Child Labour Versus Education: Poverty Constraints or Income Opportunities?*
John Cockburn
Centre for the Study of African Economies (CSAE)
and
Nuffield College
(Oxford University)
Child labour is commonly associated with poverty. However, the empirical evidence on
this link is weak. By explicitly integrating the role of household asset profiles we provide a fuller
and more nuanced explanation of child labour and schooling decisions. We use a simple agr icultural household model with a missing labour market to show how the extent and composition
of household asset portfolios simultaneously determine household income and the shadow wage
of (demand for) child labour. Child labour-increasing (-decreasing) assets are characterised by a
dominant wage (income) effect. Empirical analysis of data on rural Ethiopian households suggests that small livestock and land area are child labour-increasing, whereas oxen, bulls, ploughs,
land quality and proximity to a source of water are child labour-decreasing. Various indicators –
marginal effects, elasticities and predicted probabilities – are used to examine the impact of each
factor. We conclude that both poverty constraints and income opportunities play important roles
in the decision to send children to school or to work. We also find that work and school conflict
substantially but not entirely.
22 March, 2000
*
I would like to acknowledge financial support from the SSHRC, the Commonwealth Scholarship Program, DFID,
the ORS Awards Scheme and Nuffield College. I am also very grateful to Simon Appleton, Anthony B. Atkinson,
Bart Capeau, Ludovico Carraro, Paul Collier, Doris Cossette, Stefan Dercon, Benoit Dostie, John Hoddinott,
Bereket Kebede, Pramila Krishnan, John Muellbauer, Steve Nickell, Francis Teal and Sharada Weir for their
assistance and suggestions. All errors and omission are my own responsibility.
INTRODUCTION
The relationship between poverty and child time use decisions is complex and controversial. It is generally assumed that as household wealth increases children will be progressively
withdrawn from labour activities in favour of schooling 1 . To the extent that schooling and leisure
are normal consumer goods, demand for them will increase – and child labour supply will fall –
as income rises. To the extent that schooling is a profitable investment, i.e. the net expected
returns to schooling are greater than to child labour activities, increased wealth may encourage
schooling by relaxing household credit constraints. It may also be argued that the returns to
schooling themselves increase with household wealth, through social capital, self-employment
opportunities or other employment advantages. Empirical work has consistently failed to demo nstrate a strong relationship between child labour and income, although the link between schooling and income is well established 2 .
One possible explanation for the weak empirical link between child labour decisions and
income is that income variables are proxying omitted asset variables. Household wealth is
generally associated with greater access to productive assets. As most child labour is performed
within the household and smoothly functioning child labour markets are rare, household access
to productive assets increases the productivity of, and demand for, child labour. Indeed, land and
livestock ownership and having a family enterprise have all been shown to increase child labour
participation3 . While household income draws children out of work and into school, the productivity effect of greater asset holdings does the contrary. As we could expect the productivity
effect to be particularly strong in child labour decisions, the net effect may be weak.
This issue is of crucial policy importance where the reduction of poverty and child labour
and increased schooling are policy objectives. Recent research on poverty suggests that the most
effective manner to combat poverty is to increase the access of the poor to productive assets 4 .
According to de Janvry and Sadoulet (1996, abstract), "insufficient access to assets is the main
determinant of poverty". To the extent that assets contribute to household income and that
poverty constrains child time-use decisions, the policymaker is in a win-win situation of simultaneously reducing poverty, reducing child labour and increasing schooling. However, if increased
access to assets raises the returns to child labour sufficiently, it may instead encourage child
labour at the expense of schooling by creating profitable income opportunities. To the extent that
reduced schooling prevents the accumulation of human capital, long-term poverty alleviation
may even be compromised, creating a lose-lose situation. As the point of asset-based poverty
alleviation policies is to increase access to all assets, conflict between physical asset and human
capital accumulation must be closely monitored. Efforts should be made to identify the optimal
1
See, for example, the World Bank's position paper Fallon and Tzannatos (1998) and ILO (1996).
See Bhalotra and Heady (1998), Levison and Moe (1998), Mueller (1984), Psacharopoulos (1997), Rosenzweig
(1981) and Ravallion and Wodon (1999).
3
See Bhalotra and Heady (1998), Canagarajah and Coulombe (1997), De Tray (1983), Levison and Moe (1998),
Mergos (1992), Mueller (1984) and Rosenzweig and Evenson (1977). Further evidence of a high elasticity of child
labour relative to its returns is provided by the market child labour literature. Bhalotra and Heady (1998), Mergos
(1992), Rosenzweig (1981) and Skoufias (1994) all find a significant and strong positive relationship between child
market wage rates and child labour participation. Bhalotra (1999) finds a negative effect of wages on child labour
among the extremely poor, which she interprets as the result of a dominant income effect from children wages.
4
See also Dercon and Krishnan (1998) and Owens and Hoddinott (1999).
2
1
human capital-physical asset combination taking into account their intimate relationship via child
labour and schooling decisions.
In most cases, the types of activities performed by children are often quite different from
those performed by adults. In rural Ethiopia, the principal activities of children are fetching wood
or water and herding, whereas adult males are primarily involved in farming and adult females in
domestic work. It therefore seems likely that the effects on child labour will vary considerably
depending on the types of physical assets targeted in poverty alleviation policies. In particular,
targeting assets used in activities commonly performed only by adults may make it possible to
avoid increased child labour and reduced schooling. Furthermore, labour-saving assets such as a
nearby well or a wheelbarrow can be expected to directly reduce child labour and poverty.
A simple agricultural household model is used to examine the contrasting income and
productivity effects - and the resulting ambiguous net effect - of variations in asset holdings on
child labour supply. The analysis closely mirrors that of the backward-bending labour supply
curve in the neo-classical labour supply model. However, as only 1% of children and less than
10% of adults do any work outside the household (for wages or in-kind payment), a missing
market formulation is adopted 5 . We demonstrate how different physical assets have different
effects on child labour supply according to their degree of substitution. Increased access to
physical assets that require relatively more child labour, such as small animals, will tend to
increase child labour participation and reduce child schooling and leisure; the "income opportunities" (or substitution effect) effect will tend to dominate their "poverty constraints" (or income), effect. Consequently, the choice of assets in a poverty alleviation program is important
and should take into account the effects on child labour and schooling.
All the above-mentioned relationships are analytically ambiguous, and we thus argue the
importance of empirical analysis. We contribute to this agenda by estimating a reduced-form
child time use equation involving asset ownership and other conditioning variables likely to
influence the returns to child labour. The exceedingly low enrolment rates and high child labour
participation in rural Ethiopia, combined with extreme poverty, make it an ideal region for
examining the poverty-asset-child time allocation nexus. Like the wage coefficients in a standard
labour supply model, the coefficients on the asset variables encompass both an income ("poverty
constraint") effect and a substitution ("income opportunities") effect. As we observe the net
effect, we are able to identify, for each type of asset, which of the two dominates in each case. In
terms of our basic question, this analysis allows us to see whether access to each asset affects
child time use decisions primarily by relaxing poverty constraints on child schooling or by
creating income opportunities for child labour.
We first present the theoretical model, then the specific context of our empirical application – rural Ethiopia – and the regression results, before concluding.
5
Sadoulet and de Janvry (1995, p.149) give rural child labour as a typical example of market failure. The importance of the demographic variable in our child time use regressions below gives support to the missing market
hypothesis. See Lopez (1984) and Benjamin (1992) for formal tests of separability along these lines.
2
1. THEORY
1.1 Ambiguous net effect of physical asset holdings on child labour supply
To simplify, we consider a model of a one-member agricultural household that produces
and consumes one marketable good 6 . In terms of household decision-making, we adopt the
unitary approach and neglect intra-household bargaining and distribution issues. We contrast the
results with and without a smoothly function labour market and show that the effect of physical
asset holdings is only ambiguous in the latter case. Let us first assume that a perfect labour
market exists:
U = U(C,h)
Q = f(L; K)
PC = PQ - w(L-LH) + E = wLH + Π + E
LH + h = T
where C = consumption, h = non-work (school or leisure) time, Q = production, L = labour time
in household production, K = household assets (exogenous), P = the commodity price, w = the
market wage rate, LH = labour time of household member, E = non-labour income, Π = profits
from household production (PQ-wL) and T = time endowment. The budget and time constraints
can be combined to obtain the full-income constraint: PC = w(T-h) + Π + E
This is the familiar separable household model where production/labour demand decisions are independent of consumption/labour supply decisions. Production is a function of only
one endogenous variable, labour supply, which is itself determined independently of consumption/labour supply variables by the following profit-maximising condition: fL = W/P. The value
of profits is substituted into the full-income constraint and determines consumption and labour
supply according to the following utility maximisation condition: Uh /Uc = W/P
The separability of the model ensures that any variation in household assets (K) will only
affect labour supply through its effect on household profits and that its effects will be unambiguous. An increase in household assets will increase the marginal productivity of labour in household production leading to increased labour hiring. However, as the market wage rate is
exogenous, the only change on the consumption/labour supply side of the model will be the
resulting increase in profits. Assuming that non-labour time is a normal good, this will unamb iguously reduce the labour supply of the household member. Let us now see how the results
change in the absence of a labour market7 . As there is no hiring in or out of labour, all labour
time of the household member is devoted to household production (L=LH ):
U = U(C,h)
Q = f(L; K)
PC = PQ+E
L+h = T
6
If the produced and consumed goods are different, both their prices appear in the supply and demand functions. If
these prices are exogenous, the results will not change. Similarily, the introduction of a second purchased consumer
good, as in the standard Singh, Squire et al. (1986) model, does not fundamentally alter our results.
7
A brief list of studies involving non-separable household models include: de Janvry, Fafchamps et al. (1992),
Lambert and Magnac (1992), Nakajima (1986), Sadoulet, de Janvry et al. (1996) and Singh, Squire et al. (1986)
particularly Strauss' appendix and Lopez' chapter in this last book. See also Sadoulet and de Janvry (1995).
3
The first-order condition (Uh /U C=fL) can be represented by the point A0 in Figure 1
below. The household's production function is described by the curve Q0 . Utility maximisation
occurs at the point of tangency between the household production function (slope=fL ) and its
indifference curve U0 (slope = Uh /Uc).
If we consider the impact of an increase in the household's physical asset holdings (K),
we observe an outward shift in the household production curve. This results in a new optimum at
point A1 which, in this case, is situated to the left of point A0 implying an increase in labour time
from L0 to L1 . It is, of course, possible to redraw Figure 1 with A1 situated to the right of A0
(Figure 2). The net effect of increased physical asset holdings on labour time is thus ambiguous.
This ambiguity is the reflection of two conflicting effects: a negative "income" (more properly,
"profit") effect and a positive substitution effect. In a manner similar to the standard Slutsky
decomposition, we identify the substitution effect as the utility-constant change in labour supply
at the new equilibrium shadow wage (or marginal productivity) rate (f1 L). This is the movement
from point A to B, which unambiguously increases labour supply. By increasing the marginal
productivity of labour, increased asset holdings encourage a substitution of labour time for nonlabour time. The income effect can be measured by the movement from B to A1 . This income
effect unambiguously reduces labour time. In the case of Figure 1, the substitution effect dominates and the net effect of increased asset holdings is labour increasing. In the case of Figure 2, it
is the income effect that dominates and labour time declines.
Let us now derive these results more formally 8 . Assuming an interior solution for h and
C, The two first-order conditions (FOC) for utility maximisation are the following:
fL = Uh /UC = Z; Z=marginal rate of substitution of leisure for consumption
PC = PQ+E; P=price of household good.
As we are interested in distinguishing substitution and income effects, we first derive the effect
of a change in non-labour income by total differentiating the two FOC with respect to E:
fLL(δL/δE)=(δZ/δL)(δL/δE)+(δZ/δC)(δC/δE)
P(δC/δE) = PfL(δL/δE)+1
Substituting and simplifying, we obtain:
δL/δE = (-1/β)(1/p)(δZ/δC)
δC/δE = (1/β)((δZ/δL)-fLL), where β = Z(δZ/δC)+δZ/δL-fLL
Following Nakajima (1986), we assume that (δZ/δC) and (δZ/δL) are both positive. That
is, given the convexity of the indifference curve, the marginal rate of substitution of leisure for
consumption (the slope of the indifference curve) increases as labour supply and consumption
increases9 . We also assume fLL is negative given the concavity of the production function. Thus
the sign of the above income effects on labour supply and consumption depend on the sign of β,
which can be determined from the second-order condition (SOC) for utility maximisation:
8
9
The discussion here is based on Nakajima (1986), chpt.4.
dZ
Note:
=
dC
Uh
U
d h
UC
U CU hC − Uh U CC
dZ
UC
U U −U U
=
and
=
= C hh 2 h Ch
2
dC
dC
dh
UC
UC
d
with UC>0, Uh >0, UCC<0, Uhh <0 and
UhC=UCh >0 due to the quasi-concavity of the utility function.
4
Figure 1: Labour-increasing physical asset accumulation
C
A1
U1
-f1 L
Q0 =f(L; K0 )
Q1 =f(L; K1 )
B
A0
U0
-f0 L
-f1 L
E
h0
h1
L0
T
L1
5
Figure 2: Labour-decreasing physical asset accumulation
C
Q1 =f(L; K1 )
A1
0
0
Q =f(L; K )
U1
B
A0
-f1 L
U0
-f0 L
-f1 L
E
h0
T
L0
h1
L1
6
SOC: dfL/dL - dZ/dL <0
thus: δfL/δL-(δZ/δL)-(δZ/δC)(δC/δL) <0
or:
fLL-(δZ/δL)-Z(δZ/δC)<0
or:
-β<0 (β>0)
We can therefore conclude δL/δE <0 and δC/δE>0.
The effects of a change in physical asset holdings can be examined along similar lines.
This time we totally differentiate the FOC with respect to K:
fLL(δL/δK) + fLK = (δZ/δL)(δL/δK) + (δZ/δC)(δC/δK)
P(δC/δK) = PfL(δL/δK) + PfK
Solving and substituting the expressions for (δL/δE) and (δC/δE) and simplifying, we obtain:
δL/δK = PfK(δL/δE) + (1/β)fLK <>0
δC/δK = PfK(δC/δE) + (1/β)fLKZ >0
The first terms on the right-hand side of each equation are the income effects. The second
terms represent the substitution effects. Given fK>0, (δL/δE)<0, (δC/δE)>0, β>0, fLK>0 and Z>0,
the substitution effects are positive whereas the income effect is negative in the labour equation
but positive in the consumption equation. Clearly, the net effect on labour supply is ambiguous
and depends on the relative importance of the conflicting income and substitution effects.
In conclusion, when a smoothly functioning labour market is present, increased access to
physical assets will unambiguously increase income and reduce child labour. However, in the
absence of a labour market, we cannot predict the effects on child labour, schooling and
leisure time. Asset-based poverty alleviation policies may increase child labour at the cost
of reduced schooling and/or leisure time. In contrast, a lump-sum income transfer will
unambiguously increase income, schooling and leisure time while reducing child labour
regardless of the presence or absence of a labour market.
1.2 Differing effects of different assets
The basic conditions described by Figures 1 and 2 are valid not only for total household
production but also for any specific type of production. Let us distinguish two types of household
production (Q 1 and Q2 ) with activity-specific physical assets (K 1 and K2 ; e.g. land for farming
and livestock for herding):
U=U(C 1 ,C2 ,h)
Q1 =f(L1 ; K1 )
Q2 =g(L2 ; K2 )
P1 C1 +P2C2 =P1Q1 +P2 Q2 +E
L1 +L2 +h=T
where Li = labour time devoted to activity i. The first-order conditions become:
Uh /UC1 =fL
Uh /UC2 =gL
which can be represented by Figures 1 and 2. It is therefore possible to imagine the two figures
as representing two different activities using different physical assets. The formal model resolution is somewhat more complicated but leads to the same basic results.
7
In conclusion, the effects on child labour, schooling and leisure will depend crucially
on the type of assets owned by the household. Asset ownership policies aimed at poverty
alleviation may target specific types of assets in order to minimise child labour and maximise child schooling and leisure.
1.3 Differing effects on different household members
The results up until here have been based on a one-member household model that does
not distinguish between children and adults. However, it is straightforward to extend the model
to distinguish household members. In so doing, we show that different household members will
be affected differently by asset variations if they are not perfect substitutes. Let us consider now
a two-member household composed of one adult (superscript A) and one child (superscript C):
U=U(C 1 ,C2 ,hA,hC )
Q1 =f(LA1 ,LC1 ; K1 )
Q2 =g(LA2 ,LC2 ; K2 )
P1 C1 +P2C2 =P1Q1 +P2 Q2 +E
LA1 +LA2 +hA=TA
LC1 +LC2 +hC=TC
First-order conditions for utility maximisation in this two-member household are the following:
UhA/UC1 =fLA
UhA/UC2 =gLA
UhB/UC1 =fLB
UhB/UC2 =gLB
Each of these conditions can be represented by Figures 1 and 2. Asset accumulation that
is labour-increasing for one member may be labour-decreasing for another. Consider an extreme
case where children specialise in herding activities and adults in farming activities. Increased
livestock ownership may raise child labour activities through a dominant substitution effect
while reducing adult labour through a dominant income effect.
In conclusion, the specific characteristics of children and the types of activities they
participate will determine the time allocation effects of increased ownership of different
types of assets. Empirical analysis is required to determine which physical assets are child
labour-increasing and –decreasing and the extent of their respective impacts.
2. DATA AND SETTING
Ethiopia is a country of extremes. The second poorest nation in the World (GNP per capita
= $US 100), it also has the third highest fertility rate (seven children per woman) and the lowest
school enrolment rates (24% net primary school enrolment rate) 10 . Variations in rainfall can be
dramatic even within regions and render rural Ethiopian households particularly vulnerable.
Famines, a lengthy and ruinous civil war ending in 1991 and current tensions with Eritrea further
exacerbate the resulting climate of uncertainty. All these characteristics are even more acute in
rural areas. Although separating cause from effect is difficult, they are almost surely related to
10
World Bank (1998).
8
another lesser known but equally remarkable characteristic of Ethiopia, particularly rural Ethiopia: an extraordinarily high incidence of child labour 11 .
We study the time use of children aged 6 to 15 using data from three rounds of detailed
surveys of 1477 rural households from 15 villages throughout rural Ethiopia 12 . Practically all of
these children participate in household farm or domestic work activities and school attendance is
extremely low (18%), particularly among girls (14%). Given that practically all children do some
labour activities, we focus on the choice of a child's main activity, which better captures the
child labour-schooling trade-off. From Table 1 below, we can see that more than half of all
children (and almost 80% of 11- to 15- year old girls) have work as their main activity (Table
1)13 . Only 18% of children attend school14 . Education is not compulsory for children in Ethiopia.
Finally, a large share of children, primarily younger children, do not attend school and do not
have work as their main activity. Figure 1 provides a more detailed age profile of child activities.
Table 1: Children's main activities in rural Ethiopia
Ages 6 to 10
Ages 11 to 15
Boys
Girls
Total
Boys
Girls
Total
Work 47.5
51.4
49.5
63.5
78.1
70.9
School 15.2
10.6
12.8
31.7
18.0
24.8
Inactive 37.3
38.0
37.7
4.8
3.9
4.3
Total 100.0 100.0
100.0
100.0
100.0
100.0
Count (678)
(700) (1378)
(526)
(544) (1070)
All
Boys
54.5
22.4
23.1
100.0
(1204)
children
Girls
Total
63.1
58.9
13.8
18.1
23.1
23.1
100.0
100.0
(1244) (2448)
Figure 3: Main activities of children and young adults in rural Ethiopia
(Percentage of children with the main activity indicated, by age)
%
100
80
60
40
Main activity
Attending school
20
Work
0
Inactive
.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
16.0
18.0
20.0
Age
11
ILO (1995) and ILO (1996) provide general overviews of the child labour situation in Ethiopia.
The Centre for the Study of African Economies (CSAE) and the Economics Department of Addis Ababa University (AAU) executed the three rounds of surveys over an 18-month period beginning March 1994.
13
In the case of herding, irrespective of age and sex, child herders work 7-8 hours per day six days per week.
14
All children who attend school have school as their main activity and vice versa.
12
9
Table 2 provides a brief summary of types of activities performed by children in rural Ethiopia.
Table 2: Primary work activities of children
Boys
6-10 11-15 Total
Fetching wood/water 44.7 25.8
35.6
Herding 19.7 55.8
37.2
Farm work
2.7 10.4
6.4
Domestic work 18.9
2.1
10.7
Minding children 12.2
4.8
8.6
Family business work
1.2
.5
.9
Other
.7
.6
.6
Total 100.0 100.0 100.0
No. of observations (599) (565) (1164)
Total
32.1
19.5
23.6
16.1
1.7
5.6
1.2
100.0
(763)
Girls
11-15
17.9
30.1
44.0
2.1
.8
4.0
1.1
100.0
(375)
6-10
45.9
9.3
3.9
29.6
2.6
7.2
1.7
100.0
(388)
All
6-10
45.2
15.6
3.1
23.1
8.4
3.5
1.0
100.0
(987)
children
11-15 Total
22.7
34.2
45.5
30.2
23.8
13.2
2.1
12.9
3.2
5.9
1.9
2.8
.7
1.2
100.0 100.0
(940) (1927)
Existing studies on schooling in rural Ethiopia suggest that the income opportunities
provided by (opportunity cost of) child labour constitute a major, perhaps the principal, reason
for low school enrolment 15 . Indeed, within our sample of rural households, work was cited as the
primary reason for non-attendance in 35% of all responses and over half the responses concerning children aged 11 to 15 (Table 3). If we ignore children considered too young to attend
school, work-related reasons are cited in over half of all responses. The sex difference in children's work activities emerges clearly with boys required for farm activities and girls for other
household activities, presumably domestic work.
Table 3: Primary reason for not attending school by age group and sex
(Percent of all primary reasons given for non-attending children in sex and age group indicated)
Ages 6-10
Ages 11-15
All ages
Girls Boys Total Girls Boys Total Girls Boys Total
Required for farm activities
6
19
12
13
45
26
9
27
17
Required for other hh activities
16
7
11
40
8
27
25
7
17
Required to care for sick/elderly
0
1
1
1
0
0
0
1
1
Required to work for wages
0
0
0
1
1
1
0
0
0
All work-related reasons
22
26
24
54
54
54
34
36
35
- excluding "too young"
49
56
52
59
59
59
55
57
56
Too young
55
53
54
7
9
8
37
38
37
Too expensive
11
10
10
19
22
20
14
14
14
School availability
6
5
6
11
8
10
8
6
7
Other reasons
6
6
6
8
7
8
7
6
6
Total
100
100
100
100 100
100 100
100
100
Characterization of children according to their main activity and their household income
and asset profiles illustrates the contrasting productivity and income effects discussed in the
introduction (Table 4). That school-going children come from the highest income households is
no surprise. These households' wealth appears to be built particularly on the ownership of
15
See Rose and Al-Samarrai (1997), USAID (1994) and Weir (1997).
10
oxen/bulls, cows and tools, which are basically labour-saving assets not directly involving
children. Working children come from significantly lower income households, yet these are the
households with the highest ownership of land and small livestock, two (complementary) assets
that we could expect to increase the demand for (and returns to) child labour. Finally, inactive
children are clearly a distinct group characterised by having the lowest income levels and asset
ownership of all. These households have neither the income to send their children to school, nor
the productive assets to create income opportunities.
Table 4: Asset ownership and income by child's main activity
All kids
Boys
Girls
Work School Inactive Work School Inactive Work School Inactive
Land (ha)
2.42
1.63
1.66
2.43
1.66
1.51
2.42
1.57
1.81
Small livestock (#)
4.53
3.25
2.87
4.81
3.62
2.61
4.30
2.67
3.13
Oxen/bulls (#)
1.55
1.56
1.23
1.61
1.54
1.23
1.49
1.60
1.24
Cows (#)
2.63
3.21
1.96
2.77
3.16
1.86
2.51
3.28
2.06
Ploughs (#)
1.10
1.29
0.95
1.12
1.27
0.91
1.09
1.32
0.99
Hoes (#)
0.92
0.99
0.81
0.93
1.04
0.81
0.91
0.92
0.80
Sickles (#)
1.04
1.03
0.75
1.09
1.11
0.68
1.00
0.91
0.82
Income*
56.40
70.74
50.71 57.82
66.83
46.59
55.2
76.88
54.72
Observations
(1441)
(442)
(565)
(656)
(270)
(278)
(785)
(172)
(287)
*Income measured by real food expenditure per adult equivalent.
3. EMPIRICAL MODELLING AND RESULTS
3.1 The empirical model
Like participation decisions, main activity decisions are discrete choices, reflecting
underlying latent variables. The latent variable in the case of the main activity decision is the
unobserved level of participation. Three main activities - work, leisure or inactivity ("leisure") are considered, giving rise to a polychotomous choice framework. As these choices are mutually
exclusive - there can only be one main activity - a bivariate probit analysis of school/work
decisions, akin to Canagarajah and Coulombe (1997) or Nielsen (1998), is ruled out. Grootaert
and Patrinos (1998) assume that the participation decision-making process is sequential with the
decision to participate in schooling preceding the decision to work, although there does not
appear to be any strong reason to do so 16 .
A multinomial logit approach is a straightforward solution. The classic application of this
approach by Schmidt and Strauss (1975) considers a similar issue: the prediction of an individual's occupation. We define the probability of a child having main activity j (j=1 (work); 2
(school); 3 (inactive)) as:
a + ß jX
e j
Pj =
a + ß kX k ; j, k = 1,2,3
∑k e k
16
Unlike Grootaert and Patrinos' participation analysis, our concern with the choice of main activities implies that
the work decision branch would stem only from those not attending school.
11
We normalise α 3 =0 and β 3 =0 and take the logs of the relative probabilities with respect to P3 17 :
ln(P1 /P3 )=α 1 +β1 X
ln(P2 /P3 )=α 2 +β2 X
where X is a vector of i characteristics of each child. If a given β 1i is positive, this indicates that
the relative probability of observing work as the child's main activity, rather than inactivity,
increases with the value of Xi. This need not imply that the marginal effect of variable Xi is to
increase the absolute probability of observing work as the child's main activity; its impact on the
probability of school as main activity may be positive and overwhelming. In other words, the
sign of the β coefficients are not necessarily equal to those of the marginal effects and they must
be interpreted with care. The marginal effects in this model can be estimated according to the
following formula: M.E. = δPj/δXi = Pj(β ji-Σk Pkβ ki) k≠j
where Pj and the Pk can be evaluated at different values, typically the mean, of vector X As the
relationship between the regressors and the absolute probabilities is non-linear, marginal effects
vary according to the choice of vector X and, consequently, they will vary between individuals.
The principal explanatory variables that interest us are those affecting child labour
productivity and household income, in particular productive asset ownership and household
composition (Table 5, in bold). In rural Ethiopia, land and livestock are the principal physical
assets. Boys and girls are studied separately for several reasons. First, participation rates in the
three types of main activities vary significantly between boys and girls (Table 1). Second, boys
and girls participate in different types of labour (Table 2) and thus the impacts of access to
specific physical assets can be expected to differ. Finally, pooling restrictions are not respected 18 .
We first examine the marginal effects of each explanatory variable on the probability of a
child performing each of the three main activities. Then, in attempt to evaluate the magnitude of
these effects, we calculate elasticities and the change in these probabilities resulting from a onestandard deviation increase in the value of each explanatory variable.
3.2 Marginal effects
Using the regression coefficients from the boy and girl regressions (reported in Appendix
A), we calculate the marginal effect of each dependent variable on the probability of a "mean" or
"average" child performing each of the three main activities. As mentioned above, given the nonlinear nature of the multinomial logit model, the marginal effect varies according to its point of
evaluation. Through experimentation, it becomes clear that marginal effects vary strongly with
age. Consequently, for each sex, we calculate marginal effects at three different values: the mean
values of the regressors for all children (aged 6 to 15), for young children (aged 6 to 10) and for
older children (aged 11 to 15)19 .
17
See Greene (1997) and Maddala (1983) for introductions to the multinomial logit model.
A variant of the dummy variable approach presented in Gujarati (1970aa) and Gujarati (1970bb) is used. A first
regression is run on the pooled data with an interactive boy dummy term added for each regressor and the constant.
This unrestricted regression allows the intercept and slope coefficients to differ by sex. A second regression is run
on the same pooled data without interactive dummies, restricting the coefficients to be equal. A likelihood ratio test
is used to test the restrictions. It yields a chi-squared of 144.26 with 78 degrees of freedom, which leads to the
rejection of pooling at the 0.1% confidence level.
19
As many of the regressors are closely related to the age of the child – e.g. the number of younger and older
siblings – we use the mean values of all regressors for each age group and not simply the mean age. There was an
18
12
Table 5: Definition of variables
Variable
mainact
ln(age)
dumkid
# infants
# females
# males
# elderly
dep ratio
# younger boys
# younger girls
# older boys
# older girls
female head
age of head
yrs. school head
land owned
land fertility
land slope
perm crop
# small animals
# bull/oxen
# cows/calves
# hoes
# ploughs
# sickles
Minutes to water
Definition
Main activity of child: Work(1); School(2); Inactive(3) – dependent variable
Log of age (in years)
dummy; =1 if child of the household head
Number of infants (aged 0 to 4) in the household
Number of female adults (aged 16 to 59) in the household
Number of male adults (aged 16 to 59) in the household
Number of elderly (aged 60 and over) in the household
Dependency ratio: (# of infants, children and elderly)/(# of adults)
Number of younger boys (aged 4 to 15) in the household
Number of younger girls (aged 4 to 15) in the household
Number of older boys (aged 4 to 15) in the household
Number of older girls (aged 4 to 15) in the household
Dummy; =1 if household head is female
Age (in years) of household head
Years of formal schooling of household head
Hectares of land owned by household
Land fertility index; 1=infertile to 3=fertile
Land slope index; 1=flat to 3=steep
Dummy; =1 if household owns permanent crop plants (chat, enset, eucalyptus..)
Number of small animals (goats, sheep..) owned by the household
Number of bulls and oxen owned by the household
Number of cows and calves owned by the household
Number of hoes owned by the household
Number of ploughs owned by the household
Number of sickles owned by the household
Number of minutes to walk to nearest source of water
insufficient number of observations for carrying out age group-specific regressions for each sex. We also tried
including an age group dummy in interaction with various regressors but found that none of these interactive terms
was significant. The non-linear nature of the model and the age variable capture most of the age-specific differences.
13
Table 6: Descriptive statistics
BOYS
GIRLS
All
Aged 6-10 Aged 11-15
All
Aged 6-10 Aged 11-15
Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D.
ln(age)
2.29 0.29 2.07 0.18 2.57 0.11 2.29 0.29 2.07 0.18 2.57
0.11
0.83 0.38 0.86 0.35 0.79 0.41 0.83 0.38 0.84 0.37 0.82
0.39
dumkid
# infants
0.74 0.82 0.79 0.82 0.67 0.82 0.72 0.76 0.80 0.76 0.62
0.74
# females
1.57 0.96 1.57 0.93 1.56 1.00 1.56 0.98 1.54 0.97 1.59
0.99
# males
1.48 1.01 1.43 0.96 1.53 1.07 1.44 0.99 1.41 0.97 1.49
1.02
# elderly
0.33 0.54 0.30 0.52 0.36 0.57 0.34 0.56 0.32 0.54 0.37
0.59
dep ratio
1.88 1.14 1.91 1.18 1.84 1.09 1.90 1.20 1.96 1.21 1.81
1.18
# younger boys
0.75 0.93 0.50 0.70 1.07 1.09 0.81 0.98 0.55 0.79 1.15
1.09
# younger girls
0.79 0.93 0.54 0.72 1.12 1.07 0.76 0.92 0.53 0.74 1.05
1.04
0.52 0.79 0.76 0.90 0.21 0.46 0.52 0.82 0.77 0.94 0.20
0.48
# older boys
0.58 0.80 0.84 0.90 0.24 0.46 0.52 0.78 0.79 0.89 0.17
0.40
# older girls
0.17 0.37 0.16 0.36 0.18 0.38 0.17 0.38 0.17 0.38 0.17
0.38
female head
47.41 13.22 46.61 12.87 48.46 13.60 47.95 13.26 46.77 13.07 49.47 13.37
age of head
1.44 2.71 1.42 2.64 1.46 2.80 1.40 2.58 1.47 2.60 1.31
2.55
yrs. school head
2.04 2.71 2.00 2.21 2.10 3.24 2.16 3.88 2.10 4.31 2.24
3.25
land owned
1.72 0.67 1.74 0.66 1.70 0.67 1.75 0.65 1.72 0.65 1.78
0.66
land fertility
2.65
0.48
2.66
0.46
2.65
0.50
2.64
0.50
2.64
0.49
2.64
0.51
land slope
0.64 0.48 0.63 0.48 0.65 0.48 0.65 0.48 0.67 0.47 0.63
0.48
perm crop
4.04 6.95 3.90 6.69 4.22 7.26 3.81 6.96 3.69 6.87 3.95
7.07
# small animals
1.51 1.61 1.42 1.50 1.61 1.74 1.45 1.55 1.37 1.48 1.55
1.64
# bull/oxen
2.64
3.18
2.59
3.26
2.71
3.07
2.52
2.94
2.40
2.99
2.66
2.87
# cows/calves
0.93 1.05 0.92 1.06 0.94 1.04 0.89 0.96 0.85 0.91 0.93
1.02
# hoes
1.10 1.50 0.99 0.94 1.25 1.99 1.10 1.32 1.04 1.15 1.17
1.52
# ploughs
1.00 1.60 0.95 1.55 1.07 1.67 0.94 1.50 0.85 1.38 1.07
1.63
# sickles
18.54
20.26
18.72
20.99
18.31
19.30
17.53
15.43
17.27
15.08
17.87
15.87
Minutes to water
#Observations
1204
700
544
1244
678
526
We present marginal effects for the probability of a child working (Table 7), attending
school (Table 8) and being inactive (Table 9). We look at the influence of individual characteristics, household composition and characteristics of the household head, before focusing on the
role of productive asset ownership. Site dummies (coefficients not shown) are included for the
15 sample sites and are highly significant in several cases, indicating community effects such as
school availability perhaps.
Work and school participation both increase rapidly with age among younger children.
Direct offspring of the household are significantly more likely to attend school and, particularly
among older children, less likely to work. As we saw in Table 2, many children mind infants.
Our results indicate that the number of infants (aged 4 and under) in the household significantly
increases the likelihood of a child, particularly a young child, working. Infant minding comes
partly at the cost of child school participation.
14
Table 7: Marginal effects on the probability of having work as main activity.
All
ME
ln(age)
dumkid
# infants
# females
# males
# elderly
dep ratio
# younger boys
# younger girls
# older boys
# older girls
female head
age of head
yrs. school head
land owned
land fertility
land slope
perm crop
# small animals
# bull/oxen
# cows/calves
# hoes
# ploughs
# sickles
Minutes to water
#Obs
0.230
-0.104
0.041
0.000
0.001
-0.048
-0.004
-0.021
0.003
-0.009
-0.013
-0.057
0.001
-0.023
0.004
-0.040
0.035
0.060
0.001
-0.031
0.002
0.017
-0.035
-0.011
0.001
1204
Sig
(1%)
(2%)
(5%)
(99%)
(98%)
(23%)
(86%)
(32%)
(90%)
(72%)
(60%)
(22%)
(72%)
(0%)
(57%)
(17%)
(34%)
(31%)
(62%)
(4%)
(71%)
(32%)
(2%)
(41%)
(23%)
BOYS
Aged 6-10
ME
Sig
0.966
-0.019
0.045
0.009
0.025
-0.070
0.034
-0.009
0.031
-0.060
0.000
-0.038
0.001
-0.016
0.008
-0.026
-0.003
0.119
0.004
-0.043
0.001
0.044
-0.027
0.011
0.001
700
(0%)
(74%)
(12%)
(77%)
(41%)
(17%)
(35%)
(78%)
(30%)
(4%)
(99%)
(52%)
(66%)
(10%)
(55%)
(48%)
(96%)
(9%)
(32%)
(4%)
(95%)
(9%)
(19%)
(56%)
(47%)
Aged 11-15
ME
Sig
-0.117
-0.155
0.042
-0.005
-0.012
-0.041
-0.024
-0.028
-0.011
0.016
-0.021
-0.071
0.001
-0.029
0.003
-0.050
0.056
0.036
0.000
-0.028
0.003
0.006
-0.043
-0.024
0.001
544
(25%)
(0%)
(7%)
(85%)
(63%)
(36%)
(40%)
(20%)
(60%)
(58%)
(47%)
(18%)
(79%)
(0%)
(70%)
(13%)
(17%)
(60%)
(97%)
(10%)
(61%)
(77%)
(1%)
(11%)
(20%)
Sig
GIRLS
Aged 6-10
ME
Sig
0.416
(0%)
-0.044 (16%)
0.031
(7%)
0.001 (95%)
-0.006 (70%)
0.043 (15%)
-0.022 (22%)
0.025 (10%)
0.010 (51%)
-0.001 (94%)
0.006 (75%)
-0.037 (26%)
-0.003
(3%)
-0.017
(0%)
0.008 (26%)
-0.043
(4%)
0.056
(4%)
0.035 (37%)
0.004 (13%)
-0.010 (39%)
-0.002 (70%)
0.009 (49%)
-0.007 (52%)
0.009 (40%)
0.001 (35%)
1244
1.185
(0%)
-0.036
(52%)
0.050
(10%)
0.004
(89%)
-0.006
(83%)
0.035
(49%)
-0.011
(75%)
0.056
(6%)
0.028
(37%)
-0.018
(50%)
-0.014
(63%)
-0.056
(34%)
-0.002
(35%)
-0.013
(15%)
0.004
(50%)
-0.024
(52%)
0.045
(32%)
0.000 (100%)
0.006
(16%)
-0.017
(44%)
-0.002
(80%)
0.027
(30%)
-0.005
(82%)
0.010
(60%)
0.000
(77%)
678
All
ME
Aged 11-15
ME
Sig
0.063
-0.054
0.024
-0.001
-0.007
0.052
-0.030
0.011
0.003
0.007
0.017
-0.031
-0.003
-0.020
0.011
-0.057
0.068
0.057
0.003
-0.008
-0.002
0.001
-0.008
0.009
0.002
526
(32%)
(9%)
(14%)
(97%)
(66%)
(9%)
(6%)
(41%)
(86%)
(69%)
(40%)
(34%)
(1%)
(0%)
(25%)
(1%)
(2%)
(14%)
(24%)
(48%)
(67%)
(95%)
(34%)
(36%)
(10%)
15
Table 8: Marginal effects on the probability of child attending school.
All
ME
ln(age)
dumkid
# infants
# females
# males
# elderly
dep ratio
# younger boys
# younger girls
# older boys
# older girls
female head
age of head
yrs. school head
land owned
land fertility
land slope
perm crop
# small animals
# bull/oxen
# cows/calves
# hoes
# ploughs
# sickles
Minutes to water
#Obs
0.281
0.138
-0.030
0.006
0.015
0.024
0.028
0.024
0.016
-0.025
0.019
0.058
0.000
0.023
-0.002
0.040
-0.051
-0.011
0.001
0.018
-0.003
0.003
0.034
0.024
-0.001
1204
Sig
(0%)
(0%)
(11%)
(77%)
(44%)
(50%)
(22%)
(17%)
(36%)
(27%)
(42%)
(18%)
(85%)
(0%)
(82%)
(13%)
(12%)
(85%)
(79%)
(21%)
(58%)
(86%)
(1%)
(5%)
(21%)
BOYS
Aged 6-10
ME
Sig
0.343
0.097
-0.015
0.006
0.015
0.007
0.025
0.016
0.016
-0.027
0.014
0.036
0.000
0.015
0.000
0.025
-0.037
0.009
0.001
0.006
-0.002
0.008
0.020
0.019
-0.001
700
(0%)
(0%)
(24%)
(69%)
(28%)
(76%)
(11%)
(20%)
(20%)
(8%)
(37%)
(22%)
(94%)
(0%)
(99%)
(16%)
(9%)
(80%)
(55%)
(50%)
(59%)
(44%)
(1%)
(3%)
(25%)
Aged 11-15
ME
Sig
0.239
0.164
-0.040
0.006
0.016
0.036
0.029
0.029
0.016
-0.024
0.023
0.072
0.000
0.029
-0.003
0.050
-0.060
-0.024
0.000
0.025
-0.004
-0.001
0.042
0.027
-0.001
544
(2%)
(0%)
(8%)
(81%)
(53%)
(43%)
(30%)
(18%)
(47%)
(40%)
(44%)
(18%)
(82%)
(0%)
(76%)
(13%)
(14%)
(73%)
(90%)
(15%)
(59%)
(96%)
(1%)
(8%)
(21%)
All
ME
Sig
0.076 (10%)
0.041
(9%)
-0.015 (24%)
0.001 (93%)
0.005 (67%)
-0.040 (10%)
0.024
(5%)
-0.003 (74%)
0.001 (94%)
-0.008 (57%)
-0.016 (31%)
0.020 (43%)
0.003
(1%)
0.016
(0%)
-0.009 (25%)
0.045
(1%)
-0.052
(2%)
-0.047 (10%)
-0.002 (35%)
0.005 (58%)
0.001 (68%)
0.002 (80%)
0.006 (33%)
-0.006 (40%)
-0.001
(7%)
1244
GIRLS
Aged 6-10
ME
Sig
0.130
0.025
-0.007
0.001
0.003
-0.024
0.015
0.001
0.002
-0.007
-0.011
0.009
0.002
0.010
-0.006
0.028
-0.031
-0.032
-0.001
0.002
0.001
0.003
0.004
-0.003
-0.001
678
(0%)
(12%)
(43%)
(90%)
(71%)
(12%)
(6%)
(84%)
(73%)
(47%)
(26%)
(56%)
(2%)
(0%)
(25%)
(1%)
(3%)
(10%)
(51%)
(72%)
(72%)
(57%)
(37%)
(48%)
(6%)
Aged 11-15
ME
Sig
0.046 (46%)
0.053
(9%)
-0.021 (20%)
0.001 (95%)
0.006 (67%)
-0.051
(9%)
0.031
(5%)
-0.007 (62%)
0.000 (100%)
-0.009 (61%)
-0.019 (34%)
0.027 (40%)
0.003
(1%)
0.020
(0%)
-0.011 (25%)
0.058
(1%)
-0.067
(2%)
-0.060 (11%)
-0.003 (31%)
0.006 (55%)
0.002 (68%)
0.002 (88%)
0.008 (33%)
-0.008 (39%)
-0.002
(8%)
526
16
Table 9: Marginal effects on the probability of child being inactive.
All
ME
ln(age)
dumkid
# infants
# females
# males
# elderly
dep ratio
# younger boys
# younger girls
# older boys
# older girls
female head
age of head
yrs. school head
land owned
land fertility
land slope
perm crop
# small animals
# bull/oxen
# cows/calves
# hoes
# ploughs
# sickles
Minutes to water
#Obs
-0.511
-0.033
-0.011
-0.006
-0.016
0.024
-0.023
-0.003
-0.018
0.034
-0.006
-0.001
0.000
0.000
-0.003
-0.001
0.017
-0.049
-0.002
0.014
0.001
-0.020
0.002
-0.012
0.000
1204
Sig
(0%)
(15%)
(38%)
(66%)
(23%)
(27%)
(13%)
(81%)
(16%)
(0%)
(60%)
(97%)
(72%)
(98%)
(62%)
(97%)
(39%)
(9%)
(27%)
(12%)
(89%)
(7%)
(83%)
(14%)
(92%)
BOYS
Aged 6-10
ME
Sig
-1.310
-0.078
-0.030
-0.015
-0.040
0.063
-0.059
-0.007
-0.047
0.087
-0.014
0.001
-0.001
0.001
-0.007
0.001
0.040
-0.128
-0.005
0.037
0.001
-0.052
0.007
-0.030
0.000
700
Aged 11-15
ME
Sig
(0%) -0.122
(19%) -0.009
(35%) -0.002
(66%) -0.001
(24%) -0.004
(26%)
0.005
(14%) -0.006
(84%) -0.001
(16%) -0.004
(0%)
0.008
(63%) -0.002
(98%) -0.001
(71%)
0.000
(91%)
0.000
(62%) -0.001
(99%)
0.000
(42%)
0.004
(8%) -0.012
(27%) -0.0005
(11%)
0.003
(90%)
0.000
(6%) -0.005
(75%)
0.000
(16%) -0.003
(88%)
0.000
544
(0%)
(13%)
(43%)
(65%)
(22%)
(29%)
(13%)
(77%)
(15%)
(1%)
(58%)
(92%)
(73%)
(84%)
(63%)
(91%)
(36%)
(10%)
(28%)
(15%)
(87%)
(8%)
(92%)
(14%)
(98%)
All
ME
Sig
-0.492
(0%)
0.003
(88%)
-0.016
(18%)
-0.002
(88%)
0.001
(91%)
-0.004
(86%)
-0.002
(88%)
-0.021
(7%)
-0.011
(36%)
0.009
(36%)
0.010
(38%)
0.017
(45%)
0.000
(90%)
0.001
(75%)
0.001
(63%)
-0.002
(87%)
-0.004
(80%)
0.013
(63%)
-0.002
(26%)
0.006
(52%)
0.001
(86%)
-0.011
(27%)
0.000
(98%)
-0.002
(75%)
0.001
(33%)
1244
GIRLS
Aged 6-10
ME
Sig
-1.315
0.011
-0.043
-0.005
0.004
-0.011
-0.004
-0.057
-0.030
0.025
0.025
0.047
0.000
0.004
0.002
-0.004
-0.014
0.032
-0.005
0.015
0.002
-0.030
0.001
-0.007
0.001
678
(0%)
(85%)
(18%)
(88%)
(91%)
(83%)
(91%)
(7%)
(36%)
(37%)
(39%)
(44%)
(85%)
(69%)
(68%)
(91%)
(77%)
(65%)
(25%)
(51%)
(86%)
(27%)
(97%)
(74%)
(34%)
Aged 11-15
ME
Sig
-0.109
(0%)
0.001 (91%)
-0.003 (21%)
0.000 (87%)
0.000 (92%)
-0.001 (89%)
-0.001 (85%)
-0.005
(8%)
-0.003 (36%)
0.002 (36%)
0.002 (37%)
0.004 (46%)
0.000 (95%)
0.000 (83%)
0.000 (58%)
-0.001 (81%)
-0.001 (86%)
0.003 (60%)
-0.0004 (28%)
0.001 (53%)
0.000 (87%)
-0.003 (28%)
0.000 (100%)
-0.001 (76%)
0.0001 (31%)
526
17
There is little indication of labour complementarity or substitution between children and
adults as the numbers of males and females have no significant effect on child time use. This is
congruent with an age-based labour division where children specialise in activities such as
fetching wood/water and herding (see Table 2). The number of elderly household members
(aged 60 and over) increases the likelihood of girls working and reduces the probability of them
attending school. This suggests that girls, particularly older girls, are required to take care of
elderly household members. The results are just the contrary, although less significant, for boys.
The number and sex of other children in the household have complex effects on child
time use. The likelihood of boys working generally decreases with the number of other boys in
the household, indicating labour substitution. Whereas this results in higher probability of school
attendance when the other boys are younger, it results in higher inactivity when they are older,
suggesting a birth order effect on schooling in favour of first-borns. The effects of the number of
girls on the time use of boys are weaker. There is some evidence of a gender bias as the school
attendance increases with the number of girls in the house. The counterpart of this last effect is a
higher likelihood of girls, particularly young girls, working as the number of younger boys
increases. This greater workload applies primarily to girls who would otherwise be inactive as
the number of boys in the household does not affect the probability of attending school. Note that
the importance of these household composition effects constitutes a validation of our missing
child labour market hypothesis (Benjamin (1992)).
Children, particularly boys, are more likely to attend school and less likely to work in
female-headed households . This may be due to a lack of income-earning opportunities as
suggested by the higher but statistically insignificant likelihood of inactivity among girls, but that
should be captured by the asset variables. It may also reflect different gender attitudes 20 .
The likelihood of a girl attending school increases with the head's age while her likelihood of working falls, although no effect is noted for boys. Finally, the education of the head
significantly increases the probability of a child attending school and reduces the likelihood of
him/her working. This may reflect different attitudes of educated heads or unobserved hous ehold-specific variables affecting, for example, the returns to schooling.
Now let us turn our attention to the impact of productive assets on child time use decisions. As discussed in the theoretical section, household income and the productivity of children
(and adults) both tend to increase as access to productive assets increases. While the income
effect will tend to reduce child labour in favour of school and leisure activities, the productivity
effect will tend to increase child labour. The importance of the income effect will depend on the
income contribution of the asset itself and on the income elasticity of child time use decisions.
The productivity effect, for its part, will depend on the degree of complementarity, or substitutability, between child labour and the specific asset. Assets used in activities traditionally performed by children, such as herding, are expected to have a stronger productivity effect on child
labour than assets used in adult labour-intensive activities, such as farming. In our analysis, we
define assets broadly as all non-labour household production factors. Consequently, we include
variables such as land fertility and slope and the proximity to a source of water.
20
Appleton, Chessa et al. (1999) find that women have a stronger gender bias in favour of boys than men in Uganda.
18
Most of the asset variables have a significant (10% or 15% level) impact on at least one
age and sex group. This requires strong dominance of one of the two contrasting effects: income
and productivity. Indeed, our results indicate that child time use is sensitive to both income
opportunities and poverty constraints, although the effects vary significantly according to the
specific asset involved. Note that insignificant coefficients do not necessarily imply the absence
of income and productivity effects as they may simply offset each other so that the net effect is
insignificant. To examine this proposition, we would need to break down the two effects, which
is beyond the scope of our analysis.
Land ownership appears to increase the likelihood of a child working, principally among
older girls and young children, although these results are not significant. This may be due to the
importance of land-intensive herding activities among children. Note that past land redistribution
policies have significantly reduced intra-site variations in landholdings, which likely explains the
low significance of our results. When the 15 site dummies are not included in the regression,
land ownership has a strong and highly significant positive effect on child labour participation.
Without a larger sample of villages, we cannot clearly separate the respective role of land
ownership and unobserved site characteristics. However, in the sites with the largest landholdings (around Debre Birhan), child labour participation is very high and herding is the primary
child occupation. It is noteworthy that the increased workload for girls comes primarily at the
cost of schooling whereas this is not the case for boys. On the other hand, land quality, which
increases with land fertility and falls with land slope , reduces child labour. As high quality,
fertile and flat land is conducive to farming, which primarily draws on adult labour, rather than
herding, it is reasonable to expect its positive income effect to dominate.
Children, particularly young boys and older girls, in households with permanent crops
are more likely to work. This result is somewhat surprising as permanent crops are not particularly labour intensive compared to farming, although the labour intensity of farming applies
primarily to adults. This variable may also be proxying some unobserved site variables as
permanent crop ownership is concentrated in a few specific sites out of the 15. As in the case of
land and small animal ownership, the child labour-increasing effect of permanent crop ownership
appears to conflict with schooling for girls but not for boys.
The extent and composition of livestock ownership has markedly contrasting effects on
child time use. Children, particularly girls and young children, are more likely to work as the
number of small animals owned by the household increases. In a separate regression, we found
a negative, but insignificant, quadratic effect suggesting that at a certain level of household
income, the pull of income opportunities becomes less important. Note also that it appears that
households sacrifice the schooling of girls, especially older girls, to take advantage of these
opportunities. Whereas, in the case of boys, the increased labour comes only at the cost of
reduced inactivity, primarily among younger boys.
Bulls and oxen are important stores of wealth in the context of rural Ethiopia. They are
also used mostly by male adults for ploughing. Indeed, we find that the income effect of
bull/oxen ownership dominates the child productivity effect generating a negative net effect.
This marginal effect is particularly significant in the case of boys. Note that this reduced work19
load translates primarily into increased schooling of older children and increased inactivity
among younger children. Ownership of cows and calves has ambiguous and insignificant
effects, suggesting either that the underlying income and productivity effects are weak or that
they cancel each other out. Given the importance of herding and the importance of cows as
sources and stores of income, the latter possibility seems likely.
Among the three farm tools analysed – hoes, ploughs and sickles – the child labour
effects of ploughs closely mirrors that of bull/oxen as these two assets are used together, by male
adults, and both are important indicators of wealth. However, whereas the reduced child workload associated with bull/oxen ownership translated into increased schooling for older children
and increased inactivity for young children, in the case of ploughs it translates almost completely
into increased schooling for all children. These effects on schooling are very significant (1%
level) in the case of boys and suggest that ploughs make a more significant income contribution
than do ploughs, possibly through renting. As the alternative to ploughing is hoeing and the latter
activity is less restricted to adults, it is not surprising to find that the coefficients on the number
of hoes are the inverse of those on ploughs. It is not clear why these effects are stronger for
young children. However, even in the case of older children, the workload does not seem to
compromise school attendance, perhaps because farm work, unlike herding for example, is not
time-intensive and can be adapted to the school day. The impact of sickle ownership is quite
complex and does not provide any clear results.
Finally, as fetching wood/water is the primary work activity of children in rural Ethiopia,
an important asset is a nearby source. Indeed, we find that the probability of a child working
increases with the distance (in minutes) to the nearest source of water21 . Furthermore, it
appears that this increased workload significantly conflicts with school participation.
In conclusion to our analysis of marginal effects and as a preliminary response to the
question posed in the title of this paper, it appears that households are sensitive to both income
opportunities (child productivity) and poverty constraints in their child time use decisions.
Ownership of some assets, such as bull/oxen and ploughs, shows the potential to simultaneously
reduce poverty and child labour and increase child schooling. However, assets that are used in
typical child labour activities, such as small animals, tend to increase child labour. Our results
also suggest that child labour conflicts with schooling primarily in the case of girls. In deciding
to send boys to school, factors affecting their labour productivity do not seem to have as much
importance as for girls.
3.3 Elasticities
The magnitude of the marginal effects depends on the unit used to measure the explanatory variable. If distance to water was expressed in hours instead of minutes, the marginal effect
would be 60 times bigger. More generally, the importance of variables with low mean values is
exaggerated by marginal effect measures alone. Consequently, while the signs of the marginal
effects are reliable, it is important to take into account the mean values of the explanatory
variables (Table 6) in order to assess their importance. Elasticities present an attractive alterna21
Mason and Khandker (1995, p.21-22) also find that distance to water reduces probability of enrolment, in
Tanzania.
20
tive as they represent the percentage change in the probability of a child performing a given
activity for a one- percent change in the value of each explanatory variable. Unlike marginal
effects, elasticities are independent of the measurement unit used for the dependent and explanatory variables: Elasticity = (δPj/δXi)(X/P) = M.E.(X/P)
In the next two tables, we present the elasticities of the probabilities of a child working and that
of child attending school with respect to each of the dependent variables 22 .
Productive assets are among the strongest determinants of child time use, although many
of these elasticities have high standard deviations. Land quality, as measured by land fertility and
the negative of land slope, is shown to strongly reduce the probability of children working (Table
10), presumably due to a strong income effect and perhaps a substitution toward male adult
labour 23 . Ownership of bull/oxen and ploughs also reduces child labour, principally among boys.
A 1% increase in bull/oxen ownership, for example, would lead to a 0.11% reduction in the
probability of a young boy working. Given that the average value for oxen/bull ownership in the
households of young children is 1.4, an additional ox in each household would represent a 70%
variation in ownership that would reduce the probability of a young boy working by 8%. Of
course, the choice of one ox as the amount of variation is arbitrary. In the following section, we
will look at the effect of an increase of one standard deviation in each of the dependent variables,
thus taking into account their observed sample variability. The dummy variables for children of
the head and permanent crops also have strong effects but we will discuss them in the following
section when we examine the effects of a change in their values from zero to one.
School attendance elasticities (Table 11) are generally much higher than child labour
elasticities, reflecting in part the low initial levels of school attendance. Once again, productive
asset variables, in particular land quality and the ownership of bull/oxen and ploughs are among
the most powerful determinants. A one percent improvement in land fertility or a one percent
reduction in land slope increases school attendance by between one-third of a percent for boys
and over 0.8 percent for girls. Distant sources of water significantly reduce school attendance,
especially in the case of girls. The impact of the dummy variables will be discussed in the
following section. Girl school attendance is also shown to be strongly, and positively, correlated
to the age of the head. Perhaps older heads have had time to accumulate sufficient wealth to
finance the education of girls, which, unlike the education of boys, is perceived as a luxury.
The general conclusion to emerge from our look at elasticities is that many productive
assets have powerful effects on child time use. Among these, it appears to be those that act
primarily through income effects, reducing child labour and increasing schooling, are the most
powerful. This suggests that child time use decisions between work, schooling and inactivity are
motivated more by poverty constraints then by the income opportunities from child labour.
22
Elasticities of the probability of a child being inactive are presented in Appendix B.
We will examine the impact of the three dummy variables – for children of the household head (dumkid), femaleheaded households and households with permanent crops – further on when we consider the variation in the
predicted probabilities with these dummies equal to zero and one.
23
21
Table 10: Elasticities of the probability of having work as main activity with respect to explanatory variables
BOYS
GIRLS
All
Aged 6-10
Aged 11-15
All
Aged 6-10
Aged 11-15
Elast
s.d. Elast
s.d. Elast
s.d. Elast
s.d. Elast
s.d. Elast
s.d.
ln(age)
dumkid
# infants
# females
# males
# elderly
dep ratio
# younger boys
# younger girls
# older boys
# older girls
female head
age of head
yrs. school head
land owned
land fertility
land slope
perm crop
# small animals
# bull/oxen
# cows/calves
# hoes
# ploughs
# sickles
Minutes to water
#Obs
0.75
-0.12
0.04
0.00
0.00
-0.02
-0.01
-0.02
0.00
-0.01
-0.01
-0.01
0.04
-0.05
0.01
-0.10
0.07
0.05
0.01
-0.07
0.01
0.02
-0.06
-0.02
0.03
1204
(0.28)
(0.05)
(0.02)
(0.05)
(0.05)
(0.02)
(0.07)
(0.02)
(0.02)
(0.02)
(0.02)
(0.01)
(0.11)
(0.01)
(0.02)
(0.07)
(0.07)
(0.05)
(0.02)
(0.03)
(0.02)
(0.02)
(0.02)
(0.02)
(0.02)
3.62
-0.03
0.06
0.03
0.07
-0.04
0.12
-0.01
0.03
-0.08
0.00
-0.01
0.08
-0.04
0.03
-0.08
-0.01
0.14
0.03
-0.11
0.00
0.07
-0.05
0.02
0.03
700
(0.40)
(0.09)
(0.04)
(0.09)
(0.08)
(0.03)
(0.12)
(0.03)
(0.03)
(0.04)
(0.04)
(0.02)
(0.17)
(0.02)
(0.05)
(0.12)
(0.11)
(0.08)
(0.03)
(0.05)
(0.05)
(0.04)
(0.04)
(0.03)
(0.04)
-0.43
-0.17
0.04
-0.01
-0.03
-0.02
-0.06
-0.04
-0.02
0.00
-0.01
-0.02
0.03
-0.06
0.01
-0.12
0.11
0.03
0.00
-0.06
0.01
0.01
-0.08
-0.04
0.04
544
(0.79)
(0.13)
(0.05)
(0.12)
(0.12)
(0.05)
(0.16)
(0.07)
(0.07)
(0.02)
(0.02)
(0.03)
(0.28)
(0.03)
(0.06)
(0.17)
(0.17)
(0.14)
(0.04)
(0.08)
(0.06)
(0.06)
(0.06)
(0.05)
(0.06)
1.17
-0.05
0.03
0.00
-0.01
0.02
-0.05
0.02
0.01
0.00
0.00
-0.01
-0.16
-0.03
0.02
-0.09
0.09
0.03
0.02
-0.02
-0.01
0.01
-0.01
0.01
0.02
1244
(0.18)
(0.03)
(0.01)
(0.03)
(0.03)
(0.01)
(0.04)
(0.02)
(0.01)
(0.01)
(0.01)
(0.01)
(0.07)
(0.01)
(0.02)
(0.05)
(0.04)
(0.03)
(0.01)
(0.02)
(0.02)
(0.01)
(0.01)
(0.01)
(0.02)
4.03
-0.05
0.07
0.01
-0.01
0.02
-0.04
0.05
0.02
-0.02
-0.02
-0.02
-0.15
-0.03
0.01
-0.07
0.10
0.00
0.04
-0.04
-0.01
0.04
-0.01
0.01
-0.01
678
(0.38)
(0.08)
(0.04)
(0.08)
(0.07)
(0.03)
(0.11)
(0.03)
(0.03)
(0.03)
(0.04)
(0.02)
(0.16)
(0.02)
(0.02)
(0.10)
(0.10)
(0.08)
(0.03)
(0.05)
(0.04)
(0.04)
(0.04)
(0.03)
(0.04)
0.19
-0.05
0.02
0.00
-0.01
0.02
-0.06
0.02
0.00
0.00
0.00
-0.01
-0.19
-0.03
0.03
-0.12
0.11
0.04
0.01
-0.01
-0.01
0.00
-0.01
0.01
0.03
526
(0.19)
(0.03)
(0.01)
(0.03)
(0.03)
(0.01)
(0.03)
(0.02)
(0.02)
(0.00)
(0.00)
(0.01)
(0.07)
(0.01)
(0.03)
(0.04)
(0.04)
(0.03)
(0.01)
(0.02)
(0.01)
(0.01)
(0.01)
(0.01)
(0.02)
Table 11: Elasticities of the probability of attending school with respect to explanatory variables
All
Elast
ln(age)
dumkid
# infants
# females
# males
# elderly
dep ratio
# younger boys
# younger girls
# older boys
# older girls
female head
age of head
yrs. school head
land owned
land fertility
land slope
perm crop
# small animals
# bull/oxen
# cows/calves
# hoes
# ploughs
# sickles
Minutes to water
#Obs
3.12
0.55
-0.11
0.05
0.11
0.04
0.25
0.09
0.06
-0.06
0.05
0.05
-0.07
0.16
-0.02
0.34
-0.34
-0.03
0.01
0.13
-0.04
0.01
0.18
0.11
-0.10
1204
s.d.
(0.84)
(0.17)
(0.07)
(0.16)
(0.14)
(0.06)
(0.21)
(0.06)
(0.07)
(0.06)
(0.07)
(0.03)
(0.36)
(0.04)
(0.07)
(0.22)
(0.22)
(0.18)
(0.05)
(0.10)
(0.07)
(0.07)
(0.06)
(0.06)
(0.08)
BOYS
Aged 6-10
Aged 11-15
Elast
s.d. Elast
s.d.
GIRLS
All
Aged 6-10
Aged 11-15
Elast
s.d. Elast
s.d. Elast
s.d.
5.77
0.67
-0.10
0.07
0.17
0.02
0.38
0.07
0.07
-0.17
0.10
0.05
-0.03
0.17
0.00
0.36
-0.41
0.05
0.03
0.07
-0.04
0.06
0.16
0.14
-0.10
700
1.84
0.36
-0.11
0.02
0.07
-0.14
0.48
-0.03
0.01
-0.04
-0.09
0.04
1.30
0.23
-0.20
0.83
-0.74
-0.33
-0.08
0.07
0.03
0.02
0.07
-0.06
-0.25
1244
(0.76)
(0.21)
(0.08)
(0.18)
(0.16)
(0.06)
(0.24)
(0.05)
(0.05)
(0.10)
(0.11)
(0.04)
(0.39)
(0.04)
(0.09)
(0.26)
(0.24)
(0.19)
(0.06)
(0.11)
(0.08)
(0.08)
(0.06)
(0.07)
(0.09)
2.23
0.47
-0.10
0.04
0.09
0.05
0.20
0.11
0.06
-0.02
0.02
0.05
-0.08
0.16
-0.02
0.31
-0.30
-0.06
0.01
0.15
-0.04
0.00
0.19
0.10
-0.09
544
(0.93)
(0.15)
(0.06)
(0.14)
(0.14)
(0.06)
(0.19)
(0.08)
(0.09)
(0.02)
(0.03)
(0.03)
(0.34)
(0.04)
(0.07)
(0.20)
(0.20)
(0.16)
(0.05)
(0.10)
(0.07)
(0.07)
(0.07)
(0.06)
(0.07)
(1.13)
(0.21)
(0.10)
(0.19)
(0.18)
(0.09)
(0.24)
(0.09)
(0.09)
(0.08)
(0.09)
(0.05)
(0.51)
(0.05)
(0.18)
(0.31)
(0.31)
(0.20)
(0.08)
(0.13)
(0.09)
(0.08)
(0.08)
(0.07)
(0.14)
4.63
0.36
-0.09
0.02
0.07
-0.13
0.51
0.01
0.02
-0.09
-0.16
0.03
1.27
0.24
-0.21
0.84
-0.74
-0.36
-0.05
0.04
0.03
0.05
0.07
-0.05
-0.28
678
(1.07)
(0.23)
(0.11)
(0.20)
(0.18)
(0.09)
(0.27)
(0.07)
(0.07)
(0.12)
(0.14)
(0.05)
(0.54)
(0.06)
(0.18)
(0.33)
(0.34)
(0.22)
(0.08)
(0.13)
(0.08)
(0.09)
(0.08)
(0.07)
(0.15)
0.94
0.35
-0.10
0.01
0.08
-0.15
0.44
-0.06
0.00
-0.01
-0.03
0.04
1.31
0.21
-0.20
0.82
-0.73
-0.30
-0.08
0.08
0.04
0.01
0.08
-0.07
-0.24
526
(1.26)
(0.20)
(0.08)
(0.19)
(0.18)
(0.09)
(0.22)
(0.13)
(0.12)
(0.03)
(0.03)
(0.04)
(0.51)
(0.05)
(0.18)
(0.30)
(0.31)
(0.19)
(0.08)
(0.13)
(0.09)
(0.09)
(0.08)
(0.08)
(0.14)
23
3.4 Variable dispersion and predicted probabilities
While independent of measurement units, elasticities do not take into account the sample
dispersion of the explanatory variables. For a given elasticity, a variable with more variation will
have a large effect on child time use. In terms of policy simulations, a variable's dispersion,
measured by its standard deviation, provides a guideline to the types of variations that are
feasible and realistic to analyse. In this section, we look at the change in the predicted probability
of a child performing an activity after a one-standard deviation increase in each explanatory
variable 24 . In the cases of dummy variables – child of head (dumkid), female head and permanent crops – we look at the change in probability when the dummy goes from zero to one.
The descriptive statistics presented in Table 6 indicate that the dispersion of the explanatory variables can be quite large. Variables with particularly high standard deviations, relative to
their respective means, include: the education of the household head; the numbers of small
animals, hectares of land and tools owned by the household; and, the numbers of siblings and
elderly. The land quality variables stand out for their limited dispersion.
Neglecting the dummy variables for the moment, when we take into account the sample
variability of the regressors, education of the household head consistently emerges among the
most important determinants of child time use (Table 12). A standard deviation increase in the
education of the household head, roughly equivalent to an additional year and a half of schooling, would reduce child labour participation and increase school participation by 3 to 9 per cent,
according to the child's age group and sex. Put differently, variation in the education of household heads in the sample generates a large part of the variation in child time use decisions. This
result may reflect different attitudes of educated heads or unobserved household-specific cha racteristics affecting, for example, the returns to schooling.
The ownership of oxen/bull and tools, particularly ploughs, is revealed to strongly reduce
work participation and increase school attendance among boys. Increases in land quality and
reductions in land area emerge as the strongest determinants of school attendance for girls. An
increase in the dependency ratio is also shown to raise school attendance among girls, and reduce
labour among older girls, perhaps as it implies a greater sharing of non-adult work tasks. Finally,
the strongest factors influencing the labour participation of young girls are quite different from
older girls. A one standard deviation increase in the number of younger boys, infants and small
animals are all shown to increase labour participation among young girls by roughly 4%. This
reflects a possible gender bias and the importance of girls' child minding and herding tasks.
Among the dummy variables, being the child of the household head strongly increases the
likelihood of attending school and not working. This is particularly true for older boys, among
whom sons of the head are 14.6% more likely to attend school and 13.7% less likely to work.
Among young boys, being the son of the head increases school attendance but does not affect
work participation. For girls, like for older boys, the trade-off is between school and work. This
apparent discrimination in favour of own children can be explained either by preferences or by a
reduced commitment problem with regards to the eventual returns to the schooling investment 25 .
24
25
Behrman and Knowles (2000) use this approach in studying the effect of income on child schooling in Vietnam.
See Baland and Robinson (1998) for a discussion, and model, of the commitment problem.
Table 12: Change in predicted probabilities following a one standard deviation increase in
each explanatory variable
BOYS
All
Work
6-10 11-15
GIRLS
School
All 6-10 11-15
ln(age)
-16.8 16.4
-6.2 26.1 11.1
# infants
3.2 3.7
3.3 -2.4 -1.2
# females
0.0 0.8
-0.5 0.6 0.5
# males
-0.1 2.3
-1.3 1.6 1.4
# elderly
-2.7 -3.7
-2.4 1.3 0.4
dep ratio
-0.9 3.4
-2.8 3.3 3.1
# younger boys
-2.0 -0.7
-3.1 2.3 1.2
# younger girls
0.1 2.1
-1.3 1.5 1.2
# older boys
-1.1 -5.9
0.7 -2.0 -2.4
# older girls
-1.1 0.0
-1.0 1.6 1.3
age of head
0.8 1.2
0.7 -0.4 -0.1
yrs. school head -6.9 -4.4
-8.8 7.1 4.4
land owned
1.2 1.6
1.1 -0.4 0.0
land fertility
-2.8 -1.8
-3.4 2.8 1.8
land slope
1.5 -0.2
2.7 -2.4 -1.6
# small animals
0.9 2.7
0.0 0.4 0.7
# bull/oxen
-5.3 -6.5
-5.1 2.9 0.9
# cows/calves
0.7 0.2
1.0 -0.9 -0.7
# hoes
1.7 4.4
0.5 0.2 0.8
# ploughs
-5.7 -2.6
-9.1 5.5 2.0
# sickles
-2.2 1.4
-4.2 4.0 3.1
Minutes to water
2.2 1.7
2.5 -2.1 -1.3
Dummies
dumkid
-8.3 -0.3 -13.7 11.6 7.7
female head
-6.0 -3.9
-7.4 6.2 3.9
perm crop
6.3 12.0
3.7 -1.0 1.1
#Obs
1204 700
544 1204 700
7.9
-3.1
0.6
1.7
2.1
3.3
3.3
1.7
-1.1
1.0
-0.6
8.9
-0.9
3.5
-2.9
0.3
4.5
-1.1
-0.1
9.1
4.6
-2.5
All
Work
6-10 11-15
4.9 24.4
2.2 3.7
0.1 0.4
-0.6 -0.6
2.2 1.8
-3.0 -1.5
2.3 4.3
0.9 2.0
-0.2 -1.8
0.4 -1.3
-4.1 -2.8
-5.0 -3.8
2.6 1.3
-3.2 -1.7
2.5 2.1
2.4 4.0
-1.6 -2.5
-0.6 -0.7
0.8 2.4
-0.9 -0.6
1.2 1.4
1.1 -0.8
14.6 -4.0
7.5 -3.9
-2.4 3.8
544 1244
-3.4
-5.7
0.2
678
0.3
1.7
-0.1
-0.7
2.8
-4.0
1.2
0.3
0.3
0.6
-5.1
-6.1
3.2
-4.3
3.1
2.0
-1.3
-0.5
0.1
-1.3
1.4
2.2
School
All 6-10 11-15
4.2
-1.1
0.1
0.5
-2.0
3.3
-0.4
0.1
-0.6
-1.2
4.0
4.9
-2.9
3.4
-2.3
-1.2
0.7
0.4
0.2
0.9
-0.9
-1.9
3.2
-0.5
0.1
0.3
-1.2
2.1
0.1
0.2
-0.6
-0.9
2.4
3.0
-2.0
2.1
-1.4
-0.6
0.3
0.2
0.3
0.5
-0.5
-1.3
1.3
-1.5
0.1
0.7
-2.7
4.1
-0.7
0.0
-0.4
-0.7
5.2
6.1
-3.3
4.3
-3.0
-1.7
1.1
0.5
0.2
1.3
-1.3
-2.4
-4.8 3.6
-3.3 2.1
6.0 -5.1
526 1244
2.2
1.0
-3.5
678
4.7
2.9
-6.3
526
25
Children from female-headed household are 3.3% to 7.4% less likely to work and 1% to
7.5% more likely to attend school. Effects are, once again, strongest for older boys. Female
heads may put a bigger weight on child welfare, particularly for sons 26 , or they may have fewer
opportunities to increase income through child labour.
Finally, the fact that a child's household owns permanent crops has dramatically different
effects according to the age and sex of the child. Whereas young boys experience a 12% increase
in their likelihood of working, the effect on young girls is negligible (0.2%). This increase in the
workload of young boys does not, however, come at the expense of schooling, as it does in the
case of other children. No clear explanation of these results can be made other than the possibility that this dummy is picking up some unobserved site characteristics given the concentration of
permanent crops in specific sites.
Conclusion
From a methodological point of view, we have seen that elasticities provide a clearer
image of the importance of a given determinant than do marginal effects. In particular, the
importance of variables with low mean values tends to be inflated by marginal effects measures
taken alone, whereas variables with high mean values see their importance understated by these
measures. We have also seen that some variables with strong elasticities show little sample
variability. As a consequence, their role in explaining the variation in child time use decisions is
limited. By examining the impacts of a one-standard deviation increase in the value of each
explanatory variable, we were able to identify the most important determinants of child time use.
In particular, the importance of the head's education emerged only through this last exercise.
In terms of our core concern – the determinants of child time use – we conclude that
household assets and other household production variables do indeed play an important role in
child time use decisions. In particular, our results suggest that poverty alleviation efforts should
aim to improve access to physical assets - such as bulls, oxen, ploughs, nearby sources of water
and land fertility - that increase household income without encouraging child labour.
We also find a tight relationship between child labour and school participation with
respect to physical assets. For practically all of the explanatory variables, the sign of the labour
and schooling effects are opposite and often of similar magnitude. Thus, it appears that policies
aimed at combating child labour will simultaneously encourage schooling and vice versa27 .
Subject to the limitations of our data and analysis, our response to the question posed in
the title of this paper is that child time use decisions are driven by both poverty constraints
(income effects) and income opportunities (substitution effects). However, there seems to be a
limited number of child labour-increasing physical assets – mainly small livestock and the
proximity of water – that is, assets for which the substitution effect dominates. If a policy objective is to reduce child labour and increase child schooling, care should be taken to encourage the
accumulation of those physical assets that both increase household income and encourage
substitution away from child labour.
26
27
See footnote 20.
See Ravallion and Wodon (1999) for a study arguing the contrary.
26
Appendix A: Regression results (coefficients)
Note: Inactivity is the reference outcome
1. Work
Multinomial regression
Log likelihood = -767.60419
Number of obs
LR chi2(78)
Prob > chi2
Pseudo R2
=
=
=
=
1244
710.09
0.0000
0.3163
-----------------------------------------------------------------------------mainact |
Coef.
Std. Err.
z
P>|z|
(95% Conf. Interval)
---------+-------------------------------------------------------------------ageln1 |
5.881329
.5016874
11.723
0.000
4.898039
6.864618
dumkid | -.0909836
.2629684
-0.346
0.729
-.6063922
.424425
infant1 |
.2123975
.1454101
1.461
0.144
-.072601
.4973961
female1 |
.02223
.1490874
0.149
0.881
-.269976
.3144359
male1 | -.0213104
.1379235
-0.155
0.877
-.2916355
.2490147
old1 |
.091996
.24222
0.380
0.704
-.3827465
.5667386
dr | -.0056562
.1631864
-0.035
0.972
-.3254958
.3141834
yboys |
.2635252
.1443863
1.825
0.068
-.0194668
.5465172
ygirls |
.1371161
.1501961
0.913
0.361
-.1572628
.431495
oboys | -.1030388
.1237292
-0.833
0.405
-.3455436
.139466
ogirls | -.0993289
.1355086
-0.733
0.464
-.3649209
.166263
femhead | -.2328776
.2758934
-0.844
0.399
-.7736188
.3078636
agehead | -.0044375
.0097694
-0.454
0.650
-.0235852
.0147103
headed | -.0329191
.0430983
-0.764
0.445
-.1173903
.0515522
landown | -.0001099
.0211343
-0.005
0.996
-.0415322
.0413125
landfrt | -.0261952
.1734009
-0.151
0.880
-.3660546
.3136643
landslp |
.1162418
.2120675
0.548
0.584
-.2994029
.5318865
dumpc | -.0957636
.3200424
-0.299
0.765
-.7230352
.5315079
smlvstk |
.0246892
.0196504
1.256
0.209
-.0138249
.0632033
ttoxbll | -.0732085
.104546
-0.700
0.484
-.278115
.131698
ttcows | -.0086533
.0417312
-0.207
0.836
-.090445
.0731384
hoes |
.1356629
.1251988
1.084
0.279
-.1097222
.3810481
plows | -.0102652
.0972046
-0.106
0.916
-.2007828
.1802524
sickls |
.0370238
.0910147
0.407
0.684
-.1413617
.2154094
waterds | -.0047069
.0064394
-0.731
0.465
-.0173279
.0079141
haresaw | -2.244177
.6107996
-3.674
0.000
-3.441322
-1.047031
geblen | -.9404056
.6571828
-1.431
0.152
-2.22846
.3476491
dinki | -1.288061
.6590641
-1.954
0.051
-2.579803
.0036811
yetemen | -1.008588
.6748399
-1.495
0.135
-2.33125
.3140737
shumsha | -.7741541
.5444586
-1.422
0.155
-1.841273
.2929651
sirbana | -1.315787
.5493764
-2.395
0.017
-2.392545
-.2390294
adele | -.8402021
.5283682
-1.590
0.112
-1.875785
.1953806
korod | -.6890704
.5498202
-1.253
0.210
-1.766698
.3885575
trirufe | -.7925375
.583544
-1.358
0.174
-1.936263
.3511877
imdibir | -2.217613
.6506357
-3.408
0.001
-3.492836
-.9423907
azedebo |
1.611558
.7050634
2.286
0.022
.2296588
2.993456
adado | -2.017654
.5976344
-3.376
0.001
-3.188996
-.8463124
garagod | -.9357937
.5386658
-1.737
0.082
-1.991559
.119972
doma | -1.105466
.6361984
-1.738
0.082
-2.352392
.1414596
_cons |
-10.7834
1.335616
-8.074
0.000
-13.40116
-8.165641
27
2. School
-----------------------------------------------------------------------------mainact |
Coef.
Std. Err.
z
P>|z|
(95% Conf. Interval)
---------+-------------------------------------------------------------------ageln1 |
6.173284
.6846779
9.016
0.000
4.83134
7.515228
dumkid |
.3976839
.3604843
1.103
0.270
-.3088523
1.10422
infant1 |
.0172827
.1920481
0.090
0.928
-.3591246
.39369
female1 |
.0309813
.1858597
0.167
0.868
-.3332971
.3952597
male1 |
.0381083
.1794876
0.212
0.832
-.3136809
.3898975
old1 | -.3794905
.3435573
-1.105
0.269
-1.052851
.2938695
dr |
.2739391
.1949344
1.405
0.160
-.1081252
.6560034
yboys |
.1961638
.1794031
1.093
0.274
-.1554597
.5477874
ygirls |
.1336816
.1850874
0.722
0.470
-.2290831
.4964462
oboys | -.1861506
.188089
-0.990
0.322
-.5547983
.1824972
ogirls |
-.27164
.209637
-1.296
0.195
-.682521
.1392411
femhead |
.0220476
.3729131
0.059
0.953
-.7088487
.7529438
agehead |
.0260263
.0141298
1.842
0.065
-.0016676
.0537201
headed |
.1526585
.0486376
3.139
0.002
.0573306
.2479863
landown | -.1045012
.0941648
-1.110
0.267
-.2890608
.0800584
landfrt |
.5026939
.2413214
2.083
0.037
.0297126
.9756752
landslp | -.5002137
.3132193
-1.597
0.110
-1.114112
.1136849
dumpc | -.6388443
.4267363
-1.497
0.134
-1.475232
.1975435
smlvstk |
.0003373
.0290314
0.012
0.991
-.0565632
.0572379
ttoxbll | -.0123164
.1295307
-0.095
0.924
-.2661919
.2415591
ttcows |
.0075854
.0469921
0.161
0.872
-.0845175
.0996883
hoes |
.1487087
.1514117
0.982
0.326
-.1480527
.4454701
plows |
.0652754
.1133539
0.576
0.565
-.1568943
.287445
sickls |
-.039089
.1166583
-0.335
0.738
-.2677351
.1895571
waterds | -.0200872
.0102739
-1.955
0.051
-.0402237
.0000494
haresaw | -2.023423
.8391118
-2.411
0.016
-3.668052
-.3787938
geblen | -1.964274
1.087921
-1.806
0.071
-4.096559
.1680114
dinki | -.7744587
.8873274
-0.873
0.383
-2.513589
.964671
yetemen |
-.841587
.9037539
-0.931
0.352
-2.612912
.9297381
shumsha |
-1.31077
.7946436
-1.650
0.099
-2.868242
.2467033
sirbana | -1.049067
.749355
-1.400
0.162
-2.517775
.4196422
adele |
-3.22614
1.209696
-2.667
0.008
-5.597102
-.8551789
korod | -1.279984
.7563264
-1.692
0.091
-2.762356
.2023888
trirufe |
1.203025
.7565021
1.590
0.112
-.2796922
2.685741
imdibir | -.4634745
.8608066
-0.538
0.590
-2.150624
1.223676
azedebo |
2.252262
.8871353
2.539
0.011
.5135092
3.991016
adado | -3.490983
1.058516
-3.298
0.001
-5.565637
-1.416329
garagod |
-1.42123
.8430021
-1.686
0.092
-3.073484
.2310236
doma | -.8754275
.8600309
-1.018
0.309
-2.561057
.8102021
_cons | -15.11687
1.910176
-7.914
0.000
-18.86075
-11.373
-----------------------------------------------------------------------------(Outcome mainact==3 is the comparison group)
28
Appendix B: Elasticities of probability of being inactive with respect to explanatory variables
All
Elast
s.d.
ln(age)
-12.48 (1.21)
dumkid
-0.29 (0.20)
# infants
-0.09 (0.10)
# females
-0.10 (0.22)
# males
-0.25 (0.21)
# elderly
0.08 (0.07)
dep ratio
-0.47 (0.30)
# younger boys
-0.03 (0.11)
# younger girls
-0.16 (0.11)
# older boys
0.19 (0.06)
# older girls
-0.04 (0.07)
female head
0.00 (0.04)
age of head
-0.16 (0.43)
yrs. school head
0.00 (0.06)
land owned
-0.06 (0.13)
land fertility
-0.01 (0.28)
land slope
0.24 (0.28)
perm crop
-0.34 (0.20)
# small animals
-0.09 (0.08)
# bull/oxen
0.22 (0.14)
# cows/calves
0.02 (0.12)
# hoes
-0.20 (0.11)
# ploughs
0.02 (0.09)
# sickles
-0.13 (0.09)
Minutes to water -0.01 (0.09)
#Obs
1204
BOYS
Aged 6-10
Elast
s.d.
-8.35 (0.76)
-0.21 (0.16)
-0.07 (0.08)
-0.07 (0.16)
-0.18 (0.15)
0.06 (0.05)
-0.34 (0.23)
-0.01 (0.05)
-0.08 (0.06)
0.20 (0.07)
-0.04 (0.08)
0.00 (0.03)
-0.12 (0.32)
0.01 (0.04)
-0.05 (0.09)
0.00 (0.21)
0.17 (0.21)
-0.25 (0.14)
-0.06 (0.06)
0.16 (0.10)
0.01 (0.09)
-0.15 (0.08)
0.02 (0.06)
-0.09 (0.06)
-0.01 (0.07)
700
Aged 11-15
Elast
s.d.
-15.27 (2.81)
-0.34 (0.22)
-0.08 (0.10)
-0.11 (0.24)
-0.28 (0.23)
0.09 (0.09)
-0.51 (0.33)
-0.05 (0.17)
-0.24 (0.17)
0.09 (0.03)
-0.02 (0.03)
-0.01 (0.05)
-0.17 (0.48)
-0.01 (0.06)
-0.07 (0.14)
-0.04 (0.30)
0.28 (0.30)
-0.36 (0.22)
-0.10 (0.09)
0.24 (0.17)
0.02 (0.13)
-0.22 (0.12)
0.01 (0.11)
-0.16 (0.10)
0.00 (0.10)
544
All
Elast
s.d.
-12.27 (1.14)
0.03 (0.20)
-0.13 (0.09)
-0.03 (0.21)
0.02 (0.18)
-0.01 (0.07)
-0.04 (0.28)
-0.19 (0.11)
-0.09 (0.10)
0.05 (0.06)
0.06 (0.06)
0.03 (0.04)
0.06 (0.42)
0.02 (0.05)
0.02 (0.05)
-0.05 (0.27)
-0.06 (0.26)
0.09 (0.19)
-0.08 (0.07)
0.09 (0.14)
0.02 (0.09)
-0.11 (0.10)
0.00 (0.10)
-0.02 (0.08)
0.10 (0.10)
1244
GIRLS
Aged 6-10
Elast
s.d.
-8.12 (0.73)
0.03 (0.15)
-0.10 (0.08)
-0.02 (0.15)
0.02 (0.13)
-0.01 (0.05)
-0.02 (0.21)
-0.09 (0.05)
-0.05 (0.05)
0.06 (0.06)
0.06 (0.07)
0.02 (0.03)
0.06 (0.30)
0.02 (0.04)
0.01 (0.03)
-0.02 (0.20)
-0.06 (0.19)
0.06 (0.14)
-0.06 (0.05)
0.06 (0.09)
0.01 (0.07)
-0.08 (0.07)
0.00 (0.07)
-0.02 (0.05)
0.07 (0.07)
678
Aged 11-15
Elast
s.d.
-14.92 (2.64)
0.02 (0.21)
-0.11 (0.09)
-0.04 (0.23)
0.02 (0.20)
-0.01 (0.09)
-0.05 (0.29)
-0.29 (0.17)
-0.14 (0.16)
0.02 (0.02)
0.02 (0.02)
0.03 (0.05)
0.03 (0.47)
0.01 (0.05)
0.03 (0.05)
-0.07 (0.30)
-0.05 (0.28)
0.10 (0.20)
-0.08 (0.08)
0.10 (0.16)
0.02 (0.11)
-0.13 (0.12)
0.00 (0.11)
-0.03 (0.09)
0.12 (0.11)
526
REFERENCES
Appleton, S., I. Chessa, et al. (1999). "Are Women the Fairer Sex? Looking for Gender Differences in Gender Bias in Uganda", mimeo, Centre for the Study of African Economies,
Oxford.
Baland, J.-M. and J. A. Robinson (1998). "A Model of Child Labor", Série Recherche 206,
Centre de Recherche en Économie du Développement (Facultés Universitaires NotreDame de la Paix), Namur, Belgium.
Behrman, J. R. and J. C. Knowles (2000). “Household Income and Child Schooling in Vietnam.”
The World Bank Economic Review 13(2): 211-256.
Benjamin, D. (1992). “Household Composition, Labor Markets and Labor Demand: Testing for
Separation in Agricultural Household Models.” Econometrica 60(2): 287-322.
Bhalotra, S. (1999). "Poverty and Child Work in Rural Pakistan", presentation at IREWOC
Workshop (Amsterdam): Children, Work and Education, University of Cambridge, Cambridge.
Bhalotra, S. and C. Heady (1998). "Child Labour in Rural Pakistan and Ghana", mimeo, University of Bristol and University of Bath, Bath and Bristol.
Canagarajah, S. and H. Coulombe (1997). "Child Labor and Schooling in Ghana", Policy Research Working Paper 1844, The World Bank, Washington.
de Janvry, A., M. Fafchamps, et al. (1992). “Structural Adjustment and the Peasantry in Morocco: A Computable Household Model.” European Review of Agricultural Economics
19(4): 427-53.
de Janvry, A. and E. Sadoulet (1996). "Household Modeling for the Design of Poverty Alleviation Strategies", Working Paper 787, Department of Agricultural and Resource Econo mics (University of California at Berkeley), Berkeley.
De Tray, D. (1983). “Children Work Activities in Malaysia.” Population and Development 9(3).
Dercon, S. and P. Krishnan (1998). "Changes in Poverty in Rural Ethiopia 1989-1995: Measurement, Robustness Tests and Decomposition", White Papers 98-12, Centre for the
Study of African Economies, University of Oxford, Oxford.
Fallon, P. and Z. Tzannatos (1998). Child Labor: Issues and Directions for the World Bank.
Washington, D.C., The World Bank (Social Protection, Human Development Network).
Greene, W. H. (1997). Econometric analysis. Upper Saddle River, New Jersey, Prentice Hall.
Grootaert, C. and H. A. Patrinos (1998). The Policy Analysis of Child Labor: a Comparative
Study, unpublished.
Gujarati, D. N. (1970a). “Use of Dummy Variables in Testing for Equality between Sets of
Coefficients in Two Linear Regressions: A Note.” American Statistician 24(1): 50-52.
Gujarati, D. N. (1970b). “Use of Dummy Variables in Testing for Equality between Sets of
Coefficients in Two Linear Regressions: A Generalization.” American Statistician 24(5):
18-21.
ILO (1995). "Report of the National Workshop on Child Labour in Ethiopia", Eastern Africa
Multidisciplinary Advisory Team (EAMAT, ILO), Nazareth, Ethiopia.
ILO (1996). “Child Labour: Targeting the Intolerable.” International Labour Review 135(6):
711-712.
ILO (1996). "Report on the National Workshop on Child Labour Policy and Action Programme
for Ethiopia", Eastern Africa Multidisciplinary Advisory Team (EAMAT, ILO), Debre
Zeit, Ethiopia.
Lambert, S. and T. Magnac (1992). "Recursive or Non Recursive Agricultural Household
Decision Making: An Application to Cote d'Ivoire", Document 92-20, DELTA, Paris.
Levison, D. and K. S. Moe (1998). “Household Work as a Deterrent to Schooling: An Analysis
of Adolescent Girls in Peru.” Journal of Developing Areas 32(3): 339-356.
Lopez, R. E. (1984). “Estimating Labor Supply and Production Decisions of Self-Employed
Farm Producers.” European Economic Review 24: 61-82.
Maddala, G. S. (1983). Limited Dependent and Qualitative Variables in Econometrics. New
York, Cambridge University Press.
Mason, A. D. and S. R. Khandker (1995). "Household Schooling Decisions in Tanzania",
mimeo, The World Bank, Washington (D.C.).
Mergos, G. J. (1992). “The Economic Contribution of Children in Peasant Agriculture and the
Effect of Education: Evidence from the Philippines.” Pakistan Development Review
31(2): 189-201.
Mueller, E. (1984). “The Value and Allocation of Time in Rural Botswana.” Journal of Deve lopment Economics 15: 329-360.
Nakajima, C. (1986). Subjective Equilibrium Theory of the Farm Household. New York, Elsevier.
Nielsen, H. S. (1998). "Child Labor and School Attendance: Two Joint Decisions", Working
Paper 98-15, Centre for Labour Market and Social Research, Aarhus (Denmark).
31
Owens, T. and J. Hoddinott (1999). "Investing in Development or Investing in Relief: Quantifying the Poverty Tradeoffs Using Zimbabwe Household Panel Data", Working Paper 99-4,
Centre for the Study of African Economies, Oxford.
Psacharopoulos, G. (1997). “Child Labor Versus Educational Attainment - Some Evidence from
Latin America.” Journal of Population Economics 10(4): 377-386.
Ravallion, M. and Q. Wodon (1999). "Does Child Labor Displace Schooling? Evidence on
Behavioral Responses to an Enrollment Subsidy", mimeo, World Bank, Washington
(D.C.).
Rose, P. and S. Al-Samarrai (1997). "Household Constraints on Schooling by Gender: Empirical
Evidence from Ethiopia", IDS Working Paper 56, Institute of Development Studies,
Brighton.
Rosenzweig, M. R. (1981). "Household and Non-household Activities of Youths: Issues of
Modelling, Data and Estimation Strategies" in ILO. Child Work, Poverty and Underdevelopment. Geneva, ILO: 215-243.
Rosenzweig, M. R. and R. Evenson (1977). “Fertility, Schooling, and the Economic Contribution
of Children in Rural India: an Econometric Analysis.” Econometrica 45(5): 1065-1079.
Sadoulet, E. and A. de Janvry (1995). Quantitative Development Policy Analysis. Baltimore
(Maryland), Johns Hopkins University Press.
Sadoulet, E., A. de Janvry, et al. (1996). "Household Behavior with Imperfect Labor Markets",
Discussion paper, Department of Agricultural and Resource Economics, University of
California at Berkeley, Berkeley.
Schmidt, P. and R. P. Strauss (1975). “The Prediction of Occupation Using Multiple Logit
Models.” International Economic Review 16(2): 471-486.
Singh, I., L. Squire, et al., Eds. (1986). Agricultural Household Models: Extensions, Application
and Policy. Baltimore (Md.), Johns Hopkins University Press.
Skoufias, E. (1994). “Market Wages, Family Composition and the Time Allocation of Children
in Agricultural Households.” The Journal of Development Studies 30(2): 335-360.
USAID (1994). "Demand for Primary Schooling in Rural Ethiopia", mimeo, USAID
Weir, S. (1997). "The Determinants of School Enrollment in Rural Ethiopia", mimeo, Centre for
the Study of African Economies, University of Oxford, Oxford.
World Bank (1998). World Development Indicators. Washington (D.C.).
32