Академический Документы
Профессиональный Документы
Культура Документы
Abstract: We estimate the effects of the brutal 1991-2002 Sierra Leone civil war using unique
nationally representative household data on conflict experiences, postwar economic outcomes,
and local politics and collective action. Individuals whose households personally experienced
more intense war violence are robustly more likely to attend community meetings, more likely to
vote, more likely to contribute to local public goods, and are more aware of local politics.
Several tests indicate selection into victimization is not driving the results. The relationship
between conflict intensity and postwar outcomes is weaker at more aggregate levels, suggesting
that the war’s primary impact was on individual preferences rather than on institutions or local
social norms. More speculatively, the findings could help partially explain the rapid postwar
economic and political recovery observed in Sierra Leone and after several other recent African
civil wars.
*
John Bellows, Department of Economics, 549 Evans Hall #3880, University of California, Berkeley, CA
94720-3880, USA, jbellows@econ.berkeley.edu
Edward Miguel, Department of Economics, 549 Evans Hall #3880, University of California, Berkeley,
CA 94720-3880, USA, emiguel@econ.berkeley.edu. U.C. Berkeley and NBER
Civil war has been a prominent feature of recent history in Sub-Saharan Africa: more than two-thirds
of countries in the region experienced an episode of civil war during the past 25 years. Some
scholars have claimed these wars have played a role in the region’s disappointing recent economic
performance. For example, a recent World Bank report claims: “[t]he legacy effects of civil war are
usually so adverse that they cannot reasonably be viewed as social progress…[Civil war] has been
development in reverse” (World Bank 2003: 32). Yet the rapid postwar recovery experiences of
some African countries after brutal civil wars – notably, Mozambique and Uganda – suggest that war
need not have persistent negative economic consequences: in the decade following the end of their
wars, Mozambique and Uganda experienced annual per capita income growth of 3.9% and 4.6%,
respectively, well above the African average (United Nations 2003). This paper analyzes a novel
nationally representative dataset from postwar Sierra Leone with the goal of better understanding the
Recent research has shown that the long run effects of war on population and the economic
growth are typically minor. Studies that focus on United States bombing – including in Japan (Davis
and Weinstein 2002), Germany (Brakman et al 2004) and Vietnam (Miguel and Roland 2005) – find
few if any persistent impacts of the bombing on local population or economic performance. To the
extent that war impacts are limited to the destruction of capital, these findings are consistent with the
predictions of the neoclassical economic growth model, which predicts rapid catch-up growth
postwar. However, the neoclassical growth model has little to say about the impact of war on
institutions, politics, social norms, or individual preferences. Given the extreme trauma experienced
by civil war victims, it is plausible that effects along these human dimensions could be more
substantial and longer lasting than any impacts on capital investment levels.
1
War could also potentially generate large impacts on both national and local institutions.
Tilly (1975) argues that wars historically promoted state formation and nation building in Europe,
definition of institutions might include the social equilibrium reached by individual rational actors.
In experimental economics evidence from Honduras, Castillo and Carter (2004) find that people in
locales that experienced extensive destruction from Hurricane Mitch – which, although traumatic, is
arguably not as severe as the violence experienced by many civil war victims – shared significantly
more of the “pie” with their partner in a Dictator Game. This suggests traumatic experiences could
also have a positive impact on altruism or on local social norms regarding cooperation.
At the individual level, the experience of being a victim of war violence could also
profoundly change individual beliefs, values, and preferences. An emerging psychological literature
has documented some of these individual responses to conflict-related trauma. Studies often focus
on symptoms of post-traumatic stress syndrome (e.g. Dygrove et al 2002), but a subset of the
literature now also explores positive responses to trauma, the so-called post-traumatic growth theory
(Tedeschi and Calhoun 1996, Powell et al 2003), including changes in political action and beliefs.
For example, Israelis who survived the Holocaust are more religious, more optimistic and at the same
time more extreme in their political views (Carmil and Breznitz 1991), while Palestinians who
personally survived aerial attacks are more likely to engage in political activism (Punamaki et al
1997). One key limitation of this literature is the use of small respondent samples of unknown
representativeness.
The distinction between how individuals react to their own personal experiences versus by
observing others is critical for understanding the nature of civil war impacts in general as well as in
this study of Sierra Leone. It is also an issue that is amenable to laboratory experiments: Simonsohn
et al. (2006) find that individuals’ own personal experience (playing the Prisoner’s Dilemma and
2
other standard experiments) is far more influential in shaping subsequent game play than the first-
Many studies of the determinants of U.S. political participation have focused on how costs
and other economic factors influence the choice to vote, although it is unclear these rational choice
voting models have been successful.1 In fact, there is growing evidence on how psychological and
social factors affect political participation. Green and Garber (2004) find that subtle changes in the
framing of political messages can have major impacts on voter turn-out. U.S. parents who lost in
school choice lotteries for their children are significantly more likely to vote in subsequent school
board elections, compared to parents who won (Hastings et al 2005). In a related finding, U.S. voters
whose county suffered a seemingly random misfortune – including local floods, shark attacks, or flu
epidemics – tend to punish political incumbents in later elections (Achen et al 2004). Theoretical
explanations for this “expressive voting” are based on the assumption that victims derive some
additional utility from voting relative to non-victims. Given these positive political participation
impacts among school lottery losers and those whose town suffered a shark attack, a finding that
political activism increases among civil war victims seems intuitively plausible.
Unfortunately, the extreme scarcity of household survey data from contemporary conflict and
post conflict societies has limited research progress on questions related to the economic and political
aftermath of civil wars. One exceptional aspect of this project is the availability of high quality
nationally representative household data from Sierra Leone containing detailed information on
household experiences with war violence as well as on immediate postwar political and collective
action behaviors, in addition to the more standard socioeconomic questions. The main empirical
results focus on the individual level analysis made possible by this remarkable dataset. We also draw
1
Green and Shapiro (1994) argue these models have performed especially poorly with regards to explaining
determinants of voting behavior.
3
on a collection of other new data sources from Sierra Leone to estimate the relationship between
local conflict and postwar outcomes at the more aggregated chiefdom level.2
In our main result, we find that individuals whose households directly experienced war
violence are much more politically active than non-victims. War victims are significantly more likely
to vote (by 2.4 percentage points in our preferred specification), attend community meetings (by 5.7
percentage points), contribute to local public goods, and be knowledgeable about local politics.3 This
relationship is robust across two independently collected survey samples and multiple econometric
specifications, including a specification with village fixed effects, which compares neighbors within
the same village, but with different violence experiences, to one another. Several tests indicate that
systematic individual selection into victimization is unlikely to be driving the results; for instance,
violence effects are also strong for those too young in age to have been community leaders at the
start of the civil war, among whom conflict-related violence victimization is arguably more random
Yet two to three years after the end of the war, there are – perhaps surprisingly – no lasting
impacts on household socioeconomic status measures, including asset ownership, income earning
In contrast, we do not detect significant additional impacts of violence at the more aggregated
chiefdom level. Chiefdoms that experienced greater violence do show greater overall political
activism in certain dimensions, but these broader impacts are less robust than the individual level
effects, suggesting that the war’s primary impact was on individual preferences rather than on local
2
The chiefdoms in Sierra Leone are administrative units that were formalized by the British in the 1930s. These
colonial boundaries remain salient today as most people identify their residential location by the chiefdom. The
average chiefdom has roughly 20,000 people.
3
In a related result, Blattman (2006) finds that former child soldiers in Uganda are significantly more likely to vote
than other youth.
4
Civil war experiences are transformative for many, and our analysis suggests that one short-
run legacy is increasing individual political participation and local public good provision. As we
discuss in the conclusion, this finding echoes the observations of other scholars of Sierra Leone and
speaks to the remarkable resilience of ordinary Sierra Leoneans. More speculatively, the findings of
this paper also contribute to the recent debate on the underlying causes of Africa’s terrible recent
economic performance, and speak against claims that civil war legacies are major long-run
Sierra Leone was ravaged by a civil war that started in 1991 and lasted until January 2002. During
the war an estimated 50,000 Sierra Leoneans were killed, over half of the population was displaced
from their homes, and thousands were victims of amputations, rapes, and assaults (Human Rights
Watch 1999).
Just before the war began, Sierra Leone was the second poorest country in the world (United Nations
1993). For the preceding two decades the country had been ruled by dictators who enriched
themselves through illicit deals involving diamonds, while doing next to nothing to provide needed
services such as health care and education (Reno 1995). Partially as a result of the widespread
discontent towards the corruption and ineffectiveness of the government, a small group of rebels,
who had entered the country from Liberia in 1991, were successful in recruiting disenfranchised
youth to rise up violently against the status quo. As their numbers swelled by early 1992, these
rebels, known as the Revolutionary United Front (RUF), spread the armed conflict to all parts of the
country. Some scholars have claimed that the initial motivations of the RUF were idealistic, and that
the early rebels were guided by a strong sense of political grievances related to the failings of the
5
Another important factor in the RUF’s original motivations was access to Sierra Leone’s
diamond wealth. Mining diamonds in Sierra Leone requires no heavy machinery or technology,
since these alluvial stones sit close to the surface in dried riverbeds, so any armed group that
controlled a diamond area could extract and then sell the diamonds for large profits. All armed
groups participated to some extent in diamond smuggling during the conflict, and the control of these
diamond areas was an extremely important objective for all groups. David Keen notes that “[a]ny
battles were largely restricted to the areas with the richest diamond deposits” (Keen 2005: 212).
Additionally, since large-scale diamond smuggling was possible so long as the country remained in
chaos, profits from these “blood diamonds” represented an important incentive for armed groups to
In contrast to most popular media coverage on African civil wars, neither ethnic nor religious
divisions played a central role in the Sierra Leone conflict. The RUF rebels targeted people from
every ethnic group and throughout the country, and statistical analysis of documented human rights
violations shows that no ethnic group was disproportionately represented among RUF victims
(Conibere et al 2004). There is also no evidence that levels of civilian abuse were higher when a
particular armed faction and the community were predominantly from different ethnic groups
B. The Revolutionary United Front (RUF) and the Sierra Leone Army (SLA)
Although there were many different actors in the decade-long war, the majority of the violence was
perpetrated by the RUF: the official government truth and reconciliation commission, which
documented war atrocities reports that over 70% of all human rights abuses were committed by RUF
fighters (Conibere et al 2004). Our own analysis of the No Peace Without Justice (NPWJ) conflict
mapping project, which is a comprehensive record of all reported armed violence during the war,
similarly concludes that 75% of all attacks and battles involved the RUF as the primary fighting force
6
(NPWJ 2005). The following incidents recorded in the NPWJ report are fairly typical of the brutal
“In the early hours of 27 May 1997, the town of Karina (Biriwa Chiefdom, Bombali
District) was attacked by RUF / AFRC forces carrying guns and other weapons.
Soldiers surrounded the central mosque and killed 10 civilians celebrating the
Muslim feast of “Jonbedeh”… An unknown number of people were injured trying to
escape. RUF/AFRC forces raped an unknown number of women, and abducted 30
young civilian men and women. During the attack, numerous houses were burned
down.” (p. 133)
“RUF forces attack Koi town (Nongowa Chiefdom, Kenema District) early one morning in
mid-1994, reportedly to terrorize the inhabitants. … They fired indiscriminately and many
civilians were killed and others were wounded. The town was looted and people were forces
to carry the stolen property to Peyama.” (p. 303)
“On 11 March 1998, RUF/AFRC forces attacked the headquarter town of Jagbwema (Fiama
Chiefdom, Kono District). RUF/AFRC forces entered the town firing indiscriminately.
More than 70 houses were burnt and the town was massively looted. During the night, the
RUF/AFRC forces abducted three people, including the Town Chief, who were all later
killed. On 24 March 1998, RUF/AFRC forces coming from Jagbwema attacked Yeanoh,
shooting and killing many people.” (p. 361)
“On 26 December 1994, RUF forces attacked Mattru on the Rail (Tikonko Chiefdom,
Bo District) in the afternoon, mutilating civilians’ arms and legs. The RUF then
opened sporadic gunfire on the civilians, killing many people, looting their property
and burning down their houses. They also abducted civilian youths who they
conscripted into the RUF forces.” (p. 395)
The extent of targeting of community leaders or other opponents during RUF attacks is
important in the later analysis. It is useful here to distinguish between regions where the RUF did not
establish permanent bases and thus mainly resorted to raids like those described above, versus
regions with permanent bases that were occupied for extended periods. The ability to systematically
attack particular types of civilians is inherently greater in areas the RUF occupied relative to areas
they only briefly raided. The NPWJ report indicates that slightly more than half of all chiefdoms (86
of 152 chiefdoms) did not have permanent RUF bases during the war. In the analysis below we
restrict attention to these areas, to gauge whether estimated war violence impacts are robust to a
subsample where RUF violence against civilians was likely to be largely indiscriminate.
7
One feature of the fighting that has drawn attention from international observers is the
cooperation between the rebels and the Sierra Leone Army (SLA). These two groups coordinated
their movements in order to avoid direct battles, and at times worked out mutually beneficial profit
sharing arrangements in diamond areas. This was especially true following the 1997 coup that
formally brought the SLA and RUF together into a national coalition government called the Armed
Forces Revolutionary Council, or AFRC (Keen 2005). Some soldiers apparently fought for the SLA
by day and the RUF by night. As a result, the main victims of the violence were civilians, who were
terrorized not only by the RUF but also by the army that was supposed to protect them.
In order to protect themselves from the terror of RUF and SLA fighters, many communities
organized their own local fighting groups, which became known collectively as the Civil Defense
Forces (CDF). CDF fighters were overwhelmingly civilians and they relied primarily on local
fundraising for supplies. While there were numerous manifestations of the CDF throughout the
country, the command and organization of the CDF were often linked with traditional chiefly
authorities. For example, the largest CDF, known as the kamajors, were an extension of traditional
There are many accounts of ordinary civilians going to heroic lengths to protect themselves
from RUF attacks. One such account from Allister Sparks (2003:309), an international election
observer in the 1996 Sierra Leone presidential election, describes how the citizens of Kenema Town
“The polling stations were due to open at 7 am on 26 February, but at exactly 6.15 am
the rattle of small-arms fire broke out around the centre. … For two-and-a-half hours
the firefight raged. At times the rebels ran close past our building and we could hear
them shouting: “No election! No election!” between their bursts of AK-47 fire. Then,
indistinctly at first but gradually increasing in volume, we heard a counter-chant
coming from the direction of the town: “We want vote! We want vote!” Thousands of
people were pouring into the streets, and as the chanting crowd swelled they ran
through the town waving palm leaves. … Whether it was this display of public
8
courage or a successful counterattack by the local military was unclear, but the rebels
began to withdraw and the shooting subsided. As the observers made their way
gingerly into the town, crowds lining the streets yelled impatiently at us: “Bring the
boxes. We want vote!” The polling stations opened late, some not until the afternoon,
but electoral officials worked frantically to open extra stations, and by the time the
polls closed at 6 pm nearly every registered adult in Kenema had voted.” (p. 309)
The CDF continues to be admired within Sierra Leone for their selfless defense of civilians.
However, late in the conflict when their power and numbers had grown, some CDF units lost
discipline and they too began to abuse civilians and enter the illicit trade in diamonds, although to
The rise of the CDF is illustrative of two points raised in the introduction. First, the CDF is
an example of how war can create influential new institutions. Second, the account above presents a
concrete example of how Sierra Leonean individuals responded to war violence with an increased
desire to assert their political rights. We return to both of these points in the empirical analysis below.
Following the brutal 1999 rebel attack on Freetown, a large deployment of United Kingdom
and United Nations troops finally brought an end to the war. These foreign troops conducted a
disarmament campaign and secured a peace treaty in early 2002. Donor and non-governmental
organization (NGO) assistance has since played a major role in reconstructing physical infrastructure,
resettling internally displaced people (almost all of whom had returned home by 2003), and funding
other government expenditures. National elections for a president and members of parliament were
held in 2002, and local government elections – the first in over thirty years – in 2004.
A. Estimation Strategy
The literature reviewed in the introduction suggests that there are at least two plausible channels
9
First, there may be individual level impacts, if the trauma associated with directly
experiencing violence leads to changes in beliefs, identities, values and preferences. To investigate
this relationship, we compare postwar outcomes and behaviors across individuals that suffered from
different degrees of violence during the conflict. The preferred specification includes village (or
enumeration area) fixed effects, so the analysis essentially neighbors within the same village.4
victimization within villages is close to random at the household level. This might not hold if there
was systematic targeting by fighters along some unobserved household dimension, in which case the
observed relationship between victimization and postwar outcomes could in part be due to omitted
variable bias. We carry out several tests to examine the extent of selection into war violence below,
and these indicate that any selection bias is likely to be relatively minor.
Concerns about selection are mitigated by the specific characteristics of Sierra Leone villages
and the nature of the RUF attacks. Our surveys indicate that rural Sierra Leone villages consist
almost entirely of subsistence farmers, and there is typically no conspicuous landowning elite for the
RUF to target, as would be the case in some other societies. Additionally, Sierra Leone villages are
very small, usually consisting of a handful of inter-related extended families5, so in the specifications
including village fixed effects we effectively make comparisons across this small and relatively
homogenous collection of households. As the above accounts demonstrate, RUF attacks on civilians
were often brief, chaotic, and indiscriminate in nature, which provided little opportunity for precise
4
In the GoBifo sample (described below) villages were selected, then individuals were randomly selected for
surveys within the village. In the IRCBP sample, enumeration areas were selected first. In most IRCBP cases an
enumeration area corresponds to a single village, but in some instances one enumeration area contains two smaller
villages. In the urban settings, an IRCBP enumeration area is equivalent to a block or a neighborhood.
5
In the two districts (Bombali and Bonthe) where we have detailed data on village size, there are only 33 and 29
households per village on average, respectively.
10
According to many accounts of the war, the one group that the RUF was able to target was
members of traditional authority families (chiefs), who were well known and visible in their localities
(Keen 2005; Richards 1996). We have data at the household level on whether the households were
members of the traditional authority and thus are able to control for this characteristic in the analysis.
Other measures of socio-economic status, including education, are not robustly correlated with
victimization, suggesting that targeting along other lines was relatively rare. While we admittedly
cannot rule out that some targeting of politically active households occurred, we feel that the
observed relationships we estimate primarily reflect the impacts of war victimization on postwar
behavior, rather than selection bias, and we present several tests of this claim below.
Second, there may also be war impacts at more aggregated levels due to changes in
institutions or social norms brought on by the conflict. We investigate these effects by comparing
chiefdoms that experienced different levels of conflict intensity. In the chiefdom level analysis, we
rely on a rich set of local characteristics as explanatory variables to ensure we are isolating the effect
of violence. These controls include the number of diamond mines, roads, population density, and in
some specifications prewar socioeconomic measures. Additionally, district fixed effects are included
An important caveat of the entire empirical strategy is worth emphasizing: we focus on local
comparisons across individuals and across chiefdoms, and cannot estimate the overall national
impact of the Sierra Leone civil war. The data do no permit the estimation of national impacts
because no suitable counterfactual exists.6 This caveat is important, as the net national effect of the
war could be negative even in light of any positive local violence victimization impacts that we
estimate.
B. Individual Data
6
Liberia shares similar geography, history and culture with Sierra Leone, but Liberia was also experiencing a civil
war during this time so it cannot be used a peacetime counterfactual.
11
Data on individual experiences with war violence is extremely rare for post conflict societies, and
this has limited research progress in estimating civil war impacts. The broad collection of household
level and chiefdom level data on conflict experiences and postwar outcomes makes ours among the
In the present analysis we make use of data from two unusual household surveys that were
collected in 2005, a mere three years after the war ended. The first survey is nationally
representative7 and was conducted by the Institutional Reform and Capacity Building Project
(IRCBP).8 The second survey was conducted as baseline data for a government assistance program
called “GoBifo” and it covers only selected wards within two districts. The location of sample
enumeration areas for the IRCBP and GoBifo surveys are presented in Figures 1 and 2, respectively.
Details on these surveys, including sample sizes, can be found in the data appendix.
These two surveys contain detailed questions on household war victimization experiences.
The IRCBP survey contains the following three retrospective questions: “Were any members of your
household killed during the conflict?” “Were any members injured or maimed during the conflict?”
and “Were any members made refugees during the war?” We create a victimization index as the
average of responses to these violence related questions (Table 1, panel A); as we discuss below,
breaking the index down into its component questions does not substantively change the results. The
data also includes information on household assets, some respondent characteristics (including
education), and multiple measures of political engagement, voting, participation in collective action,
and self-expressed levels of trust and cooperation (Table 1, panels B, C, D, and E). Because the two
surveys were conducted independently and using different sampling frames, carrying out the analysis
7
The capital Freetown is excluded from the analysis. Freetown is Sierra Leone’s only large city and its local
institutions and history are quite different from the rest of the country.
8
The IRCBP is affiliated with the government of Sierra Leone and its primary role is to support the ongoing
decentralization of government services.
12
C. Chiefdom Data
We also use the number of reported attacks and battles in each chiefdom as an additional dimension
of violence. The number of attacks and battles is related to, but distinct from, the household reports
of victimization, as it also includes the battles between troops that did not involve civilians. The
2004 No Peace Without Justice (NPWJ) conflict mapping project compiled all reports by human
rights organization and the media on the location and intensity of violence during the conflict (Table
1, panel F). We construct a measure of attacks and battles from the descriptions included in this
report. The correlation across the household victimization measure and the number of attacks and
battles at the chiefdom level is moderate at 0.3. Note that the two measures of conflict-related
violence are broadly analogous to the two types of commonly used crime data, crime victimization
Additional chiefdom level data is constructed from multiple sources. The 2004 Sierra Leone
Integrated Household Survey provides data on nutrition, education and socioeconomic outcomes
(Table 1, panel G). The 2005 School Survey provides data on the quality, monitoring and funding of
local education facilities (Table 1, panel H). The 2003 Sierra Leone Data Encyclopedia provides
(Table 1, panel H). The GIS data provides information on the location of diamond mines, roads, and
population density (Table 1, panel I). The 1989 Sierra Leone Household Survey provides the only
existing data we are aware of on prewar socioeconomic conditions (Table 1, panel I).9 Further
details on data sources and variable construction are provided in the data appendix.
violence is concentrated in the eastern part of the country near Liberia, but some violence was
experienced in all regions. Figure 4 presents the residuals of the victimization index after the district
9
The sample for the 1989 household survey includes fewer than half of all chiefdoms in the country. The
documentation for the data set is incomplete, making it impossible to know how exactly this sample was chosen.
13
means have been subtracted off. This measure of local violence is effectively used in specifications
that include district fixed effects. As is apparent from the figure, subtracting off district averages
We next investigate the relationship between war intensity and the different factors that are
thought to have contributed to onset and to the length of the war. The most robust finding is that
chiefdoms with diamond mines have significantly more attacks and battles during the war. In all
specifications, including those with district fixed effects and controls for 1989 socioeconomic status,
the relationship is large, positive, and statistically significant (Table 2, regressions 4-6). Our data thus
confirms the widely held view that diamonds were related to the local intensity of fighting. Other
geographic controls, including road density, distance to Freetown (the capital) and population density
are only weakly related to both measures of violence. We find that there is no significant relationship
between diamonds and household reports of victimization (Table 2, regressions 1-3). Humphries and
Weinstein (2006: 444) similarly find no relationship between diamond mines and brutality towards
civilians, in data reported by fighting units. It appears that the fighting around diamond mines
primarily involved soldiers and did not disproportionately affect civilians in those areas.
Turning to the other factors, prewar 1989 school enrollment is negatively related to civilian
victimization (Table 2, regressions 3 and 6). This is consistent with the explanation that violence
was higher in areas with poorer public services, possibly due to more severe political grievances in
those areas (Richards 2003), or possibly fewer youth employment opportunities (Collier and Hoeffler
2004). Finally, we find that 1989 average log per capita consumption expenditures are positively
related to the number of chiefdom attacks and battles, consistent with the explanation that lootable
resources attracted armed groups (Collier and Hoeffler 2004). We do not place too much emphasis
on these 1989 data because the sample size falls to just 64 chiefdoms. Yet this is suggestive evidence
that prewar chiefdom socioeconomic conditions may be associated with later violence.
14
IV. Individual level analysis
We document that households that experienced more direct civil war victimization are significantly
more likely to be politically mobilized and engaged in local collective action than other households,
but are no different in terms of assets, religiosity or self-expressed trust postwar. These findings hold
across both the IRCBP and the GoBifo samples, using multiple measures of political mobilization,
and across many specifications, including those with village fixed effects. Before turning to these
results, we first establish that there is no strong evidence of individual selection into victimization,
We would ideally only use prewar characteristics to predict conflict victimization, but these are
unavailable at the household level. Instead we use postwar data on characteristics that are unlikely to
change as a result of the war, for instance, adult educational attainment and demographic
characteristics.
Both the chiefdom level and village level victimization indexes are positively and
significantly correlated with the household victimization index (Table 3, regressions 1, 2, 4 and 5), as
expected. More importantly for our results, in the nationally representative IRCBP sample there are
no statistically significant relationships between respondent age, gender, or education with the
household violence victimization index across the six specifications, even in the village fixed effects
specification (Table 3, regression 3). The lack of robust correlation between victimization and
observable household characteristics is our first piece of evidence that violence was not
In the smaller GoBifo sample, there is no significant correlation between respondent age or
gender, but there is a positive relationship between education and victimization (Table 3, regressions
4-6). As discussed in subsection D, the chiefdoms in the GoBifo sample nearly all contained RUF
15
bases during the war; as the presence of a permanent base made it easier for the RUF to target
community leaders or other opponents, it is perhaps not surprising that selection into violence is
more of an issue in the GoBifo sample than in the nationally representative IRCBP sample.
The one group that disproportionately suffered from war violence were traditional authority
households (i.e. chiefs’ households and extended families). Traditional authority households are
significantly more likely to be victimized across all six regressions. This lines up with many media
accounts of the war, which describe how chiefs were systematically targeted by the RUF in an
attempt to undermine the “corrupt” existing order. We account for this specific type of targeting in
the analysis by controlling for traditional authority households in all later regressions. We also
investigate below whether average chiefdom outcomes are affected by the killing of the local chief.
We begin with a detailed analysis of two specific individual behaviors: attendance at community
meetings and registering to vote / voting in either of two recent elections.10 We estimate the
relationship between household conflict victimization and these behaviors controlling for observable
community meeting attendance (Table 4) and also whether the respondent registered to vote / voted
in recent elections (Table 5), in both the IRCBP and GoBifo samples. The point estimates on
victimization are remarkably stable across specifications with district fixed effects, chiefdom fixed
effects, and village fixed effects, and are reasonably large: an increase from zero to one in the
household conflict victimization index (which corresponds to going from no violence to experiencing
all three types of violence) in our favored IRCBP village fixed effects specification is associated with
a 5.7 percentage point increase in the probability of attending a community meeting (Table 4,
10
IRCBP survey respondents were asked if they had registered to vote, while GoBifo respondents were asked if they
had voted. While not exactly the same, these two dependent variables are analyzed together in this section.
16
regression 3), on average village meeting attendance of 70%. The analogous change in household
victimization is associated with a 2.4 percentage point increase in the probability of having registered
Other determinants of meeting attendance and voting behavior that emerge in Tables 4 and 5
are sensible in the Sierra Leonean context. Women are relatively less involved in the political
process – specifically, both less likely to attend community meetings and to vote – while people from
traditional authority households attend more community meetings. Individuals with some education
are more likely to participate in community meetings, but education is only weakly related to voting
behavior. 11
Having established the robustness of the results for two outcome measures, we next expand
the analysis to consider the relationship between conflict victimization and a wider range of postwar
outcomes and behaviors. The specification shown includes village (enumeration area) fixed effects
as well as respondent controls for gender, age, education, and traditional authority household, and
We first find no robust association between conflict victimization and either working for
wages or owning a cooking stove. While the relationship between victimization and owning a radio
is negative in the IRCBP sample, it is not statistically significant for the GoBifo sample (Table 6,
panel A, rows 1, 2, and 3). Taken together, there is no clear evidence that individuals who
experienced more conflict victimization are worse off along observable socioeconomic dimensions
In contrast, those who were victims of war violence are very different in terms of political
related to attendance at community meetings and registering to vote / voting (Table 6, rows 4 and 5).
11
These determinants of voting behavior are broadly similar to data from the U.S., where historically females have
been less likely to vote (Timpone 1998). One difference is that education is positively related to voting in the U.S.,
a finding which does not find strong support in our estimates.
17
Conflict victimization is also robustly positively related to other measures of mobilization, including
membership in a school management committee and membership in a political group (Table 6, panel
Not only is political mobilization greater among those who were conflict victims, but
participation in local collective action is higher as well. The relationship between violence
victimization and participation in road brushing, a crucial locally organized activity to keep bush
paths between villages passable, is positive and statistically significant (Table 6, row 8). This is
finding is important, as it appears to confirm that increased political mobilization is producing more
local public goods, rather than just creating deadlock within communities.12 Recall that these
specifications include village fixed effects, so these collective action impacts are unlikely to simply
reflect greater local needs in war-torn areas. Nor does it seem likely that these collective action
effects just reflect the lower opportunity cost of time among conflict victims, since there is no
significant relationship between victimization and labor market activities (working for a wage or
Political knowledge of the identity of both local government and traditional officials is also
significantly higher among conflict victims, once again with large effects on the order of 9 to 11
percentage points (Table 6, rows 9 and 10), suggesting much greater engagement in local politics.
We next investigate the impact of conflict victimization and leadership for those respondents
who were too young to be prewar community leaders, those 30 years old or younger in 2005 (and 15
or younger in 1990, right before the war).13 Focusing on this demographic group reduces concerns
about selection or omitted variable bias, since these individuals could not have been targeted for
violence as a result of already being community leaders themselves prewar or in the early years of
12
Olson (1984) has noted that increased political mobilization could give rise to small, exclusive coalitions that
lobby for narrowly targeted policies that do not benefit society at large. This type of mobilization could have a
negative aggregate effect.
13
We thank Rachel Glennerster for suggesting this test.
18
the war. It is extremely rare for individuals under 20 years of age to be community leaders in our
data. Thus it is plausible that treatment effects among this subgroup primarily reflect violence
impacts rather than violence selection. This is not a perfect test, since these youths’ households may
have been targeted during the war because their parents were community leaders, for instance, and
leadership abilities may be somewhat correlated across generations. Nonetheless focusing on this
youth sub-sample should considerably weaken the link between unobserved individual prewar
Across both the IRCBP and GoBifo samples, conflict victimization effects are positive, large
and statistically significant across political mobilization measures in the youth sample (Table 6,
Panel B, columns 3 and 4). If anything, the point estimates tend to be somewhat larger in magnitude
for youths, although differences across age groups are generally not statistically significant. (Any
differential impacts may also be picking up heterogeneous violence impacts for youth.) The
robustness of the conflict victimization impacts among youths, who were too young to have been
explicitly targeted based on their role as prewar community leaders, is further suggestive evidence
The final portion of the individual analysis turns to issues of self-reported trust and
religiosity. This is of interest since the increased provision of public goods reported in Table 6 could
be a result of heightened levels of trust and cooperation across village members (what some would
call “social capital”). The experience of being victimized could plausibly both make individuals less
trustful of outsiders and more trusting of community members. However, the relationship between
victimization and levels of trust for other community members is not statistically significant in the
nationally representative IRCPB sample and the relationship is only marginally significant in the
GoBifo sample (Table 6, row 11). Similarly, there are no discernable differences in the trust of
outsiders among victims and non-victims (row 12). We also find no evidence of a statistically
significant relationship between victimization and postwar religiosity in either sample (row 13).
19
Whatever effects victimization has on political mobilization and collective action do not appear
The aggregate measures of conflict intensity, i.e. village level victimization and chiefdom
victimization, are not robustly correlated with respondent behavior. For instance, village level
victimization is never statistically significantly related to respondent meeting attendance and voting,
conditional on their own household’s victimization experience (Tables 4 and 5, regressions 1-2 and
4-5). The relationship between chiefdom level victimization and these outcomes, again conditional
on the household’s own experiences, is sometimes positive and statistically significant (for instance,
in the case of community meeting attendance, Table 4 regressions 1 and 4), but not for most other
outcome measures in Table 6 (e.g., not for voting, Table 5 regressions 1 and 4 – other regressions not
shown). The stronger household level results suggest that violence victimization impacts are mainly
driven by individual experiences and changes in preferences, rather than mainly by broader changes
to institutions and social norms, since these would be reflected in the village or chiefdom level
violence effects, and would affect both the households that directly experienced war violence as well
as other households.
It is noteworthy that differences across households within the same village are so pronounced
given the residential proximity of these households, and the fact that all of them at a minimum
witnessed extreme acts of violence during the Sierra Leone civil war, even if they did not experience
violence directly. The gap we find between those who directly experienced violence and others
provides real-world support for the experimental findings in Simonsohn et al (2006), who show in
the lab that most people’s behavior is far more responsive to their own personal experience than to
C. Other specifications
20
We next investigate the possibility of heterogeneous effects of violence victimization for different
population subgroups. When the explanatory variables for female, education, age, and traditional
authority are interacted with the household victimization index, the coefficient estimates on these
interaction terms are not generally statistically significant for the outcome measures in Table 6. The
point estimate on the interaction of violence victimization with a youth indicator is often positive,
and sometimes marginally statistically significant (as suggested by the findings in Table 6, Panel B,
columns 1-2 versus 3-4), suggesting larger political mobilization impacts among youth, but this result
is not robustly significant across samples, outcomes and specifications (regressions not shown).
It is also theoretically possible that effects differ across different types of victimization, e.g.,
physical assault versus residential displacement, if these experiences are associated with different
degrees of personal trauma. Yet, perhaps surprisingly, no single input into the conflict victimization
index is a more important determinant of postwar behaviors than the others: when the three distinct
components of the victimization index are included as separate independent variables, an F-test on
the null hypothesis that the corresponding coefficients are all equal cannot be rejected at 95%
confidence for any of the outcomes in Table 6, Panel B in the IRCBP sample (results not shown).
The main results are robust in a subsample of chiefdoms where the RUF did not establish permanent
bases, and thus where the targeting of civilian violence is likely to be more arbitrary than elsewhere,
as argued above. This analysis is conducted on the nationally-representative IRCBP survey sample.14
The main violence impact estimates remain large, positive and statistically significant for three of
four political mobilization and collective action outcomes, and marginally significant for the fourth
(Appendix Table A1, Panel B), and point estimates if anything strengthen slightly.
14
The smaller GoBifo sample contains only two chiefdoms without permanent RUF bases (according to the NPWJ
report) and thus is not suitable for this analysis.
21
We next turn to retrospective household roster data to further investigate the targeting of violence.
One specific concern for the identification strategy is that the RUF may have systematically targeted
people who were community leaders (other than the traditional authorities) or otherwise very
involved in local affairs. The analysis in Table 3, while informative regarding selection on
socioeconomic status, does not allow us to rule out targeting on unobserved leadership qualities
The retrospective household roster data was collected in the 2005 GoBifo survey, and
contains information on all household members alive in 1990 (before the war). It includes two
variables that allow us to test for targeting on leadership: whether the person was herself/himself a
victim of violence during the war, and whether she/he ever held a community leadership position, for
example, being the leader of a women’s group or a farmer’s group. We estimate the relationship
Ideally, we could observe community leadership positions during the prewar period but
unfortunately, the vague wording of the survey question left it unclear when exactly the person in
question had held a community leadership position, since the wording did not explicitly specify that
we were interested in the prewar period alone. In order to focus more precisely on prewar leaders, in
the next analysis we restrict the sample to those aged 45 or older in 1990.15 These individuals were
over 60 years old by the time of the 2005 survey, making it unlikely that they became community
leaders only after the conflict ended. It appears more likely that any community leadership role was
either before the conflict or in the early years of the conflict. This is not a perfect test, and some
endogeneity (conflict victimization affecting the subsequent likelihood of being a community leader)
remains possible. Still, to the extent the endogeneity bias is positive, it would lead us overstate the
positive correlation between victimization and community leadership, making a finding of a near
15
The results are similar if the sample is restricted to people aged 40 years or older in 1990 (not shown).
22
zero correlation between these two variables even more persuasive evidence that community leaders
Among those aged 45 or older in 1990, the community leader indicator variable is unrelated
to conflict victimization (Appendix Table A2). The point estimate is positive, but small and not
statistically significant. As before, education is also unrelated to victimization during the conflict.
Somewhat surprisingly, being a traditional authority is only weakly related to victimization in this
older subsample, suggesting that younger members of chiefly families bore the brunt of the violence,
Chiefdom level violence intensity is not robustly correlated with postwar outcomes in terms of
socioeconomic status measures, public goods provision in education, and the number of NGO
projects. Thus war violence appears to have a more decisive impact on preferences and values at the
We find no substantial lingering negative effects of the war on 2004 consumption expenditure levels
using either measure of conflict violence (the average conflict victimization, and the number of
attacks and battles). The specifications include geographic controls, district fixed effects, and finally
controls for prewar 1989 log per capita expenditures (Table 7, regressions 1-3). If anything, areas
that suffered from more violence victimization have slightly higher postwar consumption, although
effects are never statistically significant.16 In contrast, the number of diamond mines in the district is
robustly positively associated with higher local living standards in all specifications.
16
One possible partial explanation for the rapid postwar economic recovery is improved soil fertility: land was often
left fallow in areas that experienced more violence and population displacement, and this could have resulted in
temporarily higher postwar yields. This remains speculative in the absence of detailed soil data, unfortunately.
23
We next estimate the relationship between conflict and a number of chiefdom level
socioeconomic and public goods outcomes, focusing on a specification that includes all 152
chiefdoms and controls for district fixed effects and chiefdom geographic characteristics. Results are
similar for regressions including the 1989 prewar per capita log expenditure control (regressions not
shown), although the sample is considerably smaller in that case. Neither 2004 log per capita
consumption expenditures (reproducing the Table 7 result), proportion of children enrolled in school,
nor child body mass index (BMI) are significantly associated with conflict victimization in a
chiefdom (Table 8, panel A). Conflict victimization is also not significantly related to local primary
schooling outcomes, including teacher attendance, outside assistance, visits by chiefs, or local
educational attainment levels (Table 8, panels B and C). The one exception is that chiefdoms with
greater civilian victimization were significantly more likely to have successful community
fundraising for their primary schools (row 5). This suggests that conflict-affected chiefdoms have
perhaps slightly better local public goods fundraising three years after the war, consistent with the
One concern with the chiefdom level results is that chiefdoms heavily affected by the war could have
received increased amounts of NGO and donor funding in the postwar period. Not only do war
affected chiefdoms not get more NGO projects, we find that they may even receive relatively fewer
projects (Table 8, row 9). This could, in part, be due to the fact that some of the most conflict-
affected areas were not declared safe for development workers until late 2002, up to a year or more
A second issue is whether chiefdom level impacts are larger in areas where chiefs were
themselves attacked or killed in the violence. We use the NPWJ report to construct an indicator
variable for this type of violence against traditional authorities, and when this measure is included as
24
an additional explanatory variable we find that the coefficient estimate is never statistically
significant at traditional confidence levels for any of the variables in Table 8 or for those in Table 6
(regressions not shown). Thus attacks on traditional leaders do not appear to be the key drivers of the
political and collective action impacts we estimate, again reinforcing the view that individual level
VI. Conclusion
Using unique nationally representative household data for a postwar society, we find that individuals
who directly experienced violence during the recent Sierra Leone civil war are no different in terms
of postwar socioeconomic status, but they display dramatically higher levels of political mobilization
and engagement, and contributions to local public goods than non-victims. Individuals whose
households were conflict victims are more likely to attend community meetings, more likely to
register to vote / vote, more likely to participate in road brushing (maintenance), and possess more
awareness of local politics. This relationship is remarkably robust across two survey samples and
multiple specifications with different levels of control. While we cannot rule out the possibility that
omitted variable bias is playing some role – in that the types of people victimized tended to be those
who would have become postwar local leaders anyway – there is no strong evidence that more
educated people or community leaders were targeted. ,Additional tests – namely, demonstrating
robustness in the youth subsample and in chiefdoms without permanent RUF bases, where conflict-
related violence victimization is likely to be more indiscriminate or random – argue against the
hypothesis that the systematic targeting of community leaders is driving the results.
Chiefdoms that experienced more intense fighting and abuse of civilians during the conflict
are not relatively worse off three years later in terms of socioeconomic outcomes or child nutrition,
or in terms of education public goods. If anything, there is a slight indication of perhaps somewhat
better local collective action in the more affected chiefdoms, but these chiefdom level results are is
25
not as robust as impacts observed at the individual level. Taken together, it appears that the Sierra
Leone civil war had its largest impacts on preferences and values related to political activity at the
individual level.
The increased local political mobilization we document could potentially help promote future
economic development in Sierra Leone rather than hinder it. For example, we find that contributions
to a pure local public good – road brushing – are higher among war victims. These individual
contributions cannot simply be interpreted as a response to increased local problems, since the village
fixed effects control for any village-wide needs, but rather appear to reflect changes in individual
preferences and values. If this results in better provision of local public goods, it can be thought of as
The finding that the civil war was transformative resonates with the observations of other
Sierra Leone scholars. David Keen (2005: 170) has claimed that the “experience of displacement
and to some extent the exposure to aid organizations seems to have produced a heightened awareness
among many ordinary Sierra Leoneans”, and among youths in particular. Ferme also discusses the
potential to forge something positive out of the horrors of war: “[Sierra Leonans] have sometimes
turned [social instability] into a creative, though violent, opportunity to refashion themselves vis-à-
More research needs to be done to understand the legacies of civil wars in Africa, especially
since our empirical strategy only provides evidence on localized conflict impacts rather than overall
national effects. Yet, more speculatively, the finding that war victimization can increase political
mobilization and local collective action may help make sense of the rapid economic growth and
political consolidation many African countries have experienced following protracted civil wars.
The humanitarian costs of civil wars are horrific, of course, but it appears their postwar economic
26
References
Achen, Christopher H. and Larry M. Bartles. 2004. “Blind Retrospection – Electoral Responses to
Drought, Flu and Shark Attacks” Manuscript, Princeton Univ.
Blattman, Christopher. 2006. “The Consequences of Child Soldiering” Manuscript, Univ. of
California, Berkeley.
Brakman, Steven, Harry Garretsen and Marc Schramm. 2004. “The Strategic Bombing of Cities in
Germany in World War II and it Impact on City Growth.” Journal of Economic Geography 4 (1):
1-18.
Castillo, Marco, and Michael Carter. 2005. “Identifying Social Effects with Economic Field
Experiments.” Manuscript, Univ. of Wisconsin.
Carmil, Devora and Shlomo Breznitz. 1991. “Personal Trauma and World View – Are extremely
stressful experiences related to political attitudes, religious beliefs, and future orientation?”
Journal of Traumatic Stress 4(3): 393 – 405.
Collier, Paul and Anke Hoeffler. 2004. “Greed and grievance in civil war.” Oxford Economic Papers
56 (4): 563 – 595.
Conibere, Richard, Jana Asher, Kristen Cibelli, Jana Dudukovich, Rafe Kaplan, and Patrick Ball.
2004. “Statistical Appendix to the Report of Truth and Reconciliation Commission, Report of
Sierra Leone.” Human Rights Data Analysis Group, The Benetech Initiative.
Davis, Donald R., and David E. Weinstein. 2002. “Bones, Bombs, and Break Points: The Geography
of Economic Activity.” American Economic Review 92 (5): 1269-1289.
Drèze, Jean. 2002. “Militarism, Development and Democracy.” Economic and Political Weekly
April: 1171-1183.
Dyregrov, Atle, Rolf Gjestad and Magne Raundalen. 2002. “Children exposed to warfare: A
longitudinal study.” Journal of Traumatic Stress 15 (1): 59-68
Ferme, Mariane C. 2001. The Underneath of Things: Violence, History and the Everyday in Sierra
Leone. Berkeley: Univ. of California Press.
Green, Donald and Ian Shapiro. 1994. Pathologies of Rational Choice. New Haven: Yale Univ.
Press
Hastings, Justine, Thomas Kane, and Douglas Staiger. 2006. “Economic Outcomes and the Decision
to Vote: the Effect of Randomized School Admissions on Voter Participation.” NBER Working
Paper No. 11805.
Human Rights Watch. 1999. “Sierra Leone: Getting Away with Murder, Mutilation, and Rape.” New
York: Human Rights Watch.
Humphries, Macartan and Jeremy M. Weinstein. 2006. “Handling and Manhandling Civilians in
Civil War: Determinants of the Strategies of Warring Factions.” American Political Science
Review 100 (3): 429-447.
Keen, David. 2005. Conflict and Collusion in Sierra Leone. London: James Currey; New York:
Palgrave.
Miguel, Edward A. and Roland, Gerard. 2005. “The Long Run Impact of Bombing Vietnam”
Manuscript, Univ. of California, Berkeley.
Olson, Mancur. 1984. The Rise and Decline of Nations. New Haven: Yale Univ. Press.
Powell, S. et al. 2003.“Posttraumatic Growth After War: A Study with Former Refugees and
Displaced People in Sarajevo,” Journal of Clinical Psychology, 59(1): 71-83.
Reno, William. 1995. Corruption and State Politics in Sierra Leone. Cambridge and New York:
Cambridge Univ. Press.
Richards, Paul. 1996. Fighting for the Rainforest: War, Youth and Resources in Sierra Leone.
London: James Currey; Portsmouth, NH: Heinemann for the International African Institute.
27
Richards, Paul. 2003. “The Political Economy of Internal Conflict in Sierra Leone.” Working Paper
no. 21, Netherlands Institute of International Relations.
Punamaki, Raija-Leena, Samir Quota and Eyad El Sarraj. 1997. “Relationships Between Traumatic
Events, Children’s Gender, Political Activity, and Perceptions of Parenting Styles.” International
Journal of Behavioral Development 21(1): 91-109
Simonsohn, U., N. Karlsson, G. Loewenstein, and D. Ariely. 2006. “The Tree of Experience in the
Forest of Information: Overweighing Experienced Relative to Observed Information” Manuscript
Univ. of Pennsylvania.
Smith, L. Alison, Catherine Gambette, and Thomas Langley. 2004. “Conflict Mapping in Sierra
Leone: Violations of International Humanitarian Law from 1991 to 2002.” No Peace Without
Justice March.
Sparks, Allister. 2003. Beyond the Miracle: Inside the New South Africa. Chicago: Univ. of Chicago
Press.
Tedeschi, R.G., and L.G. Calhoun. 1996, “The Posttraumatic Growth Inventory,” Journal of
Traumatic Stress, 9: 455-471.
Tilly, Charles H. 1975. The Formation of National States in Western Europe. Princeton NJ:
Princeton Univ. Press.
Timpone, Richard J. 1998 “Structure, Behavior, and Voter Turnout in the United States” American
Political Science Review, 92 (1): 145-158
United Nations. 1993. Human Development Report 1993. New York: United Nations Development
Program, Oxford Univ. Press.
United Nations. 2004. Human Development Report 2004. New York: United Nations Development
Program, Oxford Univ. Press.
World Bank. 2003. Breaking the Conflict Trap: Civil War and Development Policy. Washington,
DC: World Bank.
28
Table 1: Descriptive Statistics
29
Table 1 (continued): Descriptive Statistics
Notes: Sources: (a) Institutional Reform and Capacity Building Project, 2005 Household Survey (b) GoBifo
Household Survey, 2005 (c) No Peace Without Justice Conflict Mapping Report, 2004 (d) Sierra Leone Integrated
Household Survey, 2003-2004 (e) Encyclopedia of Sierra Leone, Sierra Leone Information Systems, 2003 (f) Sierra
Leone Household Survey, 1989 (g) GIS Data, Government of Sierra Leone, 2002 (h) Sierra Leone School
Monitoring Survey, 2005
There are some differences in questions across the IRCBP and GoBifo surveys. First, some questions in the IRCBP
survey were not included in the GoBifo survey and vice versa. Second, the wording of some questions is different.
These are indicated in the table and include: (1) The IRCBP survey asks “Did you register to vote for either of the
last two elections?”; the GoBifo survey asks “Did you vote in either of the last two elections?”. (2) The IRCBP
survey asks “Are you a member of a church/mosque group?”; the GoBifo survey asks “Have you attended
church/mosque in the past month?”. (3) The IRCBP survey asks “How much do you trust members of your
community?”; the GoBifo asks three hypothetical questions that measure trust in different situations, and the
average of those three questions is the overall trust measure. (4) The IRCBP survey asks “Did you work for wages in
the past year?”; the GoBifo survey asks about specific activities that earn cash. Freetown (the capital city) is
excluded from every sample. The IRCBP sample is designed to be nationally representative. Each enumeration
area in the IRCBP sample represents a distinct village because the only major city has been excluded from the
analysis. Due to survey sampling design, there are 117 observations for the 2004 socio-economic variables (source
(d)), 64 observations for the 1989 socio-economic variables (source (f)), and 104 observations for the school survey
data (source (h)).
30
Table 2: Chiefdom Level Correlations with Conflict Intensity
Notes: Additional controls in all regressions include number of chiefdom non-diamond mines and the river density. In regressions (2), (3), (5), and (6) district
fixed effects are included for Tonkolili, Pujehun, Port Loko, Moyamba, Kono, Koinadugu, Kono, Kenema, Kambia, Bonthe, Bombali, and Bo Districts; Western
Area Rural District is the omitted district. Robust standard errors are reported. Standard errors are clustered at the district level in all regressions. Significantly
different than zero at * 90% confidence,,** 95% confidence,,*** 99% confidence. The coefficient on log per capita expenditure in column (6) is robust to
excluding Western Area Rural from the regression sample.
31
Table 3: Household Level Correlations with Conflict Intensity
Dependent Variable:
Household Conflict Victimization Index
IRCBP Sample GoBifo Sample
Explanatory Variables (1) (2) (3) (4) (5) (6)
0.0074 0.0070 0.0070 -0.0001 0.0004 0.0000
Respondent is female
(0.0078) (0.0079) (0.0081) (0.0067) (0.0068) (0.0070)
0.0003 0.0002 0.0003 0.0000 0.0000 -0.0000
Respondent age
(0.0003) (0.0003) (0.0003) (0.0002) (0.0002) (0.0002)
0.0098 0.011 0.015 0.018** 0.021** 0.019**
Respondent has any education
(0.0094) (0.010) (0.010) (0.0085) (0.0088) (0.0096)
0.033*** 0.034*** 0.028** 0.022** 0.024*** 0.025***
Traditional authority household
(0.012) (0.013) (0.013) (0.0087) (0.0089) (0.0097)
0.0083*** 0.0084*** 0.0086***
Household size in 1990
(0.0008) (0.0008) (0.0009)
0.47*** 0.28*** 0.51*** 0.48***
EA/Village conflict victimization index
(0.040) (0.054) (0.051) (0.054)
0.14*** 0.33***
Chiefdom conflict victimization index
(0.048) (0.10)
Notes: Regression (3) includes enumeration area fixed effects. Regression (6) includes village fixed effects. The enumeration area conflict victimization index
in regressions (1) and (2) is constructed as a simple average of the household conflict victimization indexes in the enumeration area excluding that household; it
has mean 0.37 and standard deviation 0.20. The village conflict victimization index in (4) and (5) is constructed analogously; it has mean 0.35 and standard
deviation 0.088. In the IRCBP sample, a traditional authority household is defined by whether the respondent has ever held a position of traditional authority.
In the GoBifo sample, a traditional authority household is defined by whether that household is a member of ruling family, which does not necessarily mean
somebody from that household has ever held a position of traditional authority. Respondent’s age has been demeaned. Robust standard errors are reported.
Standard errors are clustered at the enumeration area in all the IRCBP regressions and at the village level in all the GoBifo regressions. Significantly different
than zero at * 90% confidence,,** 95% confidence,,*** 99% confidence.
32
Table 4: Community Meetings and Conflict Victimization
Dependent Variable:
Did you attend any community meetings in the past year?
IRCBP Sample GoBifo Sample
Explanatory Variables (1) (2) (3) (4) (5) (6)
*** ** ** ** **
0.072 0.057 0.057 0.098 0.10 0.12**
Household conflict victimization index
(0.022) (0.022) (0.024) (0.045) (0.045) (0.050)
-0.13*** -0.13*** -0.13*** -0.16*** -0.16*** -0.16***
Respondent is female
(0.011) (0.011) (0.012) (0.017) (0.017) (0.018)
0.0006 0.0006 0.0004 0.0010* 0.0011* 0.0014**
Respondent’s age
(0.0004) (0.0004) (0.0004) (0.0006) (0.0006) (0.0007)
0.042*** 0.060*** 0.061*** 0.054*** 0.044** 0.049**
Respondent has any education
(0.014) (0.014) (0.015) (0.021) (0.021) (0.023)
0.091*** 0.077*** 0.071*** 0.067*** 0.056*** 0.049**
Traditional authority household
(0.015) (0.016) (0.017) (0.021) (0.021) (0.023)
0.0013 0.0010 0.0005
Household size in 1990
(0.0020) (0.0072) (0.0021)
-0.011 -0.004 -0.18 -0.17
EA/Village conflict victimization index
(0.071) (0.066) (0.13) (0.12)
0.49*** 0.67***
Chiefdom conflict victimization index
(0.11) (0.32)
Notes: Significantly different than zero at * 90% confidence,,** 95% confidence,,*** 99% confidence. Robust standard errors are clustered at the enumeration area
and village level in all IRCBP and GoBifo regressions, respectively. The EA/Village conflict victimization index is as described in the notes for Table 3. In the
IRCBP sample, a traditional authority household is defined by whether the respondent has ever been a traditional leader (chief). In the GoBifo sample, a
traditional authority household is defined by whether that household is a member of ruling family, which does not necessarily mean somebody from that
household has ever held a position of traditional authority. Respondent’s age has been demeaned.
33
Table 5: Voting and Conflict Victimization
Dependent Variable:
Did you register to vote / vote in the either of the past two elections?
IRCBP Sample GoBifo Sample
Explanatory Variables (1) (2) (3) (4) (5) (6)
*** *** *** *** ***
0.026 0.026 0.024 0.10 0.10 0.094***
Household conflict victimization index
(0.0079) (0.0082) (0.0093) (0.026) (0.026) (0.029)
-0.015*** -0.015*** -0.014*** -0.033*** -0.0364*** -0.035***
Respondent is female
(0.0049) (0.0050) (0.0052) (0.010) (0.010) (0.011)
0.0007*** 0.0007*** 0.0007*** 0.0012*** 0.0012*** 0.0013***
Respondent’s age
(0.0002) (0.0002) (0.0002) (0.0003) (0.0003) (0.0004)
-0.0050 -0.0045 -0.0021 0.011* 0.0049 0.0029
Respondent has any education
(0.0052) (0.0054) (0.0057) (0.012) (0.012) (0.013)
0.0048 0.0066 0.0077 0.017* 0.018* 0.022*
Traditional authority household
(0.0061) (0.0064) (0.0072) (0.012) (0.012) (0.014)
-0.0010 -0.0008 -0.0012
Household size in 1990
(0.0012) (0.0012) (0.0014)
0.0093 0.0073 0.076 0.076
EA/Village conflict victimization index
(0.022) (0.021) (0.061) (0.063)
0.012 0.0044
Chiefdom conflict victimization index
(0.037) (0.16)
Notes: Significantly different than zero at * 90% confidence,,** 95% confidence,,*** 99% confidence. Robust standard errors are clustered at the enumeration area
and village level in all IRCBP and GoBifo regressions, respectively. The dependent variable for the IRCBP sample is “Did you register to vote in either of the
past two elections?”, in the GoBifo sample the dependent variable is “Did you vote in either of the past two elections?”. The EA/Village conflict victimization
index is as described in the notes for Table 3. In the IRCBP sample, a traditional authority household is defined by whether the respondent has ever been a
traditional leader (chief). In the GoBifo sample, a traditional authority household is defined by whether that household is a member of ruling family, which does
not necessarily mean somebody from that household has ever held a position of traditional authority. Respondent’s age has been demeaned.
34
Table 6: Household Level Postwar Outcomes and Conflict Victimization
Notes: Each entry is from a separate OLS regression. Robust standard errors are reported. Standard errors are
clustered at the enumeration area in all the IRCBP regressions and at the village level in all the GoBifo regressions.
Significantly different than zero at * 90% confidence,,** 95% confidence,,*** 99% confidence. The specification is
analogous to regressions (3) and (6) in Tables 5 and 6: additional controls include respondent’s gender, age,
education, traditional authority household and household size in 1990. Enumeration area fixed effects are included
in all IRCBP regressions and village fixed effects in all GoBifo regressions.
35
Table 7: 2004 Log Per Capita Expenditure and Conflict Victimization
Dependent Variable:
Log per capita expenditures, 2004
Explanatory Variable (1) (2) (3)
0.53 0.42 0.
Chiefdom conflict victimization index
(0.49) (0.39) (0.63)
-0.0054 -0.0063 -0.0037
Number of attacks and battles
(0.0041) (0.0055) (0.012)
0.028*** 0.025*** 0.016**
Number of diamond mines
(0.0032) (0.0042) (0.0069)
0.29 0.80 0.69
Road density
(0.73) (0.56) (1.22)
-0.32** -0.12 0.33
Log distance to Freetown
(0.11) (0.12) (0.21)
-0.096** -0.016 -0.042
Log population density, 1985
(0.041) (0.042) (0.091)
0.034
Proportion children in school, 1989
(0.41)
0.065
Log per capita expenditure, 1989
(0.097)
Notes: Robust standard errors are reported. Standard errors are clustered at the district level in all regressions.
Significantly different than zero at * 90% confidence,,** 95% confidence,,*** 99% confidence. Due to sampling in the
2004 household survey, the sample size is smaller then the full sample of 152 chiefdoms. Additional controls in all
regressions include number of chiefdom non-diamond mines and the river density. In regressions (2) and (3) district
fixed effects are included for Tonkolili, Pujehun, Port Loko, Moyamba, Kono, Koinadugu, Kono, Kenema, Kambia,
Bonthe, Bombali, and Bo Districts; Western Area Rural District is the omitted district.
36
Table 8: Chiefdom-Level Outcomes and Conflict Victimization
Chiefdom Conflict
Victimization Index:
Dependent Variables Coefficient (std. error)
Panel A: Postwar Socio-Economic Outcomes
0.42
1. Log per capita expenditure, 2004
(0.39)
0.17
2. Proportion children enrolled in school, 2004
(0.17)
3.89
3. BMI for children, 2004
(8.94)
Panel B: School Quality Outcomes, 2005
-0.045
4. Proportion of teachers absent on day of survey
(0.16)
0.62*
5. Proportion of schools receiving financial / in-kind resources from community
(0.33)
-0.11
6. Proportion of schools receiving financial / in-kind resources from donors or NGOs
(0.39)
-0.29
7. Proportion of schools visited by a traditional authority in past year
(0.44)
Panel C: Adult Education and NGO Projects in 2004
0.038
8. Proportion of adults with any education, 2004
(0.083)
-27.89*
9. Total number of NGO projects
(16.66)
Notes: Each coefficient and standard error are from a separate OLS regression. Robust standard errors are reported.
Standard errors are clustered at the district level in all regressions. Significantly different than zero at * 90%
confidence,** 95% confidence,,*** 99% confidence.
Due to sampling, there are 117 observations in rows 1-3; 104 chiefdoms in rows 4-7; and 152 chiefdoms in rows 8-
9. The Number of NGO projects in row 9 includes all reported education, health, and agriculture NGO projects.
The specification in these regressions is equivalent to regression (2) in Table 8: additional explanatory variables
include number of attacks and battles, number of diamond mines, road density, log distance to Freetown, log
population density in 1985, number of non-diamond mines, and river density, district fixed effects are also included
(see notes in Table 8).
37
Figure 1: Location of Sample Enumeration Areas in IRCBP Sample
Notes: The IRCBP sample is nationally representative. There are a total of 539 enumeration areas in the sample.
There are 23 enumeration areas that not included above due to missing GPS coordinates. The capital – Freetown – is
not included in the sample; it is represented by a star on the map.
Notes: There are 235 villages in the GoBifo sample. The sample covers two districts: Bombali in the North and
Bonthe in the South. The sample is not designed to be representative of either district. There are 51 sample villages
not included above due to missing GPS coordinates. The capital – Freetown – is not included in the sample; it is
represented by a star on the map.
38
Figure 3: Chiefdom Conflict Victimization Index
Notes: The Conflict Victimization Index is the chiefdom average of three conflict related questions in the IRCBP
survey. Chiefdoms are shaded in deciles according to the value of the conflict index. Data is missing for
Gbonkolenken chiefdom, leaving a sample size of 151 chiefdoms.
Notes: The residuals in this figure are from a regression of the conflict index on a set of district fixed effects. Thus,
this map shows the variation being used in all of the specifications that include district fixed effects. Chiefdoms are
shaded in deciles according to the value of these residuals. Data is missing for Gbonkolenken chiefdom, leaving a
sample size of 151 chiefdoms.
39
Appendix Table A1: Chiefdoms with no RUF bases
Household Level Postwar Outcomes and Conflict Victimization
Observations 2535
EA fixed effects X
Notes: Each entry is from a separate OLS regression. Robust standard errors are reported. Standard errors are
clustered at the enumeration area in all the IRCBP regressions. Significantly different than zero at * 90%
confidence,,** 95% confidence,,*** 99% confidence. The specification is analogous to regressions (3) and (6) in
Tables 5 and 6: additional controls include respondent’s gender, age, education, traditional authority household and
household size in 1990. Enumeration area fixed effects are included in all IRCBP regressions.
40
Appendix Table A2: Individual “Selection” into Victimization
GoBifo 1990 Household Roster Data – 45 years and older in 1990
Dependent Variable
Explanatory Variable Victim of violence during the conflict
0.0011
Ever held a community leadership position
(0.027)
-0.034***
Female
(0.013)
-0.0007
Age in 1990
(0.0006)
0.017
Ever been to school
(0.024)
-0.0073
Ever held a traditional leadership position
(0.021)
R-squared 0.054
Observations 1796
Chiefdom Fixed Effects X
Mean of dependent Variable 0.07
Notes: Robust standard errors are clustered at chiefdom level. Significantly different than zero at * 90%
confidence,,** 95% confidence,,*** 99% confidence
The data come from the 2005 GoBifo retrospective household roster. A complete roster was compiled for each
respondent’s household in 1990. From that list, a random sample of five people were selected. The data consists of
information on those five people, who all lived in the respondent’s household in 1990 but do not necessarily live
with the respondent today. For reasons discussed in the text, this sample is restricted to persons over 45 years of age
in 1990, which is approximately equivalent to keeping the top quintile of the sample.
The means (std. dev) of the explanatory variables are as follows: Ever held a community leadership position 0.06
(0.24), Female 0.49 (0.50), Age in 1990 55.99 (9.75), Ever been in school 0.11 (0.31), Ever held a position of
traditional authority 0.11 (0.31). Traditional authority includes Paramount, Section, and Village chiefs. Community
leadership position includes women’s leader, youth leader, head teacher, school committee chair, imam/reverend
and master farmer. Victim of violence corresponds to “injured, maimed or killed” during the conflict.
Chiefdom fixed effects for the chiefdom of residence in 1990 are included in all regressions. Results do not change
substantially without the fixed effects. Results do not change substantially if the sample is instead restricted to
individuals over 40 years old in 1990 (not shown).
41
Data Appendix
Institutional Reform and Capacity and Building Project (IRCBP) Survey, 2005
The IRCBP overall supports the ongoing decentralization in Sierra Leone, working closely with the newly
elected Local Councils to strengthen local government. The 2005 IRCBP survey provides measures of
conflict victimization and measures of local institutional outcomes. The IRCBP survey collected
information on the provision of public services, attitudes and perceptions of local government, as well as
some demographic and socioeconomic variables. The survey was designed to be nationally representative.
Each of the 13 districts is included in the sample. Data is missing for Gbonkolenken chiefdom, which
leaves 151 chiefdoms in all . A total of 539 enumeration areas were surveyed. The sample size for all
individual level regressions is 5,138 individuals.
The GoBifo survey includes extensive household and individual data on community participation, social
capital, as well as questions on experience during the conflict. The GoBifo survey also included a
retrospective household roster. Each respondent was asked to name everybody who was living in the
same household as them in 1990. This list was to include people who are no longer living with the
respondent or are no longer alive. From that list, five people were chosen randomly and data was
collected on them. The total sample size is 13,280 individuals. For reasons discussed in the text, the
analysis included in this paper is restricted to individuals who were 45 years of age or older in 1990.
The measure used in our analysis is the number of attacks and battles that occurred within each chiefdom.
An attack is defined to be an incident in which an armed group came into a village briefly, burned houses,
raped or killed residents. It is common for attacks to be part of a larger military campaign and thus for
human rights violations to be committed on a large scale (e.g. “during these attacks RUF forces burnt
down fifty houses, killed nine people, abducted an unknown number of people and amputated a man’s
hand with an axe” p. 189). A battle is defined to be a confrontation between two armed groups (e.g. “On
25 February, the RUF made a successful counter-attack at the rutile mining site, dislodging the SLA
forces based there.” p. 430). Battles need not directly involve violence against civilians, although they
sometimes do. There were 1,995 violent incidents recoded in the NPWJ report, and 1,363 of these
incidents were classified as either an attack or a battle. To give the reader some sense of who the
perpetrators are, of the 968 recorded attacks over 95% were committed by RUF rebels and less than two
percent by CDF. The majority of the battles took place between RUF and CDF troops, with a smaller but
still substantial number also involving the SLA and ECOMOG (West African forces led by Nigeria).
42
Sierra Leone Integrated Household Survey (SLIHS), 2003-2004
Data on chiefdom-level postwar household expenditures, enrollment of children in school, and child body
mass index is available from the 2003-2004 SLIHS survey. The data collection was funded by DFID and
the World Bank, with the intent of providing more complete poverty measures for use in postwar
planning. The data was made available from the office of Statistics Sierra Leone. This national survey
was designed to be representative at the district level. As with the IRCBP survey, the large number of
households in each district allows construction of chiefdom level averages. All of the statistics used in
the present analysis are based on the cleaned sample, which included households located in 117 (out of
152) chiefdoms. Due to sampling strategy, no data was collected for the remaining 35 chiefdoms.
43