Running head: VIOLENCE AND GROUP IDENTIFICATION
Perpetrating Violence Increases Identification with Violent Groups:
Survey Evidence from Former Combatants
Rebecca Littman
Princeton University
November 21, 2016
Author Note
Thanks to Elizabeth Levy Paluck, Joel Cooper, Bethany Park, Stathis Kalyvas, members
of Elizabeth Levy Paluck’s lab, and participants in the CAPERS Conference and
Princeton’s Joint Degree Program in Social Policy for helpful comments. I also thank
Christopher Blattman and Jeannie Annan for providing full access to the datasets used in
the studies. Correspondence should be addressed to Rebecca Littman, Department of
Psychology, Princeton University, Princeton, NJ 08540. E-mail: [email protected].
VIOLENCE AND GROUP IDENTIFICATION
Abstract
Heightened group identification motivates individuals to perpetrate violence, but
can perpetrating violence – in and of itself – increase identification with violent groups?
We test this idea using archival surveys of ex-combatants. In Liberia, where many
combatants joined their violent group willingly, the data show a positive association
between perpetrating violence and identification with one’s violent group (Study 1).
Study 2a replicates and extends this finding with data from ex-combatants in Uganda
who were abducted into their group, using a natural experiment in which some abductees
were forced to perpetrate violence against loved ones while others were not. Only
perpetrating violence, not exposure to violence more generally (Study 2b), and only
violent behaviors that involve a heightened sense of personal responsibility (Study 3) are
associated with increased group identification. These findings support a cycle of violence
in which perpetrating violence increases identification with violent groups and
heightened identification increases future violent behavior.
Keywords: violence, harm, intergroup conflict, group identification, survey
2
VIOLENCE AND GROUP IDENTIFICATION
Perpetrating Violence Increases Identification with Violent Groups:
Survey Evidence from Former Combatants
A top priority for social psychologists is to understand how individuals come to
participate in collective violence such as intergroup conflict and terrorism. Despite claims
that violence is declining over time (Pinker, 2011), countries across the world
experienced more than 16,800 terrorist attacks in the year 2014 alone, resulting in over
43,500 deaths and 40,900 injuries (Miller, 2015). Another 126,059 individuals were
killed in civil conflict in 2014, the highest number of fatalities recorded in a single year
since the Rwandan genocide in 1994 (Melander, 2015). Although it is tempting to think
that most perpetrators of collective violence are sadists or psychopaths who are motivated
by an internal desire to harm others for personal pleasure or instrumental gain (Taylor,
2009; Woodworth & Porter, 2002), this is often not the case (Waller, 2002). Research
shows that even suicide bombers acting on behalf of terrorist organizations generally do
not exhibit psychopathology. Instead they tend to come from educated, middle-class
backgrounds (Atran, 2003).
Social psychologists have long been interested in the questions of how and why
individuals engage in collective violence, overcoming their personal discomfort with
physically harming others (Grossman, 1996). Motivated in part by observations of
extreme intergroup conflict and genocide, such as the Holocaust, the Rwandan genocide,
and the Israeli-Palestinian conflict, psychologists have established that heightened
identification with one’s ingroup can induce individuals to perpetrate violence on behalf
of their group. Under certain conditions, ingroup love can pave the way to outgroup hate,
making violence seem like a reasonable or even virtuous behavior (Fiske & Rai, 2014;
3
VIOLENCE AND GROUP IDENTIFICATION
Reicher, Haslam, & Rath, 2008; Sidanius & Pratto, 1999). Correlational studies and
research from the lab show that highly identified group members, particularly those
whose individual sense of self is fully integrated with that of their group, are more
supportive of and willing to engage in violent behavior on behalf of their ingroup
(Swann, Gomez, Huici, Morales, & Hixon, 2010; Swann, Jetten, Gomez, Whitehouse, &
Bastian, 2012). High identifiers may even come to associate subjective pleasure with
outgroup pain, making it possible to overcome their aversion to harming competitive
outgroup members (Cikara, 2015).
While this research demonstrates that identification with a group can motivate
ordinary people to perpetrate violence, in many cases in the real world, individuals do not
initially identify strongly with their violent group. Many violent groups are attractive for
social or economic reasons. Only after joining are individuals faced with the reality of
having to engage in violent behavior to stay in the group. For example, an individual
might join the military in order to gain transferrable skills or to receive help paying for
college in the future, rather than because of a particular ideology or pre-existing
allegiance (Kleykamp, 2006). Individuals may also be conscripted or even abducted into
violent groups, which are relatively common “recruitment” strategies employed by
militant and insurgent groups. Between 1980 and 2009, 45% of active insurgent groups
recruited through force: 32% through coercion and 13% through abduction (Cohen,
2013). Such “recruits” likely exhibit low levels of identification with their violent group
to begin with, and may even harbor negative attitudes or active resentment toward the
group. In all of these relatively common cases, the encouragement to perpetrate violence
on behalf of the group may come before the individual feels a sense of group
4
VIOLENCE AND GROUP IDENTIFICATION
identification.
In this paper, we examine whether the process of group identification leading to
engagement in violent behavior on behalf of that group also operates in the reverse
direction. That is, does perpetrating violence on behalf of a violent group increase
identification with the group? If perpetrating violence – in and of itself – increases group
identification, this has important implications for future violent behavior and may help to
explain cycles of intergroup conflict over time. Specifically, it suggests a cycle of
violence in which violent behavior increases identification with violent groups and group
identification increases future violent behavior. It also implies that individuals need not
start off as high group identifiers to eventually become loyal group members. Violence
itself may help to trigger this process, even among individuals who are resistant to
joining the group in the first place, or who join only for opportunistic reasons.
Research Hypothesis
In this paper, we propose the following hypothesis: perpetrating violence on
behalf of one’s group will increase identification with the violent group. To better
illustrate this hypothesis, imagine the extreme case of a young man living with his family
in a small village in Northern Uganda. As he sits outside of his hut one evening, his
village is attacked by members of the Lord's Resistance Army (LRA), a rebel group
terrorizing his region. A soldier forces a machete into the young man's hand, and instructs
him to aim the blade at his uncle. Either the young man swings the machete at his uncle,
or he will be harmed by the rebels holding him hostage. After the young man complies
with the soldier's request, he is taken away from his family and friends, now a new
member of the LRA. The act of killing his uncle with a machete could lead the young
5
VIOLENCE AND GROUP IDENTIFICATION
man to feel even more resentment toward the LRA for forcing him to carry out such a
heinous act. Alternatively, and perhaps surprisingly, engaging in such violent behavior
may have the opposite effect of making the young man feel more identified with the
LRA.
Existing Empirical Evidence
To date, there is very little empirical research that predicts how this young man is
most likely to react. One line of research we can look to examines the effect of
experiencing, as opposed to perpetrating, violence on attitudes towards one’s group. Two
classic studies in social psychology demonstrate that undergoing an unpleasant or violent
group initiation – personally experiencing harm in order to join a group – increases the
attractiveness of a new group (Aronson & Mills, 1959; Gerard & Mathewson, 1966). In
these studies, participants listened to a group discussion in the lab but did not interact
with other group members. Similarly, recent experimental work in the lab suggests that
collectively undergoing a painful task in a group setting (small groups of 2-6 university
students) increases self-reported level of identification with the group (Bastian, Jetten, &
Ferris, 2014). However, the only study that investigates this effect using groups outside
of the lab (Dutch sororities) found that undergoing a more severe initiation actually
decreased the attractiveness of the group (Lodewijkx & Syroit, 1997). Additionally, it is
important to note that since all of these studies involved experiencing, as opposed to
perpetrating, violence, it is unclear whether the same psychology applies.
The one previous paper to examine the association between aggression and group
identification finds that engaging in aggressive acts against one’s former ingroup is the
strongest predictor of commitment to one’s new group (Toosi, Masicampo, & Ambady,
6
VIOLENCE AND GROUP IDENTIFICATION
2014). While this finding is an important first step, the paper looks at players who switch
teams in professional basketball and in a live-action tag game. Therefore the aggression
outcomes are not physical violence of the kind we see in situations of conflict. The
researchers suggest that their findings could extend to more extreme scenarios, such as
when civilians become armed insurgents and even to the horrific situation in which child
soldiers are forced to kill a family member. This is what we aim to test in this paper,
using archival data from ex-combatants who fought for violent groups in Liberia and
Uganda and engaged in extreme forms of physical violence, including forced violence
against family members.
Theorizing the Path from Violence to Group Identification
Now let us return to the case of the young man in Northern Uganda who is
abducted into the LRA and forced to perpetrate violence against his uncle. Why do we
hypothesize that engaging in this violent behavior would increase his identification with
the violent group? While there are many possible mechanisms that likely work in tandem,
we will focus on two mechanisms stemming from the aversive and counternormative
nature of violence: (1) individuals increase their group identification in order to justify
the distress (negative physiological arousal) they feel after perpetrating violence, and (2)
individuals fear rejection from meaningful others outside of the violent group, and
therefore increase their identification with the group that supports and rewards their
violent behavior.
The Aversive Nature of Violent Behavior
Research from moral and clinical psychology shows that individuals generally
exhibit an aversion to harming others and will avoid causing harm even at personal cost
7
VIOLENCE AND GROUP IDENTIFICATION
(Crockett, Kurth-Nelson, Siegel, Dayan, & Dolan, 2014; Grossman, 1996). Previous
research finds that when individuals – particularly those who are inexperienced with
violent behavior – do harm others, they tend to experience high levels of psychological
and physiological distress (Maguen et al., 2010; Miller & Cushman, 2013; Miller,
Hannikainen, & Cushman, 2014). This was first demonstrated in the field of social
psychology by Milgram's (1965) electric shock experiments. Although a number of
participants complied with the experimenter's request to shock another individual until he
was presumably gravely injured or dead, Milgram describes the tension that many
participants exhibited as they carried out the task.
More recent lab studies reveal that individuals experience distressing
physiological reactions when they make moral decisions about harming others or
simulate violent actions (Cushman, Gaffey, Gray, & Berry Mendes, 2012; Navarrete et
al., 2012). In terms of psychological distress, the act of killing another human being in
combat has been linked to symptoms of posttraumatic stress disorder (PTSD) and
depression, even when controlling for other war experiences and exposure to violence
(MacNair, 2002; Maguen et al., 2010).
What are the consequences of engaging in aversive behaviors? According to the
theory of cognitive dissonance, when individuals feel distress after engaging in a
behavior that violates personal beliefs or normative standards, they are motivated to
reduce this distress in order to restore a positive view of the self (Festinger, 1957; Stone
& Cooper, 2001). Capitalizing on the aversive nature of violent behavior, early research
used interpersonal harm as a precipitating event through which to study dissonance (e.g.,
Brock & Buss, 1962; Glass, 1964; Lerner & Matthews, 1967). Researchers explored how
8
VIOLENCE AND GROUP IDENTIFICATION
harming others under classic conditions of dissonance-arousal could lead perpetrators to
trivialize the suffering of their victim (Brock & Buss, 1962) or act less friendly toward
their victim (Glass, 1964). In an intergroup context, derogating the target of one’s violent
behavior may extend to derogating the entire outgroup as a way of reducing dissonance
(Cooper & Mackie, 1983; Glasford, Dovidio, & Pratto, 2009).
In addition to derogating one’s victim or trivializing their suffering, we
hypothesize that individuals may reduce distress after engaging in violent behavior by
increasing their identification with the violent group. This is consistent with Festinger’s
(1957) original observation that social groups can be vehicles for reducing cognitive
dissonance. When individuals perpetrate violence on behalf of a group, they generally
receive support and encouragement for their behavior from the other members of their
violent group. The experience of receiving positive feedback from their group for the
aversive violent behavior may help to provide a justification for their actions, thus
reducing their distress and in turn increasing identification with the violent group. For
example, individuals may tell themselves that they must really love their group if they
were willing to harm someone on behalf of the group. They may also simply feel more
positive about the group after receiving support for their aversive behavior from other
group members. In short, individuals may heighten their identification with the violent
group to help reduce their discomfort after perpetrating violence.
The Counternormative Nature of Violent Behavior
In most societies, people are socialized from a young age with the norm of “do no
harm.” Even in violent contexts, at least some of an individual’s family and community
members likely still believe that harming others is not acceptable behavior, and therefore
9
VIOLENCE AND GROUP IDENTIFICATION
do not endorse the individual’s engagement in violent behavior. In this case, we theorize
that engaging in violent behavior on behalf of a group might increase identification with
the violent group through its negative consequences on the individual’s other group
memberships. For example, a young man abducted by the LRA in Uganda may feel like
his family now sees him as a rebel, and does not approve of his violent behavior. In fact
many young people in Uganda abducted by the LRA felt that they “couldn’t go home”
(Otim, 2009). If an individual perceives a lack of social support for their violent behavior
from meaningful others outside of the violent group, this can lead to disidentification
with other social identities (Becker, Tausch, Spears, and Christ, 2011; Glasford, Pratto, &
Dovidio, 2008; McKimmie et al., 2003). In the case of violent groups, this might lead
individuals to disidentify from others who are perceived to be less supportive of violent
behavior, such as the individual’s family and community members back home.
Not only is this perceived lack of support for violent behavior expected to lead to
disidentification with other social groups, it is also expected to lead to increased
identification with the violent group. When individuals feel that others are rejecting them,
this heightens the activation of group constructs and social identities, and leads
individuals to judge their own group as more meaningful and cohesive than other groups
(Knowles & Gardner, 2008). Having fewer other social identities also leads individuals to
identify more strongly with their existing ingroup (Grant & Hogg, 2012). On a practical
level, some violent groups may make it difficult for individuals to leave the group,
threatening to harm the individual or their loved ones if they try to escape. In this case,
the individual may have no other identities available to them apart from the violent
group, causing them to cling to their one available social identity.
10
VIOLENCE AND GROUP IDENTIFICATION
Present Research: Survey Evidence from Former Combatants
In this research, we test the hypothesis that perpetrating violence on behalf of a
group increases group identification. A strong test of this hypothesis, of course, presents
an ethical and methodological challenge. Researchers would need to randomly assign
individuals who belong to the same group to either engage in violent or non-violent
behavior on behalf of the group. Undoubtedly, researchers cannot ethically randomly
assign individuals to engage in extreme forms of violent behavior in the lab or in the
field. While it may be possible to simulate similar psychological processes in the lab
using less extreme manipulations of interpersonal harm, a prior step is to demonstrate that
this phenomenon exists among members of actual violent groups who injure or kill others
on behalf of their group.
In this paper, we use archival data – surveys with former combatants in Liberia
and Uganda – to test the hypothesis that engaging in violent behavior on behalf of a
group increases identification with the violent group in a set of enormously consequential
real world cases. Additionally, in Northern Uganda we study a tragic natural experiment
in which some members of the Lord’s Resistance Army (LRA) were quasi-randomly
assigned to perpetrate violence against loved ones while other group members were not
assigned to perpetrate violence. This natural experiment takes place in one of the most
unlikely contexts in which to find a relationship between violent behavior and group
identification. The LRA was almost entirely comprised of youth abducted from their
homes and forced with the threat of death to join and stay in the violent group. This is a
case in which it should be extremely difficult to produce any level of group identification,
and thus it provides a strict test of the hypothesis. Additionally, while it is not possible to
11
VIOLENCE AND GROUP IDENTIFICATION
causally test the mechanisms through which perpetrating violence increases group
identification using these archival datasets, we are able to conduct a valuable within
dataset test of the distress-based mechanism theorized above.
Study 1: Violent Behavior and Group Identification in Liberia
Study 1 uses survey data from former combatants in Liberia to examine whether
there is a positive association between perpetrating violence and group identification,
controlling for individual differences in war experiences and demographic characteristics.
Study Context
Liberia, a small West African nation of roughly 3.5 million people, experienced
two long and brutal civil wars between 1989 and 2003. In 1989, following a decade of
increasing political tensions and human rights abuses, Charles Taylor's National Patriotic
Front of Liberia (NPFL) began a revolt, which quickly escalated into a civil war. Taylor's
government proved to be little different from that of the previous regime, and by 1999
two anti-Taylor groups emerged to fight the government. The civil wars in Liberia left
nearly 150,000 people dead, many of them civilians, and led to the complete breakdown
of law and order (Ellis, 1999).
Description of Data
The archival data used in this study come from surveys with 1,099 former
combatants (93.4% male), collected by Blattman and Annan (2016) from August through
October 2009, approximately six years after the end of Liberia’s civil crisis. The data
were collected as part of the registration process for an ex-combatant agricultural training
program run by the non-governmental organization Action on Armed Violence (Blattman
12
VIOLENCE AND GROUP IDENTIFICATION
& Annan, 2016; see Appendix 1 for more information on the survey and data collection
procedures).
Analytic Approach
In this study, we test whether there is a positive association between the violence
that ex-combatants report committing during the war and their violent group
identification using ordinary least squares (OLS) regression. We predict group
identification by violence perpetrated on behalf of the group, controlling for the
individual differences in demographic characteristics and war experiences described
below, and calculating robust standard errors.
Measures
Independent variable: Violent behavior. In a section on war experiences that
appeared approximately halfway through the survey, respondents were asked: “How
many violent acts were you forced to commit: none, few, some, or plenty?"
Dependent variable: Group identification. The dependent variable is an index
of identification with one’s violent group, the respondent’s fighting faction during
Liberia’s civil war. After discussing their war experiences with the interviewer,
respondents were asked a series of questions on their experience in their violent group.
This included two questions that tap into different dimensions of identification: “Did you
enjoy being a member of this faction?” and “Was there a time when you really believed
in this faction with your heart?” Answer options were on a four-point scale of “not at all,”
“small,” “some,” or “plenty.” Responses to these two questions were averaged to create a
group identification index.
Covariates. We control for a number of variables to account for individual
13
VIOLENCE AND GROUP IDENTIFICATION
differences in demographic characteristics and war experiences: age, gender, religion (an
indicator for whether the respondent is Christian), and ethnicity (indicators for the five
major ethnic groups represented in the sample). Additionally, we control for individual
differences in exposure to different types of violence using indices of violence witnessed,
violence personally experienced, and violence experienced by loved ones (see Appendix
2 for a full list of the questions in these indices). We also control for a number of other
war experience variables such as length of time the respondent spent in their violent
group, whether they were given a gun to carry, and whether they ever had other soldiers
under their command (see Table 1 for the full list of war experience covariates).
Results
Respondents in this sample come from eleven fighting groups active during
Liberia’s civil conflict. The mean length of time spent in one’s group was 27.6 months
(sd = 29.4), and the mean age at joining the group was 19.6 years old (sd = 6.78). Over
half of respondents reported being abducted into their fighting group (57%), while the
remaining 43% reported joining voluntarily. Turning to engagement in violent behavior,
66.4% of respondents reported perpetrating no violent acts, 17.9% reported “few” acts
perpetrated, 5.7% “some” acts, and 10% “plenty” acts. When describing their level of
identification with their fighting group, about 40% of respondents reported that they
enjoyed being in their violent group and that they really believed in the group with their
heart, at least a small amount of the time (see Table 1 for more descriptive statistics).
14
VIOLENCE AND GROUP IDENTIFICATION
Table 1: Descriptive statistics for the Liberian ex-combatant sample
Index of group identification (0-3)
Enjoyed being a member of the violent
group (0-3)
Believed in the group with their heart (0-3)
Index of violence perpetrated (0-1)
Index of violence personally experienced (0-1)
Index of violence witnessed (0-1)
Index of violence experienced by loved ones
(0-1)
Length of time in violent group (months)
Age when joined violent group
Carried a gun while in violent group
On the frontlines while in violent group
Main role in violent group was as a soldier
Held rank in violent group
Held leadership position in violent group
Abducted into violent group
Most common violent groups:
Member of LURD
Member of NPFL
Member of MODEL
Member of GoL
Member of LPC
Mean
0.88 (1.14)
N
1099
0.86 (1.21)
1099
0.90 (1.24)
0.20 (0.33)
0.49 (0.24)
0.78 (0.25)
1099
1077
1077
1077
0.51 (0.21)
1077
27.6 (29.4)
19.6 (6.8)
79%
18%
71%
30%
42%
57%
1094
1090
1096
1099
1099
1099
1099
1099
28%
27%
12%
12%
8%
1099
1099
1099
1099
1099
Note: This table includes descriptive statistics for the independent variable,
dependent variable, and all war experience covariates used in the regression analysis for
Study 1.
Association between perpetrating violence and group identification. In this
sample of former combatants from eleven violent groups in Liberia, there is a significant
positive association between engaging in violent behavior and level of identification with
the violent group. These results hold when controlling for individual differences in war
experiences and demographic characteristics (β=0.304, CI = 0.092 – 0.515, p=0.005), as
well as when excluding all covariates (β=0.273, CI = 0.057 – 0.490, p=0.013; see
15
VIOLENCE AND GROUP IDENTIFICATION
Appendix 4 Table 1 for the full regression results). Additionally, the findings remain
consistent when analyzing the two items in the group identification index separately
(β=0.372, CI = 0.148 – 0.595, p=0.001 for enjoy; β=0.236, CI = 0.005 – 0.467, p=0.045
for believe).
Discussion
Survey data from former members of violent groups in Liberia provides initial
evidence of a positive association between perpetrating violence and identification with
one’s violent group. However, this sample has a number of limitations, which are
characteristic of the correlational data obtained in this literature more generally. Chief
among them are issues of causality. First, there is an issue of directionality; the
association between perpetrating violence and group identification could be explained in
full by more highly identified group members engaging in more violent behavior. For
example, highly identified group members may seek out opportunities to engage in
violence and may be more likely to comply when ordered to carry out violent acts.
Second, since perpetration of violence is not randomly assigned, we cannot rule out the
possibility that the positive association between group identification and violent behavior
is driven by a third, unobserved variable. In other words, we cannot say with certainty
that the relationship between violent behavior and group identification is causal. In the
next set of studies, we use survey data from former combatants in Uganda that addresses
these limitations.
16
VIOLENCE AND GROUP IDENTIFICATION
Study 2a: Violent Behavior and Group Identification among Members of the Lord’s
Resistance Army in Uganda
In Study 2a, we replicate and extend the findings from Study 1, using survey data
from former combatants who were abducted into the Lord’s Resistance Army (LRA) in
Northern Uganda. We first test whether the positive association between perpetrating
violence and group identification replicates in this sample, and then we examine whether
quasi-random assignment to perpetrating violence against loved ones increases group
identification, allowing for a causal interpretation of the data.
Study Context
Northern Uganda was plagued by a state of internal conflict for nearly 20 years.
After failing to garner widespread support for his movement, Joseph Kony and his rebel
group the LRA turned against the local population. The LRA came to rely on the
abduction of children and young adults to fill its ranks. Abductees served as porters,
soldiers, and sex slaves, commonly forced to loot, carry heavy loads, participate in battles
with government forces, and mutilate or kill citizens. In extreme cases, the LRA
leadership forced abductees to injure or kill members of their own communities or
families (Blattman & Annan, 2010a; Eichstaedt, 2009). Estimates of the number of
abducted youth in northern Uganda range from 52,000 to 75,000 (Blattman & Annan,
2010b; Pham, Vinck, & Stover, 2008). The majority of abductees eventually escaped and
returned to their communities of origin (Beber & Blattman, 2013).
Description of Data
The survey data used in this study come from interviews with 351 male excombatants in Uganda who were abducted into the LRA. The data were collected as part
17
VIOLENCE AND GROUP IDENTIFICATION
of the Survey of War Affected Youth (SWAY) Uganda in the Kitgum and Pader districts
of Northern Uganda between September 2005 and March 2006 (Blattman & Annan,
2010a). The sample is limited to individuals abducted for at least two weeks because
questions on group identification were only asked of those respondents.
There are a number of aspects of this dataset and of the situation in Northern
Uganda more broadly that make it particularly well suited to examining the relationship
between violent behavior and group identification. First, we can reasonably assume that
respondents in this sample did not identify with Kony or the LRA at the time of their
abduction. We are able to make this assumption because in the case of Northern Uganda,
support for Joseph Kony and the LRA was non-existent among the local Acholi
population throughout the war (Acker, 2004). As this sample is tragically comprised of
individuals who were abducted from the local population into the LRA, any increase in
respondents’ level of group identification was triggered after joining the LRA.
Additionally, it is often difficult to draw generalizable conclusions about the
psychology of violence from combatant datasets because members of violent groups may
differ from the general population in important ways. For example, individuals who are
particularly drawn to violence may choose to join violent groups and are thus
overrepresented in such groups relative to the general population. In the case of Northern
Uganda, however, Beber and Blattman (2013) have shown that abduction by the LRA
was largely indiscriminate. In other words, abduction was essentially “randomly
assigned” among individuals within the same age range and location. Thus, we can feel
confident that this is a relatively representative and non-clinical sample in terms of
propensity to engage in violent behavior, increasing the generalizability of the findings.
18
VIOLENCE AND GROUP IDENTIFICATION
Along similar lines, individuals who choose to leave violent groups or who are
killed may differ from those who choose to remain combatants. This particular dataset
was designed to avoid the problem of missing data from those who never returned home
after their time in the LRA or those who did not survive combat. The survey sampling
procedure involved recreating household rosters from before the war and randomly
sampling from this pre-war population (Beber & Blattman, 2013). Thus, the researchers
were able to collect basic information on respondents who were abducted by the LRA
and never returned from captivity, on those who were no longer living in Northern
Uganda at the time of the survey, and on those who died. The final sample used in this
study is weighted to account for these dead or missing abductees.
Quasi-random assignment to violent behavior. Most importantly for the causal
test of our hypothesis, the situation in Northern Uganda presents a natural experiment in
which some newly abducted members of the LRA were quasi-randomly assigned to
perpetrate violence against loved ones while other group members were not. During the
initial abduction, some individuals were forced to injure or kill family members or
friends, before they had a chance to develop identification with the LRA through other
means. While it is conceivable that violence carried out against loved ones could occur
after the abduction incident, this would be incredibly rare as abductees were kept far
away from their home communities in order to reduce the chance that they would escape
from the LRA and return home. Family members abducted into the LRA at the same time
were kept apart for this same reason (Baines, 2014).
In Appendix 3, we provide statistical support for the claim that forced violence
perpetrated against loved ones was quasi-randomly assigned among new LRA members.
19
VIOLENCE AND GROUP IDENTIFICATION
First we compare conditional and unconditional differences in means on a number of prewar household characteristics to examine whether individuals forced to harm loved ones
are observably different than those who were not forced to harm loved ones. Next we
calculate the predicted probability that an abductee was forced to perpetrate violence
against loved ones based on pre-war data, using a logistic regression of forced violence
on indicators for year and location of birth, as well as pre-war household characteristics.
These analyses suggest that harming loved ones was quasi-randomly assigned,
conditional on the abductee’s location, age, and the size of their household before the
war.
One threat to the claim that violence was quasi-randomly assigned is that certain
LRA commanders may have been more likely to make their abductees harm loved ones,
and these particular commanders may also have been more effective at engendering
group identification. However, there is no significant relationship between harming loved
ones and level of violence in one’s LRA unit (β =-0.071, p=0.502), suggesting that the
effects are not driven by the dynamics of particularly violent units.
Analytic Approach
First we replicate the analysis in Study 1 examining whether there is a positive
association between violent behavior and group identification, controlling for individual
differences in household characteristics and war experiences. To do so, we use weighted
least squares regression, in which we predict level of identification with the LRA by an
index of violence perpetrated. Observations are weighted by their inverse attrition and
inverse absentee probabilities to account for abductees who died or never returned
(Horvitz & Thompson, 1952). Regressions control for the pre-war household and war
20
VIOLENCE AND GROUP IDENTIFICATION
experience covariates described below, as well as for the location of the survey and the
year in which the respondent was abducted.
Second, we extend the findings from Study 1 by testing whether there is a causal
relationship between violent behavior and group identification, such that perpetrating
violence increases identification with the violent group. In order to examine whether
perpetrating violence causes an increase in level of identification with the violent group,
one must be able to reasonably assume that engaging in violent behavior is randomly
assigned at the individual level. As described above, we argue that the conflict in
Northern Uganda represents a natural experiment in which some group members were
quasi-randomly assigned to perpetrate violence against loved ones on behalf of the LRA,
while other group members were not. To estimate the causal effect of violent behavior (in
this case forced violence against loved ones, a subset of the broader violent behavior
index we use in the correlational analysis) on group identification, we use a similar
regression model, replacing the index of violent behavior with an indicator for whether
the respondent perpetrated violence against loved ones.
Measures
Violent behavior. To form our first independent variable and test whether
perpetrating violence is correlated with group identification, we combine questions on six
types of violent acts that an individual may have perpetrated: killing a family member or
friend, beating or “cutting” (e.g., with a machete) a family member or friend, killing a
civilian, beating or “cutting” a civilian, killing an opposing soldier in battle, and raping
someone. Respondents indicated whether they ever engaged in each behavior while they
were with the LRA (yes / no), and responses to these six questions were averaged to
21
VIOLENCE AND GROUP IDENTIFICATION
create the violent behavior index. To test the causal hypothesis that perpetrating violence
increases group identification, the second independent variable is an indicator for whether
the respondent ever harmed a loved one, either killing or beating a family member or
friend.
Group identification. The dependent variable in this study is an index tapping
into different aspects of group identification while the respondent was a member of the
LRA (not their current level of identification with the LRA at the time of survey).
Respondents were asked whether they: felt allegiance toward the LRA and its leader,
wanted to stay in the LRA and not escape, and felt like an important part of their unit.
Answer options were on a four-point scale of “never,” “rarely,” “sometimes,” or “often.”
Responses to these three questions were averaged to create the group identification index.
Covariates. In the regressions we control for individual differences in war
experiences, such as exposure to violence, length of longest abduction (log of the number
of months), and experiences in the LRA. A full list of war experience covariates can be
found in Table 2. We also control for individual differences in pre-war household
characteristics. These variables come from a survey conducted with members of the
respondent’s household, in which they were asked questions about what their household
looked like in the year 1996. That particular year was chosen by the survey designers
because it pre-dated 85% of abductions and was easily remembered as the first
presidential election since 1980 (Beber & Blattman, 2013). Examples of these pre-war
household characteristics include parents’ level of education and household assets (see all
pre-war household covariates in Appendix 4 Table 2).
22
VIOLENCE AND GROUP IDENTIFICATION
Table 2: Descriptive statistics for the Ugandan ex-combatant sample
Index of group identification (0-3)
Felt allegiance to the violent group (0-3)
Felt like an important part of their unit (0-3)
Preferred to stay in the violent group (0-3)
Index of violence perpetrated (0-1)
Harmed a loved one on behalf of the LRA
Supports the use of violence in everyday life (0-1)
Forced to abuse the bodies of dead persons
Forced to betray a family member or friend
Index of violence personally experienced (0-1)
Index of violence experienced by loved ones (0-1)
Index of violence witnessed (0-1)
Index of violence in LRA unit (0-3)
Time spent in the LRA (months)
Age at longest abduction
Number of years since returning from LRA
Allowed to keep/sleep with the gun in LRA
Ever rewarded for job well done in LRA
Held rank or led other soldiers in LRA
Received indoctrination in LRA
Mean
1.15 (0.73)
1.18 (1.24)
1.79 (1.19)
0.45 (0.99)
0.14 (0.23)
17%
0.18 (0.23)
26%
9%
0.56 (0.24)
0.41 (0.27)
0.75 (0.18)
1.06 (0.59)
14.80 (17.91)
16.49 (4.69)
4.11 (3.09)
53%
6%
23%
46%
N
337
326
325
310
348
348
351
348
348
351
350
351
334
351
351
351
351
351
351
351
Note: This table includes descriptive statistics for the independent variable, dependent
variable, and all war experience covariates used in the regression analysis for Study 2a.
Observations are weighted to account for respondents selected to be in the sample who
could not be located, never returned from the LRA, or are no longer alive (inverse
attrition and absentee probabilities).
Results
The mean length of abduction in Uganda was 14.8 months (sd = 17.91), and
nearly all respondents reported eventually leaving the LRA by escaping (94.8% escaped,
2.7% were released, and 2.3% were rescued). However, when respondents were asked to
report their level of group identification while they were with the LRA, identification was
relatively high (mean = 1.15, sd = 0.73). Nearly three-quarters reported that they felt like
23
VIOLENCE AND GROUP IDENTIFICATION
an important part of their unit either “rarely,” “sometimes,” or “often.” Additionally,
about half reported that they felt allegiance to Kony and the LRA at least “rarely,”
“sometimes,” or “often.” Extreme violent behavior was also not uncommon in this
sample. Figure 1 shows the percentage of respondents in the sample who reported
engaging in each of the six violent behaviors while they were with the LRA (see Table 2
for more descriptive statistics).
Percentage of Sample
30
20
10
0
a
Be
iv
tc
n
ilia
ilia
Ki
iv
ll c
n
ldi
Ki
o
ll s
er
at
Be
ed
lov
e
on
ll
Ki
ne
do
e
lov
pe
Ra
Figure 1: Percentage of Ugandan LRA ex-combatants who reported engaging in
each violent act. Percentages are weighted to account for respondents selected to be in
the sample who could not be located, never returned from the LRA, or are no longer alive
(inverse attrition and absentee probabilities).
24
VIOLENCE AND GROUP IDENTIFICATION
Association between perpetrating violence and group identification. First we
test whether there is a positive association between perpetrating violence and group
identification, replicating the results from Study 1 in this different sample and context. As
shown in Figure 2, engaging in violent behavior is positively associated with group
identification, even though all individuals in this sample were abducted into the LRA (see
Appendix 4 Table 2 for full regression results). The relationship between perpetrating
violence and group identification remains statistically significant with and without the
inclusion of pre-war household and other war experience covariates (β=0.607, CI=0.184
– 1.029, p=0.005 with covariates; β=0.728, CI=0.381 –1.074, p=0.000 without
covariates).
Does perpetrating violence increase group identification? Next, we test
whether quasi-random assignment to violent behavior – harming loved ones on behalf of
the LRA – increases identification with the violent group. As shown in Figure 2,
engaging in violent behavior against family members or friends significantly increases
identification with the LRA, compared to being in the LRA but not engaging in violence
against loved ones (β=0.331, CI=0.105 – 0.556, p=0.004 with covariates; β= 0.389,
CI=0.191 – 0.587, p = 0.000 without covariates; see Appendix 4 Table 2 for regression
results, with and without covariates).
25
VIOLENCE AND GROUP IDENTIFICATION
3
Group Identification (0−3)
Group Identification (0−3)
3
2
1
2
●
1
●
0
0
1
2
3
4
5
Number of Violent Acts Perpetrated
6
Did not harm
loved ones
Harmed
loved ones
Figure 2: Perpetrating violence on behalf of the LRA increases group identification.
In the first panel, the black line shows predicted values from a weighted least squares
regression using the number of violent acts perpetrated to predict group identification,
controlling for individual differences in pre-war household characteristics and war
experiences, as well as for location of the survey and year of abduction. The regression is
weighted to account for respondents selected to be in the sample who could not be
located, never returned from the LRA, or are no longer alive (inverse attrition and
absentee probabilities). The shaded area is the 95% confidence interval around the
predicted values. The second panel shows predicted values and 95% confidence intervals
from a weighted least squares regression estimating the effect of harming loved ones on
group identification, with the same set of controls and weights.
Discussion
Study 2a shows that perpetrating violence is positively associated with group
identification among a sample of abductees in Northern Uganda, and provides evidence
26
VIOLENCE AND GROUP IDENTIFICATION
of a causal relationship between violent behavior – harming loved ones on behalf of the
LRA – and heightened group identification. This is a particularly remarkable finding in
the context of Northern Uganda since all ex-combatants in the sample were abducted into
the LRA. Yet these individuals still exhibit the predicted effect that perpetrating violence
increases identification with the LRA, a group they likely despised prior to their
abduction. We can also feel confident that this is not solely a feature of the extreme
situation of abduction and forced violence against family members in Uganda, since the
same pattern of results emerged in the Study 1 context in which many combatants joined
their group willingly and violence against loved ones was a rare event.
One alternative explanation for these findings is that greater exposure to violence
more generally may increase group identification. In the next study, we explore this
alternative explanation by testing whether different types of exposure to violence are
associated with increased group identification.
Study 2b: Exposure to Violence and Group Identification
Previous research has shown that high levels of identification with any kind of
group occur most frequently in contexts where individuals are reminded of uncertainty
and mortality (Castano, Yzerbyt, Paladino, & Sacchi, 2002; Hogg, Hohman, & Rivera,
2008). In particular, when individuals are confronted with the threat of their own
mortality, they cling to their ingroup (Castano, 2004). Thus, it is possible that the positive
association between perpetrating violence and group identification found in Studies 1 and
2a is actually part of a broader phenomenon in which greater exposure to violence
heightens group identification. We test this idea in Study 2b by examining the association
between different types of exposure to violence and identification with the LRA.
27
VIOLENCE AND GROUP IDENTIFICATION
Description of Key Variables
Exposure to violence. As independent variables, we use three indices of exposure
to violence: violence personally experienced (8 items), violence personally witnessed (7
items), and violence in the respondent's LRA unit (5 items). The violence personally
experienced index includes questions such as: “Someone shot bullets at you or your
home” and “You received a severe beating to the body by someone.” Respondents
indicated whether each of the 8 items ever happened to them, and then responses to all of
the questions were averaged to create the index. The violence personally witnessed index
includes questions such as: “You witnessed a killing” and “You witnessed the setting of
houses on fire with people inside.” Respondents indicated whether each of the 7 items
ever happened to them, and then responses to all of the questions were averaged to create
the index.
Finally, the index of violence in LRA unit measures how often one’s particular
unit engaged in different kinds of violence, including questions such as: “Did the fighters
in your unit ever mutilate civilians, such as cutting off their limbs, lips, noses or ears?”
and “When new abductees arrived, were other abductees ever forced to beat or kill
them?” Responses were measured on a four-point scale of frequency ranging from
“never” to “often,” and answers to the five questions were averaged to create the index. A
full list of the items that make up these indices can be found in Appendix 2.
Violent behavior. In this study, we compare exposure to violence to the same
violent behavior index used in Study 2a, an index of six types of violent acts that an
individual may have perpetrated on behalf of the LRA.
28
VIOLENCE AND GROUP IDENTIFICATION
Group identification and covariates. The dependent variable in this study is the
same group identification index described in Study 2a. We also control for the same set
of war experience and pre-war household covariates, as well as for location of the survey
and year of abduction.
Results
Exposure to violence and group identification. Exposure to violence is not
statistically significantly related to group identification in this sample of former
combatants from Northern Uganda (see Figure 3). Personally experiencing violence and
violence in one’s LRA unit are negatively but not statistically significantly associated
with group identification (β=-0.347, CI= -0.771 – 0.078, p=0.109 for violence personally
experienced; β=-0.076, CI= -0.230 – 0.079, p=0.337 for violence in LRA unit; see
Appendix 4 Table 4 for full regression results). Personally witnessing violence is
positively but not statistically significantly associated with group identification (β=0.359,
CI= -0.159 – 0.877, p=0.173).
Additionally, we tested whether this finding replicates in the sample of former
combatants from Liberia, and consistent with the results from Uganda we find that
personally experiencing violence and witnessing violence are not statistically
significantly related to group identification (β=0.049, CI= -0.268 – 0.366, p=0.761 for
violence personally experienced; β=0.169, CI= -0.074 – 0.412, p=0.172 for violence
personally witnessed).
29
Group Identification (0−3)
VIOLENCE AND GROUP IDENTIFICATION
3
3
2
2
1
1
0
1
2
3
4
5
6
0
Group Identification (0−3)
Violent Acts Perpetrated
3
3
2
2
1
1
0
1
2
Violence in LRA Unit
2
4
6
8
Violence Personally Experienced
3
0
1
2
3
4
5
6
Violence Witnessed
Figure 3: Perpetration of violence, not exposure to violence, is positively associated
with group identification. Black lines are predicted values from a weighted least squares
regression estimating the effect of exposure to violence on group identification,
controlling for individual differences in pre-war household characteristics and war
experiences, as well as for location of the survey and year of abduction. The regression is
weighted to account for respondents selected to be in the sample who could not be
located, never returned from the LRA, or are no longer alive (inverse attrition and
absentee probabilities). The shaded areas are 95% confidence intervals around the
predicted values.
30
VIOLENCE AND GROUP IDENTIFICATION
Discussion
In this study, we show that exposure to violence is not positively associated with
group identification, ruling out the alternative explanation that the positive association
between violent behavior and group identification found in Studies 1 and 2a is driven by
exposure to violence more generally. Perpetration of violence seems to be necessary for
violence to lead to heightened identification with the violent group.
While Studies 1 and 2 have demonstrated that perpetrating violence on behalf of
one’s group is positively associated with group identification, they have not yet provided
insight into the mechanisms through which perpetrating violence increases group
identification. In the next study, we use this same dataset of ex-combatants from
Northern Uganda to examine whether feelings of distress after violent behavior play a
role in heightened group identification.
Study 3: Personal Responsibility for Harm and Group Identification
As discussed above, the physical act of perpetrating violence tends to be an
aversive experience that leads to physiological distress (Cushman, Gaffey, Gray, & Berry
Mendes, 2012; Navarrete et al., 2012). One theorized mechanism is that individuals
reduce their violence-induced distress by increasing their identification with the violent
group, thus justifying their violent behavior. While the ideal test of this mechanism
would involve measuring distress immediately after an individual engages in violent
behavior, this is nearly impossible when studying the perpetration of extreme violence
outside of the lab. However, it is still possible to explore this mechanism based on
accumulated knowledge from previous research showing what types of behaviors are
most likely to induce dissonance.
31
VIOLENCE AND GROUP IDENTIFICATION
According to cognitive dissonance research, individuals are more likely to
experience distress when they feel a sense of personal responsibility for the consequences
of their behavior (Collins & Hoyt, 1972; Cooper & Fazio, 1984). Even when an
individual does not actively choose to engage in a behavior, such as an abductee being
forced to harm a loved one by their LRA commander, they can experience dissonance
when they feel a sense of personal responsibility for their actions (Cooper, 1971). For
example, in a study using a virtual reality set-up to create an immersive experience of the
classic trolley problem, participants exhibited more physiological distress when they felt
personally responsible for sacrificing the life of one person in order to save five other
people than when they achieved the same outcome but did not feel a sense of personal
responsibility (Navarrete et al., 2011).
In this study, we provide evidence for a distress-based mechanism by comparing
the effect of perpetrating violence to other wartime behaviors that are similarly aversive
but involve less personal responsibility for the victim’s suffering or death. When an
individual has their own finger on the trigger, and the consequences of their actions are
severe, it becomes difficult to absolve oneself of responsibility for the harm. This remains
true even if another individual gives the order to carry out the violent act. Thus, in the
extreme case of an abductee beating or killing a family member or friend, the individual
would likely still feel a sense of personal responsibility for their actions and the harm
they caused to their loved ones. In line with past research, we would expect the direct
perpetration of violence to induce distress, which the individual would be motivated to
justify.
32
VIOLENCE AND GROUP IDENTIFICATION
In this study, we compare perpetrating violence to the following two behaviors:
abusing dead bodies and betraying loved ones, leading to their possible injury or death.
While both abusing dead bodies and betraying loved ones are highly aversive behaviors,
they involve less personal responsibility for the victim’s pain and suffering than the direct
perpetration of harm. In both cases, the abductee does not have their own finger on the
trigger; they are not directly harming the victim. While abusing a dead body is considered
to be a taboo in Northern Uganda and would likely be an extremely negative experience,
the fact remains that the abductee is not responsible for killing the victim. In the case of
betrayal, the abductee may not be sure whether any harm resulted from their actions.
Even if harm does occur, they are not directly responsible for the victim’s injuries. Under
these conditions, it is easier for the abductee to put at least some of the responsibility for
the harm on their commander. Therefore the behavior is not likely to induce such high
levels of distress as the direct perpetration of violence, and the abductee will not need to
justify their behavior by increasing their identification with the LRA (Bandura, 1999).
Description of Key Variables
Harming dead bodies and betrayal. For the independent variables in this study,
we use survey questions regarding whether the respondent (the abductee) was ever forced
to step on or otherwise abuse the bodies of dead persons (26% engaged in this behavior
while with the LRA), and whether they were forced to betray a family member or friend,
placing them at risk of death or injury (9% engaged in this behavior).
Violent behavior. The third independent variable in this study, violent behavior,
is the same indicator used in Study 2a for whether the respondent ever harmed a loved
one, either killing or beating a family member or friend.
33
VIOLENCE AND GROUP IDENTIFICATION
Group identification and covariates. The dependent variable is the same group
identification index described in Study 2a. Additionally we control for the same set of
war experience and pre-war household covariates, as well as for location of the survey
and year of abduction.
Results
Betrayal, abusing dead bodies, and group identification. In this study we
compare the effect of behaviors that involve a sense of personal responsibility over the
consequences of one’s actions, which are theoretically expected to induce more distress,
to those that do not involve a sense of personal responsibility. As shown in Figure 4, only
the direct perpetration of harm (killing or beating loved ones) is associated with a
significant increase in level of identification with the LRA (β=0.331, CI= 0.105 – 0.556,
p=0.004; see Appendix 4 Table 5 for regression results). As predicted, betraying a family
member or friend, leading to the risk of injury or death, is not associated with heightened
group identification (β=-0.160, CI= -0.474 – 0.153, p=0.315). Abusing dead bodies is
associated with a marginally significant decrease in group identification (β=-0.205, CI= 0.420 – 0.009, p=0.061).
34
Group Identification (0−3)
VIOLENCE AND GROUP IDENTIFICATION
3
3
3
2
2
2
●
1
●
●
●
1
●
0
0
Did not harm
loved ones
Harmed
loved ones
1
●
0
Did not betray
loved ones
Betrayed
loved ones
Did not abuse
dead bodies
Abused
dead bodies
Figure 4: Behaviors that involve direct, not indirect, personal responsibility for harm
increase identification with the violent group. Point estimates are predicted values
from a weighted least squares regression estimating the effects of direct harm, indirect
harm, and abusing dead bodies on group identification, controlling for individual
differences in pre-war household characteristics and war experiences, as well as for
location of the survey and year of abduction. The regression is weighted to account for
respondents selected to be in the sample who could not be located, never returned from
the LRA, or are no longer alive (inverse attrition and absentee probabilities). Error bars
are 95% confidence intervals.
Discussion
While it is impossible to directly measure the level of distress that these former
combatants experienced when they harmed others on behalf of the LRA, we find that
only violent behaviors that involve feelings of personal responsibility over the harmful
consequences – behaviors that are theoretically expected to induce greater levels of
dissonance (Collins & Hoyt, 1972; Cooper & Fazio, 1984) – are positively related to
group identification. These findings provide suggestive evidence for the mechanism that
35
VIOLENCE AND GROUP IDENTIFICATION
heightened group identification is serving as a justification for the distress individuals
experience when harming others on behalf of their violent group.
General Discussion
In this series of studies, we tested the idea that when individuals perpetrate
violence on behalf of a group, they will increase their identification with the violent
group. We tested this proposition in a set of enormously consequential real world cases,
using archival data from surveys with former combatants in Liberia and Northern
Uganda. In the context of Liberia, where many combatants joined their violent group
willingly, the data show a positive association between perpetrating violence and group
identification. In the context of Uganda, where all respondents were abducted into the
LRA and some abductees were quasi-randomly assigned to perpetrate violence against
loved ones while others were not, perpetrating violence increased identification with the
LRA.
Exploring the Mechanisms
Justifying violence through group identification. In Study 3, we provided
suggestive evidence for a dissonance-based mechanism; individuals feel distress when
they perpetrate violence, and they reduce this distress by increasing their identification
with the violent group, thus justifying the violence. An implication of this mechanism is
that individuals may not need to become more supportive of violence in order to reduce
their distress. In other words, individuals may continue to engage in the aversive violent
behavior over time without changing their attitude or moral judgment about the behavior
itself. Instead, they can justify their continued perpetration of violence by conceptualizing
36
VIOLENCE AND GROUP IDENTIFICATION
it is a necessary “evil,” crucial to achieving their group’s goals or to ensuring the survival
of their ingroup.
In the case of violent behavior, this point has important implications for issues of
ex-combatant reintegration and post-conflict reconstruction. If individuals have not
become more supportive of harming others while perpetrating violence (i.e., they still
hold the general attitude that harming others is undesirable), then interventions need not
focus on changing attitudes surrounding violence. Instead, interventions should focus on
helping individuals to disidentify from their violent group and to develop productive
alternative social identities that promote peaceful resolution to problems. This point
should also be true more broadly for other violent or antisocial groups.
Disidentification with other social groups. We theorized an additional – and
likely complementary – mechanism that we were not able to test given the limitations of
these archival datasets; engaging in violent behavior on behalf of a group increases
identification with the violent group through its negative consequences on the
individual’s other group memberships. While we were not able to test this mechanism
using the archival dataset from Uganda, we would like to highlight a popular radio
program developed by a station in Gulu, the largest town in Northern Uganda, which
suggests this mechanism was at play. In December 2003, the radio station 102 Mega FM
launched a program known locally as “Dwog Paco,” which translates to “Come Back
Home.” The show features former LRA combatants who come on the air and talk about
their experiences while with the LRA. Importantly, these former combatants appeal to
individuals who are still out there with the LRA, assuring them that they will not be
killed if they surrender, and that they will receive forgiveness from their communities
37
VIOLENCE AND GROUP IDENTIFICATION
(Palmer, 2004). As one former child soldier described: “When I started listening to my
fellow colleagues whom we were in the bush with [Sudan], it gave me the confidence to
come back and join my friends at home” (Otim, 2009, p. 2). The program was so
successful in encouraging combatants to leave the LRA that commanders attempted to
ban junior soldiers from listening to the radio (Kimani, 2007; Otim, 2009).
Implications of Increased Group Identification
The finding that violent behavior increases group identification has important
implications for future violent behavior and may help to explain cycles of intergroup
conflict over time. As discussed earlier, heightened identification with one’s group can
motivate individuals to engage in violent behavior on behalf of the group. If perpetrating
violence increases group identification, then individuals who engage in violent behavior
are expected to become more willing to engage in future violent behavior, suggesting a
cycle of violence and group identification (Littman & Paluck, 2015). This also helps to
explain in part how low identifiers, such as those forced to join a group through
conscription or abduction, can eventually become high identifiers who are willing to
harm others on behalf of their group. Perpetrating violence – in and of itself – may help
trigger this process from reluctant or opportunistic joiner to loyal group member.
As for patterns of intergroup conflict over time, social identities are particularly
salient in contexts of conflict, and serve as a strong influence over combatants’ decision
to follow or flout humanitarian law (Castano, Leidner, & Slawuta, 2008). Additionally,
group identification influences perceptions of responsibility for conflict and the
magnitude of harm experienced by the groups involved in conflict; stronger group
identifiers tend to attribute more responsibility for the instigation and consequences of
38
VIOLENCE AND GROUP IDENTIFICATION
violent conflict to the outgroup, while also perceiving that more harm has been inflicted
on the ingroup and less on the outgroup (Bilali, Tropp, & Dasgupta, 2012). Glorification
of the ingroup, an element of group identification, also leads to increased dehumanization
of the victims of violence and influences decisions about the punishment of perpetrators
and reparations to victims (Castano, 2008; Leidner, Castano, Zaiser, & Giner-Sorolla,
2010).
Limitations and Future Directions
All of the data used in these studies are based on self-report, which may be
subject to desirability bias and underreporting of sensitive behaviors and attitudes
(Blattman & Miguel, 2010). These studies also use retrospective surveys, in which
respondents are asked questions about their experiences in and identification with the
violent group once they have already left the group. Additionally, the measures of violent
behavior used in these surveys have some limitations. Ideally, the data would include
information about the number of times an individual engaged in a particular violent
behavior and at which point in time during their group membership the violence
occurred.
Considering the temporal aspect of violent behavior is important for future
research because individuals may internalize the normative standards of their violent
group and come to see violence as necessary and virtuous (Fiske & Rai, 2014; Reicher,
Halsam, & Rath, 2008). Individuals may also become physiologically desensitized to
violence over time (Carnagey, Anderson, & Bushman, 2007; Engelhardt, Batholow, Kerr,
& Bushman, 2011), and therefore not feel as much distress or arousal when perpetrating
violence. Under these conditions, engaging in violent behavior may no longer induce
39
VIOLENCE AND GROUP IDENTIFICATION
distress. However, the prevalence of posttraumatic stress disorder in former members of
violent groups (e.g., Cesur, Sabia & Tekin, 2013; MacNair, 2002; Maguen et al., 2009)
suggests that violent behavior may never be fully justified and may continue to induce
feelings of distress over time.
Another avenue for future research is to explore how violent behavior relates to
different dimensions of group identification. The questions used in this research tap into a
number of elements of group identification, including: enjoyment, belief in the group,
allegiance to the group and its leader, a desire to stay in the group, and feeling like an
important member of the group. When analyzing the items separately, all but a desire to
stay in the group show a positive association with perpetrating violence (see Appendix 4
Tables 1 & 3). While we believe it is a strength of the paper that the association between
violence perpetrated and group identification emerges when using questions that tap into
these different elements of identification, future work should carefully design
questionnaires to measure theoretically-derived constructs such as ingroup attachment
and glorification.
Given the inherent limitations of correlational and quasi-experimental research,
future work should attempt to build on this survey data from former combatants to
explore the relationship between violent behavior and group identification in an
experimental setting in which perpetration of violence can be randomly assigned and
mechanisms can be tested in more depth. However, it may be difficult – if not impossible
– to explore this research question in the lab. Researchers cannot ethically ask
participants to engage in intentional interpersonal harm that causes bodily injury or death.
Instead they can have participants simulate harm or engage in much less extreme versions
40
VIOLENCE AND GROUP IDENTIFICATION
of aggressive behavior or social and psychological harm. The underlying psychology may
be different when engaging in these permissible actions as opposed to extreme violence
on behalf of an actual violent group.
In conclusion, this series of studies shows how to pose theoretically driven
psychological questions of difficult to study phenomena in the world, such as war
violence, using archival data. We find that perpetrating violence on behalf of a violent
group increases group identification, even among a sample of individuals who were
abducted into a group they despised and forced to carry out heinous acts of violence
against loved ones. Thus, violent behavior can trigger group identification, leading to
cycles of future violence and intergroup conflict. These findings pave the way for future
research in the laboratory and the field to dive deeper into the phenomenon, for example
examining mechanisms and exploring different types of violent behavior and groups.
41
VIOLENCE AND GROUP IDENTIFICATION
References
Acker, F. V. (2004). Uganda and the Lord’s Resistance Army: The new order no one
ordered. African Affairs, 103(412), 335-357.
Aronson, E. & Mills, J. (1959). The effect of severity of initiation on liking for a group.
The Journal of Abnormal and Social Psychology, 59(2), 177-181.
Atran, S. (2003). Genesis of suicide terrorism. Science, 299(5612), 1534–1539.
Baines, E. (2014). Forced marriage as a political project: Sexual rules and relations in the
Lord's Resistance Army. Journal of Peace Research, 51(3), 405-417.
Bandura, A. (1999). Moral disengagement in the perpetration of inhumanities.
Personality and Social Psychology Review, 3(3), 193-209.
Bastian, B., Jetten, J., & Ferris, L. J. (2014). Pain as social glue: Shared pain increases
cooperation. Psychological Science, 25(11), 2079-2085.
Beber & Blattman (2013). The logic of child soldiering and coercion. International
Organization, 67(1), 65-104.
Becker, Tausch, Spears, & Christ (2011). Committed Dis(s)idents: Participation in radical
collective action fosters disidentification with the broader in-group but enhances
political identification. Personality and Social Psychology Bulletin, 37(8), 11041116.
Bilali, Tropp, & Dasgupta (2012). Attributions of responsibility and perceived harm in
the aftermath of mass violence. Peace and Conflict: Journal of Peace Psychology,
18(1), 21-39.
Blattman, C., & Annan, J. (2010a). The consequences of child soldiering. The Review of
Economics and Statistics, 92(4), 882-898.
42
VIOLENCE AND GROUP IDENTIFICATION
Blattman, C., & Annan, J. (2010b). On the nature and causes of LRA abduction: What
the abductees say. In T. Allen & K. Vlassenroot (Eds.), The lord's resistance
army: Myth and reality (pp. 132-155). Zed Books.
Blattman, C. & Annan, J. (2016). Can employment reduce lawlessness and rebellion? A
field experiment with high-risk men in a fragile state. American Political Science
Review, 110(1), 1-17.
Blattman, C., & Miguel, E. (2010). Civil War. Journal of Economic Literature, 48(1), 357.
Brock, T. C., & Buss, A. H. (1962). Dissonance, aggression, and evaluation of pain. The
Journal of Abnormal and Social Psychology, 65(3), 197-202.
Carnagey, N. L., Anderson, C. A., Bushman, B. J. (2007). The effect of video game
violence on psychological desensitization to real-life violence. Journal of
Experimental Social Psychology, 43(3), 489-496.
Castano E. (2004). In case of death, cling to the ingroup. European Journal of Social
Psychology, 34(4), 375-384.
Castano (2008). On the perils of glorifying the in-group: Intergroup violence, in-group
glorification, and moral disengagement. Social and Personality Compass, 2(1),
154-170.
Castano, E., Leidner, B., & Slawuta, P. (2008). Social identification processes, group
dynamics and the behaviour of combatants. International Review of the Red
Cross, 90(2), 259-271.
43
VIOLENCE AND GROUP IDENTIFICATION
Castano, E., Yzerbyt, V., Paladino, M., & Sacchi, S. (2002). I belong, therefore, I exist:
Ingroup identification, ingroup entitativity, and ingroup bias. Personality and
Social Psychology Bulletin, 28(2), 135-143.
Cesur, R., Sabia J. J., & Tekin, E. (2013). The psychological costs of war: Military
combat and mental health. Journal of Health Economics, 32(1), 51-65.
Cikara, M. (2015). Intergroup schadenfreude: Motivating participation in collective
violence. Current opinion in behavioral sciences, 3, 12-17.
Cohen, D. K. (2013). Explaining rape during civil war: Cross-national evidence (19802009). American Political Science Review, 107(3), 461-477.
Collins, B. E., & Hoyt, M. F. (1972). Personal responsibility-for-consequences: An
integration and extension of the “forced compliance” literature. Journal of
Experimental Social Psychology, 8(6), 558-593.
Cooper, J. (1971). Personal responsibility and dissonance: The role of foreseen
consequences. Journal of Personality and Social Psychology, 18(3), 354-363.
Cooper, J., & Fazio, R. H. (1984). A new look at dissonance theory. In L. Berkowitz
(Ed.), Advances in experimental social psychology (pp. 229–262). Hillsdale, NJ:
Erlbaum.
Cooper, J., & Mackie, D. (1983). Cognitive dissonance in an intergroup context. Journal
of Personality and Social Psychology, 44(3), 536-544.
Crockett, M. J., Kurth-Nelson, Z., Siegel, J. Z., Dayan, P., & Dolan, R. J. (2014). Harm to
others outweighs harm to self in moral decision making. PNAS, 111(48), 1732017325.
44
VIOLENCE AND GROUP IDENTIFICATION
Cushman, F., Gray, K., Gaffey, A., & Mendes, W. B. (2012). Simulating murder: The
aversion to harmful action. Emotion, 12(1), 2-7.
Eichstaedt, P. (2009). First kill your family: Child soldiers of Uganda and the Lord's
Resistance Army. Chicago Review Press.
Elliot, A. J., & Devine, P.G. (1994). On the motivational nature of cognitive dissonance:
Dissonance as psychological discomfort. Journal of Personality and Social
Psychology, 67(3), 382-294.
Ellis, S. (1999). The Mask of Anarchy: The Destruction of Liberia and the Religious
Dimensions of an African Civil War. New York University Press.
Engelhardt, C. R., Bartholow, B. D., Kerr, G. T., & Bushman, B. J. (2011). This is your
brain on violent video games: Neural desensitization to violence predicts
increased aggression following violent video game exposure. Journal of
Experimental Social Psychology, 47(5), 1033-1036.
Festinger, L. (1957). A theory of cognitive dissonance. Evanston, IL: Row, Peterson.
Fiske, A. P., & Rai, T. S. (2014). Virtuous violence: Hurting and killing to create,
sustain, end, and honor social relationships. Cambridge University Press.
Gerard, H. B. & Mathewson, G. C. (1966). The effects of severity of initiation on liking
for a group: A replication. Journal of Experimental Social Psychology, 2(3), 278287.
Glasford, Dovidio, & Pratto (2009). I continue to feel so good about us: In-group
identification and the use of social identity-enhancing strategies to reduce
intragroup dissonance. Personality and Social Psychology Bulletin, 35(4), 415427.
45
VIOLENCE AND GROUP IDENTIFICATION
Glasford, Pratto, & Dovidio (2008). Intragroup dissonance: Responses to ingroup
violation of personal values. Journal of Experimental Social Psychology, 44(4),
1057-1064.
Glass, D. (1964). Changes in liking as a means of reducing cognitive discrepancies
between self-esteem and aggression. Journal of Personality, 32(4), 531–549.
Gosling, P., Denizeau, M., & Oberlé, D. (2006). Denial of responsibility: A new mode of
dissonance reduction. Journal of Personality and Social Psychology, 90(5), 722733.
Grant, F., & Hogg, M. A. (2012). Self-uncertainty, social identity prominence and group
identification. Journal of Experimental Social Psychology, 48(2), 538-542.
Grossman, D. (1996). On killing. Boston, MA: Little, Brown.
Hogg, M. A., Hohman, Z. P., & Rivera, J. E. (2008). Why do people join groups? Three
motivational accounts from social psychology. Social and Personality Psychology
Compass, 2(3), 1269-1280.
Horvitz, D. G., & Thompson, D. J. (1952). A generalization of sampling without
replacement from a finite universe. Journal of the American Statistical
Association, 47(260), 663-685.
Kimani, M. (2007). Broadcasting peace: Radio a tool for recovery. Africa Renewal
Online. Retrieved from http://www.un.org/africarenewal/magazine/october2007/broadcasting-peace-radio-tool-recovery.
Knowles, M. L., & Gardner, W. L. (2008). Benefits of membership: The activation and
amplification of group identities in response to social rejection. Personality and
Social Psychology Bulletin, 34(9), 1200-1213.
46
VIOLENCE AND GROUP IDENTIFICATION
Kleykamp, M. A. (2006). College, jobs, or the military? Enlistment during a time of war.
Social Science Quarterly, 87(2), 272-290.
Leidner, Castano, Zaiser, & Giner-Sorolla (2010). Ingroup glorification, moral
disengagement, and justice in the context of collective violence. Personality and
Social Psychology Bulletin, 36(8), 1115-1129.
Lerner, M. J., & Matthews, G. (1967). Reactions to suffering of others under conditions
of indirect responsibility. Journal of Personality and Social Responsibility, 5(3),
319-325.
Littman & Paluck (2015). The cycle of violence: Understanding individual participation
in collective violence. Advances in Political Psychology, 36, 79-99.
Lodewijkx, H. F. M., & Syroit, J. E. M. M. (1997). Severity of initiation revisited: Does
severity of initiation increase attractiveness in real groups? European journal of
social psychology, 27(3), 275-300.
MacNair, R. M. (2002). Perpetration-induced traumatic stress in combat veterans. Peace
and Conflict: Journal of Peace Psychology, 8(1), 63-72.
Maguen, S., Metzler, T. J., Litz, B. T., Seal, K. H., Knight, S. J., Marmar, C. R. (2009).
The impact of killing in war on mental health symptoms and related functioning.
Journal of Traumatic Stress, 22(5), 435-443.
Maguen, S., Metzler, T. J., Litz, B. T., Seal, K. H., Knight, S. J., & Marmar, C. R. (2010).
The impact of reported direct and indirect killing on mental health symptoms in
Iraq war veterans. Journal of Traumatic Stress, 23(1), 86-90.
47
VIOLENCE AND GROUP IDENTIFICATION
Martens, A., Kosloff, S., Greenberg, J., Landau, M., & Schmader, T. (2007). Killing
begets killing: Evidence from a bug-killing paradigm that initial killing fuels
subsequent killing. Personality and Social Psychology Bulletin, 33(9), 1251-1264.
McKimmie et al. (2003). I’m a hypocrite, but so is everyone else: Group support and the
reduction of cognitive dissonance. Group Dynamics: Theory, Research, and
Practice, 7(3), 214-224.
Melander, E. (2015). Organized violence in the world 2015: An assessment by the
Uppsala conflict data program. Retrieved from the Uppsala Conflict Data
Program website: http://www.pcr.uu.se/digitalAssets/61/61533_1brochure2.pdf.
Milgram, S. (1965). Some conditions of obedience and disobedience to authority. Human
Relations, 18(1), 57-76.
Miller, E. (2015). Overview: Terrorism in 2014. Retrieved from National Consortium for
the Study of Terrorism and Responses to Terrorism website:
https://www.start.umd.edu/pubs/START_GTD_OverviewofTerrorism2014_Aug2
015.pdf.
Miller, R. & Cushman, F. (2013). Aversive for me, wrong for you: First-person
behavioral aversions underlie the moral condemnation of harm. Social and
Personality Psychology Compass, 7(10), 707-718.
Miller, R. M., Hannikainen, I. A., & Cushman, F. A. (2014). Bad actions or bad
outcomes? Differentiating affective contributions to the moral condemnation of
harm. Emotion, 14(3), 573-587.
48
VIOLENCE AND GROUP IDENTIFICATION
Navarrete, C. D., McDonald, M. M., Mott, M. L., & Asher, B. (2012). Virtual morality:
emotion and action in a simulated three-dimensional “trolley problem”. Emotion,
12(2), 364-370.
Otim, P. W. (2009). An interactive media: Reflections on Mega FM and Its
Peacebuilding Role in Uganda. Beyond Intractability. Retrieved from
http://www.beyondintractability.org/casestudy/otim-interactive.
Palmer, J. (2004) The power of radio helps to end Uganda’s long war. The Christian
Science Monitor. Retrieved from http://www.csmonitor.com/2004/0921/p07s01woaf.html.
Pham, P. N., Vinck, P., & Stover, E. (2008). The Lord's Resistance Army and Forced
Conscription in Northern Uganda. Human Rights Quarterly, 30, 404-411.
Pinker, S. (2011). The better angels of our nature: The decline of violence in history and
its causes. Penguin UK.
Reicher, S., Haslam, S.A., & Rath, R. (2008). Making a virtue of evil: A five-step social
identity model of the development of collective hate. Social and Personality
Psychology Compass, 2(3), 1313-1344.
Sidanius, J., & Pratto, F. (1999). Social dominance: An intergroup theory of social
hierarchy and oppression. NewYork, NY: Cambridge University Press.
Stone, J., & Cooper, J. (2001). A Self-Standards Model of Cognitive Dissonance. Journal
of Experimental Social Psychology, 37(3), 228-243.
Swann, W. B., Gomez, A., Huici, C., Morales, J. F., & Hixon, J. G. (2010). Identity
fusion and self-sacrifice: Arousal as a catalyst of pro-group fighting, dying, and
helping behavior. Journal of Personality and Social Psychology, 99(5), 824-841.
49
VIOLENCE AND GROUP IDENTIFICATION
Swann, W. B., Jetten, J., Gomez, A., Whitehouse, H., & Bastian, B. (2012). When group
membership gets personal: A theory of identity fusion. Psychological Review,
119(3), 441-456.
Taylor, K. (2009). Cruelty: Human evil and the human brain. Oxford, UK: Oxford
University Press.
Toosi, N. R., Masicampo, E. J., & Ambady, N. (2014). Aggressive acts increase
commitment to new groups: Zombie attacks and blocked shots. Social
Psychological & Personality Science, 5(3), 286-292.
Waller, J. (2002). Perpetrators of genocide: An explanatory model of extraordinary
human evil. Journal of Hate Studies, 1(1), 5-22.
Woodworth, M., & Porter, S. (2002). In cold blood: Characteristics of criminal homicide
as a function of psychopathy. Journal of Abnormal Psychology, 111(3), 436–445.
50
VIOLENCE AND GROUP IDENTIFICATION
Appendix 1: Information on Surveys and Data Collection
Liberia Dataset
For the Liberia ex-combatant dataset, the surveys were conducted through
Innovations for Poverty Action, an international research organization, from August
through October 2009. The data was collected in face-to-face interviews by trained
Liberian researchers in Liberian English, the local dialect. More info can be found at
http://www.poverty-action.org/study/ex-combatant-reintegration-liberia.
It is important to note that the respondents in this sample represent a subset of the
population of former combatants in Liberia; all of them expressed interest in attending a
residential agricultural training program, many of them were engaged in illicit economic
activities at the time of the survey, and 73% did not go through Liberia's formal
reintegration process after the war. Therefore, the results may not generalize to a broader
population of former combatants. Also, the measure of violent behavior used in this
analysis is not very precise, as respondents simply reported whether they engaged in
none, few, some, or plenty violent acts. Former combatants may interpret this question in
different ways depending on their experience with violence.
Uganda Dataset
For the Uganda ex-combatant dataset, the surveys were conducted as part of the
Survey of War Affected Youth (SWAY) Uganda in the Kitgum and Pader districts of
northern Uganda between September 2005 and March 2006. The data was collected in
face-to-face interviews by trained Ugandan researchers in Luo, the local language spoken
in Northern Uganda. More information about SWAY Uganda, including sampling
51
VIOLENCE AND GROUP IDENTIFICATION
procedures, survey instrument, and raw data, is available at
http://chrisblattman.com/data/sway/.
52
VIOLENCE AND GROUP IDENTIFICATION
Appendix 2: Survey Questions in Violence Exposure Indices
Liberia Dataset
Index of Violence Personally Experienced:
1. Did someone shoot bullets at you?
2. Did someone attack you with a cutlass or other weapon?
3. Were you forced to have sex with someone?
4. Were you on the frontline or witness battles?
5. Did someone you know betray you or point you out to soldiers?
6. Did you receive a serious beating to the body by non-family members?
7. Did you receive a serious physical injury in a battle or attack?
Index of Violence Witnessed:
1. Did you see someone get beaten or tortured?
2. Did you see someone get killed?
3. Did you see someone forced to have sex with someone else?
Index of Violence Experienced by Loved Ones:
1. Was your wife or husband killed during the war?
2. Was your child killed during the war?
3. Was another family member or close friend killed?
4. Did a family member die without proper burial?
5. Did you lose the person who was supporting you?
Uganda Dataset
Index of Violence Personally Experienced:
1. Someone took or destroyed your personal property
53
VIOLENCE AND GROUP IDENTIFICATION
2. Someone shot bullets at you or your home
3. You received a severe beating to the body by someone
4. Someone attacked you with a panga or other weapon
5. You were tied up or locked up as a prisoner
6. You received a serious physical injury in a battle or rebel attack
7. You were forced to carry heavy loads or do other forced labor
8. Someone you know betrayed you and put you at risk of death or injury
Index of Violence Experienced by Loved Ones:
1. A parent was murdered or died violently
2. A parent disappeared or was abducted
3. Another family member or friend was murdered or died violently
4. Another family member or friend disappeared or was abducted
5. A family member received a serious physical injury from combat or a landmine
Index of Violence in Respondent’s LRA Unit:
1. When new abductees arrived in this unit, were they ever beaten? If yes: often,
sometimes, or rarely?
2. When new abductees arrived, were other abductees ever forced to beat or kill
them? If yes: often, sometimes, or rarely? If yes: often, sometimes, or rarely?
3. Did the fighters in your unit ever mutilate civilians, such as cutting off their limbs,
fingers, lips, noses or ears? If yes: often, sometimes, or rarely?
4. Were abductees in your unit ever forced to desecrate dead bodies? If yes: often,
sometimes, or rarely?
5. Were abductees in your unit ever forced to carry body parts? If yes: often,
54
VIOLENCE AND GROUP IDENTIFICATION
sometimes, or rarely?
Index of Violence Personally Witnessed:
1. You heard gun fire regularly
2. You witnessed an attack by the LRA or battle with UPDF
3. You witnessed beatings or torture of other people
4. You witnessed a killing
5. You witnessed a massacre
6. You witnessed the setting of houses on fire with people inside
7. You witnessed the rape or sexual abuse of a woman
Attitudes toward the Use of Violence in Everyday Life:
1. If a man insulted your neighbor, would it be acceptable for your neighbor to
seriously beat him?
2. If a man was caught stealing from one of your neighbors, would it be acceptable
for your neighbor to seriously beat him?
3. If a man’s wife were to burn his supper, would it be acceptable for him to beat
her?
4. If a man’s wife were to argue with him or talk back, would it be acceptable for
him to beat her?
55
VIOLENCE AND GROUP IDENTIFICATION
Appendix 3: Quasi-Random Assignment to Family Violence
To assess whether being forced to harm a loved one is quasi-random, we first
compare unconditional and conditional differences in means on pre-war household
characteristics between abductees who were forced to perpetrate violence against loved
ones, and those who were not forced to do so. As shown in Appendix 3 Table 1, the only
pre-war household variable that is significantly related to forced violence when looking at
the conditional means is household size in 1996. Those with larger households were less
likely to be forced to perpetrate violence against family and friends, potentially because
larger households were more likely to be able to fight back.
Appendix 3 Table 1: Difference in means between respondents who harmed loved
ones and those who did not harm loves ones
Pre-treatment Covariate
Age
Father was a farmer in 1996 (0/1)
Household size in 1996
Landholdings in 1996
In the top 10% of landholdings (0/1)
Number of cattle in 1996
Number of other livestock in 1996
Owned a plow in 1996 (0/1)
Father was uneducated (0/1)
Father's years of schooling
Mother was uneducated (0/1)
Mother's years of schooling
Father died before 1996 (0/1)
Mother died before 1996 (0/1)
Orphaned before 1996
Unconditional
Means
Conditional
Means
-1.47 [0.74]*
0.00 [0.05]
-1.24 [0.53]*
12.27 [5.63]*
0.19 [0.07]**
7.14 [6.52]
1.20 [4.71]
-0.09 [0.11]
-0.01 [0.04]
0.25 [0.50]
-0.04 [0.11]
0.38 [0.49]
0.10 [0.08]
0.01 [0.05]
-0.01 [0.02]
-0.652 [0.63]
0.00 [0.04]
-1.40 [0.50]**
-0.13 [1.95]
0.04 [0.02]
5.23 [7.18]
1.18 [2.30]
-0.01 [0.06]
-0.03 [0.04]
-0.46 [0.38]
0.01 [0.03]
0.13 [0.16]
0.06 [0.07]
0.04 [0.05]
-0.02 [0.03]
Note: Observations are weighted to account for respondents selected to be in the sample
who could not be located, never returned from the LRA, or are no longer alive (inverse
attrition and absentee probabilities).
56
VIOLENCE AND GROUP IDENTIFICATION
Next, we calculate the predicted probability that an abductee was forced to
perpetrate violence against family or friends based on pre-war data, using a logistic
regression of forced violence on indicators for year and location of birth, as well as prewar household characteristics. Consistent with the analysis in Blattman & Annan (2010a)
showing that abduction is quasi-random, we plot the distribution of predicted
probabilities of forced violence against family or friends based on pre-war data. Predicted
probabilities come from a logistic regression of forced violence on indicators for year and
camp of birth, as well as pre-war household characteristics. As shown Appendix 3 Figure
1, the predicted probabilities for those forced to perpetrate violence against loved ones
and those who did not perpetrate violence against loved ones overlap a great deal. The
difference in the two distributions of predicted probabilities is driven exclusively by year
and location of birth. Including other pre-war household covariates in the model does not
significantly alter the distributions. These findings are consistent with Blattman &
Annan’s (2010a) results for abducted vs. non-abducted respondents, suggesting that
forced violence against family members and friends is similarly quasi-random.
57
VIOLENCE AND GROUP IDENTIFICATION
Family Violence
10
5
0
% of Respondents
15
No Family Violence
0
.1
.2
.3
.4
.5
.6
.7
.8
.9
1
0
.1
.2
.3
.4
.5
.6
.7
.8
.9
1
Predicted Probability of Forced Violence
Age & location dummies only
All pre-treatment covariates
Appendix 3 Figure 1. Predicted probability of forced violence against loved ones.
Observations are weighted to account for respondents selected to be in the sample who
could not be located, never returned from the LRA, or are no longer alive (inverse
attrition and absentee probabilities).
58
VIOLENCE AND GROUP IDENTIFICATION
Appendix 4: Regression Tables
Appendix 4 Table 1: Perpetrating violence and group identification in Liberia
(1)
Index of
group
identification
(2)
Index of
group
identification
(3)
Enjoyed
being a
member
(4)
Enjoyed
being a
member
(5)
Really
believed in
the group
0.304**
[0.108]
0.049
[0.161]
0.273*
[0.110]
0.372**
[0.114]
0.063
[0.173]
0.330**
[0.115]
0.236*
[0.118]
0.035
[0.179]
VARIABLES
Index of Violence
Perpetrated (0-1)
Index of violence
personally
experienced (0-1)
Index of violence
witnessed (0-1)
Index of violence
experienced by
loved ones (0-1)
Log of length of
time in violent group
Age when joined
violent group
Carried a gun while
in violent group
On the frontlines
while in violent
group
Main role in violent
group was as a
soldier
Held rank in violent
group
Held leadership
position in violent
group
Abducted into
violent group
Age of respondent
Gender of
respondent
Christian
Ethnic group: Gola
Ethnic group: Kpelle
0.169
[0.124]
-0.203
[0.160]
0.197
[0.137]
-0.322
[0.169]
0.142
[0.137]
-0.085
[0.175]
0.101**
[0.035]
-0.092***
[0.024]
-0.035
[0.096]
0.040
[0.098]
0.098**
[0.038]
-0.109***
[0.028]
-0.076
[0.101]
0.042
[0.109]
0.105**
[0.039]
-0.075***
[0.022]
0.005
[0.103]
0.038
[0.110]
0.226*
[0.088]
0.223*
[0.092]
0.229*
[0.097]
0.144
[0.117]
0.273**
[0.104]
0.106
[0.125]
0.274*
[0.111]
0.181
[0.129]
0.272*
[0.115]
-0.674***
[0.073]
0.081***
[0.024]
0.012
[0.108]
-0.012
[0.080]
-0.351**
[0.119]
-0.040
[0.080]
-0.679***
[0.079]
0.100***
[0.027]
-0.005
[0.108]
-0.020
[0.087]
-0.341**
[0.126]
-0.053
[0.085]
-0.668***
[0.080]
0.062**
[0.022]
0.029
[0.127]
-0.003
[0.089]
-0.360*
[0.140]
-0.027
[0.088]
(6)
Really
believ
ed in
the
group
0.216
[0.119]
59
VIOLENCE AND GROUP IDENTIFICATION
Ethnic group: Kru
Ethnic group: Mano
Ethnic group: Sapo
0.148
[0.130]
-0.152
[0.127]
0.432**
[0.152]
0.156
[0.139]
-0.124
[0.136]
0.524**
[0.167]
Observations
1,054
1,074
1,054
1,074
R-squared
0.546
0.439
0.503
0.403
Note: Robust standard errors are in brackets. *** p<0.001, ** p<0.01, *p<0.05.
0.140
[0.141]
-0.181
[0.139]
0.341*
[0.167]
1,054
0.503
1,074
0.406
60
VIOLENCE AND GROUP IDENTIFICATION
Appendix 4 Table 2: Perpetrating violence and group identification in Uganda
(1)
Index of group
identification
(2)
Index of group
identification
(3)
Index of group
identification
(4)
Index of group
identification
Index of Violence Perpetrated (0-1)
0.607**
[0.215]
0.728***
[0.176]
---
---
Ever harmed a loved one on behalf
of the LRA (0/1)
---
---
0.331**
[0.115]
-0.339
[0.214]
0.461
[0.257]
-0.052
[0.079]
0.391*
[0.163]
-0.205
[0.109]
-0.160
[0.159]
0.030
[0.052]
-0.121*
[0.056]
-0.014
[0.033]
0.332**
[0.111]
0.210
[0.206]
0.234
[0.161]
0.247*
[0.096]
0.111*
[0.056]
0.021
[0.011]
-0.038
[0.020]
-0.196*
[0.091]
0.074
[0.145]
0.389***
[0.100]
VARIABLES
Index of violence personally
experienced (0-1)
Index of violence witnessed during
the war (0-1)
Index of violence in LRA unit (03)
Index of violence experienced by
loved ones (0-1)
Forced to step on or abuse dead
bodies (0/1)
Forced to betray a family member
or friend (0/1)
Log of length of time in violent
group
Age when joined violent group
Years since returning from violent
group
Carried a gun while in violent
group (0/1)
Ever rewarded for a job well done
(0/1)
Ever held rank in LRA or led other
soldiers (0/1)
Received indoctrination in LRA
(0/1)
Age of respondent
Father’s years of education
Mother’s years of education
Father died before start of the war
(1996 or earlier)
Mother died before start of the war
(1996 or earlier)
-0.347
[0.216]
0.359
[0.263]
-0.076
[0.079]
0.340*
[0.173]
-0.230*
[0.111]
-0.154
[0.156]
0.028
[0.051]
-0.124*
[0.058]
-0.015
[0.033]
0.296**
[0.109]
0.200
[0.207]
0.262
[0.162]
0.250*
[0.096]
0.116*
[0.057]
0.019
[0.011]
-0.033
[0.019]
-0.207*
[0.091]
0.072
[0.144]
61
VIOLENCE AND GROUP IDENTIFICATION
Father was a farmer in 1996
Household land in 1996 (acres)
Household size in 1996
Number of cattle in 1996
Number of livestock in 1996
Number of plows in 1996
0.057
[0.138]
0
[0.002]
-0.008
[0.014]
0.001
[0.001]
-0.003
[0.002]
0.060
[0.111]
0.064
[0.138]
0
[0.002]
-0.007
[0.014]
0.001
[0.001]
-0.003
[0.002]
0.060
[0.112]
Observations
331
334
331
334
R-squared
0.807
0.757
0.807
0.754
Note: The regression also controls for location of the survey and year of abduction. It is weighted to
account for respondents selected to be in the sample who could not be located, never returned from the
LRA, or are no longer alive (inverse attrition and absentee probabilities). Robust standard errors are in
brackets. *** p<0.001, ** p<0.01, * p<0.05.
62
VIOLENCE AND GROUP IDENTIFICATION
Appendix 4 Table 3: Perpetrating violence and the dimensions of group identification
in Uganda
(1)
Allegiance to
the LRA and
its leader
(2)
Felt like
important part
of unit
(3)
Wanted to stay
in the LRA
and not escape
Ever harmed a loved one on behalf
of the LRA (0/1)
0.486*
[0.198]
0.390*
[0.193]
Index of violence personally
experienced (0-1)
-0.466
[0.412]
0.816
[0.484]
-0.199
[0.146]
-0.495
[0.311]
-0.222
[0.187]
-0.157
[0.231]
0.168*
[0.078]
-0.004
[0.088]
0.049
[0.042]
0.108
[0.187]
0.788*
[0.368]
0.075
[0.249]
-.223
[0.167]
-0.003
[0.089]
-0.012
[0.022]
-0.022
[0.032]
-0.267
[0.158]
0.034
[0.224]
-0.010
[0.410]
0.013
[0.157]
-0.243
[0.359]
0.383
[0.409]
-0.017
[0.104]
0.344*
[0.236]
-0.123
[0.140]
-0.091
[0.205]
0.014
[0.074]
-0.173*
[0.073]
-0.058
[0.047]
0.264
[0.154]
0.382
[0.305]
-0.251
[0.225]
0.314*
[0.142]
0.177*
[0.074]
0.054**
[0.019]
-0.055*
[0.025]
-0.302*
[0.124]
0.220
[0.246]
VARIABLES
Index of violence witnessed during
the war (0-1)
Index of violence in LRA unit (03)
Index of violence experienced by
loved ones (0-1)
Forced to step on or abuse dead
bodies (0/1)
Forced to betray a family member
or friend (0/1)
Log of length of time in violent
group
Age when joined violent group
Years since returning from violent
group
Carried a gun while in violent
group (0/1)
Ever rewarded for a job well done
(0/1)
Ever held rank in LRA or led other
soldiers (0/1)
Received indoctrination in LRA
(0/1)
Age of respondent
Father’s years of education
Mother’s years of education
Father died before start of the war
(1996 or earlier)
Mother died before start of the war
(1996 or earlier)
0.024
[0.441]
0.037
[0.130]
1.078***
[0.293]
-0.309
[0.197]
-0.283
[0.294]
-0.047
[0.086]
-0.137
[0.088]
-0.018
[0.056]
0.637***
[0.186]
-0.436
[0.302]
0.814
[0.284]**
0.617***
[0.159]
0.117
[0.088]
0.030
[0.022]
-0.030
[0.033]
-0.133
[0.161]
0.029
[0.254]
63
VIOLENCE AND GROUP IDENTIFICATION
Father was a farmer in 1996
Household land in 1996 (acres)
Household size in 1996
Number of cattle in 1996
Number of livestock in 1996
Number of plows in 1996
0.029
[0.260]
-0.002
[0.003]
-0.042
[0.021]
0.001
[0.002]
0.003
[0.005]
-0.289
[0.180]
-0.159
[0.216]
0
[0.002]
0.001
[0.020]
0.002
[0.002]
-0.005
[0.004]
0.038
[0.191]
0.279
[0.171]
0.002
[0.002]
0.014
[0.022]
-0.001
[0.001]
-0.006
[0.003]
0.472**
[0.162]
Observations
331
334
331
R-squared
0.807
0.757
0.807
Note: The regression also controls for location of the survey and year of abduction. It is weighted to
account for respondents selected to be in the sample who could not be located, never returned from the
LRA, or are no longer alive (inverse attrition and absentee probabilities). Robust standard errors are in
brackets. *** p<0.001, ** p<0.01, * p<0.05.
64
VIOLENCE AND GROUP IDENTIFICATION
Appendix 4 Table 4: Exposure to violence and group identification in Uganda
VARIABLES
Index of Violence Perpetrated (0-1)
Index of violence personally
experienced (0-1)
Index of violence witnessed during the
war (0-1)
Index of violence in LRA unit (0-3)
Index of violence experienced by loved
ones (0-1)
Forced to step on or abuse dead bodies
(0/1)
Forced to betray a family member or
friend (0/1)
Log of length of time in violent group
Age when joined violent group
Years since returning from violent
group
Carried a gun while in violent group
(0/1)
Ever rewarded for a job well done (0/1)
Ever held rank in LRA or led other
soldiers (0/1)
Received indoctrination in LRA (0/1)
Age of respondent
Father’s years of education
Mother’s years of education
Father died before start of the war
(1996 or earlier)
Mother died before start of the war
(1996 or earlier)
Father was a farmer in 1996
(1)
Index of Group
Identification
(2)
Index of Group
Identification
0.607**
[0.215]
-0.347
[0.216]
0.359
[0.263]
-0.076
[0.079]
0.340*
[0.173]
-0.230*
[0.111]
-0.154
[0.156]
0.028
[0.051]
-0.124*
[0.058]
-0.015
[0.033]
0.296**
[0.109]
0.200
[0.207]
0.262
[0.162]
0.250*
[0.096]
0.116*
[0.057]
0.019
[0.011]
-0.033
[0.019]
-0.207*
[0.091]
0.072
[0.144]
0.057
[0.138]
0.717***
[0.201]
-0.403
[0.230]
0.576*
[0.277]
-0.083
[0.081]
0.099
[0.184]
65
VIOLENCE AND GROUP IDENTIFICATION
Household land in 1996 (acres)
Household size in 1996
Number of cattle in 1996
Number of livestock in 1996
Number of plows in 1996
0
[0.002]
-0.008
[0.014]
0.001
[0.001]
-0.003
[0.002]
0.060
[0.111]
Observations
331
331
R-squared
0.807
0.763
Note: The regression also controls for location of the survey and year of
abduction. It is weighted to account for respondents selected to be in the sample
who could not be located, never returned from the LRA, or are no longer alive
(inverse attrition and absentee probabilities). Robust standard errors are in
brackets. *** p<0.001, ** p<0.01, * p<0.05.
66
VIOLENCE AND GROUP IDENTIFICATION
Appendix 4 Table 5: Responsibility for harm and group identification in Uganda
VARIABLES
Ever harmed a loved one on behalf of
the LRA (0/1)
Forced to step on or abuse dead bodies
(0/1)
Forced to betray a family member or
friend (0/1)
Index of violence personally
experienced (0-1)
Index of violence witnessed during the
war (0-1)
Index of violence in LRA unit (0-3)
Index of violence experienced by loved
ones (0-1)
Log of length of time in violent group
Age when joined violent group
Years since returning from violent
group
Carried a gun while in violent group
(0/1)
Ever rewarded for a job well done (0/1)
Ever held rank in LRA or led other
soldiers (0/1)
Received indoctrination in LRA (0/1)
Age of respondent
Father’s years of education
Mother’s years of education
Father died before start of the war
(1996 or earlier)
Mother died before start of the war
(1996 or earlier)
Father was a farmer in 1996
(1)
Index of Group
Identification
0.331**
[0.115]
-0.205
[0.109]
-0.160
[0.159]
-0.339
[0.214]
0.461
[0.257]
-0.052
[0.079]
0.391*
[0.163]
0.030
[0.052]
-0.121*
[0.056]
-0.014
[0.033]
0.332**
[0.111]
0.210
[0.206]
0.234
[0.161]
0.247*
[0.096]
0.111*
[0.056]
0.021
[0.011]
-0.038
[0.020]
-0.196*
[0.091]
0.074
[0.145]
0.064
[0.138]
(2)
Index of Group
Identification
0.446***
[0.108]
-0.087
[0.102]
-0.172
[0.152]
67
VIOLENCE AND GROUP IDENTIFICATION
Household land in 1996 (acres)
Household size in 1996
Number of cattle in 1996
Number of livestock in 1996
Number of plows in 1996
-0.000
[0.002]
-0.007
[0.014]
0.001
[0.001]
-0.003
[0.002]
0.060
[0.112]
Observations
331
334
R-squared
0.807
0.756
Note: The regression also controls for location of the survey and year of
abduction. It is weighted to account for respondents selected to be in the sample
who could not be located, never returned from the LRA, or are no longer alive
(inverse attrition and absentee probabilities). Robust standard errors are in
brackets. *** p<0.001, ** p<0.01, * p<0.05.
68