Poverty and Child Soldier Recruitment: A Disaggregated Study of African Regions

Poverty and Child Soldier Recruitment:
A Disaggregated Study of African Regions
Abstract (794 characters):
In the popular debate, poverty is often identified as the cause of child soldier recruitment. The
argument suggests that economic deprivation and few viable life choices push children into
recruitment for armed conflict. The poverty argument has rarely been tested systematically, and
statistical results are inconclusive. Previous analyses potentially suffer from two
methodological problems: ecological fallacy and selection on the dependent variable. We meet
these shortcomings in previous tests of the poverty–child soldier nexus by introducing new data
that geographically disaggregates recruitment and poverty. Using a cross-sectional research
design for all sub-national regions in Africa in the period 1990-2004, we find some evidence
that the poorest regions are more subjected to child soldier recruitment. However, other factors,
such as the existence of refugee camps seem to outperform the poverty explanation.
Characters text (with spaces):
Characters graphics:
Characters (total):
71,470
8,800
80,270
1. Introduction
Child soldiers frequently appear in armed conflicts around the world, and the problem is
particularly endemic in Africa. Children are recruited by both government and rebel armed
forces, and often serve in paramilitary, militia or self-defense groups backed by state authorities
which might not conscript children themselves (CSUCS 2004; Achvarina/Reich 2006). In DRC,
for example, child soldiers were serving at the front lines with all armed groups, in some cases
representing up to 35% of the troops (UN 2003). Despite increased efforts of the international
community to combat child soldiering1, children are being recruited and re-recruited for
conflicts. This dire trend calls for action, not only on the improvement and strengthening of the
international norms and programs, but also on the systematic investigation of the root causes of
child soldier recruitment.
The bulk of existing literature on child soldier recruitment consists of non-empirical
academic works and NGOs reports, usually based on interviews with a handful of children who
have been involved in combat. This literature often cites poverty as the cause of child
soldiering. There also exist a few larger systematic surveys of ex-combatants in Africa (e.g.
Blattman 2007 on Uganda; Pugel 2007; Bøås/Hatløy 2008 on Liberia; and Humphreys/
Weinstein 2004 on Sierra Leone). These studies provide unique insights into the individual
motivation and methods of recruitment, but as they focus on single conflicts, they cannot be
used for cross-country comparisons. To our knowledge, Achvarina and Reich (2006) is the
single existing cross-country study comparing different conflicts on the causes of child soldier
recruitment. They found that national measures of poverty cannot explain child soldier
participation. Rather, they argue that refugee camps better explain child soldier recruitment
rates, as children in refugee camps are easily accessible targets for armed forces seeking
recruits. Hence, child soldiers will constitute a larger percentage of belligerent forces where
camps are relatively vulnerable to raids. A methodological caveat here is that this study is based
1
International protocols have been signed, monitoring has been initiated in several countries, and a practice of
naming and shaming was put into place by the UN Security Council.
1
on nation-wide measures which run a high risk of ecological fallacies. For example, poverty is
spatially clustered within countries, and even in societies with low levels of overall inequality,
some regions are richer than others (Buhaug/Rød 2006). Studies that utilize nation-wide
measures of poverty cannot account for sub-national variations. As the level of economic
welfare (as well as child soldier recruitment) might vary significantly within countries, using
national aggregate measures of development, such as GDP per capita, might not reveal the true
impact of poverty on child soldier recruitment. The case of Uganda is an illustrative example.
The conflict in the northern part of the country has been notorious for recruitment of children
into the Lords Resistance Army (LRA). This part of the country is significantly less developed
than the south (Younger 2004). National aggregate poverty measures might therefore obscure
the true relationship between poverty and child soldier recruitment. The map of Uganda (right)
in Figure 1 shows variations in regional rates of both infant mortality (darker regions have
higher infant mortality) and child soldier recruitment (hatched), and the map of Chad (left) also
illustrates regional variations in terms of household assets, where darker regions (relatively
deprived compared to country average) correspond to a large degree the regions in which
recruitment took place (hatched).
[Figure 1 about here]
In this study we test the relationship between poverty and child soldier recruitment using
all sub-national regions in Africa as the units of analysis.2 This allows us to better account for
geographical variations in recruitment, poverty and other factors, and avoids some of the
problems of the ecological fallacy of aggregate measures and selection bias. Whereas
individual-based survey data on combatants, non-combatants and their poverty levels would be
the best source of data for testing the poverty-child soldier recruitment nexus, no such crossnational data currently exists. We therefore use the next best option of disaggregating by subnational region. We introduce new data on regions affected by child soldier recruitment and, by
2
Our spatio-temporal domain covers the African continent, i.e. all the first-level administrative units, 690 in
total, in the period 1990–2004.
2
means of GIS (Geographic Information Systems), we link these data with geo-referenced data
on regional absolute and relative poverty.
We do find some support for a relationship between absolute poverty (measured as
infant mortality) and child soldier recruitment. With regard to relative deprivation (or interregional inequality) we find no significant effect. Regions that are poorer than the country
average are no more at risk of child soldier recruitment than those regions above the country
average. However, the positive relationship between refugee camps and child soldier
recruitment found by Achvarina and Reich (2006) is upheld despite our disaggregated design.
One clear recommendation for the policy community is therefore to aim at protecting refugee
camps from recruitment raids by armed groups.
The remainder of the article is organized as follows. In Section 2 we summarize general
theories of recruitment and armed conflict and offer a literature review on the causes of child
soldiering, focusing on the role of poverty. In Section 3 we present the data and research
design. Section 4 summarizes the results of our empirical tests. Finally, in Section 5, we
conclude and suggest an agenda for future research.
2. Poverty and Child Soldier Recruitment
Most civil wars occur in relatively poor countries, and socioeconomic status has for long been
assumed to be associated with involvement in violent conflict. The direct link between
economic development and domestic peace has proven to be among the most robust findings in
recent large-N country-level studies of civil war (see Hegre/Sambanis 2006). However,
although there is agreement on this empirical relationship there is no consensus on the
theoretical explanation for it. Fearon and Laitin (2003) maintain that GDP per capita is a proxy
for state capacity, indicating that richer states are better able to monitor the population and
conduct effective counterinsurgencies. But, what motivates a person to risk his life in armed
conflict? The recruitment literature brings the poverty argument closer to the micro level by
suggesting that low development provides motivation for violence due to low opportunity costs
as well as a potential for private gains from looting (Doyle/Sambanis 2000; Gates 2002). The
recruitment costs are lower when the alternative means of income are low, in situations of
under-employment and poverty. If people have no other viable means to ensure a life sustaining
3
income, the threshold for joining an army (be it government or rebel) is presumed to be low.
Consequently, Collier and Hoeffler (2004) claim that it is easier to maintain a rebellion in poor
countries than in richer countries.3 The arguments for why poverty and inequality should matter
for child soldier recruitment parallel many of the explanations for recruitment of adults.
However, while inhibiting other qualities than adult soldiers and thus being attractive targets for
certain types of armies, children are thought to be particularly vulnerable to being forcibly
recruited or kidnapped by armies.
Most studies of child soldiers, while disagreeing about the significance of poverty’s
impact, generally admit that it matters to some extent. For example, the earliest comprehensive
book on child soldiers by Goodwin-Gill and Cohn (1994) identifies poverty as a factor without
assigning it a greater value than to other variables. Honwana (2006: 28) considers poverty one
of the main push factors behind child motivation to join armed groups beside migration,
political ideology, or the “mutability of youth”. Stronger statements about the relationship
between poverty and child recruitment have been offered by Graça Machel (1996: 11), who
concludes that “the children most likely to become soldiers are those from impoverished and
marginalized backgrounds” along with the ones that are unaccompanied. Two authors
independently claim that the economic factor is “a particularly strong” explanation for child
soldiering, compared to other explanations that include proliferation of small and cheap
weapons and the changing nature of warfare (Singer 2005: 38, 55; McManimon 1999). Brett
and Specht (2004: 14, original emphasis) argue that poverty “is perhaps the most obvious
common feature of child soldiers generally, which is one of the reasons why it is frequently
identified as the cause of child soldiering”.
There is currently no well-defined consensus in the literature on the mechanisms that
link poverty to child soldiering. To outline different potential mechanisms theoretically, we
distinguish between voluntary and forced recruitment, as poverty could feature as a factor in
3
The opportunity costs of joining a rebellion have been proxied with a variety of indicators in the existing literature. Collier
and Hoeffler (2004) use the rate of economic growth per capita and the secondary school enrolment rate for males. Esty et al.
(1995, 1998) and Goldstone et al. (2005) use infant mortality as a proxy for development and thereby opportunity costs of
potential soldiers. However, all these studies have used aggregate country averages of opportunity costs for recruitment rather
than localized indicators of poverty as indices of recruitment costs, although poverty and wealth tend to be spatially clustered
within countries.
4
both forms of recruitment. Both recruitment forms point in the direction of a positive
relationship between poverty and recruitment.4 Whereas poverty-based economic motivation
(both greed and grievance-based motivation as well as pure struggle for survival) is vital for
understanding voluntary recruitment, forced recruitment is often a question of protective
capabilities, which is often also a function of poverty. In the next two sections we discuss how
poverty may play out in voluntary and forced conscription of children.
2.1 Voluntary Recruitment
Children quite often join armed struggles without pressure being exerted upon them and may
actually look for military groups themselves to offer their services. In one ILO study, 64% of
all former child soldier informants from the DRC, Burundi, Rwanda and Congo reported
joining an armed group on the basis of personal decision as opposed to being directly forced to
do so (ILO 2003: 26).
Voluntary recruitment for armed conflict requires some level of motivation. GoodwinGill and Cohn (1994) propose three different scenarios of how poverty affects the motivation to
join an armed group. In line with the classical literature on the relationship between poverty and
conflict, they label these causal mechanisms as: grievance (“social and economic injustice
motivates adults and children to take up arms, sometimes with a long-term vision of affecting
change”); greed (“to obtain a subsistence wage”), and survival (to get food for the day)
(Goodwin-Gill/Cohn 1994: 23).
The concept of ‘grievance’ is usually based on the logic of relative poverty, or
inequality. Most traditional works on inequality and conflict relate to the theory of relative
deprivation (see Gurr 1970). This premise suggests that while absolute poverty may lead to
apathy and inactivity, comparisons with others in the same society who do better can lead to
frustration and antagonism which again may result in violence to redress inequality. A
continuation of this argument is to see grievance-induced discontent due to a group’s
marginalization as a determinant of mobilization for violent political struggle. ‘Grievance’
factors have been largely dismissed by the large-N country-level studies, which find no link
4
We use this dichotomy here as a useful analytical distinction in the theoretical discussion of mechanisms
leading to child soldier recruitment. In real life situations, the distinction might not be so clear cut.
5
between economic inequality and conflict (Fearon /Laitin 2003; Collier/Hoeffler 2004). Østby
(2008), however, argues that such dismissal of grievance factors may be premature, because the
above studies address economic inequality between individuals while ignoring inequalities
between groups. Case studies suggest that what matters for conflict are so-called ‘horizontal
inequalities’, or inequalities that coincide with identity-based cleavages (Stewart 2000, 2002).
In brief, as conflict as usually fought between groups, not individuals, inequalities based on
cultural cleavages may facilitate recruitment and mobilization for armed conflict.
Andvig (2006) argues that grievance as motivation for joining a rebellion works in the
same manner for both children and adults. Children tend to equate violence with power and the
reasons given for enlistment include not only peer pressure and opportunity to engage in
looting, but also political commitment and ethnic loyalties (Stewart/Boyden 2001). For
example, if a child belongs to a group or region which is relatively economically deprived and
where schooling opportunities are low, this may lead to frustration and a sense of unfairness
which in turn may influence the child’s willingness to become a soldier in order to try to change
the status quo. Furthermore, identity-based groups – the ones that share the same ethnic,
religious or regional affiliation – also tend to have stronger group cohesion than other types of
groups (see e.g. Guichaoua 2006; Stewart 2000, 2002). Coupled with the evidence from child
psychology and empirical studies about children’s “greater tendency towards altruism and
bonding to a group” (Andvig/ Gates 2006: 7; Harbaugh/Krause 1999), the strong cohesion of
identity-based groups is an additional attractive factor in the decision of children to join a rebel
group. Arguably, social pressure and ideological propaganda can also persuade children to
enroll with armed groups (ILO 2003: 25). This corresponds well to Wessels (1998: 639) who
argues that “issues of identity, nationalism, and ideology may also loom large” in children’s
decision to participate in armed struggle.
Like adults, children can also be driven by greed. Rebellions provide opportunities to
loot and get access to financial resources, including salaries for soldiering. Gates (2002: 128)
argues that “faced with dismal conditions at home, involving poverty, boredom, or, in some
areas, no family” children might have fewer reservations to join an armed group. In other
words, children might voluntarily join armies due to perceived prospects that look brighter than
poverty or boredom, which may in part stem from lack of educational opportunities. GoodwinGill and Cohn’s (1994) third scenario, the motivation of survival – that is the decision to join an
6
army as the best option for a child to secure food or basic security – can be hard to distinguish
from a greed drive. Orphaned children may be particularly susceptible to the greed motivation
as the groups of armed adults might become the only substitute for parental care in terms of
food provision, perceived security guarantee, and a mere establishment of a missing category of
an adult-child relationship in an orphan’s life (Brett/McCallin 1996; Singer 2005; UNICEF
2002). At the same time, studies with aggregate variables using national measures of orphans
did not find that particular variable to be significant in explaining the variation in child soldier
rates across different African conflicts (Achvarina/Reich 2006). 5
Still, parental protection might not be a guaranteed condition even for children with live
parents, with the most extreme cases being parents who voluntarily give away their children to
rebels due to greed or ideological motivations, often because a family member is already in the
military (ILO 2003: 36). Impoverished parents sometimes send their children to armed groups
in exchange for minor soldier's wages that go directly to the family (Machel 1996: 12). Such
‘volunteering’ includes “parents who encourage their daughters to become soldiers if their
marriage prospects are poor” (Machel 1996: 12). Alternatively, children can become de facto
child soldiers if the whole family moves with armed forces for economic reasons, or they can
be recruited because a family member is already in the military.
Why would a military organization recruit children as soldiers? Army commanders in
Africa have reported several reasons, such as children being easily manipulated and efficient
cheap fighters, with a better performance of certain tasks such as scouting (ILO 2003). From
the perspective of commanders and army leaders, recruiting underage soldiers can decrease the
cost and ease of recruitment, particularly of impoverished children. With respect to voluntary
recruitment, any army that wants to conscript soldiers needs to be able to offer some level of
benefits, be it food or payment. Gates (2002) sees the participation constraint as essentially a
comparison of the utility offered by a rebel group (or possibly any army) compared to some ex
5
Ideally, we would have preferred to have tested whether regions with larger numbers or ratios of orphaned
children are more prone to child soldier recruitment. Unfortunately, however, such data seem to be non-existent on
the sub-national level. One potential proxy for orphan rates could be regional HIV/AIDS figures for which the best
source seems to be USAID’s HIV/AIDS Surveillance Database (http://hivaidssurveillancedb.org/hivdb/).
However, very few countries have reliable data broken down to regional figures.
7
ante outside option. He argues that children offer a higher possibility for rebel groups to meet
the so called reservation level of benefits that a recruit demands in order to join, as this level is
proposed to be lower for children than adults. In addition, children might mobilize only for a
promise of future delivery of benefits. For example, in Liberia, children from marginalized
economic groups were promised free access to education after the end of the war. This promise
was enough to convince some of them to join Charles Taylor’s armed forces.6 In DRC and
Congo, former child soldiers have also reported that they joined to receive payment or a job
after the war (ILO 2003: 30). The prospect of even marginal payment is a strong incentive for
children to enlist in situations when their parents are missing or finding it hard to provide basic
food security. Hence, child soldiers mean cheap labor for rebels with limited resources. It is
quite intuitive to build on this argument and suggest that poorer children offer even cheaper
labor and thus better alternatives for recruiters, especially the ones that abstain from forced
conscription.
2.2 Forced Recruitment
The above reasoning has pertained only to voluntary recruitment of child soldiers. Often,
however, participation is forced at gunpoint. This practice was widespread among LRA in the
northern Uganda (Blattman 2007) and RUF in Sierra Leone (CSUCS 2001). ILO found that
about 21% of the child soldiers sampled in four African countries (Burundi, Congo, DRC,
Liberia) reported to have been abducted, and 15% forced (ILO 2003).7 How can poverty matter
when children are taken by force? We argue that the less privileged will typically have the least
resources to defend their families and children, and hence provide an easy prey. From the
recruiters’ point of view, poorer communities typically have less means of protection (due to
insufficient infrastructure, economic resources, or lower priority of government protection
policies) and therefore might be more attractive destinations for recruitment. Singer (2005: 4)
7
This information was obtained during personal interviews by one of the authors with Liberian former child
soldiers.
7
By abduction the ILO report refers to “situations in which children have been taken forcibly or under threat of
arms”; whereas forced recruitment is defined to refer to “cases in which the child did not have a choice. This could
be because of moral pressure or the obligation to enlist.”
8
argues that targeted children are “usually from special risk groups: street children, the rural
poor, refugees, and others displaced”. He explains this logic by introducing the concept of
“efficient recruiting sweeps” pertinent to these four special risk groups in particular. Some
literature by practitioners also notes that “in all conflicts, children from wealthier and more
educated families are at less risk” of forced recruitment as they are either left “undisturbed” or
“released if their parents can buy them out” or “sent out of the country to avoid the possibility
of forced conscription.” (Machel 1996: 12). In other words, defense capabilities or protection
provision can be intimately linked to absolute poverty and increase likelihood of recruitment
among the poor and also linked to relative poverty between sub-national regions, where the less
well off regions are more likely recruitment grounds for child soldiers.
This can seem contrary to the argument of Ethan Bueno de Mesquita (2005) that
terrorist operatives have relatively high educational attainment and economic opportunity.
Bueno de Mesquita convincingly argues that whereas individuals from societies’ worst-off
socioeconomic groups are most likely to join a terrorist organization, the terrorist organization
screens the volunteers for quality and target the most competent, richer and better educated
candidates. However, unlike terrorist organizations that have a need for masterminds who are
more likely to succeed at the demanding tasks required of a terrorist operative, we do not
believe that the same logic applies for the strategic recruitment of child soldiers for armed
conflict. A typical African conflict does not require particularly skilled operatives. Child soldier
recruitment mainly takes place in under-developed countries, characterized by limited counterintelligence that can often be performed by children and prevalence of primitive but effective
AK-47 rifles easily handled by children. Under such conditions it should be more strategic for
African rebel leaders and commanders to recruit a large number of children. As argued by
Blattman (2007: 1): “rebel leaders have an incentive to recruit any civilians that are expected to
yield some military benefit”. Furthermore, indoctrination and disorientation is likely to be more
successful with poor low-educated children.
--In the above theoretical framework we have discussed the impact of both absolute and
relative poverty. In sum, the absolute level of poverty is expected to affect both voluntary and
forced recruitment of child soldiers either because of the lack of alternative viable survival
strategies or due to negligible defense against forced recruitment. From the perspective of
9
recruiters, the poorer the children, the easier it is to recruit them, either by force or voluntarily.
This leads to our first hypothesis:
H1: Higher levels of absolute poverty in a region increase the likelihood of recruitment of child
soldiers.
Whereas the previous literature on child soldiers has focused almost exclusively on the impact
of absolute poverty, we argued in the above discussion that relative poverty (deprivation) could
also play a role regarding voluntary as well as forced recruitment of child soldiers. If children
are motivated and mobilize to redress economic grievances, the relatively poor should also,
more often than the relatively privileged, engage in child soldiering, especially if the relative
deprivation is a result of systematic discrimination between particular identity groups, such as
ethnic or regional groups. Furthermore, the logic of forced conscription of children in poor
communities can also be viewed as an outcome of external processes that come as a result of
relative poverty. In other words, if a government or rebel group wants to recruit children by
force, the rational choice would arguably be to target regions that fall below the country
average, and the more the region falls below this average the more at risk the region would be
of being seen as a good recruitment ground. In line with this reasoning we propose the
following hypothesis:
H2: The relatively poorer (more deprived) regions in a country will be more at risk of child
soldier recruitment than the ones that are relatively better off.
3. Data and Research Design
Since civil wars are often quite local, nation level indicators to explain either the location of
conflicts or child soldier recruitment might therefore often become misleading or, at best,
irrelevant. Hence, the present study is among the emerging efforts at geographically
‘disaggregating the study of civil war’, or investigating the causes of conflict below the national
level (see e.g. Buhaug/Lujala 2005; Buhaug/Rød 2006; Hegre/Raleigh 2005; Østby et al. 2008;
Raleigh/Urdal 2005). Some regions of a country might experience more recruitment than
10
others, and the causal factors used to explain this recruitment also vary geographically. As the
next-best option to individual survey data8, we rely on indicators of localized socio-economic
status at the regional level. Our units of analysis are sub-national regions in Africa representing
first-level administrative units (regions/provinces) according to ESRI’s (1998) definition. The
total number of observations in the dataset adds up to 690 regions in 52 countries (i.e. all the
sub-national regions in Africa in the period 1990–2004). Due to some missing observations on
certain variables the tests range from 354 and up to a maximum of 688 observations.
Geographical Information Systems (GIS) software, allows us to combine spatial data on
regional welfare and child soldier recruitment.
3.1 Dependent variable: Child Soldier Recruitment
Our dependent variable, child soldier recruitment, is dichotomous, coded as 1 if there were
reports of child soldiers recruited in the region in the period 1990–2004, and 0 if we could not
find any such reports.9 We cover recruitment of voluntary and forced nature from home and
displaced (including refugee) communities of children alike.10 We rely on the existing
operational definition of child soldiers from the United National Children’s Fund (UNICEF,
undated) which is used in the field to collect the data. According to this definition, child soldier
is “any child—boy or girl—under 18 years of age, who is part of any kind of regular or
irregular armed force or armed group in any capacity, including, but not limited to, cooks,
porters, messengers, and anyone accompanying such groups other than family members”
8
No such comparative geo-referenced survey data across cases currently exists.
9
Ideally, we would like to have a scale measure of the magnitude of the recruitment or a ratio variable of child
soldiers to the total number of soldiers, but due to data constraints this has not yet been possible. Reliable timeseries data would also have been preferred due to potential endogeneity problems regarding the relationship
between IMR and child soldier recruitment. However, since IMR figures refer to the death of infants (under 1 year
of age), and child soldiers refer to older children, this problem should at least be less serious. Although IMR
figures are of course likely to be higher resulting from conflict (as would be the case with any other measures of
poverty), it is not obvious that such figures should be higher in conflicts including child soldiers than in conflicts
where child soldier recruitment does not occur.
10
It is not possible to distinguish between forced or voluntary recruitment in the analyses. For most of the
observations it is not clear what type of recruitment was prevalent, and in many instances both types of recruitment
were going hand in hand by the same armed groups or forces, and a clear distinction can be hard to establish.
11
(UNICEF, undated: 4). One controversy surrounding UNICEF’s definition deals with the
established benchmark of 18 years old as a minimum age for recruitment. The point is often
raised by many observers that this number is driven by international conventions based on
western norms and does not make sense in the context of African countries where, according to
some, the age of adulthood is often set at a much lower level and in some countries hardly
reaches 15 years old. Still, due to a lack of available data on recruitment broken down by age,
we use the date collected according to the UNICEF definition.11
Out of the 690 regions in our sample, nearly 42.9% (296 regions in 28 countries)
experienced conflict in the 1990–2004 period, and 10.6% (73 regions in 17 countries)
experienced child soldier recruitment. Of these 73 regions, 66 were in a conflict zone and 7
regions were not. The data on recruitment is coded based on systematic evaluations of country
reports of child soldiers issued by NGOs, international organizations, governments, academics,
and even military organizations.12 The map in Figure 2 shows which regions experienced child
soldier recruitment (hatched) and conflict (shaded).
[Figure 2 about here]
3.2 Absolute and Relative Poverty
To test our two hypotheses we need to operationalize two different concepts of poverty:
absolute and relative poverty. We use data from two different sources to generate these
measures. The first is geo-referenced disaggregated data on infant mortality rates (IMR) from
the CIESIN data project at Columbia University, which covers 52 African countries.13
According to the norm, CIESIN defines regional, annual IMR as the number of children who
die
11
before
their
first
birthday
for
every
1,000
live
births,
Another problem with the UNICEF definition of child soldiers is that it does not distinguish between different
tasks that children are taken to perform. However, in our article this distinction is not critical as our poverty
argument is mostly supply-based and deals with the question of vulnerability of certain groups of children for
recruitment with the demand for children by armed groups viewed as already given.
12
The coding and geo-referencing of this information has been done by the authors. IMR rates for southern
Sudan have been made available by Theisen and Brandsegg (2007).
13
i.e.
CIESIN online at: http://sedac.ciesin.columbia.edu/povmap/ .
12
 * 1000 . Infant mortality has been used as
IMR =  Deaths under age 1 in year
Live births in year 

an alternative to GDP per capita or similar measures in quantitative studies in the conflict
literature (see e.g. Esty et al. 1995, 1998; Goldstone et al. 2005; Urdal 2006) and elsewhere, as
the two are typically very highly correlated and believed to capture the same phenomenon of
general development. One criticism of using IMR has been that as countries cross a certain
threshold of wealth there is often little variation on such a basic measure of development as
infant mortality. However, as we are investigating Africa this is less problematic than in other
studies, as the countries in question are in large part defined as low-income. Also, IMR is
theoretically closer to the relationship we want to test, as it better captures poverty-related
factors identified that lead to child soldiering, such as food shortages. The map in Figure 3
shows the coverage of the CIESIN data of infant mortality rates overlaid with child soldier
recruitment (hatched). A simple visual investigation of this map reveals what seems to be a
pattern of the poorer regions (i.e. the higher IMR figures) experiencing more child soldier
recruitment, than the richer regions. However, this relationship will be tested further in the
statistical analysis.
[Figure 3 about here]
Our second data source used to construct measures of absolute and relative poverty is
geo-referenced information from the Demographic and Health Surveys (DHS) from 22 African
countries14 conducted during the period 1986-2001. In a DHS, a sample of households is
selected throughout the entire country, and women between the ages of 15 and 49 are
interviewed about health, nutrition, household welfare and other issues. The sample design is a
probabilistic two-stage sample, in which enumerated areas (EAs) are randomly selected with
probability proportional to their size. Several DHS Surveys include detailed information about
the geographical location of each EA. This allows us to couple local-level socioeconomic
information from the surveys with the geographically recorded data on the location of child
14
These countries are: Benin, Burkina Faso, Cameroon, Central African Republic, Chad, Cote d’Ivoire,
Ethiopia, Ghana, Guinea, Kenya, Liberia, Madagascar, Malawi, Mali, Namibia, Niger, Nigeria, Senegal, Uganda,
Tanzania, Togo, and Zimbabwe.
13
soldier recruitment. Here, we use individual-level information from each EA to aggregate
measures of regional welfare to the first-level administrative units (regions) that constitute our
units of analysis.
We use the DHS surveys to generate two indicators of absolute regional socioeconomic
welfare (poverty). First, a household asset index is generated on the basis of the following
variables from the DHS surveys: v119-v125 (dummies for whether or not each household has
electricity, a radio, a television, a refrigerator, a bicycle, a motorcycle and/or a car). Our second
indicator, education years, is based on the variable v133 (highest years of education
completed).15 With a cross-sectional research design, there is a potential problem of conflict
affecting education levels in the population, particularly in cases of high intensity and long
lasting conflicts. For most of the included conflict cases in our sample, however, the DHS
surveys were conducted prior to conflict outbreak. For example the DHS in Guinea was
conducted in 1999 and conflict broke out in 2000, in Liberia DHS was conducted in 1986 and
conflict broke out in 1989, and in Niger a DHS was conducted in 1992, and two conflicts broke
out in 1994 and 1996 respectively. In some cases the problem of endogeneity might also be less
pronounced because conflict was short lived (i.e. the coup in Togo in 1991, which probably in
itself did little direct harm to the education provision) or minor (i.e. conflict incidences in Mali
in 1990 and 1994, and the on-again off-again conflict in Casamance in Senegal between 1990
and 2003 which remained minor in terms of fatalities).16 Also, since the DHS data report the
highest numbers of years of education completed based on a sample of adult females, there
should be a relatively high degree of inertia in these measurements even with conflict taking
place prior to the recorded education attainment.
15
One could argue that once children are recruited as child soldiers they drop out of the educational system,
which implies an unclear direction of causality. Note, however, that education level as measured here refers to the
average education level of women aged 15-49, i.e. largely the adult population. Hence, this should reduce potential
endogeneity problems.
16
The case of Uganda potentially causes problems here, as the DHS survey was conducted in 2000-2001 whereas the civil
conflict in the country had been ongoing since the late 1980s, however mostly restricted to the northern region. Ethiopia also
represents a potentially problematic case, as there had been long lasting and severe conflict going on in the country prior to the
conduction of DHS surveys in 1992. Chad is another case of conflict ongoing since the late 1980s and DHS surveys conducted
in 1996.
14
To evaluate the impact of spatial inequalities, we measure inequality, or regional
relative deprivation (RRD), as the relative performance of each region compared to the overall
performance of the country on both the assets indicator and the education indicator, using the
following formula:
  M A A 
RRD = −1 ln ∑ i1 i 2  
  i =1 M  
where M is the maximum number of household assets, A1 refers to mean asset score of a
given region and A2 is the corresponding mean score of the country as a whole. This provides a
continuous variable ranging from –.76 (lowest level of relative deprivation) to 1.81 (highest
level of relative deprivation). Note that the value ‘0’ indicates perfect equality, whereas
negative values of RRD refer to relative privilege of the region in question. The measure of
educational relative deprivation is generated similarly. For each of the regions, the scores on the
various inequality measures were copied to the remaining years in the period 1990–2004.
3.3 Control Variables
Although poverty is often identified by the NGO community as the cause of child soldiering,
Achvarina and Reich (2006) introduce another factor which they argue largely outperform the
poverty explanation: the degree of access to refugee/IDP camps gained by the belligerent
parties in conflicts. Children gather in refugee camps in great numbers. Refugee camps are
supposed to be protected under international laws and protocols, but protection is often, in
practice, uneven or nonexistent. This lack of physical protection of camps provides an incentive
that will likely increase the probability of successful raids by armed factions seeking recruits.
However, children in refugee camps may also voluntarily become recruits, motivated by the
prospects of a better future compared to life in the refugee camp. Since security is often a
problem for refugee camps, regions with a refugee camp may be more likely to have
experienced child recruitment than regions without such camps. For our geographically
disaggregated study we use information on the location of refugee camps to determine which
sub-national regions had such camps in the period under investigation. The information on the
location of refugee camps has been collected by the United Nations High Commissioner for
Refugees (UNHCR) and geo-referenced (point data with latitude and longitude) by Weidmann
et al. (2007). To adapt this to our dataset structure, we have used GIS to determine in which
15
sub-national regions the camps were situated. The refugee camp data includes 710 camps,
distributed over 160 regions. From this we created a dummy control variable indicating if the
region had a refugee camp (1) or not (0).
As our sample consists of sub-national regions, irrespectively of whether they
experienced an armed struggle or not, we need to control for conflict factors. We test three
conflict related controls: how many years the region was exposed to conflict, how intense the
conflict was in terms of battle-related casualties, and whether the neighboring region(s) had a
conflict,. It is intuitively plausible that recruitment of child soldiers is associated with the
presence or absence of conflict in the same area. The information about location of conflict
zones (as well as conflict duration) is based on a version of the Uppsala/PRIO Armed Conflict
Dataset (ACD, Version 3-2005b; Gleditsch et al. 2002)17 which includes data on the spatial
location of battle zones depicted as GIS-generated conflict polygons (Buhaug/Rød 2006).
Children living near a conflict zone are exposed to extreme insecurity and a climate of fear
which might increase the likelihood of their voluntary enrollment. Likewise, proximity to the
conflict zone increases the accessibility of recruiters to these children. To control for this
potential spatial correlation we include a dummy for whether there was a conflict in the
neighboring region (1) or not (0). As a conflict drags on, the availability of male recruits may
drop, and the need for new fighters can lead recruiters to conscript children. Thus, an ILO
report (2003: 25) on four conflicts that involved child soldier recruitment states, based on the
cases of Burundi, DRC, Congo and Rwanda, states that “the longer the conflict lasts, the greater
the risk of recruiting soldiers that are younger and younger”. To control for conflict duration we
include a count variable of the number of years of conflict in the region during the period 1990–
2004. Demand for recruits and minors may also intensify with intense conflicts – those ones
that experience high level of fighting and face a big number of battle deaths as a result. We
17
Available at: http://www.prio.no/CSCW/Datasets/Armed-Conflict/UCDP-PRIO/Old-Versions/3-2005b/. This
dataset covers every armed conflict between a state government and an organized opposition group with at least 25 battlerelated deaths per year.
16
control for this by including a log of battle-related deaths (Lacina/Gleditsch 2005) for each civil
conflict in the time covered by our data.18
Finally, recruitment of children could occur in remote, less densely populated areas. For
example, children in rural areas of Uganda were more at risk of recruitment because there was
less protection and security offered by the government and recruiters could go into small
villages and more or less unhindered kidnap children for combat (Blattman 2007). The level of
development might also differ between areas of various density of population. We therefore
include as a control a measure of the log of the population density in the region to isolate the
effects of poverty and inequality from this potentially confounding factor. On the other hand it
is also conceivable that recruitment is more likely in regions with a larger pool of potential
recruits due to the consideration of efficiency of conscription sweeps. We therefore also include
a measure of the log of total regional population. Descriptive statistics for all variables as well
as a correlation matrix are provided in Appendices A1 and A2.
We analyze a dichotomous dependent variable for whether or not there was any child
soldier recruitment in a region in the period 1990–2004, using a logit regression model. Since
regions are likely to be somewhat interdependent, standard error estimates are clustered by
country.
4. Empirical Analyses
Table 1 summarizes the results of multivariate logit regressions of child soldier recruitment and
absolute poverty. Model 1 in Table 1 is the baseline model, and shows that the existence of a
refugee camp in the region, conflict intensity, and conflict in neighboring region(s) are strong
predictors of child soldier recruitment. One of the strongest effects is from the existence of a
refugee camp. A region with a refugee camp is almost six times more likely to experience child
soldier recruitment than a region with no such camps (2.2% vs. 12.6% risk). Conflict intensity
also affects child soldier recruitment in the expected direction. Higher numbers of battle related
18
As battle-related deaths are not available by sub-national regions, we use the death figures for a conflict as a whole.
In regions with overlapping conflicts we assign the higher intensity figure.
17
deaths are associated with a significantly higher likelihood of child soldier recruitment being
reported.19 The number of years of conflict does not have a significant relationship with
recruitment risk when conflict intensity (battle deaths) is accounted for. However, including
both intensity and conflict years in the same model is problematic, as there is a high correlation
of 0.85 between these variables in the model 1 sample, which probably accounts for the
insignificant result for conflict years.20 We report the conflict intensity variable in the
subsequent models, as the explanatory power of the models increases with this measure.
However, we have also run models with conflict years instead of conflict intensity as robustness
check. This does not alter our main findings (see Appendix A3).21 The neighborhood effect is
positive and significant, as expected, and the odds for seeing child soldier recruitment in
regions with conflict in a neighboring region is more than fourfold that of the odds of such
recruitment in regions that do not border conflict regions. Neither the log of the population size
nor population density have any statistically significant effect.
[Table 1 about here]
In the following models in Table 1 we keep the control variables that proved significant
in the baseline model, to keep the number of independent variables to a minimum, avoid
multicollinearity, and keep as large a sample as possible. A likelihood-ratio test confirms that
dropping the insignificant controls is possible without losing explanatory power. In Model 2 we
add our measure of absolute poverty measured by household assets from the DHS surveys.
19
We also test a different specification of conflict intensity by using dummies for low intensity conflicts (never
reaching 1,000 battle deaths in any year) and high intensity conflict (reaching 1,000 battle deaths in at least one of the conflict
years) with no conflict serving as a reference category. The results are similar: regions with high intensity conflicts are
significantly more likely to have reports of child soldier recruitment than regions without any conflict (with a probability of
16.8%).
20
Otherwise multicollinearity does not pose a problem in the reported models as the correlation matrix A1 in
Appendices indicates.
21
Whereas conflict years might be interpreted as a form of duration measure, it does not necessarily capture the actual
length of armed struggle, as often there are interruptions in conflict over time (unless there has been conflict in the region in all
the years included). Also, there is no necessary relationship between conflict duration and intensity, as Lacina (2006: 285) finds
that there is no significant relationship between deaths per year and conflict duration, although a longer conflict duration
increases the likelihood of a higher total death count for the conflict as a whole.
18
Contrary to our expectation, it seems as the higher the level of household assets in the region,
the more likely is child soldier recruitment. However, the effect of the asset variable is only
significant at the 10% level. This unexpected relationship could indicate that our summary
measure for economic wellbeing relating to the ownership of various household assets does not
fully capture the general structural economic differences in all countries. Goods other than
household assets may be more important indicators of economic distribution in many
developing countries, like for example land tenure. Alternatively, we suspect that the result is
driven by sample effects, given that we only have data for 22 countries on this variable. In
Model 3 we see that absolute poverty in terms of education years has no significant effect on
recruitment.
Model 4 includes a much larger sample of regions, covering all of the African regions
(N=688). Here, we find a positive effect of high infant mortality rates on child soldier
recruitment, significant at the 5% level. In other words, the larger sample result does provide
some evidence that poverty (measured here as IMR) has a positive effect on child soldier
recruitment, as suggested by Hypothesis 1. In a region with a refugee camp, a conflict in the
neighboring region and mean number of battle deaths, the probability of child recruitment
increases from about 5 % to more than 38,5% if infant mortality shifts from the lower 10th
percentile to the higher 90th percentile. This finding, however, is not robust to sample change.
Running Model 4 in Table 1 with the smaller sample from Models 2 and 3 (22 countries) does
not give a significant result on the IMR variable, and this term neither yields a significant
relationship when added to Models 2 and 3 in Table 1. The non-finding on IMR for the reduced
sample could indicate a sample bias. The correlation matrix (Appendix A2) indicates that the
three measures of absolute poverty are correlated in the expected way, and that particularly
education levels and infant mortality rates are highly correlated (–0.66). This is understandable,
as both variables relate to basic access to the social system.22 However, whether the reduced
sample is skewed in a way that influences the results can only be determined with increased
availability of geo-referenced DHS data in the future.
22
The two variables IMR and (female) education have also been found to be highly correlated in other studies, such as
Brockerhoff & Hewett (2000) who found that educational attainment of mothers is among the most important factors explaining
child survival.
19
The controls in models 2–4 perform largely in the expected way. Recruitment of child
soldiers is more likely in regions with refugee camps than regions without such camps, and
more likely in the regions with higher intensity. Running the models in Table 1 with alternative
specifications of conflict intensity does not alter this picture.23 However, conflict in a
neighboring region is positive but insignificant in models 2-4.
Figure 4 below shows the predicted probabilities of seeing child soldier recruitment in a
region based on Model 4 in Table 1. The graph also shows how the impact of having one or
more refugee camps in the region affects the recruitment likelihood at different rates of infant
mortality. The poorer the region (in terms of IMR), the higher the probability of child soldier
recruitment overall, but the effect is stronger for regions with refugee camp(s).
[Figure 4 about here]
In Table 2 we test our measures of relative deprivation or regional inequality. We find
that there is no significant difference between the relatively deprived regions and the more
privileged regions within countries. This is contrary to our expectation in Hypothesis 2. One
explanation for this non-finding on relative deprivation might be that recruitment also occurs
across international borders, one example being recruitment of Liberian children to the civil war
in Côte d’Ivoire. In future research a test of the impact of relative deprivation on child soldier
recruitment could be to investigate deprivation relative to neighboring regions rather than
comparing regions to the country average. Another possibility is that recruitment is more ad
hoc, depending on where actual battles are taking place. If battles occur in more developed
regions (i.e. for control over the capital) this could explain the non finding on relative
deprivation. In future research, one possibility is to test this possibility with battle location data
(Raleigh/Hegre, 2005) once most African conflicts have been coded.
[Table 2 about here]
23
We run a measure of the number of battle deaths in the peak year of conflict. We also run the models with dummies
for low intensity conflict (less than 1,000 battle deaths in any given year) and high intensity conflict (1,000 battle deaths in at
least one of the conflict years) with no conflict as reference category. The findings are consistent.
20
The refugee camp variable keeps its strong and significant effect on the probability of
child soldier recruitment in all models in Table 2. The presence of one or more refugee camps
in a region strongly increases the likelihood of recruitment of child soldiers. Not surprisingly,
both the dummy for regions with low or high intensity conflict exhibit an increased likelihood
of child soldier recruitment than non-conflict regions, whereas, like in Table 1, conflict in a
neighboring region does not show consistent significance in the models testing relative poverty.
5. Conclusion
Poverty is frequently offered as one of the main explanations for child soldiering in the popular
debate, and strong statements have been made about this claimed relationship in the academic
literature. However, the only systematic cross-national test we are aware of (Achvarina/Reich
2006) did not find a significant link between these factors. In our disaggregated study of
African regions, we do find some evidence for a positive impact of poverty measured in terms
of infant mortality rates, on child soldier recruitment, a relationship which can be masked in
national level studies. Regions with higher infant mortality rates are more likely to have
experienced recruitment of children as soldiers than regions with lower infant mortality rates,
controlling for several other factors. However, our alternative poverty measures -- education
and household assets -- did not yield any significant findings. This might be due to a sample
effect as disaggregated data on household assets and education levels are available only for 22
countries. Neither did any of our measures of relative poverty show any significant relationship
with child soldier recruitment, lending no support to our second hypothesis that recruitment will
take place more frequently in regions that are relatively worse off than other regions in a
country. As mentioned elsewhere, recruitment across national borders and/or possible
dependence on ad-hoc recruitment at the location of battles could help explain the absence of
empirical support for a relative deprivation–recruitment relationship.
Refugee camps can function as honey-pots and easy targets for groups wanting to recruit
children for combat. We found strong support for this relationship in our disaggregated study,
and the refugee camp effect indeed seems to outperform the poverty explanation of child soldier
recruitment, in line with the finding of Achvarina and Reich (2006). Being in a conflict zone,
21
proximity to conflict zones, and particularly the intensity of conflict also seem to be good
predictors of child soldier recruitment.
The research on child soldier recruitment, poverty, and spatial inequalities, as well as
the efforts to disaggregate studies of civil war are all picking up pace. The combined efforts of
these fields of research therefore show great promise for advances in future research. Access to
reliable data on both poverty figures and child soldier recruitment for a large sample of cases is
one of the most evident challenges for future research. In this study we are only able to cover
the African continent, or a sub-sample thereof. Moving from the national to the regional level
of analysis is a great improvement, but does not necessarily fully solve the problem of
ecological fallacy. In other words, we cannot be sure whether the poorer children within each
region are the more prone to becoming soldiers. Hence, future research could consider
disaggregating the relationship further, e.g. down to the district level, or simply compare
individual survey data from several countries. Furthermore, in the analyses presented in this
article the poverty measures are constant over time due to data constraints, meaning that at
present we can only perform a cross-sectional analysis. Ideally, the data on child soldier
recruitment should also offer more detail with regard to magnitude, i.e. the amount of children
recruited at various locations and at various times. However, the existing competing theoretical
arguments, conceivable logic, and evidence do not suggest that there is a single stage of conflict
that should be particularly at risk of child soldier recruitment. It is rather plausible that the
practice seems to manifest itself at any stage of the conflict. The identifiable patterns of child
recruitment can include the end of the conflict, the beginning of the conflict (e.g. NPFL in
Liberia)24, intermittent recruitment at conflict peaks (e.g. LURD in Liberia) (CSUCS, 2004), or
even regular recruitment throughout the conflict. The timing of child soldier recruitment can be
investigated further in future research should time variant recruitment data by location become
available for a cross country sample. For the purpose of uncovering the true causes of child
soldier recruitment, the ideal dataset would be based on survey data of a representative sample
of children from several countries, including both children who became soldiers and children
24
Laurent-Désiré Kabila of the Democratic Republic of Congo (DRC) enrolled thousands of children for his initial military
campaign against the Mobutu government in 1996 and 1997 (Human Rights Watch, 2001).
22
who did not (for a more thorough discussion on this see Ames 2008). However, no such data
currently exists.
From our findings one clear policy implication stands out. Although poverty might
certainly affect why children join armed groups and why they cannot protect themselves against
forced recruitment, the most efficient policies to stop child soldier recruitment are likely to be
those that specifically focus on protecting refugee camps from intrusion by armed groups.
Bibliography
Achvarina, Vera and Simon Reich, 2006: ‘No Place to Hide: Refugees, Displaced Persons, and
the Recruitment of Child Soldiers’, International Security 31(1): 127–164.
Ames, Barry, 2008: ‘Methodological Problems in the Study of Child Soldiers’ unpublished
manuscript.
Andvig, Jens Christopher, 2006: ‘Child Soldiers: Reasons for Variation in their Rate of
Recruitment and Standards of Welfare’, NUPI paper 704.
Andvig, Jens Christopher & Scott Gates, 2006: ‘Recruiting Children for Armed Conflict’, paper
presented at the APSA Meeting in Chicago, IL, 31 August–3 September.
Blattman, Christopher, 2007: ‘The Causes of Child Soldiering: Theory and Evidence from
Northern Uganda’, paper presented at the International Studies Association, Chicago, IL,
28 February–3 March.
Bøås, Morten and Ann Hatløy, 2008: ‘”Getting In, Getting Out”: Militia Membership and
Prospects for Re-Integration in Post-War Liberia’, Modern African Studies 46(1): 33–55.
Brett, Rachel and Margaret McCallin, 1996: Children: The Invisible Soldiers. Växjö, Sweden:
Rädda Barnen.
Brett, Rachel and Irma Specht, 2004: Young Soldiers: Why They Choose to Fight, ILO: Lynne
Rienner.
Brockerhoff, Martin and Paul Hewett, 2000: ‘Inequality of Child Mortality among Ethnic
Groups in Sub-Saharan Africa’, Bulletin of the World Health Organization 78(1): 30–41.
Bueno de Mesquita, Ethan, 2005: ‘The Quality of Terror’, American Journal of Political
Science 49(3): 515–530.
23
Buhaug, Halvard and Jan Ketil Rød, 2006: ‘Local Determinants of African Civil Wars, 1970–
2001. Political Geography 25(3): 315–35.
Buhaug, Halvard and Päivi Lujala, 2005: ‘Accounting for Scale: Measuring Geography in
Quantitative Studies of Civil War’, Political Geography 24(4): 399–418.
Collier, Paul & Anke Hoeffler, 2004: ‘Greed and Grievance in Civil War’, Oxford Economic
Papers 56(4): 563–95.
CSUCS (Coalition to Stop the Use of Child Soldiers), 2001: Child Soldiers. Global Report
2001 (http://www.child-soldiers.org/library/global-reports).
CSUCS (Coalition to Stop the Use of Child Soldiers), 2004: Child Soldiers. Global Report
2004 (http://www.child-soldiers.org/document_get.php?id=966).
Doyle, Michael W. and Nicholas Sambanis, 2000: ‘International Peacebuilding: A Theoretical
and Quantitative Analysis’, American Political Science Review 94(4): 779–801.
ESRI, 1998. World sub Country Administrative Units 1998:
(http://www.cdc.gov/EpiInfo/documents/shapes.doc).
Esty, Daniel C.; Jack A. Goldstone, Ted R. Gurr, Pamela T. Surko, and Alan N. Unger, 1995:
‘Working Paper: State Failure Task Force Report’, McLean, VA: Science Applications
International Corporation.
Esty, Daniel C; Jack A. Goldstone, Ted R. Gurr, Pamela T. Surko, Alan N. Unger, & RS Chen,
1998: ‘The State Failure Task Force Report: Phase II Findings’, McLean, VA: Science
Applications International Corporation.
Fearon, James D. and David D. Laitin, 2003: ‘Ethnicity, Insurgency, and Civil War’, American
Political Science Review 97(1): 75–90.
Gates, Scott, 2002: ‘Recruitment and Allegiance: The Microfoundations of Rebellion’, Journal
of Conflict Resolution 46(1): 111–30.
Gleditsch, Nils Petter Peter; Wallensteen, Michael Eriksson, Margareta Sollenberg, and Håvard
Strand, 2002: ‘Armed Conflict 1946-2001: A New Dataset’, Journal of Peace Research
39(5): 615–37.
Goldstone, Jack; Robert H. Bates, Ted R. Gurr, Michael Lustik, Monty G. Marshall, Jay
Ulfelder, and Mark Woodward, 2005: ‘A Global Forecasting Model of Political
Instability’, paper presented at the Annual Meeting of the Political Science Association,
Washington, DC, September 1–4.
24
Goodwin-Gill, Guy and Ilene Cohn, 1994: Child Soldiers: The Role of Children in Armed
Conflict. New York: Clarendon.
Guichaoua, Yvan, 2006: ‘Why do Youths Join Ethnic Militias? A Survey on the Oodua
People’s Congress in Southwestern Nigeria’, unpublished paper, Centre for Research on
Inequality, Human Security and Ethnicity, University of Oxford.
Gurr, Ted Robert, 1970: Why Men Rebel. Princeton, NJ: Princeton University Press.
Harbaugh, William T. and Kate Krause, 1999: ‘Children's contributions in public good
experiments: The development of altruistic and free-riding behaviors,’ University of
Oregon Department of Economics Working Papers.
Hegre, Håvard and Nicholas Sambanis, 2006: ‘Sensitivity Analysis of Empirical Results on
Civil War Onset’, Journal of Conflict Resolution 50(4): 508–35.
Hegre, Håvard and Clionadh Raleigh, 2005: ‘Population Size, Concentration, and Civil War: A
Geographically Disaggregated Analysis’, paper presented at the conference Mapping the
Complexity of Civil Wars, Zürich, 15–17 September (http://www.icr.ethz.ch/mccw/
papers/hegre.pdf).
Honwana, Alcinda, 2006: Child Soldiers in Africa. Philadelphia, PA: University of
Pennsylvania Press.
Human Rights Watch, 2001: Democratic Republic Of The Congo Reluctant Recruits: Children
And Adults Forcibly Recruited For Military Service In North Kivu, Vol 13(3).
Humphreys, Macartan and Jeremy M. Weinstein, 2004: ‘What the Fighters Say: A Survey of
Ex-Combatants in Sierra Leone’, CGSD Working Paper 20.
International Labour Organization (ILO), 2003: Wounded Childhood: the Use of Children in
Armed Conflict in Central Africa. Geneva, Switzerland: ILO.
Lacina, Bethany Ann & Nils Petter Gleditsch, 2005: ‘Monitoring Trends in Global Combat: A
New Dataset of Battle Deaths’, European Journal of Population 21(2–3): 145–165.
Machel, Graça, 1996: ‘Impact of Armed Conflict on Children’, UNICEF Report
(http://www.unicef.org/graca/).
McManimon, Shannon, 1999: ‘Use of Children as Soldiers’, Foreign Policy in Focus 4(27)
(http://www.fpif.org/briefs/vol4/v4n27child_body.html).
Østby, Gudrun, 2008: ‘Polarization, Horizontal Inequalities and Violent Civil Conflict’,
Journal of Peace Research 45(2): 143–162.
25
Østby, Gudrun; Ragnhild Nordås, and Jan Ketil Rød, 2008: ‘Regional Inequalities and Civil
Conflict in Sub-Saharan Africa, forthcoming in International Studies Quarterly.
Pugel, James, 2007: What the Fighters Say: A survey of Ex-Combatants in Liberia. Monrovia,
Liberia: United Nations Development Programme.
Raleigh, Clionadh and Håvard Hegre, 2005: ‘Introducing ACLED: An Armed Conflict
Location and Event Dataset’, unpublished manuscript (http://new.prio.no/CSCWDatasets/Data-on-Armed-Conflict/ACLED---Armed-Conflict-Location-and-Event-Data/).
Raleigh, Clionadh and Henrik Urdal, 2007: ‘Climate Change, Environmental Degradation and
Armed Conflict’, Political Geography 26(6): 674–694.
Singer, Peter W., 2005: Children at War. New York: Pantheon.
Stewart, Frances, 2000: ‘Crisis Prevention: Tackling Horizontal Inequalities’, Oxford
Development Studies 28(3): 245–62.
Stewart, Frances, 2002: ‘Horizontal Inequalities: A Neglected Dimension of Development’,
Working Paper Number 81, Queen Elizabeth House, University of Oxford.
(http://www2.qeh.ox.ac.uk/research/qehwplist.html?jorseries=WPS&jorcode1=2002).
Stewart, Frances and Jo Boyden, 2001: ‘Policy to Protect Children From and During War’, in
Harnessing Globalisation for Children: A Report to UNICEF (http://www.uniceficdc.org/research/ESP/globalization/index.html)
Theisen, Ole Magnus and Kristian Bjarnøe Brandsegg, 2007: ‘The Environment and Non-State
Conflicts’, paper presented at the Fifteenth Norwegian National Conference in Political
Science, Rica Nidelven Hotel, Trondheim, 3–5 January.
Tomz, Michael, Jason Wittenberg and Gary King, 2003: CLARIFY: Software for Interpreting
and Presenting Statistical Results. Stanford University, University of Wisconsin, and
Harvard University (http://gking.harvard.edu/clarify/clarify.pdf).
UN, 2003. Thirteenth report of the Secretary-General on the United Nations Organization
Mission
in
the
Democratic
Republic
of
the
Congo.
21
February
2003.
(http://daccessdds.un.org/doc/UNDOC/GEN/N03/249/65/IMG/N0324965.pdf?OpenElem
ent)
UNICEF, 2002: Adult Wars, Child Soldiers. Bangkok: UNICEF
(http://www.unicef.org/emerg/AdultWarsChildSoldiers.pdf).
UNICEF, undated: ‘Fact Sheet: Child Soldiers.
26
(http://www.unicef.org/protection/childsoldiers.pdf).
Weidmann, Nils B.; Patrick Kuhn, and Varja Nikolic, 2007: ‘Refugees as Local Catalysts of
Conflict? A Statistical Assessment’, paper prepared for the Annual Convention of the
International Studies Association, Chicago, IL, 28 February–3 March.
Wessels, Michael G., 1998: ‘Review: Children, Armed Conflict and Peace’, Journal of Peace
Research 35(5): 635–46.
Younger, Stephen D., 2004: ‘Growth and Poverty Reduction in Uganda, 1992–1999: A
Multidimensional Analysis of Changes in Living Standards’, paper presented at the
conference “Growth, Poverty Reduction and Human Development in Africa”, CSAE,
Oxford, 21–22 March.
27
Table 1. Logit Regression of Child Soldier Recruitment and Absolute Poverty,
African Regions 1990-2004
Refugee camp
Battle deaths (log)
Conflict in neighboring region
Conflict years
Regional population (log)
Regional pop. density (log)
(1)
2.003***
(3.89)
0.271***
(3.47)
1.454*
(1.85)
-0.018
(0.29)
0.072
(0.34)
-0.025
(0.22)
Household assets
(2)
2.609***
(5.69)
0.279***
(3.21)
1.251
(1.49)
(3)
2.442***
(5.46)
0.321***
(3.12)
1.117
(1.38)
3.653*
(1.94)
Education years
0.190
(1.14)
Infant mortality rates (log)
Constant
Regions
Countries
Pseudo R2
(4)
1.598***
(3.97)
0.226***
(3.56)
0.725
(0.93)
-6.808**
(2.39)
626
49
.325
-6.273***
(6.33)
354
22
.394
-6.028***
(5.60)
354
22
.389
1.537**
(2.28)
-11.753***
(3.50)
688
52
.340
Robust z statistics in parentheses, absolute numbers. Huber-White clustering on country. * significant at 10%;
** significant at 5%; *** significant at 1%.
28
Table 2. Logit Regression of Child Soldier Recruitment and Relative Poverty
Deprivation, African Regions 1990–2004
Refugee camp
Battle deaths (log)
Conflict in neighboring region
Relative deprivation (household assets)
(5)
(6)
(7)
2.456***
(5.68)
0.280***
(2.97)
2.456***
(5.68)
-0.142
(0.39)
2.421***
(5.52)
0.281***
(2.97)
2.421***
(5.52)
1.923***
(4.27)
0.231***
(3.35)
1.923***
(4.27)
Relative deprivation (education years)
-0.217
(1.07)
Relative deprivation (infant mortality)
Constant
Regions
Countries
Pseudo R2
-5.118***
(7.40)
354
22
.378
-5.070***
(7.29)
354
22
.379
0.179
(0.17)
-5.312***
(7.75)
688
52
.299
Robust z statistics in parentheses, absolute numbers. Huber–White clustering on country. * significant at 10%;
** significant at 5%; *** significant at 1%.
29
Figure 1. Poverty and Child Soldier Recruitment in Chad (left) and Uganda (right)
Darker areas relatively deprived in terms of household assets
The darker the region, the higher the infant mortality rates. Hatched:
compared to country mean. Hatched: Child soldier recruitment.
Child soldier recruitment.
30
Figure 2. Child Soldier Recruitment and Conflict Zones, Africa, 1990–2004
31
Figure 3. Infant Mortality Rates (CIESIN) and Child Soldier Recruitment, 1990–
2004
32
Predicted probability of child soldier recruitment
0
.2
.4
.6
Figure 4. Predicted Probabilities of Child Soldier Recruitment Based on Infant
Mortality Rates (CIESIN) and Refugee Camp(s) in Region, 1990–2004
0
50
100
Infant mortality rates
150
200
Note 1: Upper solid line indicates the existence of one or more refugee camps in the region; lower solid
line refers to no refugee camp in the region. Solid lines are the local polynomial regression fits calculated with
bandwidth of 10. Dashed lines depict 95% confidence bands. The graph is based on Model 4, Table 1. The
confidence bands are relatively close to the non-parametric regression fits. This indicates the statistical significance
of the impact of having one or more refugee camps in the region on the recruitment likelihood at different rates of
infant mortality.
33
APPENDIX
A1. Descriptive statistics, all variables used in analysis.
Variable
Dependent variable
Child solder recruitment
Absolute Poverty
Household assets
Education years
Infant Mortality Rate (IMR) (log)
Relative Deprivation
Rel. Depr. Household assets
Rel. Depr. Education yrs.
Rel. Depr. IMR (log)
Controls
Refugee camp in region
Battle deaths (log)
Conflict in neighboring region
Conflict years
Region population size (log)
Region population density (log)
N (regions)
Mean
690
0.106
354
354
688
0.218
2.902
4.288
354
354
688
0.199
0.396
–0.015
690
690
690
690
626
626
0.232
3.772
0.670
2.922
13.037
3.407
Min
Max
0
1
0.116
2.496
0.613
0.03
0.01
2.08
0.65
9.67
5.20
0.437
0.805
0.179
–0.76
–1.07
–1.61
1.81
5.21
0.58
0
0
0
0
8.07
–2.84
1
11.91
1
15
16.36
10.78
Std. Dev.
4.577
4.653
1.429
1.967
34
A2. Correlation Matrix
Child
soldier
recruitment
Child soldier recruit.
Household assets
Education years
IMR
RD (assets)
RD (educ.)
RD (IMR)
Refugee camp
Battle deaths
Conflict in neighbor reg.
Conflict years
Regional pop.
Regional pop. density
1
.106
–.017
.148
–.001
–.082
.003
.434
.392
.229
.389
–.057
.028
Assets
1
.487
–.311
–.772
–.386
–.093
–.028
–.053
–.211
–.139
–.263
.055
Educ.
years
1
–.676
–.480
.587
–.190
.030
–.279
–.323
–.284
–.039
–.037
IMR
1
.228
.193
.431
.148
.419
.387
.392
.128
.222
RD
(assets)
1
.576
.142
–.044
.115
.188
.164
.086
–.112
RD
(educ.)
1
.199
–.161
–.012
.040
.060
.004
–.085
RD
(IMR)
1
.034
.100
.140
.100
–.070
–.089
Refugee
camp
1
.205
.143
.267
.158
.028
Battle
deaths
1
.539
.810
–.006
.019
Conflict
in
neighbor
region
1
.412
.105
.028
Conflict
years
1
–.008
–.059
Reg.
pop.
1
.670
N=302
35
A3. Logit Regression of Child Soldier Recruitment and Absolute Poverty,
African Regions 1990–2004. Table 1 replacing conflict intensity with conflict
years.
Refugee camp
Conflict in neighboring region
Conflict years
Regional pop. (log)
Regional pop. density (log)
Household assets
Education years
(A1)
1.981***
(3.76)
2.552***
(2.63)
0.135***
(2.72)
0.179
(0.71)
–0.085
(0.67)
(A2)
2.465***
(6.05)
2.234***
(2.78)
0.166*
(1.72)
(A3)
2.306***
(6.42)
1.996***
(2.76)
0.172
(1.62)
(A4)
1.553***
(3.86)
1.604**
(2.28)
0.127***
(3.17)
4.101**
(2.38)
0.089
(0.65)
Infant mortality rates (log)
1.805**
(2.52)
Constant
–8.020**
–6.315***
–5.418***
–12.914***
(2.41)
(6.53)
(6.78)
(3.60)
Regions
626
354
354
688
Countries
49
22
22
52
Pseudo R2
.282
.341
.322
.307
Robust z statistics in parentheses. Huber-White clustering on country. * significant at 10%; ** significant at
5%; *** significant at 1%
6
`