The Consequences of Child Soldiering

TL;DR: This article examined the case of Uganda, where rebel recruitment methods provided exogenous variation in conscription and found that schooling falls by nearly a year, skilled employment halves, and earnings drop by a third.

Abstract: Little is known about the impacts of military service on human capital and labor market outcomes due to an absence of data as well as sample selection: recruits are self-selected, screened, and selectively survive. We examine the case of Uganda, where rebel recruitment methods provide exogenous variation in conscription. Economic and educational impacts are widespread and persistent: schooling falls by nearly a year, skilled employment halves, and earnings drop by a third. Military service seems to be a poor substitute for schooling. Psychological distress is evident among those exposed to severe war violence and is not limited to ex-combatants.

Summary

Introduction

• THIS paper assesses the impact of combat on the humancapital of Ugandan youth, the consequences for lifetime labor market performance, and lessons for the economic recovery of civil war–torn countries.
• To overcome this problem, the authors conducted a survey during the war in northern Uganda, where for 20 years, an unpopular rebel group has forcibly recruited tens of thousands of youth.
• The results suggest that the largest and most pervasive impact of abduction is on education and earnings, largely due to time away from civilian schooling and work experience.
• Unfortunately their counterfactual— nonabducted youth in the war zone—does not help to identify the impact of war on noncombatants.

A. War and Abduction in Northern Uganda

• Historically Uganda’s economic power rested in the south, while political and military power came from the north (Omara-Otunnu, 1994).
• Several Acholi guerrilla forces resisted the takeover, but settled for peace or were defeated by 1988.
• LRA activity was initially low, but in 1994 and 1995, in response to Uganda’s support for Sudanese rebels, the government of Sudan began supplying Kony with weapons and territory on which to build bases.
• Youth were typically taken by roving groups of ten to twenty rebels during night raids on rural homes.

B. Current Practice and Evidence

• One Ugandan study compared abducted to nonabducted youth in the war zone and concluded that abducted youth were more anxious and depressed, more hostile, less prosocially active, and less confident (MacMullin & Loughry, 2002).
• Conclusions remain uncertain, however, because few studies consider endogeneity problems and measuring trauma and distress across cultures is challenging at best.
• Some require little capital or skill (such as collecting firewood), while others require a little capital (hawking goods), moderate capital (a bicycle taxi), or substantial capital and skills .
• This labor market is dynamic, and as youth accumulate skills and funds, they shift to more productive work.

III. Data and Measurement

• In 2005 and 2006 the authors conducted phase 1 of the Survey of War Affected Youth (SWAY), a survey of 741 males born between 1975 and 1991 in one of eight rural subcounties in the districts of Kitgum and Pader.
• The authors randomly sampled 1,100 households from U.N. World Food Programme lists compiled in 2002.
• The authors chose the year 1996 because it was easily recalled as the date of the first election since 1980 and because it predates 85% of local abductions.
• A sample of 870 surviving male youth was drawn from this retrospective roster.
• First, war experiences are self-reported and retrospective.

IV. Empirical Strategy

• The authors develop a comparison group of nonabducted youth in the war zone who for largely exogenous reasons were never abducted.
• This counterfactual is a crucial one if the authors are interested in addressing reintegration gaps—that is, closing any inequality between combatants and noncombatants, or providing reparations beyond those received by other war-affected populations.

A. Dealing with Endogenous Selection into the Armed Group

• The standard solution is the counterfactual approach, where a relevant control group is found and the average treatment effect (ATE) is estimated by taking the difference in the outcomes of the treated and controls (Rubin, 1974; Imbens & Wooldridge, 2008).
• First, self-selection into the armed group was nonexistent in the subcounties the authors surveyed.
• Abducted and nonabducted youth differ only in mean year of birth and mean prewar household size.
• Moreover, abduction levels varied over the course of the war, so youth of some ages were more vulnerable to abduction than others.

B. Dealing with Selective Attrition and Survival

• The tracking success rate of this study meets or exceeds the rates achieved by several gold standard panel surveys in poor countries (Thomas, Frankenberg & Smith 2001; Baird, Hamory & Miguel, 2008; Humphreys & Weinstein, 2008).
• Attrition patterns vary by treatment status: abductees are half as likely to be unfound migrants, twice as likely to have perished, and comprise all of those who did not return from abduction.
• Moreover, those who remain with the armed group are likely missing more education and work experience than the average returnee, also contributing to potential underestimation of their economic and educational ATEs.
• Accordingly, the authors employ a method of sensitivity analysis proposed by Lee (2005) whereby best-case and worst-case scenarios for differential attrition are constructed by trimming the distribution of the outcome in the group with less attrition (in this case, the nonabducted).

C. Estimation

• Assuming conditional unconfoundedness, the most efficient and consistent approach to ATE estimation is a weighted least squares (WLS) regression with weighting on the inverse of a nonparametric estimate of the propensity score (Hirano, Imbens, & Ridder, 2003).
• 4 Data on attritors usually come from a prior round of survey data, while in this case, the data are provided by families.
• The availability of data on current activity, however, is a distinct advantage over attrition-correction methods that rely on data from prior surveys.
• Note that the ATE estimated in equation (1) will be biased if abduction has positive or negative externalities on nonabducted youth.
• The authors address this concern after seeing the results.

A. Average Treatment Effects

• Abduction ATEs for ten outcomes are listed in table 3.
• Each entry in column 1 represents the coefficient on an abduction indicator, column 2 lists the mean level of each outcome among nonabducted youth, and column 3 calculates the proportional impact of the ATE relative to the nonabducted mean.
• This literacy ATE seems large given the ATE in schooling.
• If abduction is associated with the propensity to be employed or earn nothing, the authors will conflate the direct impact of abduction on wages with the indirect effects on the type of people employed (Heckman, 1979; Lee, 2005).
• Abducted youth are 11 percentage points more likely to be in the top quartile of the distress index—a 49% increase relative to the nonabducted.

C. Externalities from Abduction

• Or abductions and displacement could lead to positive externalities as nonabducted populations are urbanized in displacement camps (closer to schools) or displaced to towns (closer to markets).
• Table 4 displays the coefficient on abduction intensity in regressions of each outcome on intensity, year and location of birth, and prewar controls.
• Yet the authors see no difference in social support between abducted and nonabducted youth.
• Nonabducted youth were more likely to migrate, and those who did earned higher wages than their rural counterparts.

D. Child versus Adult Combatants

• The results, displayed in table 5, suggest little significant difference in outcomes for children versus adult abductees.
• Moreover, even if a youth actively avoids traumatic memories, this repression will often manifest itself in other symptoms.
• As a result, the authors would expect to see greater educational and labor market impacts for child soldiers than adult ones.
• One reason the authors see little variation in outcomes by age is that younger abductees are more likely to return to school on return: two-thirds of youth abducted before the age of 18 return to school versus less than a third of those abducted over that age.

E. Sensitivity Analysis

• A remaining concern is the potential for unobserved selection and bias.
• Only year and location of birth—the primary determinants of selection by the armed group—meet or cross this hypothetical threshold.
• Each þ represents a prewar covariate, plotted according to its additional explanatory power for treatment assignment (on the horizontal axis) and its explanatory power for the outcome (vertical axis), which in this case is educational attainment.
• The ATEs under the worst-case scenario are generally closer to 0 and less than robust than the untrimmed ATEs.
• The results imply that under austere and implausible dramatic selection, abduction still has the predicted effect on outcomes, albeit at a lower level of statistical significance.

VI. Heterogeneous Treatments and the Channel of Impact

• This approach, however, obscures the diversity of experiences and the true treatment received.
• In particular, abduction length ranged from 1 day to 10 years, and violence varied dramatically.
• When treatment is heterogeneous, the binary ATE can be interpreted as the average per unit effect along a response function mapping treatment exposure to outcomes (Angrist & Imbens, 1995).
• One might prefer, however, to estimate the entire response function, considering abduction length or violence the ‘‘true’’ treatment.
• Even so, estimating the (potentially biased) relationship between their measures of treatment exposure and outcomes is useful for understanding the underlying causal channels and is more easily generalized.

A. Psychosocial Outcomes

• The evidence suggests that youth who exhibit the most serious symptoms of psychosocial distress are generally those who experienced the greatest war violence, as discussed in detail by Annan and Blattman (2009).
• The indices of violence are linear and additive, based on selfreported indicators for six different acts witnessed by the youth (for example, rape and killings of others), six acts inflicted on the youth himself (for example, beatings, imprisonment), and five acts on his family (for example, abduction, war injury, killing).
• After controlling for violence, longer abductions are not robustly associated with higher distress.
• Turning to social support, their index is decreasing in abduction length.
• An obvious possibility is that longer lengths of time away reduce social and family ties.

B. Educational and Labor Market Outcomes

• Time away from human capital accumulation rather than violent trauma may account for the persistent educational effects of abduction.
• For a discussion of estimation bias regarding schooling, see Card (1999) and for health, see Strauss and Thomas (1998).
• The relationship between wages and violence does not appear to be driven by injuries; controlling for an injury indicator or an indicator for being in the top quartile of distress (regressions not shown).
• This failure to observe a relationship between abduction length and wages is puzzling, and suggests that a reduction in education and experience may not be the only causal channel by which abduction affects labor market outcomes.

VII. Conclusion

• New data and a tragic natural experiment provide some of the first estimates of the nature, magnitude, and distribution of the effects of civil war on youth.
• With distress and aggression concentrated, more targeted and specialized psychosocial services are probably needed for those who have experienced the most violence (abducted or not).
• The authors interpret their findings as the incremental effect of conscription in communities already subjected to the horrors of war and abduction, an important quantity of interest if they wish to target aid and development assistance to the people who need them most.
• Child labor or an event like the Cultural Revolution can offer the ability to acquire useful human and physical capital, muting the long-term economic impact.
• For this research to be accurate and comparable, greater attention ought to be paid to representative samples, accounting for attrition, and the careful identification of comparison groups.

Figures

TABLE 2.—COMPARISON OF MEANS

FIGURE 5.—IMPACT OF RELAXING THE ASSUMPTION OF UNCONFOUNDEDNESS

TABLE 5.—VARIATION IN THE AVERAGE TREATMENT EFFECT BY AGE OF ABDUCTION

FIGURE 3.—DISTRIBUTIONS OF THE PREDICTED PROBABILITY OF ABDUCTION BASED ON AGE AND LOCATION ALONE VERSUS ALL PRETREATMENT COVARIATES (BY ABDUCTION STATUS)

TABLE 4.—SPILLOVER EFFECTS OF ABDUCTION INTENSITY

TABLE 8.—THE CORRELATES OF WAGES AND THE DECOMPOSITION OF THE WAGE ATE

TABLE 7.—ESTIMATES OF THE IMPACT OF ABDUCTION LENGTH AND VIOLENCE ON OUTCOMES

TABLE 3.—ESTIMATES OF THE AVERAGE TREATMENT EFFECT OF ABDUCTION

FIGURE 1.—MAP OF UGANDA AND SURVEY FIELD SITE

TABLE 6.—TREATMENT EFFECT BOUNDING FOR SELECTIVE ATTRITION

FIGURE 4.—QUANTILE DIFFERENCES IN THE DISTRIBUTION OF PSYCHOLOGICAL DISTRESS BETWEEN ABDUCTED AND NONABDUCTED YOUTH

FIGURE 2.—DISTRIBUTION OF ABDUCTIONS BY AGE AT THE TIME OF ABDUCTION

Full text

THE CONSEQUENCES OF CHILD SOLDIERING
fn16
Christopher Blattman and Jeannie Annan*
Abstract—Little is known about the impacts of military service on human
capital and labor market outcomes due to an absence of data as well as
sample selection: recruits are self-selected, screened, and selectively sur-
vive. We examine the case of Uganda, where rebel recruitment methods
provide exogenous variation in conscription. Economic and educational
impacts are widespread and persistent: schooling falls by nearly a year,
skilled employment halves, and earnings drop by a third. Military service
seems to be a poor substitute for schooling. Psychological distress is evi-
dent among those exposed to severe war violence and is not limited to ex-
combatants.
I. Introduction
T
HIS paper assesses the impact of combat on the human
capital of Ugandan youth, the consequences for life-
time labor market performance, and lessons for the eco-
nomic recovery of civil war–torn countries. Civil conflict
has afflicted a third of all nations and two-thirds of Africa
since 1991 (Marshall & Gurr, 2005). The recovery of chil-
dren and young adults is a critical concern in these postwar
economies; lost education can take years to regain, and phy-
sical and psychological trauma may be long-lasting. Some
of these conflicts involve up to a third of male youth in
active combat, many of whom are children under age 18.
With so many millions of young ex-combatants, injuries to
human capital could hinder a nation’s productivity and
growth for decades. Moreover, any impact of military ser-
vice on inequality, aggression, and alienation could threaten
a nation’s long-term stability.
The effects of combat are uncertain. The dominant view
holds that these youth are traumatized, violent, social par-
iahs. Speaking at a 2007 U.N. conference, the French for-
eign minister warned that young ex-combatants are ‘‘a time
bomb that threatens stability and growth, ... lost for peace
and lost for the development of their countries’’ (BBC,
2007). Former child soldiers in particular are ‘‘damaged,
uneducated pariahs,’’ says a New York Times editorial
(‘‘Armies of Children,’’ 2006). A growing body of ethno-
graphic evidence portrays another view, however, finding
that resilience rather than disabling psychological trauma is
the norm (Shepler, 2005; Boothby, Crawford, & Halperin,
2006; Wessells, 2006).
Virtually no representative data or well-identified causal
estimates exist to judge either set of claims (Blattman &
Miguel, 2010). A small literature has found large and persis-
tent earnings and mortality gaps between veterans and non-
veterans in the United States and Europe (Hearst, Newman,
& Hulley, 1986; Angrist, 1990, 1998; Angrist & Krueger,
1994; Imbens & van der Klaauw, 1995). For instance,
Angrist (1990) finds that white American males conscripted
into the Vietnam War saw a 15% reduction in their long-
term earnings due to work experience lost, while Costa and
Kahn (2010) use data on U.S. Civil War veterans to link
wartime stress to higher mortality later in life. The impacts
of combat, however, are not uniformly bad; both U.S. stu-
dies, for instance, find that military service increased the
human capital and lifetime earnings of black Americans—
sad evidence of their otherwise poor alternatives.
Unfortunately, these findings do not generalize easily to
the developing countries where most civil wars rage. They
also draw on data sets with a limited range of outcomes,
and so the full nature of the impact (and the causal channel)
is hard to see. The sole survey-based study of the impact of
civil war combat comes from a pioneering survey of Sierra
Leonean ex-combatants by Humphreys and Weinstein
(2004, 2007). They find that increased exposure to violence
is associated with lower community acceptance but not with
employability. Without a noncombatant comparison group,
however, our understanding of the impact of military ser-
vice remains incomplete.
One reason we know so little about these impacts is the
paucity of data in war zones. To overcome this problem, we
conducted a survey during the war in northern Uganda,
where for 20 years, an unpopular rebel group has forcibly
recruited tens of thousands of youth.
A second challenge in identifying the causal effects of
military service is selection: ex-fighters are usually a select
group, including those who chose to join and those screened
by the armed force. The ideal research design would be one
where rebel participation was randomly or exogenously
assigned. We argue that forced recruitment in Uganda
resembles just such a terrible case. Under the assumption
that abduction is conditionally unconfounded, causal
impacts can be estimated using noncombatants of the same
year and place of birth as counterfactuals. Naturally, unob-
served sources of selection and survival could bias the
results. Hence we explicitly model the sensitivity of the
treatment effects to unobserved selection to show that mod-
erate to large amounts of selective abduction and attrition
cannot account for the causal effects or change our general
conclusions.
Received for publication May 1, 2008. Revision accepted for publica-
tion March 6, 2009.
* Blattman: Yale University, Political Science and Economics; Annan:
Yale University, School of Medicine and CIRA.
Macartan Humphreys, Edward Miguel, Chalmer Thompson, and Jeremy
Weinstein deserve special thanks for their input and guidance. For com-
ments, we also thank David Albouy, Tim Allen, Matias Cattaneo, Bryan
Graham, Chang-Tai Hsieh, Guido Imbens, Pamela Jakiela, Seema Jaya-
chandran, Stathis Kalyvas, Kory Kroft, David Lee, David Leonard, David
Lynch, Lauren Morris-MacLean, Devin Pope, Gerard Roland, Amos Saw-
yer, Thomas Sexton, Harvey Weinstein, our anonymous referees, and
numerous seminar participants. For data collection, we thank Roger Hor-
ton, Okot Godfrey, and our field research assistants. Robert Blair provided
excellent research assistance. Logistical support was supplied by AVSI
Uganda and UNICEF Uganda. Military escorts were provided by the
Uganda People’s Defense Force. Financial support was primarily received
from UNICEF, the John D. and Catherine T. MacArthur Foundation, the
U.S. Institute of Peace, and the Harry Frank Guggenheim Foundation.
The Review of Economics and Statistics, November 2010, 92(4): 882–898
Ó
2010 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology

The results suggest that the largest and most pervasive
impact of abduction is on education and earnings, largely
due to time away from civilian schooling and work experi-
ence. This educational deficit impedes labor market suc-
cess: while abducted youth are just as likely to be
employed, they are half as likely to be engaged in skilled
work and earn wages that are lower by a third.
The data also support a growing body of ethnographic
evidence that finds ex-soldiers to be socially and psycholo-
gically resilient. Community acceptance of former abduc-
tees is high, and they report similar levels of social support
as do nonabductees. Abductees also exhibit little difference
in aggression. Finally, we highlight evidence that formerly
abducted youth do report 15% more symptoms of emotional
distress, in large part because of exposure to more extreme
violence (Annan & Blattman, 2009). Considering the high
incidence of violence received and perpetrated, this result
speaks to a remarkable resilience among youth, even among
those youth reporting the highest levels of distress, a con-
clusion bolstered by in-depth interview evidence (Annan,
Brier, & Aryemo, 2008).
These results challenge the conventional assumptions
about ex-combatants. Unfortunately our counterfactual—
nonabducted youth in the war zone—does not help to iden-
tify the impact of war on noncombatants. Rather, our
approach assesses the added impact of military service on
youth already in a war zone. This incremental effect is
important in order to address postwar gaps in reintegration,
especially when so many policymakers and aid agencies
appear to be ignoring the largest gaps. We conclude with
policy lessons for ex-combatant reintegration and postcon-
flict recovery.
II. Background
A. War and Abduction in Northern Uganda
Historically Uganda’s economic power rested in the
south, while political and military power came from the
north (Omara-Otunnu, 1994). In 1986, however, southern
rebels overthrew a government and army dominated by a
northern ethnic group, the Acholi. Several Acholi guerrilla
forces resisted the takeover, but settled for peace or were
defeated by 1988. A handful of these fighters refused to set-
tle, however, and gathered under a spiritual leader named
Joseph Kony to form the Lord’s Resistance Army (LRA;
Doom & Vlassenroot, 1999; Allen, 2005). Like many other
armed group leaders in Africa, Kony is widely believed to
possess spiritual powers. He claims to seek a spiritual
cleansing of Uganda and theocratic rule.
The decision to continue fighting was an unpopular one,
however, and the LRA received little public support. With
few recruits and no material resources, the LRA immediately
took to looting homes and abducting youth to maintain sup-
plies and force. The Acholi populace, after three years of
such abductions and looting, began to join a government-
sponsored defense militia in 1990. To punish this betrayal
and dissuade further collaboration, in 1991 Kony ordered the
widespread killing and mutilation of Acholi civilians, further
alienating the population (Behrend, 1999; Branch, 2005).
LRA activity was initially low, but in 1994 and 1995, in
response to Uganda’s support for Sudanese rebels, the gov-
ernment of Sudan began supplying Kony with weapons and
territory on which to build bases. Sudan’s support invigo-
rated the LRA, and rebel attacks and abductions escalated
dramatically after 1996.
Abduction was on a large scale and seemingly indiscri-
minate; 60,000 to 80,000 youth are estimated to have been
abducted (Annan, Blattman, & Horton, 2006; Pham, Vinck,
& Stover, 2007), and more than a quarter of males currently
aged 14 to 30 in our study region were abducted for at least
two weeks (table 1). Most were abducted after 1996 and
from one of the Acholi districts of Gulu, Kitgum, or Pader
(figure 1).
Youth were typically taken by roving groups of ten to
twenty rebels during night raids on rural homes. Adolescent
males appear to have been the most pliable, reliable, and
effective forced recruits, and so were disproportionately tar-
geted by the LRA (Beber & Blattman, 2008). Youth under
age 11 and over 24 tended to be avoided, as seen in figure
2, and had a high probability of immediate release. Lengths
of abduction ranged from 1 day to 10 years, averaging 8.9
months in our sample (table 1). Youth who failed to escape
were trained as fighters and, after a few months, received a
gun. Two-thirds of abductees were forced to perpetrate
a crime or violence. A third eventually became fighters, and
a fifth were forced to murder soldiers, civilians, or even
family members in order to bind them to the group, reduce
their fear of killing, and discourage disobedience.
Eighty-four percent of abductees eventually escaped,
usually in an unsupervised moment such as the heat of bat-
tle. The remainder perished, as no more than 1,000
abducted youth (about 1% of all abductees) are thought to
remain with the LRA at this time. An amnesty has been
granted to all ‘‘returnees’’ and community acceptance rates
are high; fewer than 2% report insults or fear from their
community and family.
B. Current Practice and Evidence
The focus on trauma in the wake of war and disaster is
pervasive, and youth postconflict programs concentrate
heavily on psychosocial care (Cohn & Goodwin-Gill, 1994;
Machel, 1996; ILO, 2003; CSUCS, 2005; Wessells, 2006).
In Uganda, foreign aid has concentrated on psychosocial
programs for former abductees, a focus driven by abundant
anecdotal evidence of social rejection and trauma. For
example, one youth we interviewed is haunted by being
forced to kill his brother: ‘‘I started dreaming of him a week
after the incident, and at times I would see him during the
day. How I beat him would all re-surface.’’ Frequently
reliving such events through nightmares or flashbacks is a
883THE CONSEQUENCES OF CHILD SOLDIERING

common symptom following traumatic events, and night-
mares are one of the most commonly reported symptoms of
distress in the sample (Annan & Blattman, 2009).
Some psychological studies emphasize high levels of trau-
matic stress among child soldiers. One Ugandan study com-
pared abducted to nonabducted youth in the war zone and
concluded that abducted youth were more anxious and
depressed, more hostile, less prosocially active, and less con-
fident (MacMullin & Loughry, 2002). A second study identi-
fied clinical posttraumatic stress in 97% of abductees (Der-
luyn et al., 2004). Both studies, however, used nonrandom
convenience samples and did not discuss potential selection
effects. The validity of their conclusions is thus questionable.
In contrast, other literature on child and refugee mental
health emphasizes the resiliency of victims of traumatic
events, as well as the concentration of disabling trauma in a
minority (Miller & Rasco, 2004; Wessells, 2006). Conclu-
sions remain uncertain, however, because few studies con-
sider endogeneity problems and measuring trauma and dis-
tress across cultures is challenging at best. Hollifield et al.
(2002) review 394 studies of war trauma and conclude that
most ‘‘are either descriptive or include quantitative data
from instruments that have limited or untested validity and
reliability.’’
Presurvey qualitative fieldwork suggested that disabling
distress was the exception. Forty youth were selected for in-
depth interviews and assessments, including multiple inter-
views of the youth and his family, friends, teachers, and co-
workers. These interviews suggested that aggression was
low and that social reintegration and psychological resili-
ence were widespread (Annan et al., 2008). Symptoms of
distress that interfered with daily functioning, a defining ele-
ment of psychological disorder, seemed to be concentrated
in a minority and were not limited to former abductees.
Rather, in the minds of many of abducted youth and their
families, the interruption of education and employment was
of greater concern. Youth complained of difficulty reenter-
ing the school system, creating an education gap that lim-
ited their options in the labor market. According to one
elder, ‘‘The youth who have not been abducted are engaged
in different activities like business and vocational work like
carpentry, because they had the opportunity to acquire the
different skills.’’
Youth earn income mainly through small entrepreneurial
activities. Some require little capital or skill (such as col-
lecting firewood), while others require a little capital
(hawking goods), moderate capital (a bicycle taxi), or sub-
stantial capital and skills (tailoring). This labor market is
dynamic, and as youth accumulate skills and funds, they
shift to more productive work. One youth began making
charcoal from discarded wood. With his profits, he pur-
chased a bicycle and began a taxi service, and with these
profits, he educated himself and later opened a small store.
Qualitatively, abduction appears to interrupt this accumula-
tion of skills and capital and thus stalls productive employ-
ment.
TABLE 1.—DESCRIPTION OF KEY VARIABLES:WAR EXPERIENCES AND POSTWAR OUTCOMES
Sample Mean
Variable Name Description All Abducted
Not
Abducted
Number of
Observations
War experiences
Months abducted Length of the respondent’s longest abduction, in months 8.9 [15.6] 462
Age of abduction Age (in years) at the time of the respondent’s longest abduction 15.3 [4.7] 462
Index of violence experienced Sum of 17 indicators of violence witnessed, received, or upon own family 5.0 [3.1] 7.2 [2.7] 3.2 [2.1] 738
Index of violence perpetrated Sum of 8 indicators of violence perpetrated by the respondent (self-reported) 0.7 [1.4] 1.5 [1.8] 0.1 [0.3] 738
Education and labor market outcomes
Educational attainment Highest level of education obtained (including tertiary and vocational training) 7.4 [3.0] 7.1 [2.9] 7.6 [3.0] 741
Indicator for functional literacy Indicator equaling 1 if a respondent reports being able to read a book or a newspaper
in any language
0.75 [0.43] 0.69 [0.46] 0.80 [0.40] 741
Indicator for any work in past month Indicator equaling 1 if days employed were greater than zero 0.64 [0.48] 0.69 [0.46] 0.61 [0.49] 741
Indicator for capital or skill-intensive work Indicator equaling 1 if the main occupation is a profession, a vocation, or a small business 0.10 [0.30] 0.08 [0.26] 0.12 [0.32] 741
Daily wage (in Uganda shillings) Gross cash earnings in the past month divided by days employed; 237 observations are undefined 3,221 8,621 2,498 4,941 3,915 11,018 504
Psychosocial outcomes
Index of psychological distress Sum of 19 survey questions on symptoms of depression and traumatic stress 4.0 [2.4] 4.2 [2.5] 3.8 [2.2] 741
Indicator for top 25% of distress index Indicator equaling 1 if the psychological distress index exceeds a score of 5 (the top quartile) 0.27 [0.4] 0.32 [0.5] 0.23 [0.4] 741
Index of social support Sum of 14 questions on concrete social support received from family and friends in past month 5.5 [2.4] 5.5 [2.4] 5.5 [2.5] 741
Indicator for hostility Indicator equaling 1 if reported being one of four hostile behaviors 0.07 [0.3] 0.07 [0.3] 0.07 [0.3] 741
Indicator for a physical fight Indicator equaling 1 if the respondent reported being in a physical fight in the past 6 months 0.07 [0.3] 0.07 [0.2] 0.07 [0.3] 741
Sample means weighted by inverse sampling and inverse attrition probabilities. Standard deviations are in brackets.
884 THE REVIEW OF ECONOMICS AND STATISTICS

FIGURE 1.—MAP OF UGANDA AND SURVEY FIELD SITE
885THE CONSEQUENCES OF CHILD SOLDIERING

III. Data and Measurement
In 2005 and 2006 we conducted phase 1 of the Survey of
War Affected Youth (SWAY), a survey of 741 males born
between 1975 and 1991 in one of eight rural subcounties in
the districts of Kitgum and Pader.
1
To minimize attrition
from migration and mortality, we tried to identify a repre-
sentative sample of youth living in the eight subcounties
before the conflict. We randomly sampled 1,100 households
from U.N. World Food Programme lists compiled in 2002.
Ninety-three percent of these households were found, and
Acholi enumerators worked with household heads to
develop a roster of household members in 1996.
2
We chose
the year 1996 because it was easily recalled as the date of
the first election since 1980 and because it predates 85% of
local abductions. A sample of 870 surviving male youth
was drawn from this retrospective roster. Abductees were
oversampled.
Of surviving males, 41% had moved since 1996, and
enumerators attempted to track all migrants; 741 (84%)
were located, including all nonmigrants and 70% of
migrants. We interviewed the families of all 129 unfound
youth for data on abduction experiences and current out-
comes and collected demographic data on the 349 youth
who had died or not returned from abduction.
The 741 youth who completed the survey provided data
on their war experiences as well as current well-being and
outcomes. Key variables are described in table 1. Two
aspects of these data are noteworthy. First, war experiences
are self-reported and retrospective.
3
Second, the measures
of violence, social integration, hostility, and distress are
additive indices of questions commonly used for measuring
psychosocial well-being in conflict zones. Each is described
in more detail below.
IV. Empirical Strategy
We develop a comparison group of nonabducted youth in
the war zone who for largely exogenous reasons were never
abducted. We can interpret the difference in outcomes as
the incremental effect of conscription in communities sub-
jected to abduction. This counterfactual is a crucial one if
we are interested in addressing reintegration gaps—that is,
closing any inequality between combatants and noncomba-
tants, or providing reparations beyond those received by
other war-affected populations. We discuss the effects of
war on nonabducted youth in section VII.
A. Dealing with Endogenous Selection into the Armed
Group
The fundamental empirical problem we face is that we
cannot observe an ex-combatant’s well-being in the absence
of abduction. The standard solution is the counterfactual
approach, where a relevant control group is found and the
average treatment effect (ATE) is estimated by taking the
difference in the outcomes of the treated and controls
(Rubin, 1974; Imbens & Wooldridge, 2008). The estimated
ATE is only as reliable as the counterfactual, of course, and
it will be unbiased only when abduction and potential out-
comes are independent.
In the case of ex-combatants, we are concerned that cur-
rent differences are the result of prewar traits that led to
selection into the armed group. To deal with such potential
endogeneity, we look for situations where participation in
the armed group is independent of outcomes conditional on
observed prerecruitment variables. LRA abduction presents
just such an unlikely case.
Interviews with LRA leaders suggest that the most com-
mon types of selection are not present. First, self-selection
into the armed group was nonexistent in the subcounties we
surveyed. The LRA’s murder and mutilation of civilians in
1991 destroyed what little support the group ever enjoyed,
FIGURE 2.—DISTRIBUTION OF ABDUCTIONS BY AGE AT THE TIME OF ABDUCTION
The bars represent a probability mass function for age at the time of longest abduction, and so sum
to 1. The data include absentee youth and youth who have since died or did not return from abduction
(collected from the household survey).
1
Subcounties include 25 to 100 villages and range from 10,000 to
40,000 people. The eight subcounties represent roughly 10% of each dis-
trict and 5% of all Acholi. These clusters were not selected randomly
since poor security limited our team to subcounties that could be reached
within a 90-minute drive from Kitgum and were visited regularly by our
military escorts. We selected subcounties of varying size that seemed
representative of the region as a whole.
2
These lists are an approximate census of the population since the
entire population in each subcounty was displaced and receiving food aid.
Based on interviews with local leaders, we believe that most households
left family members in the camp, in part to receive the food aid. Hence
these households’ youth would be included in the sample frame. We esti-
mate 5% of households left entirely and so do not enter the sample frame,
introducing unknown selection.
3
We took several measures to guard against youth misrepresenting
themselves as abductees (in the hopes of aid). We emphasized the absence
of any link between the study and aid. Abduction data were also collected
separately from the household head, and irregularities were investigated.
Finally, the survey asked more than 200 detailed questions on abduction.
Only 5% of abductees raised suspicion, and reclassifying these has no
material impact on our conclusions.
886 THE REVIEW OF ECONOMICS AND STATISTICS

Frequently asked questions

Q1. What are the contributions in "The consequences of child soldiering" ?

The authors examine the case of Uganda, where rebel recruitment methods provide exogenous variation in conscription. 

Q2. What are the future works mentioned in the paper "The consequences of child soldiering" ?

The authors interpret their findings as the incremental effect of conscription in communities already subjected to the horrors of war and abduction, an important quantity of interest if they wish to target aid and development assistance to the people who need them most. It remains a crucial area for future research. In the meantime, it is worth noting that the incremental impact of military service may be more important than the gap between war- and nonwar-affected populations, at least in Uganda. National survey data suggest that while the impact of the war on wealth may have been large, the educational impact has been small: nonabducted youth in the war zone have levels of education and literacy that are similar to or even greater than that in other parts of Uganda. 

Q3. How many youth were selected for indepth interviews and assessments?

Forty youth were selected for indepth interviews and assessments, including multiple interviews of the youth and his family, friends, teachers, and coworkers. 

Q4. What did the elder say about the youth who had not been abducted?

According to one elder, ‘‘The youth who have not been abducted are engaged in different activities like business and vocational work like carpentry, because they had the opportunity to acquire the different skills. 

Q5. What could be the effect of abductions and displacement?

Or abductions and displacement could lead to positive externalities as nonabducted populations are urbanized in displacement camps (closer to schools) or displaced to towns (closer to markets). 

Q6. How many years of schooling is associated with an increase in literacy?

For instance, looking at all youth in the sample, moving from 6 to 7 years of schooling is associated with a 22 percentage point increase in literacy. 

Q7. How is the best-case scenario bound calculated?

The best-case scenario bound is calculated by dropping nonabductees with the lowest values of the outcome and calculating the trimmed ATE. 

Q8. How many abducted youths did they survive?

The remainder perished, as no more than 1,000 abducted youth (about 1% of all abductees) are thought to remain with the LRA at this time. 

Q9. What is the impact of child labor on education and earnings in both Vietnam and Tanzania?

Causal estimates of the impact of agricultural child labor on education and earnings in both Vietnam and Tanzania suggest that a doubling of hours employed reduces school enrollment by nearly a third and educational attainment by 6% (Beegle, Dehejia, & Gatti, 2004, 2006). 

Q10. How long does the correlation between abductions and education be?

Long abductions are strongly correlated with losses in education and literacy: each year of abduction is associated with 0.54 years less education and a 9 percentage point reduction in literacy. 

Q11. How many violent acts did the average abductee experience?

The average abductee reported 4 more violent acts experienced and 1.4 more perpetrated than nonabductees (table 1), implying a distress impact of 0.98. 

Q12. How can the authors determine the relative influence of each component of human capital in the abduction wage gap?

14The authors can obtain a rough measure of the relative influence of each component of human capital in the abduction wage gap by multiplying the earnings function coefficients by the respective abduction ATEs (table 8, columns 2 and 3).15 Education, as the strongest determinant of wages (and a principal casualty of abduction), appears to be the most significant channel by which abduction reduces wages, representing 55% of the reduced-form ATE for log wages (column 4).