos and

Psacharopoulos (1995), who, for Paraguay (in contrast to Peru), find a positive effect of the

number of siblings (not household size) on the probability of combining work and school relative

to the probability of simply attending school. Also, Amin, Quayes, and Rives (2004) find a

significantly positive effect of household size on all groups of boys (rural/urban, younger/older)

and all girls except urban older girls, for whom they find no effect, though they do not control for

household composition.

3.5 Household Composition

Household composition effects refer to the age and gender structure of the household.

Additional compositional effects that may be taken into account are whether both parents are

alive and whether they are present in the household (or have, for example, migrated away for

work). We would expect that the absence of a parent would create economic hardship and

increase child labor, especially if the absent parent was the primary wage earner (usually the

male).

Given that pre-school children are too young to work, and that an increase in their

number is thus equivalent to a lump-sum reduction in full income (an income-dilution effect), we


would expect from the theoretical model that full income raises the probability of full-time work,

lowers that of full-time study, and has ambiguous effect on that of part-time work. According to

the theoretical model, an increase in the number of school-age children, holding full income

constant, raises the probability of part-time work, and lowers that of full-time study, but has no

effect on full-time work.

Grootaert (1999) finds no clear evidence of sibling effects in Côte d’Ivoire although

Coulombe (1998), using the same data, finds that the number of children under 6 raises work

participation for older children. In Vietnam, both the number of siblings under 6 and the number

of school-age siblings (6-15 years) raise the probability of school-age children working only and

the probability of children working and studying for both 1993 and 1998 (relative to full-time

study), except siblings under 5 had no effect on working and studying in 1998 (Rosati and

Tzannatos 2006). Ilahi (1999) finds no household composition effects on child labor in Peru.

Using the same data, Sasaki and Temesgen (1999) confirm that the number of children in the

household does not affect full-time work participation of children in Peru but they find it does

increase the probabilities of doing school and work and being idle, relative to full-time study.

On the other hand, estimates of binary probit models in Ray (2000) suggest a positive effect of

the number of siblings on work probabilities in Peru. In Brazil, Emerson and Souza (2008) find

a positive effect of the number of children on the probability of child labor, and a negative effect

on the probability of school participation.

The presence of younger siblings discourages work participation amongst girls in rural

Ghana, household composition having no effect on the work hours of Ghanaian boys (Bhalotra

and Heady 2000). The same study finds that the presence of younger boys (under 10) in the

household reduces the work participation of both boys and girls aged 10-14 in rural Pakistan,


whereas the presence of little girls in the household has no effect. Ray (2000) uses the same

Pakistan data as Bhalotra and Heady and, aggregating over the sibling terms, finds no effects of

number of siblings on child labor. Kruger (2007) finds that in Brazil, the number of 0-5 year old

children has a positive effect on child labor and a negative effect on child schooling, and that the

number of 6-14 year old children has a positive effect on child labor, but also a positive effect on

schooling for girls (no effect on schooling for boys). Cigno, Rosati and Tzannatos (2000) find

that having both younger siblings (0-6) and siblings in one’s own age group (6-12) raises the

probability of working of school-age children in rural India. Similarly, the number of 0-6 year

old siblings raises the probability of work relative to school-only in Peru (Patrinos and

Psacharopoulos 1997). Ersado (2005) finds no effect on the number of children under 5 on the

probability that a child works in Peru or in rural Nepal, and a positive effect on this probability in

Zimbabwe and urban Nepal. He finds no effect of children under 5 on the probability of

schooling in rural Peru and Nepal, and urban Zimbabwe, and a significantly negative effect in

urban Peru and Nepal, and rural Zimbabwe.

Using data from Colombia and Bolivia respectively, Cartwright (1999) and Cartwright

and Patrinos (1999) find that having older brothers and sisters reduces the probability that a

younger child works. Canagarajah and Coulombe (1999) who use the same Ghana data find that

the number of adult males in the household has a significantly positive effect on the work

participation of 11-14 year old children in rural and urban areas, though there is no effect for 7-

10 year-olds. They find that the numbers of siblings and other compositional variables have no

effect. The presence of men and (especially) women over 60 reduces the probability that a girl in

Pakistan works, there being no effects on Pakistani boys or on Ghanaian children. Overall, the

effects of household composition are gender-specific and they are stronger in Pakistan than in


Ghana (Bhalotra and Heady 2000). Bhalotra and Heady also find that the number of women

over 60, though decreasing child labor, also decreases child participation in school for boys and

girls in rural Ghana, and has no effect on boys and girls in rural Pakistan (2003). In Brazil,

Kruger (2007) finds that the number of people in the household over 65 has a significantly

negative effect on child labor participation for both girls and boys, though no effect on child

schooling rates.

3.6 Gender of Household Head

The prevalence of female-headed households varies considerably across countries. It

tends to be greater in sub-Saharan Africa than in Asia. For example, it is 30% in rural Ghana as

compared with 3% in rural Pakistan (Bhalotra and Heady 2000). Most of the studies that include

female headship in their econometric model also include a measure of household income. If

female headship significantly raises child labor participation at a given level of income, then it

must indicate a degree of vulnerability of the household that is not picked up by household

income. Bhalotra and Tzannatos (2003) postulate that this could be the result of a female-headed

household’s borrowing ability or, more generally, its ability to deal with a crisis, its perception of

the range of job alternatives available to it, or its assessment of its human capital. The result is

also consistent with women being less altruistic towards children than men, but empirical

evidence indicates this is not the case (e.g. Rubalcava, Teruel, and Thomas (2009) find that

women allocate more resources toward investment in the future, Cardosa and Souza (2004) find

that cash transfers to women have a larger positive effect on schooling than transfers to men).

Support for the hypothesis that children of female-headed households are more likely to

work and less likely to be in school is found for Paraguay in Patrinos and Psacharopoulos (1995)


and for rural (but not urban) Côte d’Ivoire in Grootaert (1999). Bhalotra and Heady (2003) find

a positive effect of female headship on the labor participation rates of boys and girls in rural

Pakistan and for girls in rural Ghana (with no effect for boys in rural Ghana). They find no

effect of female headship on child schooling for any of the groups. Amin, Quayes, and Rives

(2004) find that in Bangladesh, female headship is positively correlated with child labor for most

of the groups of children (except for rural older boys, where there is no effect). Ersado (2005)

finds no effect of female headship on schooling or labor for the majority of the children in Nepal,

Peru, and Zimbabwe. He does find a negative effect of female headship on child labor in urban

Zimbabwe, but also a negative effect on schooling in rural Nepal and urban Zimbabwe.

Canagarajah and Coulombe (1999) do not separate the data by gender and they find that the

indicator for female headship is insignificant. Although he does allow gender-specific effects,

Ilahi (1999) finds no role for female headship in Peru.

Ray (2000) finds no relationship between child labor and female headship for children in

Peru and Pakistan, but does find a positive relationship between female headship and schooling

for girls in Pakistan (no effect for the boys in Pakistan or children in Peru). This is consistent

with empirical evidence indicating that the higher altruism of mothers is often focused more on

girls than boys (e.g. Duflo (2003) finds that grandmothers give more of their pension to their

grandchildren than grandfathers, and more to granddaughters than grandsons).

3.7 Marital Status of Household Head

As with female headship, the effect of the marital status of the household head on child

labor and schooling cannot come from differences in income levels, since income is controlled

for in most of the empirical studies. The other possible reasons might be similar to those for


female headship (such as borrowing ability), and we might also expect that female headship and

the head being single are correlated. Since most studies do not control for the head being single,

female headship might have a positive effect on child labor because many female heads are also

single, and hence have an additional burden in their ability to borrow money. Also, if the effect

of parents’ characteristics differ according to the parent’s gender, e.g. if the mother’s education

has a larger effect on child labor and schooling than the father’s education, the effect of the

parent being single would depend on the gender of the remaining parent. We would also expect

that these gender differences would remain even if the household head is not the biological

parent of the child (e.g. an uncle or grandparent).

Cardosa and Souza (2003), one of the few studies that controls for the marital status of

the household head, find that in Brazil the absence of the father has a significantly negative

impact on schooling and on child labor for boys but no effects for girls. The absence of the

mother has a significantly negatively correlation with schooling for boys (though less of an effect

than the father’s absence), a significantly negatively correlation with schooling for girls, and a

significantly positively correlation with labor for boys with no effect on labor for girls. These

results indicate that effects do change depending on the gender of the remaining parent, and that

these effects also change based on the gender of the children. According to these results, boys

are generally more affected by the parent being single than girls, though the magnitude and even

the direction of the effect will depend on the gender of the remaining parent.

3.8 Age of Household Head

This is an indicator of the stage of the lifecycle that the household is at. If the oldest male

reports as head, this variable may also indicate whether the child lives in a vertically extended


household, with grandparents. If the equation also includes a full set of age-gender variables that

reflect household composition, the age of the household head has a less clear meaning and a

weaker role to play (Bhalotra and Tzannatos 2003). Perhaps because of this fact, most studies do

not include this in their model. Those studies that do include it and find it significant do not have

full controls for household composition, e.g. Nielsen (1998), Ray (2000), Cardoso and Souza

(2003), Ersado (2005), and Emerson and Souza (2008).

3.9 Relation to Head

Households in developing countries are large and complex and often contain not just

vertical but also horizontal extensions (Bhalotra and Tzannatos 2003). As a result, nephews,

nieces, sisters-in-law, and grandchildren may be counted amongst children along with sons and

daughters of the head of household. Additionally, in sub-Saharan Africa, there is a high

prevalence of child fostering and orphans. Assuming that the head plays the primary role in

decisions regarding child labor, an interesting hypothesis is that the children of the household

head are preferred and hence less likely to work.

Cockburn (2001) investigates this variable in probit estimations for work and school in

Ethiopia and finds that children of the household head are more likely to attend school. In

contrast, Bhalotra and Heady (2003) find that children of the head are more likely to be in work

in rural Pakistan but in rural Ghana, sons are less likely to be in work (no effect for daughters).

They also find no effect on schooling for sons in rural Pakistan or for sons and daughters in

Ghana, but a negative effect on schooling for daughters in rural Pakistan. However, Blunch and

Verner (2001), also analyzing data from Ghana, find that being the child of the head is positively

correlated with child labor for rural boys, negatively correlated for urban girls, and has no effect


on rural girls or urban boys. Jensen and Nielsen (1997) find that in Zambia having a non-

biological relation to the head of household negatively affects the probability of attending school.

Given the increasing proportion of orphaned children in Africa (Subbarao, Plangemann, and

Mattimore 2001), it is important to investigate whether outcomes are different for children living

with adult caretakers other than their parents. Based on data from Uganda, Bishai et al. (2003)

show that biological relatedness is a strong predictor of the quality of care offered to children.

Evidence from the Demographic and Health Surveys for 10 countries in sub-Saharan

Africa in which households were interviewed between 1992 and 2000 shows that orphaned

children in Africa live, on average, in poorer households and are significantly less likely than

other children to be enrolled in school. The lower school enrolment of orphans as compared

with other children is not explained by their greater average poverty since orphans are less likely

to be in school than non-orphans with whom they co-reside. This suggests that distant relatives

and unrelated caregivers invest less in orphaned children than in their own children or closer

child relatives (Case, Paxson and Ableidinger 2004).

3.10 Child Gender

The effect of a child’s gender on their labor and schooling varies widely by country, and

has a great deal to do with the cultural norms of that country. These can affect parents’ attitudes

towards their children, the returns to education, the opportunity cost of education—all of which

in turn affect child labor and schooling decisions. There is also a great deal of evidence that the

effects of other variables on child labor and schooling change according to gender, indicating

that the determinants for male and female child labor and schooling should be considered

separately.


Liu (1998) finds that the probability of engaging jointly in schooling and market work is

significantly higher for boys than for girls in Vietnam, while the probability of engaging jointly

in school and house work is higher for girls than for boys. Cartwright and Patrinos (1999) in

Bolivia find that boys are more likely to work full time than are girls. In Colombia, Cartwright

(1999) finds that boys are more likely to work than girls, but girls are more likely to be working

full time (as compared to combining work and school). Rosati and Tzannatos (2006) use a

multinomial logit model to show that females are more likely to be working full time (compared

to full-time study), and that they are just as likely to be combining work and school (relative to

full-time study). Ersado (2005) finds that girls are more likely to work in Nepal and Zimbabwe.

For children in Pakistan (Ray 1998), Peru (Ray 1998, Ersado 2005), Paraguay (Patrinos

and Psacharopoulos 1995), Ecuador (Sasaki 2000), and for older children (aged 12-14) in

Bangladesh (Amin, Quayes, and Rives 2004), it is found that girls are less likely to work than

boys. Results from Côte d’Ivoire (Grootaert 1999, Coulombe 1998) are slightly different—while

girls are less likely to engage in work and schooling activities than only work, they are more

likely to undertake household work. Deb and Rosati (2004) find that girls are more likely to be

idle (neither work nor school) in Ghana and India, but assert that this may just reflect the fact

that girls are expected to perform household chores (which are not picked up by their surveys).

Some studies find no significant differences in gender patterns of work. In Ghana,

Canagarajah and Coulombe (1999), find that there is no significant difference in the probability

of being economically active between male and female children. However, using more recent

data, Blunch and Verner’s (2000) estimation shows that Ghanaian girls are slightly more likely

to work than boys. In Zambia, Nielsen (1998) finds no significant difference in participation

rates between boys and girls. Amin, Quayes, and Rives (2004) find no difference in child labor


for younger girls and boys (aged 5-11) in Bangladesh.

Ray (1998) finds that males attend school more than females in Pakistan. In Ghana,

Canagarajah and Coulombe (1999) find that boys are more likely than girls to attend school.

Similarly, school enrollment is higher for boys than for girls in Zambia (Nielsen 1998), Côte

d’Ivoire (Grootaert 1999, Coulombe 1998), Nepal and Zimbabwe (Ersado 2005).

However, in some countries in Latin America (Colombia, Paraguay, Nicaragua), studies

find that girls are much more likely to go to school than boys. Boys often leave school after

completing the basic primary cycle while girls continue schooling for a few more years. This

finding is consistent with the higher labor force participation of boys mentioned earlier.

However, Ersado (2005) finds that in Peru, despite females being significantly less likely to

work, they are significantly less likely to go to school. In Vietnam, Liu (1998) finds that there is

no gender difference in the predicted probability of falling in the category of “school only” –

there is no discrimination against girls with respect to educational opportunities.

3.11 Child Age

The theoretical model would expect older children to be more likely to engage in labor

activities (especially wage work) as the returns to participating in the labor market are likely to

be higher, raising the opportunity cost of the child’s time. Also, as the child ages, they are less

likely to be required to attend school by compulsory schooling laws, which usually set the

minimum age to quit school at around 14 or 15 in developing countries. With compulsory

schooling laws and diminishing returns to education, children are less likely to go to school as

they get older as well. However, a quadratic effect should also be allowed, since very young

children are probably less likely to go to school as well.


In Bangladesh, Amin, Quayes, and Rives (2004), using a linear term for age, find that

child labor increases with age for rural and urban older boys (aged 12-14), rural younger boys

(aged 5-11), and urban younger girls, but has no effect for urban younger boys, urban older girls,

and rural older and younger girls. In Ray’s (1998) study on Peru and Pakistan, participation rates

in labor activities increase with age in both countries. In both countries, data show that child

labor increases with age – though in Pakistan older girls are less likely to participate in the labor

force – perhaps as a result of cultural factors that militate against girls working – especially

against them being engaged in market work. However, girls are likely to remain engaged in

household work as they grow older. For the case of Columbia, Cartwright (1999) finds that the

probability of children working increases with age. In urban Bolivia, Cartwright and Patrinos

(1999) find that age increases the probability that a child will work (full time or a combination of

work and school). Similar results with respect to age are found for Côte d’Ivoire (Grootaert

1999), Paraguay (Patrinos and Psacharopoulos 1995), the Philippines (Sakellariou and Lall

1999), Turkey (Tunali 1997), Ecuador (Sasaki 2000), Bangladesh (Ravallion and Wodon 1999),

Brazil (Emerson and Souza 2008), and Ghana and Pakistan (Bhalotra and Heady 2003).

Rosati and Tzannatos (2006) in Vietnam use a multinomial logit model and find that age

has a quadratic (concave) effect on the probability that a child works only and the probability

that a child works and studies, relative to study only. According to Ray (1998), the school

enrollment rate is 90 percent for children aged 6 years in Peru, and it peaks at 9 years of age (98

percent) and then steadily falls to 62 percent by 17 years of age. In Pakistan, the school

enrollment rate starts at 65 percent for children aged 10 year and peaks at 11 years of age (70

percent). It steadily falls to 40 percent by 17 years of age. This quadratic relationship between

attendance and age is also demonstrated by Rosati and Rossi’s (2003) analysis for Pakistan and


for Nicaragua. Coulombe (1998) finds that for Côte d’Ivoire, there is a quadratic relationship

between school enrollment and age, with enrollment peaking at 11 years. Liu (1998) allows a

quadratic relationship in her multinomial logit model for Vietnam. She finds that the probability

of schooling increases with age till the age of 11 and then falls slightly. Similar results are

obtained in a study of Bangladesh (Ravallion and Wodon 1999) and Ghana (Canagarajah and

Coulombe 1999, Bhalotra and Heady 2003). Other studies that have used a linear variable for

age usually find a negative relationship between schooling and age, including Patrinos and

Psacharopoulos (1997) in Peru, Ersado (2005) in Nepal, Peru, and Zimbabwe, and Bhalotra and

Heady (2003) in Pakistan.

3.12 Region Effects

Within countries, rural areas support a higher incidence of child labor than do urban areas

for nearly all of the empirical studies surveyed here (the one exception being Bangladesh, in

Amin, Quayes and Rives (2004), where urban areas have significantly higher labor rates). Since

most of the studies control for household income, a higher percentage of poverty in rural areas is

not a sufficient reason. Bhalotra and Tzannatos (2003) offer other possible reasons: relatively

weak school infrastructure and lower rates of technical change in rural areas may discourage

school attendance. Children may also be more easily absorbed into the informal economies of

rural areas, on account of the prevalence of self-employment, relatively low skill requirements in

agricultural work, and the greater degree of market imperfection in rural regions. As with child

gender, there is considerable evidence that the determinants of child labor and schooling change

depending on whether the region is rural or urban, indicating that regressions should be done

separately for these two groups.


3.13 Community Infrastructure Effects

Besides the importance of the presence and quality of schools on child labor and

schooling decisions, other measures of community infrastructure, such as access to public

transport, safe drinking water, and electricity, often play a role in whether a child will work

and/or go to school. We would expect communities with better, more established infrastructure

to have higher school attendance since the costs would be less (if the school is closer or if there

is public transportation) and/or the benefits greater (if the quality of the school is better).

Examining child labor and schooling in rural areas of Ghana and Pakistan, Bhalotra and

Heady (2003) find that the presence of a girls’ primary school has no effect on children in Ghana

or Pakistan, but a boys’ primary school increases the probability that girls will work in Pakistan

with no effect for boys. Surprisingly, the presence of a girls’ primary school significantly lowers

the probability that a girl in Pakistan will attend school but raises the probability that a boy in

Ghana will attend school (no effect for boys in Pakistan or girls in Ghana). The presence of a

middle school and/or a secondary school significantly lowers the probability of working for boys

and girls in Ghana (with no data available for Pakistan), but has no effect on child schooling

rates. The availability of public transport also lowers the probability of work in both countries,

but only for girls. Public transport also raises the probability of schooling for boys in Pakistan,

but lowers it for boys and girls in Ghana (with no effect on girls in Pakistan). Blunch and Verner

(2001), also analyzing data from Ghana, find that the distance to the nearest primary school is

significantly correlated with child labor for rural children, but not urban children; the same result

is true of distance to the nearest secondary school. This is probably the result of the scarcity of

schooling in rural areas due to their geographic isolation. This could also help explain why rural

areas have a higher incidence of child labor and lower child schooling rates than urban areas


Ersado (2005) finds that in rural Nepal, the number of schools in the area is positively

correlated with schooling, but has no effect on child labor (no data for number of schools

available for urban Nepal, or for Peru and Zimbabwe, the other two countries he surveys). Also,

having electricity is positively correlated with schooling for urban children in Peru and

negatively correlated with child labor for rural children in Nepal (no effect for Zimbabwe or for

other groups in Peru and Nepal); bad water storage is positively correlated with child labor in

rural Nepal and rural Zimbabwe, but also positively correlated with schooling for rural

Zimbabwe and urban Nepal. In Ray (2000), having electricity lowers the probability of child

labor for boys in Peru, but raises it for girls in Pakistan, with no effect on girls in Ghana and boys

in Pakistan. Having electricity also significantly raises the probability of girls in Peru and boys

in Pakistan going to school, with no effect on boys in Peru or girls in Pakistan. Bad water

storage is significantly positively correlated with child labor for boys in Pakistan, and negatively

correlated for girls in Peru, with no effect for girls in Pakistan and boys in Peru; it is also

significantly negatively correlated with child schooling for boys and girls in Pakistan, but has no

effect on schooling for children in Peru.


4. Methodology

My selection of which community, household, and child characteristics to consider is

based on previous studies on child labor, detailed in depth in the previous section, the theoretical

model, and on what data is available. The data set I am using comes from the Philippines’ 2001

Survey on Children 5-17 Years Old (SOC), a nationwide survey which provides estimates on the

number of Filipino working children between ages 5 and 17 years, their characteristics and those

of the households to which they belong. The 2001 SOC is a collaborative effort between the

Philippines’ National Statistics Office (NSO) and the International Labor Organization’s

International Program on the Elimination of Child Labor (ILO-IPEC). It has a national sample

of 26,964 sample households, 17,454 of which were found to have members whose ages were 5-

17 years old. Out of these, 17,444 households (99.9%) responded (NSO, p. 5).

The data set has data on sex, age, and education for each individual in the household and

household data on gross monthly income, expenditures, land ownership, access to drinking water

and electricity, and participation in governmental assistance programs. For children ages 5-17,

they also included data on what the child does on a day to day basis—worked, looked for work,

studied, did housekeeping and/or was idle. There is also a second survey for all children who

said that they had worked (not including housekeeping) in the last twelve months, which asked

detailed questions about what kind of work they were doing—how many hours, whether it was

risky or not, its effect on health and school, etc.

Much of the previous literature uses the probability that a child has worked in the last

twelve months as their measure of child labor, and I wanted to see if the results changed at all

based on a narrower definition of child labor. The International Labor Organization defines

child labor as work that is “mentally, physically, socially or morally dangerous and harmful to


children” and/or “interferes with their schooling: by depriving them of the opportunity to attend

school; by obliging them to leave school prematurely; or by requiring them to attempt to

combine school attendance with excessively long and heavy work” (ILO-IPEC, p. 16).

However, the ILO also admits that “whether or not particular forms of ‘work’ can be called

‘child labour’ depends on the child’s age, the type and hours of work performed, the conditions

under which it is performed and the objectives pursued by individual countries” (ILO-IPEC, p.

16).

With such a vague definition, I tried to be as conservative as possible in my definition of

child labor. If a child’s work was hazardous, in bad conditions, at night, caused emotional or

mental stress, or entailed heavy physical labor, I defined that as child labor, regardless of the

child’s age, under the reasoning that these conditions were “harmful to physical and mental

development” (ILO-IPEC, p. 16). The latter half of the definition, i.e. that work must not

interfere with schooling, depends a great deal on the child’s age, though the ILO specifies that

only light work can be allowed before the age of completion for compulsory schooling, which in

the Philippines is fourteen. I included in my definition of child labor any child that was fourteen

or under and not attending school, as well as those fourteen and under working nine hours or

more per day (regardless of how many days per week they worked) and/or working five hours or

more per day for more than two days per week, on the grounds that long hours would interfere

with their schooling.

Unfortunately, the survey only asked children who said they were working if they

attended school. Because of this, I could not construct a variable for whether a child was in

school unless I limited the sample to those working children. Though the question of what

causes a child to go to school when already working is an interesting one, I wanted to be able to


use the whole sample of children in my analysis. To do this, I used the probability that a child

studies, which is a question they ask all children ages 5 to 17. This is not a perfect proxy for

school attendance, but it still has similar social and economic consequences. In areas with poor

school quality, studying might actually be a better measure of human capital accumulation than

school attendance alone.

The independent variables I chose are based on previous literature and the theoretical

model. On the household level, I have included monthly expenditure, which in this dataset is

given in six levels rather than as a continuous variable. Following previous literature, I am using

expenditures rather than incomes, since agricultural incomes are volatile, making measured

income sensitive to the reference period of the survey. I also included the region of the

household (rural or urban), and if the household has land. Unfortunately, the survey did not

include questions about the size of the land, but evidence that houses with land have higher child

labor rates than those without also supports Bhalotra and Heady’s hypothesis that one of the

causes of child labor is the lack of a labor market (described in Section 3.1.1). Instead of only

controlling for household size, I have followed some of the previous literature and accounted for

the composition by dividing the household into different age groups and gender. The household

composition variables I am looking at are the total number of children under 5, the total number

of children aged 5 to 17, the total number of adults aged 18 to 59, and the total number of adults

who are 60 or older in the household. All of these groups are also divided by gender.

I have accounted for various characteristics of the household head, including gender, age,

marital status, and his or her education level. Though previous studies consider characteristics of

both parents, my model assumes that the household head, regardless of relation to the child,

makes the decisions regarding child activities. Over 15% of children in the survey are not the


biological children of the head, indicating that the parents of the children are not always the main

decision makers in households in the Philippines. I have also interacted whether the head is

single with whether the head is female to see if being single affects a female-headed household

more than a male-headed household.

The characteristics for individual children I am considering are age, gender, and whether

the child has a strong biological relationship to the household head, i.e. is the son/daughter,

brother/sister, or grandchild, as opposed to more distant relatives or completely unrelated

children. Following previous literature, I have added a quadratic term for age. Though the

theoretical model also focuses on child wage, there is no reliable data available in the dataset,

and the issue of endogeneity also makes it difficult to interpret the effect of wage. There is only

wage data for children who work, so wage, at least as it is measured here, is not an appropriate

proxy for the opportunity cost of labor. The closest proxy is whether or not the household has

agricultural land, since owning land would increase the marginal product of labor (and therefore

the opportunity cost of education) since the children can most likely work more easily and

efficiently on their own land.

For community characteristics, I am using access to drinking water and electricity, and

also a variable for whether the household is aware of a community organization in their area to

see if there are positive effects from local organizations. There is no data on the availability,

quality, or costs of schools, unfortunately, so I have to use access to drinking water and

electricity and the presence of a community organization as signals of the community’s

infrastructure, of which schools are a part. I have also added a variable for whether a family is

aware of government welfare programs to examine the possible effects of government programs

designed to help poor families. I interacted this variable with the gender of the household head


based on previous literature indicating that cash transfers to women generally benefit the family

more, and generally benefit girls more than boys (Carvalho 2000; Duflo 2003; Cardosa and

Souza 2003; and Rubalcava, Teruel, and Thomas 2009).

Following the majority of the previous literature, I am using a multinomial logit model

with three outcomes: engaged in child labor only, engaged in child labor and studying, and

studying only. Since in the previous literature many of the factors affect child labor differently

depending on the gender of the child and the region in which they live (rural or urban), I ran the

estimations for the group as a whole, then separately for rural and urban children and boys and

girls, and then again for four groups—urban girls, urban boys, rural girls, and rural boys. All the

estimations also control for fixed effects of the different regions (16 total regions in the survey).

The model I am going to estimate is:

Activity = B0 + B1• (Male) + B2• (Age) + B3• (Age Squared) + B4• (2nd Income Quantile)

+ B5• (3rd Income Quantile) + B6• (4

th Income Quantile) + B7• (5th Income Quantile) +

B8•(6th Income Quantile) + B9• (Land) + B10• (Rural) + B11• (Electricity) + B12• (Access

to Water) + B13• (# Male Children 0-5) + B14• (# Female Children 0-5) + B15• (# Male

Children 5-17) + B16• (# Female Children 5-17)+ B17• (# Male Adults 18-59) + B18• (#

Female Adults 18-59) + B19• (# Male Adults >60) + B20• (# Female Adults >60) + B21•

(Biological Relationship) + B22• (Female Head) + B23• (Single Head) + B24•

(Single*Female Head) + B25• (Age of Head) + B26• (Head Education 1) + B27• (Head

Education 2) + B28• (Head Education 3) + B29• (Head Education 4) + B30• (Head

Education 5) + B31• (Welfare) + B32•(Welfare*Female Head) + B33• (Community

Organization)


For easier reference, I have included a list with variable names and descriptions.

Name Description

Activity Categorical variable equal to 0 if a child is only engaged in child labor, 1 if engaged in both child labor and studying, and 2 if studying only

Male Dummy variable equal to 1 if the child is male, and 0 if female Age Age of the child (years)

Age Squared Age of the child squared (years squared) 2nd Income Quantile Dummy variable equal to 1 if household has average monthly

expenditure of PhP2000-2999 (baseline is households spending less than PhP2000)

3rd Income Quantile Dummy variable equal to 1 if household has average monthly expenditure of PhP3000-4999

4th Income Quantile Dummy variable equal to 1 if household has average monthly expenditure of PhP5000-9999

5th Income Quantile Dummy variable equal to 1 if household has average monthly expenditure of PhP10000-14999

6th Income Quantile Dummy variable equal to 1 if household has average monthly expenditure of over PhP15000

Land Dummy variable equal to 1 if household has agricultural land

Rural Dummy variable equal to 1 if household is in a rural region

Electricity Dummy variable equal to 1 if household has electricity

Access to Water Dummy variable equal to 1 if household’s main source of drinking water is a community water system or a tubed/piped well (as opposed to dug well, river, rain)

# Male Children 0-5 Number of male children under 5 in the household

# Female Children 0-5 Number of female children under 5 in the household # Male Children 5-17 Number of male children aged 5-17 in the household # Female Children 5-17 Number of female children aged 5-17 in the household # Male Adults 18-59 Number of male adults 18-59 in the household # Female Adults 18-59 Number of female adults 18-59 in the household # Male Adults >60 Number of male adults 60 or older in the household # Female Adults >60 Number of female adults 60 or older in the household Biological Relationship Dummy variable equal to 1 if child is the household head’s son,

daughter, brother, sister, or grandchild

Female Head Dummy variable equal to 1 if household head is female Single Head Dummy variable equal to 1 if household head is single Single*Female Head Dummy variable equal to 1 if household head is single and female Age of Head Age of the household head (years) Head Education 1 Dummy variable equal to 1 if household head’s highest level of

education is elementary graduate (baseline is no school or not enough school to graduate elementary)

Head Education 2 Dummy variable equal to 1 if head’s highest level of education is high school (not graduating)

Head Education 3 Dummy variable equal to 1 if head’s highest level of education is


high school graduate Head Education 4 Dummy variable equal to 1 if head’s highest level of education is

college (not graduating) Head Education 5 Dummy variable equal to 1 if head’s highest level of education is

college graduate and above Welfare Dummy variable equal to 1 if household knows about governmental

welfare programs Welfare*Female Head Interaction term of household head being female and knowing about

welfare Community Organization Dummy variable equal to 1 if household knows about a community

organization in their area


5. Results

Table 1 shows the estimation results for the full sample, and separate results for rural and

urban, and boys and girls, while Table 2 shows the results for four separate groups: rural boys,

rural girls, urban boys, and urban girls. From Table 1, we can see that males are more likely to

be engaged in child labor (part time and full time) relative to studying than females, and the

effect is strong in both rural and urban areas. Children in rural areas are also more likely to work

both part time and full time than those in urban areas though the effect is stronger for boys. This

is consistent with the majority of previous empirical evidence. However, being in a rural area,

though it does affect the probability a girl will labor and study, has little effect on whether a girl

will labor full time or not.

Overall, the results show strong support for the Luxury Axiom in rural areas, i.e. as

households become richer, their children are less likely to be engaged in child labor (both part

time and full time) than to only study. Income is not as strong a determinant in urban areas,

especially for girls. In fact, for urban girls, children are actually more likely to be engaged in

part-time child labor than only studying as income increases, though the effect on labor only is

insignificant. However, since the income data is top-coded, the effects of income might actually

be understated in the results. Households in the bottom five (out of six) income quantiles,

accounting for more than 90% of the surveyed households, spend less than PhP15,000 a month,

which is approximately equal to $9184 a year (in 2001 PPP terms) (see Table 11 in the

Appendix). Considering the average household has over six people in it, the per capita

expenditure is quite low for the majority of survey respondents. Even at these low levels of

income, an increase in income still reduces child labor (part time and full time) for rural children

and for urban boys (though the effect on part-time labor is not significant).


The significance of income for rural children and the insignificance for urban children is

consistent with Ersado’s (2005) study of Peru, Nepal, and Zimbabwe and Canagarajah and

Coulombe’s (1999) study of Ghana. The positive effect of income on the probability that an

urban girl studies and labors relative to only studies could be due to Bhalotra and Heady’s (2003)

hypothesis that richer households with a larger amount of land use more child labor. Though

there is a control in this model for whether the household has agricultural land, there is no

measure of the size of that land, and merely owning land does not seem to be correlated with

income level (see Table 11 in the Appendix). However, it seems strange that this effect would

be present for girls but not boys and in urban areas but not rural areas. This could also result

from the endogeneity of income, since children’s income contributes to household income.

Endogeneity problems would lead to understatements of the effect of income or even positive

effects, since total household income would increase as the child worked more. Endogeneity

also explains why households in the second income quantile do not generally have significantly

less labor than those in the bottom quantile, and actually have significantly more in the case of

urban girls. These households might have been in the bottom quantile of income if their children

were not working.

The strong significance of electricity and the slightly weaker significance of access to

drinking water, in reducing labor both part-time and full-time for all groups indicate that

community infrastructure is important. Of course, these are also possible measures of wealth,

and might even be better measures of wealth, since they are not endogenously determined.

Electricity and access to water could also be correlated with other aspects of community

infrastructure, like public transport and schools. The presence of schools and transportation

would decrease the costs of schooling. The theoretical model predicts that full-time work will


decrease as educational expenses decrease. The effect on part-time work depends on the relative

sizes of the income and substitution effects. The income effect, i.e. that lower educational costs

mean increased income, would lead to children working less due to the increase in income. The

substitution effect would lead to children studying less since the decrease in educational costs

means that students can get more human capital for a decreased amount of studying time. In the

results, the probability of full-time work is significantly lower in the presence of electricity and

access to drinking water, as the model predicts. Since part-time work also decreases, the income

effect must dominate the substitution effect. This assumes that electricity and access to drinking

water are related to schooling costs.

Owning agricultural land is strongly correlated with working and studying rather than just

studying. If we assume that the opportunity cost of education is higher for children in

households with land (since they can more easily and efficiently work their own land), the

theoretical model would predict that full-time work would increase and full-time studying would

decrease. Though there is evidence to support the latter, full-time work does not increase

significantly. Moreover, for rural girls, land has no effect on labor at all. Since the land measure

only indicates the presence of land, rather than the size of the land, it is possible that many

households have small plots (the fact that many of the households with land are quite poor

supports this supposition, see Table 11). With such a small plot, there may be enough labor to

productively work on the land without having to take a child out of school, and so having land

does not inhibit the children from working and studying.

Unsurprisingly, age is positively correlated with engaging in child labor, often with a

quadratic effect. It is interesting to note, though, that age is insignificant in determining whether

girls (both rural and urban) are engaged in child labor full time. Household composition,


income, and especially the biological relationship with the household head are more important

determinants of the probability a girl will work full time. Having a strong biological relationship

with the head is negatively correlated with working full time for all groups, but the effect is

stronger for girls, and the strongest for urban girls. This is consistent with the hypothesis that the

household head is less altruistic to distant relatives and foster children, and this effect is

particularly strong for girls.

The effects of household composition are different among the different groups, though

for all groups there is evidence that a higher number of children in the household leads to more

child labor. The presence of younger children has a stronger effect in most cases, as we would

expect from the theoretical model. Pre-school children are too young to work, and so an increase

in their number, holding income constant, is equivalent to a reduction in full income (an income-

dilution effect). According to the theoretical model, a decrease in income will raise the

probability of full-time work, and lowers that of full-time study. These predictions are consistent

with the results. An increase in the number of school-age (5-17) children, though also lowering

total income per household member, should have less of an effect since they can earn income as

well. The results are consistent with these predictions in general, as the effects of the total

number of school-age children are negligible or at least smaller than the effects of pre-school

children for all of the groups. The number of school-age children can still have a significant

effect, especially if a child’s siblings, though school-age, are much younger than the child.

The effect of the number of adults is variable, and depends on the age and gender of the

adult as well as the region and gender of the child. Older female adults play the most important

role in reducing child labor and increasing studying, though young and middle-aged female

adults are important as well. The number of male adults, on the other hand, is generally less


significant in reducing child labor, and for urban girls and boys, is even positively correlated

with child labor. This may be reflective of different levels of altruism in males and females or

their different views on the benefits of schools versus labor. The age of the household head is

mostly insignificant, as expected, since the average age of the household is already accounted for

by the controls for household composition.

Female headship plays a mostly insignificant role, as does the head being single and the

interaction of the two. Though female headship appears highly significant for rural boys, all of

the female heads are single in households where rural boys work full time, making these results

difficult to interpret. When I excluded the interaction term from the estimation, female headship

had an insignificant effect. A single male head does lower the probability of rural boys working

part time (relative to only studying), while a single female head raises it. A single head, male or

female, raises the probability that rural girls are engaged only in child labor.

The education level of the household head is strongly negatively correlated with the

probability of being engaged only in child labor for all groups, and also with the probability of

part-time work for all groups except rural girls. This is consistent with most of the previous

literature. Since income is already controlled for, this cannot only reflect that household heads

with more education are richer. However, since expenditures are only a proxy for income, the

years of education of the household head might capture some permanent income effects.

Moreover, more educated heads might have a better knowledge of the returns to education and/or

be in a better position to enable their children to exploit the earning potential acquired through

education. In this case, the head’s education could be a proxy for returns to education.

The effect of government programs, specifically welfare and community organizations,

seems relatively weak, and only present for rural boys. In fact for urban boys, knowledge of


welfare programs is actually positively correlated with working. This weak effect is most likely

due to the small number of people who actually know about these programs (see Table 11 in the

Appendix). Those who do know about the programs are generally more well off. Similarly,

community organizations are generally in areas that are relatively richer, reducing the impact

they might otherwise have on child labor and schooling (Table 11).


6. Conclusions

I have found strong support in the Philippines that income and child labor are correlated.

Programs aimed at promoting economic growth and alleviating poverty should help reduce child

labor and increase school attendance. Equally important though, is community infrastructure—

electricity, water access, schools, public transport. These strongly reduce the probability that

child will be engaged full time in child labor and also increase the probability that a child will

study full time for all groups of children. This is especially important when one considers the

intergenerational effects. Children who grow up to be household heads will be more likely to

allow their children to study full time rather than work if they have more education. If, on the

other hand, children work full time or cannot successfully study in combination with strenuous

labor, their educational level will most likely be the same or worse than their parents. As we can

see from the results, lower education levels lead to more child labor, even when controlling for

household income. If we also assume that less education leads to less income, the negative

effects of working rather than studying will be passed down from generation to generation.

Government welfare programs, on the other hand, seem to do little to alleviate child

labor, probably due to the fact that the poorest are generally unaware of them. Similarly, it

seems that community organizations are generally built in relatively richer areas and hence are

not particularly helpful to the most poverty-stricken. Strengthening government communication

with the poor could make these programs more effective. These results also reveal that children

who do not have a strong biological attachment to the household head are much more likely to

work and less likely to study. This indicates that children in the care of adults that are unrelated

or distantly related are often mistreated compared to biological children. Government programs


designed to help these children would also reduce the overall incidence of child labor, especially

for urban girls.


7. Appendix

Descriptive Statistics for Child Characteristics

Table 3 – Percentage of Children in Each Income Quantile

Full Sample Rural Boys Rural Girls Urban Boys Urban Girls

Income Quantile % of Children % of Children % of Children % of Children % of Children

Bottom (<2000) 6.88 9.40 9.26 3.88 3.87

2nd (2000-2999) 13.33 18.52 18.18 8.44 8.28

3rd (3000-4999) 27.85 32.43 32.06 23.81 22.98

4th (5000-9999) 29.61 26.69 26.93 32.95 32.14

5th (10000-14999) 13.31 8.32 8.74 17.71 18.41

Top (≥15000) 9.02 4.64 4.83 13.21 14.32

Total 100.00 100.00 100.00 100.00 100.00

Observations 37326 8167 7630 10982 10547

*All income levels given in Philippine Pesos (PhP)

**In PPP terms, $1 in 2001 was equal to PhP 19.6

Table 4 – Percentage of Children by Education Level of Head


Highest Education Level of Head % of

Children % of

Children % of

Children % of

Children % of

Children Did not graduate elementary school 22.97 33.30 33.56 15.23 15.37

Elementary Graduate 20.26 25.46 24.93 16.37 16.92

High School Level 12.61 12.86 12.48 13.16 11.96

High School Graduate 22.08 17.30 17.61 25.71 25.23

College Level 11.46 6.88 7.10 14.77 14.72

College Graduate/Postgraduate 10.60 4.20 4.31 14.75 15.80

Total 100.00 100.00 100.00 100.00 100.00

Observations 37326 8167 7630 10982 10547

Table 5 – Summary of Continuous Variables


Variable Mean SD Mean SD Mean SD Mean SD Mean SD

Age 11.25 3.38 11.25 3.28 11.14 3.27 11.29 3.44 11.29 3.45

# Male Children 0-5 0.30 0.55 0.31 0.57 0.31 0.55 0.29 0.55 0.29 0.55

# Female Children 0-5 0.28 0.54 0.30 0.55 0.30 0.56 0.26 0.52 0.26 0.54

# Male Children 5-17 1.61 1.17 2.16 1.05 1.17 1.04 2.06 1.08 1.03 1.03

# Female Children 5-17 1.51 1.13 1.08 1.01 2.08 1.03 0.99 1.01 1.99 1.01

# Male Adults 18-59 1.36 0.84 1.32 0.79 1.33 0.79 1.37 0.86 1.39 0.87

# Female Adults 18-59 1.36 0.79 1.25 0.67 1.24 0.65 1.44 0.86 1.44 0.85

# Male Adults >60 0.09 0.29 0.09 0.29 0.09 0.28 0.09 0.29 0.10 0.30

# Female Adults >60 0.12 0.33 0.10 0.31 0.10 0.32 0.13 0.35 0.13 0.35

Total Household Size 6.62 2.19 6.60 2.01 6.62 2.05 6.63 2.29 6.64 2.31

Age of Head 45.47 10.82 44.85 10.26 44.93 10.58 45.83 11.02 45.96 11.15

Total Observations 37326 8167 7630 10982 10547


Descriptive Statistics for Child Characteristics (Cont’d)

Table 6 – % of Children (5-17) Who Answered Yes


Male 51.30

Land 28.43 41.40 40.77 18.98 19.30

Rural 42.32

Electricity 77.54 59.13 61.11 90.17 90.53

Access to Water 80.76 71.70 71.04 87.49 87.79

Biological Relation 95.69 97.07 96.74 95.66 93.91

Female Head 10.46 7.97 8.32 12.12 12.21

Single Head 12.41 10.41 10.73 13.74 13.78

Single and Female Head 9.50 7.29 7.65 10.95 11.03

Welfare 6.85 6.47 6.83 6.80 7.21

Welfare and Female Head 0.68 0.47 0.43 0.83 0.86

Community Organization 11.00 11.17 11.22 10.54 11.21

Total Observations 37326 8167 7630 10982 10547

Table 7 – Child Activity

% of Children Engaged in Activity

Child Activity Full Sample Rural Boys Rural Girls Urban Boys Urban Girls

Labor 12.18 21.49 10.55 11.09 7.30

Study 95.86 91.81 97.55 95.84 97.80

Labor Only 4.14 8.19 2.45 4.16 2.20

Study Only 87.82 78.51 89.45 88.91 92.70

Labor and Study 8.05 13.30 8.10 6.93 5.10

Observations 37326 8167 7630 10982 10547


Child Labor versus Child Work Table 8 - % of Children Engaged in Child Work and Child Labor % of Total Children (5-17)

Child Activity Full Sample Rural Boys Rural Girls Urban Boys Urban Girls

Work 16.05 27.48 15.19 13.85 10.11

Labor 12.18 21.49 10.55 11.09 7.30

Observations 37326 8167 7630 10982 10547

Table 9 - % of Children 5-17 Engaged in Work Defined as Child Labor

% of Working Children (5-17) Who Answered Yes

Hazardous work? 19.74 26.78 13.29 28.40 9.47

In bad conditions? 30.58 32.09 30.37 35.70 35.65

Heavy labor? 26.49 39.39 17.60 34.45 10.88

Stressful? 35.25 44.39 35.81 39.32 27.30 Worked nights often? 20.88 13.73 20.02 27.55 37.80

Observations 6524 2244 1159 1521 1066

Table 10 - % of Children Aged 5-14 Engaged in Work Defined as Child Labor

% of Working Children (5-14) Who Answered Yes Not attending school? 17.91 22.26 9.53 22.59 11.49

>9 hours per day 4.92 4.24 3.78 6.00 6.54

>5 hours/day for >2 days/week

14.88 17.29 8.47 17.57 13.47

Observations 3132 1249 661 717 505

Household Characteristics by Income Level

Table 11 – Household Level Characteristics of Each Income Quantile

Income Quantile # Households % of Total % Own Land % Aware of

Welfare % with Community

Organization

Bottom (<2000) 1200 6.88 28.75 4.25 6.58

2nd (2000-2999) 2325 13.33 30.88 4.95 8.04

3rd (3000-4999) 4859 27.85 28.79 6.36 9.49

4th (5000-9999) 5165 29.61 26.52 7.76 11.66

5th (10000-14999) 2322 13.31 27.39 7.71 12.70

Top (≥15000) 1573 9.02 28.67 8.84 17.55

Total 17444 100.00 28.20 6.84 10.89

*All income levels given in Philippine Pesos (PhP)

**In PPP terms, $1 in 2001 was equal to PhP 19.6


8. Reference List

Amin, S., M.S. Quayes and J.M. Rives. 2004. “Poverty and Other Determinants of Child Labor in Bangladesh,” Southern Economic Journal, June 70(4): pp. 876-892.

Basu, Kaushik, and Pham Hoang Van. 1998. “The Economics of Child Labor,” American Economic Review, June 88(3): pp. 412-27.

Behrman, J. and J. Knowles. 1999. “Household Income And Child Schooling In Vietnam,” The World Bank Economic Review, May 13(2): pp. 211-56.

Bhalotra, Sonia and Christopher Heady. 2000. “Child Farm Labor: Theory and Evidence,” London School of Economics. Development and Distribution Series Discussion Paper 24, July.

Bhalotra, Sonia and Christopher Heady. 2003. “Child Farm Labor: The Wealth Paradox,” World Bank Economic Review, 17(2): pp. 197-227.

Bhalotra, Sonia and Zafiris Tzannatos. 2003. “Child Labor: What Have We Learnt?” The World Bank Human Development Network. Social Protection Discussion Paper 0317, September.

Bhatty, Kiran. 1998. “Educational Deprivation in India: A Survey of Field Investigations,” Economic and Political Weekly, July 33(28): pp. 1858-1869.

Bishai, David, El Daw Suliman, Heena Brahmbhatt, Fred Wabwire-Mangen, Godfrey Kigozi, Nelson Sewankambo, David Serwadda, Maria Wawe, and Ron Gray. 2003. “Does Biological Relatedness Affect Survival?” Demographic Research, May 8(9): pp. 261-278.

Blunch, Niels-Hugo and Dorte Verner. 2001. “Revisiting the Link between Poverty and Child Labor: The Ghanaian Experience,” Centre for Labour Market and Social Research. Working Paper 01-03, April.

Bourguignon, F., F. Ferreira and P. Leite. 2002. “Ex-ante Evaluation of Conditional Cash Transfer Programs: the Case of Bolsa Escola,” The World Bank. Policy Research Working Paper Series 2916.

Canagarajah, S. and H.S. Nielsen. 2001. “Child Labor in Africa: A Comparative Study,” Annals of the American Academy of Political and Social Science, May 575, pp. 71-91.

Canagarajah, Sudharshan and Harold Coulombe. 1999. “Child Labor and Schooling in Ghana,” The World Bank. Policy Research Working Paper 1844.

Cardoso, Eliana and Andrea Souza. 2004. “The Impact of Cash Transfers on Child Labor and School Attendance in Brazil,” Vanderbilt University, Working Papers 0407, April.

Cartwright, K. 1999. “Child Labor in Columbia.” In The Policy Analysis of Child Labor: A Comparative Study, ed. C. Grootaert and H.A. Patrinos, 63-102. New York: St. Martin’s Press.


Cartwright, K. and H.A. Patrinos. 1999. “Child Labor in Urban Bolivia.” In The Policy Analysis of Child Labor: A Comparative Study, ed. C. Grootaert and H.A. Patrinos, 103-130. New York: St. Martin’s Press.

Carvalho, Irineu. 2000. “Household Income as a Determinant of Child Labour and School Enrolment in Brazil: Evidence from a Social Security Reform,” PhD diss. Massachusetts Institute of Technology.

Case, Anne, Christina Paxson, Joseph Ableidinger. 2004. “Orphans in Africa: Parental Death, Poverty, and School Enrollment,” Demography, August 41(3): pp. 483-508.

Cigno, Alessandro & Furio Camillo Rosati. 2000. “Why do Indian Children Work, and is it Bad for Them?” Institute for the Study of Labor (IZA). Discussion Papers No. 115.

Cigno, Alessandro, Furio Camillo Rosati, and Zafiris Tzannatos. 2001. “Child Labor, Nutrition and Education in Rural India: An Economic Analysis of Parental Choice and Policy Options,” The World Bank Human Development Network. Social Protection Discussion Paper No. 0131, December.

Cockburn, J. 2001. “Income Contributions of Child Work in Rural Ethiopia,” Mimeo, University of Laval, October. < http://www.crefa.ecn.ulaval.ca/cahier/0117.pdf>

Coulombe, H. 1998. “Child Labor And Education In Côte d’Ivoire,” The World Bank. Background Paper.

Dayioglu, Meltem. 2006. “The Impact of Household Income on Child Labour in Urban Turkey,” Journal of Development Studies, August 42(6): pp. 939-56.

Deb, Partha, and Furio Camillo Rosati. 2004. “Determinants of Child Labor and School Attendance: The Role of Household Unobservables,” ILO, World Bank, and UNICEF. Understanding Children’s Work Working Paper, April.

Duflo, Esther. 2003. “Grandmothers and Granddaughters: Old-Age Pensions and Intrahousehold Allocation in South Africa,” World Bank Economic Review, 17(1): pp. 1-25.

Dumas, C. 2007. “Why Do Parents Make Their Children Work? A Test of the Poverty Hypothesis in Burkina Faso,” Oxford Economic Papers-New Series, April 59(2): pp. 301-329.

Edmonds, Eric V. 2005. “Does Child Labor Decline with Improving Economic Status?” Journal of Human Resources, Winter 40(1): pp. 77-99.

Edmonds, Eric V. 2006. “Child Labor and Schooling Responses to Anticipated Income in South Africa,” Journal of Development Economics, December 81(2): pp. 386-414.

Emerson, Patrick M. and Andre Souza. 2008. “Birth Order, Child Labor, and School Attendance in Brazil,” World Development, September 36(9): pp. 1647-64.


Ersado, L. 2005. “Child Labor and Schooling Decisions in Urban and Rural Areas: Comparative Evidence from Nepal, Peru, and Zimbabwe,” World Development, March 33(3): pp. 455-480.

Grootaert, C. 1999. “Child Labor in Côte d’Ivoire: Incidence and Determinants,” The World Bank. Policy Research Working Paper Series No. 1905.

Ilahi, Nadeem. 2001. “Children’s Work and Schooling: Does Gender Matter? Evidence from the Peru LSMS Panel Data,” The World Bank. Policy Research Working Paper No. 2745, December.

International Labour Organization – International Programme on the Elimination of Child Labour (ILO-IPEC). 2004. Child Labour: A Textbook for University Students. Geneva, Switzerland: International Labour Office.

Jensen, Peter and Helena Skyt Nielsen. 1997. “Child Labour or School Attendance? Evidence from Zambia,” Journal of Population Economics, October 10(4): pp. 407-24.

Kruger, Diana I. 2007. “Coffee Production Effects on Child Labor and Schooling in Rural Brazil,” Journal of Development Economics, March 82(2): pp. 448-63.

Liu, A.Y. 1998. “School Children’s Participation Behaviour in Vietnam: An Empirical Analysis,” Paper presented at the Twelfth Annual Conference of the European Society for Population Economics, Amsterdam, Netherlands.

National Statistics Office (NSO) and ILO-IPEC. 2003. 2001 Survey on Children 5-17 Years Old: Final Report. Manila, Philippines: International Labor Office.

Nielsen, H. S. 1998. “Child Labor And Schooling In Zambia,” The World Bank. Working Paper.

Patrinos, H.A. and G. Psacharopoulos. 1995. “Educational Performance and Child labor in Paraguay,” International Journal of Educational Development, 15(1): pp. 47-60.

Patrinos, H.A. and G. Psacharopoulos. 1997. “Family Size, Schooling And Child Labor In Peru - An Empirical Analysis,” Journal of Population Economics, October 10(4): pp. 387-406.

Ravallion, Martin and Quentin T. Wodon. 1999. “Does Child Labor Displace Schooling? Evidence on Behavioral Responses to an Enrollment Subsidy,” The World Bank. Policy Research Working Paper No. 2116, May.

Ray, Ranjan. 1998. “Analysis of Child Labour in Peru and Pakistan: A Comparative Study,” Mimeo, University of Tasmania, Australia.

Ray, Ranjan. 2000. “Child Labour, Child Schooling and their Interaction with Adult Labor: Empirical Evidence for Peru and Pakistan,” The World Bank Economic Review, May 14(2): pp. 347-71.


Rosati, Furio Camillo and Mariacristina Rossi. 2003. “Children's Working Hours and School Enrollment: Evidence from Pakistan and Nicaragua.” World Bank Economic Review, 17(2): pp. 283-95.

Rosati, Furio Camillo and Zafiris Tzannatos. 2000. “Child Labor In Vietnam,” The World Bank. Working Paper No. 14.

Rosati, Furio Camillo and Zafiris Tzannatos. 2006. “Child Labour in Vietnam,” Pacific Economic Review, 11(1): pp. 1-31.

Rubalcava, Luis, Graciela Teruel, and Duncan Thomas. 2009. “Impacts of the Oportunidades Program: Investments, Time Preferences, and Public Transfers Paid to Women,” Economic Development and Cultural Change, April 57(3): pp. 507-38.

Sakellariou, Chris and Ashish Lall. 1999. “Child Labor in the Philippines.” In The Policy Analysis of Child Labor: A Comparative Study, ed. C. Grootaert and H.A. Patrinos, 131-51. New York: St. Martin’s Press.

Sasaki, M. and T. Temesgen. 1999. “Children in Different Activities: Child Labor and Schooling in Peru,” Mimeo, The World Bank.

Schady, Norbert and Maria Caridad Araujo. 2006. “Cash Transfers, Conditions, School Enrollment, and Child Work: Evidence from a Randomized Experiment in Ecuador,” The World Bank. World Bank Policy Research Working Paper No. 3930, June.

Subbarao, K., A. Mattimore and K. Plangemann. 2001. “Social Protection Of Africa’s Orphans And Vulnerable Children: Issues And Good Practice Program Options,” The World Bank. Africa Region Human Development Working Paper, August.

Tunali, I. 1997. “Young Hands at Work: Child Labor in Turkey,” Mimeo, Koc University.

the determinants of child labor and schooling in the ... · the determinants of child labor and...

Documents