How very massive atrocities end:
A dataset and typology
, Chad Hazlett†
Understanding how the most severe mass atrocities have historically come to an end may
aid in designing policy interventions to more rapidly terminate future episodes. To facilitate
research in this area, we construct a new dataset covering all 43 very large mass atrocities
perpetrated by governments or non-governments since 1945 with at least 50,000 civilian
fatalities. This article introduces and summarizes these data, including an inductively
generated typology of three major ending types: those in which (i) violence is carried
out to its intended conclusion (37%); (ii) the perpetrating force is driven out of power
militarily (26%); or (iii) the perpetrators shift to a diﬀerent strategy no longer involving
mass atrocities against civilians (37%). We ﬁnd that international actors play a range
of important roles in endings, often involving encouragement and support for changes in
strategy that reduce mass killings. Endings could be attributed principally to armed foreign
interventions in only four cases, three of which involved regime change. Within the cases
we study, no ending was attributable to a neutral peacekeeping mission.
Keywords: mass atrocity, genocide, conﬂict termination, civil war
∗Fletcher School of Law and Diplomacy
†Assistant Professor, UCLA. email@example.com (Corresponding).
Mass atrocities – widespread and systematic violence against civilians – have proven easier to
condemn than to stop. Yet these atrocities do eventually stop. Valuable lessons for halting future
atrocities may be learned by studying past terminations, but little systematic work has sought
to characterize how even the largest atrocities have ended. Instead, nearly all work on atrocity
endings has examined the eﬀects of third party interventions, such as sanctions or peacekeeping
missions, on the scale or duration of violence. These studies provide important insights, but do
not illuminate the variety of ways in which atrocities end, especially when endings are unrelated
to outside intervention. This article aims to advance and facilitate research on the empirical
question of how very large atrocities have ended. We introduce a dataset describing the 43
mass atrocities involving 50,000 civilian fatalities since 1945. Our focus on these very massive
atrocities ensures the availability of data required to satisfy our coding rules. Consequently, we
speak only to how these ‘worst’ atrocities end. Because these events are so large as to require a
degree of coordination and capacity, they may diﬀer in important ways from smaller atrocities,
not to mention other violent phenomena such as terrorist campaigns or criminal activities.
In addition to providing data on each atrocity and how it ended, we provide complete case
studies and narrative descriptions of each ending. We also oﬀer a framework that organizes
these endings into three types. First, atrocities end as planned when perpetrators achieve their
aims, often by eliminating the threat they believed was posed by the targeted group. Second,
atrocities can end when perpetrators were prevented from achieving their aims, notably through
military defeat. The third ending type, labeled strategic shift, occurs when perpetrators retain
power and capacity, and continue to pursue their initial goals, but through means that no longer
rely upon mass killing of civilians. Such a shift may be a response to rising (military, political,
or economic) costs of continuing the atrocities, failure to achieve the intended goals, or shifts in
political power and preferences within the perpetrating organization.
Of the 43 mass atrocities we study, 16 (37%) ended as planned. These atrocities tend to
be shorter, averaging 3.9 years as compared to 7.3 for the other two groups. The proportion
of atrocities ending this way appears to be decreasing in recent decades. Atrocities ending by
strategic shift account for 16 (37%) of the cases, and show an increase in the proportion ending
this way in recent decades, commensurate to the decline in as planned endings. Finally, endings
by defeat account for the remaining 11 (26%) of cases, and have remained relatively static.
This work has several implications of interest to policymakers. First, we note that no atrocity
reaching the 50,000 fatality scale has ever come to an end due to an armed, foreign-intervention
that remained neutral. In contrast, 11 atrocities have come to an end due to defeat of the
perpetrator in armed attacks by either foreign or domestic forces. Critically, however, such non-
neutral interventions do not always succeed—and are sometimes implicated in further atrocities.
Second and more optimistically, the increasing share of atrocities ending through strategic shift
suggest that international actors have opportunity to encourage less violent endings by raising the
political, ﬁnancial, or other costs of perpetrating mass atrocities, by bolstering more moderate
actors, or by supporting mediation designed to resolve core conﬂict issues by means other than
2 Background and theory
2.1 Eﬀects of interventions
Most scholarly research on endings has examined the impact of particular policy interventions on
the scale or duration of violence. For example, Krain (2014) ﬁnds no impact of economic sanc-
tions on atrocities. Others ﬁnd regimes become more abusive or repressive following sanctions
(Peksen, 2012; Davenport & Appel, 2014). Despite the increasingly widespread use of targeted
sanctions (Biersteker et al., 2013), there is little systematic evidencee of their eﬀectiveness (Tour-
inho, 2015). Similarly, the deterrence eﬀect of criminal prosecutions remains largely untested
empirically (Cronin-Furman, 2013). Further, any general deterrence eﬀect of such accountability
is not likely to explain the ending to any particular atrocity that has already reached a large
In research on diplomatic interventions, clear expressions of disapproval by multilateral or-
ganizations and NGOs have been argued to decrease the severity of violence against civilians
(DeMeritt, 2012; Krain, 2012). Krain (2012) further argues that condemnation of atrocities by
outsiders may prompt small modiﬁcations in perpetrator behavior, but rarely leads to fundamen-
tal changes such as atrocity endings. Mediation may hasten endings of civil conﬂicts (Svensson
& Wallensteen, 2013), which is promising, though as demonstrated here conﬂict endings and
atrocity endings are often distinct.
Finally, armed intervention hostile to perpetrators, widely discussed as a possible cause
of atrocity endings, may hasten the conclusion of genocide and mass atrocities (Krain, 2005;
DeMeritt, 2015). Other works, however, argue that such interventions can cause the weakened
party to engage in higher violence against civilians (Wood, Kathman & Gent, 2012).
2.2 Directly studying endings
Even precise and unbiased estimates for the eﬀects of many possible policy interventions would
say little about how atrocities actually end, especially when there is no obvious outside interven-
tion. Outside of earlier phases of this research project (Conley & de Waal, 2014; Conley, 2016),
we know of only one other work seeking to understand how mass atrocities end (Bellamy, 2015).
In a ﬁrst important diﬀerence, our work considers the mass atrocity as the unit of analysis,
whereas Bellamy (2015) uses civil conﬂicts during which these atrocities (often but not always)
occur as the unit. We also include cases regardless of whether the perpetrator was the state,
a non-state group, or a foreign state, while Bellamy (2015) uses only state-led atrocities. More
centrally, our work is the ﬁrst we know of to propose a typology of atrocity endings. Neverthe-
less we regard these projects as two views on a similar question, complementing each other and,
fortunately, convergent in their main conclusions.
2.3 A framework for very massive atrocity endings
We now present a framework, arrived at inductively, that generate the three-way coding of
endings used in our dataset.
Endings as planned.The ﬁrst and simplest type of ending occurs when the perpetrating
organization, which had presumably engaged in mass atrocity to pursue some goal, achieves that
goal. Atrocities are then unnecessary. A common example of this occurs when regimes use mass
atrocity as a weapon against insurgency (e.g. Valentino, Huth & Balch-Lindsay, 2004) or to
protect against a threat they believe is posed by a given group.
Endings in defeat.Atrocities can themselves produce new grievances, damage institutions,
and empower new violence entreprenuers, all of which create new problems that may prolong
violence. At the same time, committing and sustaining atrocities requires material resources and
political support. Either can be depleted, through eﬀorts of an opposition, by the perpetrator’s
defeat or loss of power, or by exhaustion. In principle, a category of endings by ‘capacity-loss’
is therefore possible. In our own data, however, we observe endings due to a loss of capacity
by the perpetrator only in cases where that perpetrator has been defeated. Thus, we label the
second category of ending type as simply defeat.
Endings in strategic shift.Finally, when perpetrators have neither been defeated nor
achieved their goals, but reduce their use of mass atrocity in favor of a diﬀerent strategy, we label
the ending as a strategic shift. Perpetrators ostensibly see mass atrocity as their favored choice
to solve a problem (e.g. to combat a threat) when they began an atrocity. This may change over
time, leading to adoption of strategies that no longer involve mass atrocities against civilians at
the threshold used here. Such shifts can have many origins, including more eﬀective resistance
from the targeted group, the rise of more moderate political leaders, international pressure, the
changing price to buy out armed opponents, or other reasons that ‘draining the sea’ (Valentino,
Huth & Balch-Lindsay, 2004) comes to be seen as less eﬀective or more costly than an alter-
native. The new strategy may involve a decision by the perpetrator to restrain themselves and
deescalate violence, e.g. by shifting emphasis to a strategy of co-optation or concession, with
or without the aid of a mediated solution. Alternatively, the strategy may instead shift to one
that still employs high levels of violence, but no longer directed against civilians, e.g. through
a focus on targeted assassinations or defeating an opponent on the battleﬁeld rather than by
attacking the civilian communities that support them.
Our case-selection criteria are notable for: (i) the high threshold deployed (50,000 civilian fa-
talities), (ii) the otherwise broad inclusion rules regarding perpetrators, with both state and
non-state actors as perpetrators, domestic and foreign government as perpetrators; and (iii)
using the atrocity as the unit of observation, rather than using civil conﬂicts as the unit. We
elaborate on these and other choices here.
Our concern is strictly with very massive atrocities, deﬁned as widespread and systematic killing
of unarmed people (civilians or prisoners of war) within a single country resulting in at least
50,000 civilian fatalities in cases post-1945. We also include conditions in which people were
under the direct control of perpetrators, held in camps and prisons, and denied the means for
We use a threshold of 5,000 civilian deaths in a given year, with onsets marked by the
ﬁrst year above this level, and endings marked as the ﬁnal year at this threshold under a single
perpetrator, when followed by two consecutive years below this level. If more than 5,000 civilians
are killed by a separate perpetrator group within the subsequent two years, this is coded as a
subsequent atrocity. The overall threshold of 50,000 fatalities was driven principally by the
requirements for data accuracy, which we found not to be fulﬁlled across the universe of cases
at a lower threshold. We also do not use population size to scale this requirement, i.e. we focus
on largest atrocities, not the atrocities that kill the largest proportions of a population.
While the state has historically been the actor most likely to perpetrate mass atrocities
(Ulfelder & Valentino, 2008), non-state actors can also be perpetrators, and we see no reason to
exclude them. In fact, non-state actors are coded as the primary perpetrator in 16% of cases,
and as a secondary perpetrator (another party responsible for at least 5,000 civilian fatalities)
in 23% of cases. Our dataset also includes cases where the perpetrator was a state engaged in
killing civilians outside its own territory.
Also important to our deﬁnition is the dissociation of mass atrocities from possibly co-
occurring civil wars. Mass atrocities very often occur during multi-sided armed conﬂicts, which
has led some investigators to examine only atrocities that occur during conﬂicts. Yet atrocities
are distinct from conﬂict, often with diﬀerent beginning and end dates, and diﬀerentiable causes
for both beginning and ending. Our choice to include very massive atrocities regardless of
ongoing conﬂict proves itself empirically: almost a quarter of our cases of mass atrocity occur
outside of civil conﬂicts. Table VI in the Online Appendix describes and compares these datasets.
The universe of cases was established by drawing on diverse existing lists and datasets of
atrocity and genocides plus additional research. Case studies for each atrocity episode in the
dataset drew on expert qualitative analysis to outline the context for instability, describe the scale
and pattern of atrocities, present the evidence base for fatality ﬁgures, and detail terminations.
This process is inherently subjective and not all experts will agree on all conclusions. We
consulted outside experts to help with these decisions and obtained in-depth reviews on diﬃcult
cases including the Africa cases, Iraq, Indonesia and Vietnam. In addition, this dataset grew
from an earlier edited volume Conley (2016) as well as a sequence of four conferences convening
outside experts who had been assigned to review these cases.
Our analytic strategy assumes that the research team could attribute the cause of each
ending with reasonable accuracy. Though it can be diﬃcult to learn the eﬀects of causes in
social science applications where those causes cannot be randomized, in many individual cases,
we found arguably convincing evidence for the cause of the atrocity ending. We liken this
to forensic investigations such as autopsies or traﬃc accident investigations, wherein human
reasoning is often a powerful tool for attributing a cause or set of contributing causes to a
particular death or accident. That said, other investigators may come to diﬀerent conclusions.
We aim to assist challenges to our ﬁndings by making our case study materials available and
coding nine of the endings we found to be more diﬃcult with potential secondary ending types
(see Table IV in the Online Appendix).
Our approach also presumes we can know enough about perpetrators’ goals to determine if
they have been achieved as planned. State actors were the perpetrator in 37 (86%) of cases, with
the remaining led by non-state actors.1In principle, perpetrating organizations encompass many
individuals with diverse aims. Yet, in each case, we found it reasonable to summarize the main
purpose animating a period of high-level, sustained atrocities that were achieved at some eﬀort
and expense. Such goals were often arguably clear from context, e.g., when atrocities occurred
as part of a counterinsurgency campaign, targeting the communities perceived to be supporting
the insurgency. In many cases, this was also accompanied by explicit speech evidence from those
directing the atrocities. We note that lower level atrocities or other violent events may not allow
for clear understandings of perpetrators’ goals in this way, but the resources and coordination
required at the 50,000 fatality level lend themselves to an identiﬁable central purpose behind
1Note however that in 10 (23%) cases, we code a secondary perpetrator killing at least 5000 civilians.
3.2 Other variables
Finally, we collected a number of other variables to facilitate future research on these atrocities.
•number of subsequent years with more than 5,000 civilians killed;
•estimated number of fatalities;
•number of actors known to have killed more than 5,000 civilians;
•whether the violence involved popular participation or was conducted only by members of
specialized organized groups;
•dates of the associated multi-sided armed conﬂict (if present), as coded by the Uppsala
Conﬂict Data Program (UCDP/PRIO Armed Conﬂict Dataset version 19.1, Pettersson,
H¨ogbladh & ¨
•whether the perpetrator remained in power and normalized relations with the target
group(s) thereafter (if the regime was the primary perpetrator);
•leadership change: if same regime remains intact at the ending, but a key leader is shifted
out of power, whether through elite coup, natural death or democratic change;
•military defeat: none, domestic, or foreign;
•type of foreign military intervention: protection mission (including those deployed after
end of atrocity), intervention mandated to defeat perpetrator, or both; and
•withdrawal of foreign forces: if a withdrawal of forces occurred during the atrocities, and
whether it was associated with a reduction in atrocities.
All variables are described fully in the data dictionary available at http://www.prio.org/jpr/datasets.
4 Results: Ending types
4.1 Endings as planned
We coded atrocities as ending as planned when the perpetrator’s dominant goal at the time
of initiating mass atrocities was (a) reasonably clear or explicitly stated and (ii) achieved un-
changed, after which (iii) atrocities stopped. In 16 cases (37%), we coded as planned endings.
As might be expected, these atrocities tend to be shorter, lasting an average of 3.9 years.
Table I of the Online Appendix provides a narrative summary of all cases ending as planned,
providing a brief rationale for each coding. Seven of these cases involve new or emerging regimes
committing atrocities while coming to and consolidating power (dates indicate ending): China
(1953); D.R. Congo (1997); Ethiopia (1985); India (1947); Indonesia (1965); Liberia (1996);
and Yugoslavia (1948). In Indonesia, Suharto did not formally accede to power until 1968,
he did control the military-led violence against Communists while also purging supporters of
Sukarno from key government oﬃces. In Ethiopia, the military regime committed atrocities as
it consolidated power (1976), then continued mass violence as means of counterinsurgency.
Seven of the remaining nine cases in this category involved atrocities committed during
counterinsurgencies: Burundi (1972); Guatemala (1983); Indonesia (1980); Iraq (1988); Nigeria
(1970); Sri Lanka (2009); and Darfur, Sudan (2004). The ﬁnal two cases are more unusual: the
USSR’s responsibility for the deaths of German prisoners of war while in detention following
World War II, and Chinese repression during Great Leap forward.
As a concrete example, we describe our coding of the Guatemalan atrocities (1981-1983)
associated with the civil war (1963-1995). Violence against civilians met our criteria during
military oﬀensives targeting the Mayan population in 1981-1983. By 1984, much of the civilian
population in areas targeted by the government was eﬀectively under the control of the govern-
ment, leading to the as planned ending. We note that military leaders ousted President Rios
Montt in favor of Oscar Humberto Mejia Victores, and that political violence continued through
mid-1984, though not exceeding 5,000 deaths each year after 1983. Despite these complexities—
the continuation of some violence and the leadership change—we attribute the ending of major
violence against civilians with the military’s own perception of successful counterinsurgency.
In 11 cases (26%) the primary perpetrators were ousted through (military) defeat before achiev-
ing their goals, thus ending the atrocities. Narrative summaries of each such ending is provided
in Online Appendix Table II. These atrocities lasted on average 6.7 years. Seven such cases in-
volved defeat by predominantly domestic forces: Afghanistan (1988), Angola (2002), Indochina
(1955), Rwanda (1994), Somalia (1991), Uganda (1986), and Vietnam (1975). Three involved
defeat primarily by international forces: Bangladesh (1971), Cambodia (1965), and Uganda
(1979). In Sierra Leone (1999), both domestic and international forces played important roles
in defeating the Revolutionary United Front.
While a decisive military victory may suggest a clear atrocity ending, this is often not the
case. A secondary perpetrator continued atrocities above the 5,000 annual civilian deaths after
the primary perpetrator was defeated in ﬁve of these cases—Afghanistan, Indochina, Rwanda,
Somalia and Uganda (1979). Rwanda is an interesting example. Atrocities ended in the main
with military defeat of the Interim Government. Revenge killings continued after the Rwandan
Patriotic Front took control, with evidence suggesting they exceeded the 5,000 fatality level, but
not reaching the 50,000 level within Rwanda required for a new episode in this study. A separate
case captures subsequent killing in Zaire starting in 1996.
A complicated but illustrative example is Uganda, where two atrocities occurred—both end-
ing in defeat—but with the ﬁrst event creating the conditions of violence that led to the second.
The ﬁrst period under General Idi Amin ended when the previous president, Milton Obote re-
turned from exile with the support of the Tanzanian military (1979). However, violence then
surged a second time following an insurgency led by Yoweri Museveni and the National Resis-
tance Army (NRA) in what became known as the Ugandan Bush War. Eventually, Obote was
overthrown in a coup led by Basilio Olara-Okello in 1985; several leadership changes quickly fol-
lowed as the regime disintegrated. Violence against civilians only clearly fell below our threshold
in January 1986 when Museveni’s NRA deﬁnitively defeated the national army.
In the remaining cases of this type, military defeat of the primary perpetrator immediately
ended the atrocities: Angola (2002), Bangladesh (1971), Cambodia (1979), Uganda (1986) and
Vietnam (1975). Atrocities may even end before the complete defeat, as the perpetrator loses
capacity, as in Sierra Leone (1999) where the killing of civilians at the 5,000 per year level ended
in 1999 as the insurgent RUF became incapacitated, though complete defeat did not occur until
4.3 Strategic shift
The third ending type accounts for the remaining 16 (37%) of cases. In endings by strategic shift,
the perpetrator’s original purpose for violence is not achieved and the perpetrator is not defeated,
rather they decide to shift away from policies that lead to mass civilian fatalities. These cases
form a coherent group in that perpetrators were neither defeated nor did they achieve their
aims and yet they shifted strategy, however these cases can be quite heterogeneous in both
the causes for shifting and the chosen strategy. Future research using these data may with to
explore possible disaggregations of endings in this category, for example into those that result
in apparent concession or deescalation and those that involve re-purposing violence towards
non-civilians targets rather than an end to violence.
Narrative summaries of each case coded this way can be found in Online Appendix Table III.
The average duration for atrocities of this type was 7.75 years. In several cases where the primary
perpetrators were foreign forces (Algeria/France, Angola/Portugal, Cambodia/US, and Mozam-
bique/Portugal), domestic pressure in the perpetrators’ home country forced a withdrawal or
change in use of force. During anti-colonial conﬂicts in Mozambique and Angola, Portuguese
armed forces faced steady resistance and the conﬂicts stalemated. A coup in Portugal then led
to an about face on colonial policy, provoking retreat and an end to atrocities in both cases.
In Cambodia, U.S. Congressional oversight and progress towards a conclusion in Vietnam con-
tributed to decisions to end bombings. Similarly, domestic pressure contributed to France’s
change in colonial policies in Algeria, despite military successes.
Stalemated conﬂict provides the backdrop for many cases in this category. It appears to be the
primary cause for declines in atrocities in Korea (1953), Russia (1996), and Sudan (2005). This
comports with arguments that a ‘mutually hurting stalemate’ (Zartman, 1989) may motivate
actors to seek deescalation and a political solution as a form of strategic shift.
Changing allegiances among various armed forces can also generate this ending, and appear
to pave the way for a shift in strategy away from atrocities in Algeria (1998), Iraq (2009), and
Burundi (2000). Likewise, internal changes in political coalitions led to shifts in atrocities in
the peacetime cases of Colombia (1953) and E. Guinea (1979), resulting in a new political pact
(Colombia) or coup (E. Guinea).
Finally, international pressure appears to have contributed to a decline in atrocities in El
Salvador (1985), Burundi (2000), DRC (2003), and Mozambique (1992). International pressure
also seems to have altered patterns of post-war atrocities in Poland, as part of violent expulsion
of the German population (1946).
Burundi provides an example of how the interaction of stalemate, re-alignments and interna-
tional pressure can contribute to decisions that decrease the killing of civilians. First, in previous
periods of atrocities (1972) and in the ﬁrst year of this period (1993) the Tutsi-dominated govern-
ment forces overwhelmed targeted Hutu civilians. However, by 1994, multiple opposition armed
groups arose and were able to hold territory. A decline in violence is visible in the data start-
ing in 1998, coinciding with international condemnation of abuses, coupled with international
conﬂict mediation. Violence slowly declined in parallel as rebel groups fragmented, engaged in
peace talks, and momentum shifted towards a political resolution.
Despite important diﬀerences in our universe of cases and analyses, our ﬁndings converge with
those of Bellamy (2015) in two key ways. First, external armed intervention is not the princi-
pal form of ending. Second, Bellamy ﬁnds that ‘around half [of atrocities] end only when the
perpetrators themselves decide to end atrocities, usually because they have accomplished their
goals’ (p565). Most closely related, we ﬁnd one set of atrocities ended when perpetrators de-
cide that they have accomplished their goals (as planned). We distinguish these from another
set wherein perpetrators shift policies (strategic shift), accounting together for 74% of atrocity
endings. Finally, in terms of policy recommendations, Bellamy argues ‘it is more important to
think in terms of what can be done to shape the perpetrators’ incentive structures or encourage
internal dissent within the perpetrating elite.’ In our framework, this suggests attempting to
bring about an end through strategic shift, which we also advocate.
5.1 Trends over time
The decades since 1945 have seen enormous changes in the international system, in superpower
policies that may lead to support for both insurgencies and brutal counterinsurgencies, and in
international norms and institutions designed to curb mass atrocities. While the dataset is small,
plotting the relative proportion of each ending type over time provisionally suggests shifts in the
dominant ending types over time (Figure 1). Endings as planned may have declined slightly,
though this is far from clear. What is more suggestive is the sharp rise of endings by strategic
shift since approximately 1990, with a commensurate drop in endings by defeat. An interesting
question we cannot answer is whether the apparent drop in endings by defeat and perhaps as
planned, and the rise in endings by strategic shift is due in any part to the strengthening of the
international norms and institutions seeking to end atrocities.
Figure 1. Types of ending by year
1950 1960 1970 1980 1990 2000 2010
0.0 0.2 0.4 0.6 0.8 1.0
Atrocity end year
Trends in the probability of each ending type by year of ending (cubic polynomial ﬁt). Given the small number
of events spread out over more than half a decade, we do not place uncertainty estimates on these trends.
5.2 Atrocities versus conﬂict
While armed conﬂict and atrocities are clearly distinct, most prior research has used conﬂicts
as the basis for coding atrocities. In our dataset, there are 10 (23%) cases in which we code
an atrocity but no multi-sided armed conﬂict is coded by UCDP. Second, the dates ascribed to
conﬂicts and atrocities when they do overlap often diﬀer: start dates for atrocities and conﬂicts
varied (by at least a calendar year) in 65% of cases, and end dates in 59%. In particular, conﬂict
often outlasted atrocities (18, or 42%, of cases). The plurality of these cases ended in strategic
shift, though in others, lower levels of ﬁghting continued after atrocities ended.
5.3 Armed intervention
None of the very large scale atrocities studied here ended due to the intervention of a neutral,
protection-oriented peacekeeping mission.2Online Appendix Table V reviews every case in which
military intervention occurred.
Foreign interventions aiming to defeat the perpetrator appear to be responsible for endings
in Uganda (1979), Bangladesh (1971), and Cambodia (1979). In Sierra Leone, a mixture of
domestic and international forces defeated the principal perpetrator.
UN peacekeeping forces were deployed while atrocities occurred in Angola (2002) and Rwanda
(1994), with no clear impact on the atrocity endings. In Darfur (Sudan, 2004), the AU and later
hybrid AU-UN mission may have played a role, which we capture as a secondary coding (Online
Appendix IV). However, we argue that the primary reduction in atrocities resulted from an as
planned conclusion. Overall, these conclusions are consistent with those of DeMeritt (2015) and
Krain (2005), who ﬁnd that only interventions hostile to the perpetrators succeed in ending
Peacekeeping missions may be successful in achieving other aims. For example, missions
deployed after violence has dropped (Burundi 2000, DRC 2003, Mozambique 1992, Somalia
1991, and Sudan 2005) did not directly cause reductions, but may have played important roles
in the mediation process, prevented resurgences of violence, or otherwise contributed to stability
(Hultman, Kathman & Shannon, 2016). Critically, peacekeeping missions deployed to conﬂicts
while fatality rates are lower, could prevent cases from ever reaching the 50,000 death threshold.
Finally, though all eﬀective cases of armed intervention attempted to defeat the perpetrators,
not all such interventions were beneﬁcial. Endings can be attributed to the withdrawal of
international forces, including several wars of independence (Algeria, Angola, Indochina, and
Mozambique), and later cases (Afghanistan, Vietnam, DRC 1998-2003, and DRC 1996-1997).
Adding complication is the case of Iraq (2003-2009), where a decline in violence provided political
space and rationale for the reduction of international forces.
2It is possible of course that the discussion of such a deployment, even if it did not occur, could have had an
impact on endings, particularly by strategic shift.
5.4 Lower-scale events
The reader may be interested in the outcomes for cases below our very high fatality threshold
of 50,000. Without being able to speak to the universe of cases at a lower threshold, we do note
that several lower threshold cases end as planned in contexts where a targeted group was small
to begin with, and was eﬀectively crushed by a state. Examples include: Myanmar’s attacks
on Karen separatists in 1951; Laotian suppression of Meo tribesmen in 1965; and Pakistani
suppression of Baluchi separatists in 1977.
Of particular interest are well-known cases of interventions designed to confront the perpetra-
tor: NATO-led missions in Bosnia-Herzegovina (1995) and Kosovo (1999). We code atrocities in
Bosnia, where an estimated 35,000 civilians were killed, as ending through strategic shift, emerg-
ing from battleﬁeld setbacks (caused by domestic and international military action), mediation,
and fractures in the perpetrating coalition of the Serbian and Bosnian Serb forces. The politi-
cal organizations associated with both the Serbian and Bosnian forces, despite some leadership
changes, remained in power even while falling short of their wartime aims (Conley, 2016).
We can only speculate as to whether the coding framework described here would apply at a
lower threshold. That conﬂicts would end as planned, by defeat of the perpetrator, or when the
perpetrator changes strategies seems a reasonable starting point for categorizing endings at any
scale. However, that framework largely presumes that there is a dominant perpetrating organi-
zation, whose decisions (or defeat) can bring about at least the primary ending of atrocities. For
the largest atrocities, this premise is sustainable because the capacity and coordination required
to achieve such extraordinary levels of violence requires a perpetrator with a degree of coordina-
tion at scale, as seen in every case coded here. Lowering the threshold would begin to introduce
more ‘Hobbesian’ cases in which the presumption of a dominant perpetrator is unrealistic (e.g.
Libya, 2011). In such cases, endings may appear as temporary re-alignments of factions that
enable decreases in violence (strategic shifts) or as defeat of some of these organizations at the
hands of others (defeat) – but these endings are often brief and incomplete. Indeed, even the
notion of a primary ending at a given point in time becomes problematic when there are many
perpetrating groups and violence have many recurrences.
5.5 Policy implications
A ﬁrst implication for policymakers regards the role of armed intervention. In our data, no
atrocity reaching the 50,000 fatality level has so far been ended by a neutral, peacekeeping
intervention. Biased interventions seeking to instead defeat the perpetrator can be eﬀective
– though they come with the risk of contributing to atrocities as well, not to mention the
implications for future stability.
Second, strategic shift endings, instruct that shifts in strategy away from killing civilians
do not require complete defeat of the perpetrator. While they sometimes involve a leadership
change within the perpetrating organization, even this is not always the case. Instead, if pol-
icymakers can ﬁnd ways to increase the costs of ongoing atrocity to the perpetrators, and/or
strengthen the hand of those elites who would prefer a diﬀerent approach, then such a shift
may be achievable. Speciﬁcally, policies might usefully (1) support opportunities for moderation
within regimes committing these acts, (2) impose political or economic costs on perpetrators and
seek to strengthen the hand of moderates, and (3) continue and expand eﬀorts by third party
actors to encourage ceaseﬁres, negotiations, and settlements, possibly with the aid of incentives
The dataset, codebook, and R scripts for analyses shown here, along with the Online appendix,
can be found at http://www.prio.org/jpr/datasets.
We thank Alex de Waal and Darin Christensen for their insightful comments, as well as Claire Q.
Smith, Fanar Haddad, Noel Twagiramungu, and Roddy Brett for contributions to individual case
studies. Additional thanks are due to student researchers who contributed to the project, a full
list of whom is noted in the data repository, with special acknowledgement to Ross Weistroﬀer.
We thank the World Peace Foundation at the Fletcher School for ﬁnancial support of this
Bellamy, Alex J (2015) When states go bad the termination of state perpetrated mass killing.
Journal of Peace Research 52(5): 565–576.
Biersteker, Thomas; S Eckert, Marcos Tourinho & Zuzana Hud´akov´a (2013) The eﬀectiveness of
united nations targeted sanctions, ﬁndings from the targeted sanctions consortium. Graduate
Institute for International Studies, Geneva.
Conley, Bridget (2016) How Mass Atrocities End: Studies from Guatemala, Burundi, Indonesia,
the Sudans, Bosnia-Herzegovina, and Iraq. Cambridge: Cambridge University Press.
Conley, Bridget & Alex de Waal (2014) Setting the agenda for evidence-based research on ending
mass atrocities. Journal of Genocide Research 16(1): 55–76.
Cronin-Furman, Kate (2013) Managing expectations: International criminal trials and the
prospects for deterrence of mass atrocity. International Journal of Transitional Justice 7(3):
Davenport, Christian & Benjamin Appel (2014) Stopping state repression: An examination of
spells, 1976-2004. Available at SSRN 2485195.
DeMeritt, Jacqueline HR (2012) International organizations and government killing: Does nam-
ing and shaming save lives? International Interactions 38(5): 597–621.
DeMeritt, Jacqueline HR (2015) Delegating death military intervention and government killing.
Journal of Conﬂict Resolution 59(3): 428–454.
Hultman, Lisa; Jacob D Kathman & Megan Shannon (2016) United nations peacekeeping dy-
namics and the duration of post-civil conﬂict peace. Conﬂict Management and Peace Science
Krain, Matthew (2005) International intervention and the severity of genocides and politicides.
International Studies Quarterly 49(3): 363–388.
Krain, Matthew (2012) J’accuse! does naming and shaming perpetrators reduce the severity of
genocides or politicides? 1. International Studies Quarterly 56(3): 574–589.
Krain, Matthew (2014) The eﬀects of diplomatic sanctions and engagement on the severity of
ongoing genocides or politicides. Journal of Genocide Research 16(1): 25–53.
Peksen, Dursun (2012) Does foreign military intervention help human rights? Political Research
Quarterly 65(3): 558–571.
Pettersson, Ther´ese; Stina H¨ogbladh & Magnus ¨
Oberg (2019) Organized violence, 1989–2018
and peace agreements. Journal of Peace Research 56(4): 589–603.
Svensson, Isak & Peter Wallensteen (2013) Talking peace: International mediation in armed
conﬂicts. In: From Civil War to Strong Peace in Africa: The Thomas Ohlson Memorial
Conference, 18-20 April 2013, Uppsala.
Tourinho, Marcos (2015) Towards a world police? the implications of individual un targeted
sanctions. International Aﬀairs 91(6): 1399–1412.
Ulfelder, Jay & Benjamin Valentino (2008) Assessing risks of state-sponsored mass killing. Avail-
able at SSRN 1703426.
Valentino, Benjamin; Paul Huth & Dylan Balch-Lindsay (2004) “draining the sea”: mass killing
and guerrilla warfare. International Organization 58(02): 375–407.
Wood, Reed M; Jacob D Kathman & Stephen E Gent (2012) Armed intervention and civilian
victimization in intrastate conﬂicts. Journal of Peace Research 49(5): 647–660.
Zartman, I William (1989) Ripe for resolution: Conﬂict and intervention in Africa. Oxford:
Oxford University Press on Demand.
BRIDGET CONLEY, b. 1973, PhD in Comparative Literature (Binghamton University, 2001);
Research Director, World Peace Foundation (2011–) ; Associate Research Professor, The Fletcher
School, Tufts University (2011- ). Most recent books in English: Memory from the Margins:
Ethiopia’s Red Terror Martyrs Memorial Museum (Palgrave, 2019); and (ed) How Mass Atroc-
ities End: Studies from Guatemala, Burundi, Indonesia, the Sudans, Bosnia-Herzegovina, and
Iraq (Cambridge, 2016).
CHAD HAZLETT, b. 1979, PhD in Political Science (MIT, 2014). Assistant Professor in
Statistics and Political Science, UCLA (2014- ). Current research interests: causal inference
with observational data, civil war and mass atrocity, and eﬀects of violence on attitudes.
Table I. As planned endings
Case Brief Description
Burundi (1972) Burundian Army targeted Hutu leadership and crushed opposition, achieving its aim of
suppressing resistance. The President appointed a new prime minister and government tasked
with restoring order.
Following brutal civil war, Communist Party consolidated control and implemented a new
national agenda. Violence decreased as the central government ﬁnished these campaigns and
re-centralized the authority to kill. Persecution continued principally by placing dissenters into
The worst of the violence ended after the government wound down the Communist Revolution’s
central campaigns by the end of 1968. Members of Red Guard and other extremist
organizations were crushed by military and sent for rural re-education to restore stability.
Rwanda and the Alliance for Democratic Forces for the Liberation of Congo-Zaire invaded Zaire,
targeting Hutu civilians in refugee camps and as they ﬂed. The period ended with the overthrow
of the Zairean government of Mobutu Sese Seko. Subsequent violence began in 1998 and its
treated as a separate case.
The military regime solidiﬁed its position with crucial military support from the Soviet Union,
crushing and dismantling the urban political opposition. Its brutal counterinsurgency strategy,
including resettlement programs, ended partially due to success and international pressure.
The government’s counterinsurgency was largely successful, after which violence declined.
Systematic massacres ended nationwide by August of 1983 with the military coup against Rios
Montt, though political violence continued through mid-1984 at lower levels.
(1946-1947) Partition violence primarily ended as mass migration was completed. However, people continued
to migrate at lower levels and with considerably less violence over subsequent years.
The military gained power, and through a widespread purge, crushed opposition and signaled its
willingness to use massive violence if needed. Under presidential orders, General Suharto
dissolved the opposition Communist Party of Indonesia (PKI), blacklisting former members from
Large-scale atrocities subsided as the Timorese resistance lost ground and civilians returned to
villages under Indonesian military control (1978), leading to violence more targeted against
Timorese leadership. By March 1979 the Indonesian military declared East Timor ‘paciﬁed’ and
ended operations. Signiﬁcant incidents of killing occurred but were not sustained after 1980.
The August 1988 conclusion to the Iraq-Iran war was followed by the ﬁnal phase of the Anfal
campaign, lasting several days after which the last areas were declared free of Kurdish saboteurs.
On September 6, 1988, the Iraqi regime declared its victory by announcing a general amnesty for
all Kurds and large-scale oﬀensives ended. Lower levels of persecution and killing continued
through April 23, 1989, when the regime considered its objectives to have been accomplished.
By 1996, Charles Taylor’s insurgent forces had achieved military dominance. With the war
formally ending in an internationally mediated peace process, atrocities were no longer necessary.
Taylor subsequently won national elections in May 1997.
Atrocities ended with the collapse of the separatist Biafran state and its military defeat by the
Nigerian government, which declared a policy of ‘no victor, no vanquished.’
Post-1945, POWs were used for labor, causing many deaths. Killing the POWs was not a goal
but was not prevented. As projects were completed and Soviet leaders decided it was politically
beneﬁcial to release POWs. We thus code this as planned because the POWs fulﬁlled the
political and labor goals the government had for them.
Case Brief Description
(2009) The government of Sri Lanka deﬁnitively defeated the separatist LTTE militarily. Lower levels of
lethal violence in IDP camps more slowly concluded as the number of IDPs decreased.
The principal drop in atrocities, marked by the N’djamena Humanitarian Ceaseﬁre Agreement in
April 2004, occurred when the government achieved its primary military objectives and the army
exhausted its supplies, limiting its capacity. (Note: A possible secondary cause was international
political pressure – particularly from the African Union – to gain greater humanitarian access,
but it is diﬃcult to know if this would have been possible without the military ob jectives being
The decline in atrocities occurred initially as opposing armed and political groups, and civilians
associated with them, were identiﬁed and deemed under control. By 1948, Tito’s regime
Table II. Endings by defeat of perpetrator
Violence declined signiﬁcantly with the Soviet withdrawal. Subsequent violence
over 5,000 annual civilians deaths was caused by warlords and the Taliban; data is
poor for this period, but it is possible that the trend continued until the Taliban
consolidated control in 1998.
By August 1999 the government decided on a military solution to defeating
UNITA and quickly regained several of UNITA’s core territories, killing its leader,
Savimbi in February 2002, leading to a ceaseﬁre and concession of defeat.
Violence signiﬁcantly declined following the French withdrawal after it was
militarily defeated. Subsequent violence over 5,000 annual civilian killings
occurred until the new governments in the north and south consolidated control in
The overwhelming number of deaths – 500,000 to 800,000 – were perpetrated by
the Interim Government and ended when the RPF defeated it. Subsequent killings
continued into 1995, including at least 5,000 civilians killed by the RPF at Kibeho
IDP camp, in eﬀorts to establish the new regime’s control, and during cross
border incursions by remnants of the earlier regime, concluding when the new
regime consolidated power; subsequent RPF killings in Zaire are treated in a
At the end of 1990, the United Somali Congress launched an oﬀensive on
Mogadishu, and on 27 January 1991, President Siad Barre ﬂed the capital.
Subsequent violence was perpetrated by various Somali warlords thereafter,
through international intervention, and likely only falling below 5,000 annual
deaths in 1993.
(1991-1999) Atrocities ended as the RUF neared defeat in early 2000, largely due to escalated
ﬁghting supported by UK forces.
(1980-1986) Violence fell below our threshold when Museveni’s NRA took control of the
country in January 1986.
(1971-1979) Backed by the Tanzanian army, the Ugandan National Liberation Front overthrew
Idi Amin on April 11, 1979.
U.S. forces negotiated an exit under the Paris Peace Accords in January 1973.
Atrocities continued with ongoing ﬁghting until both the war and atrocities were
deﬁnitively ended when Saigon fell in April 1975.
By October 1971, the war between Pakistan and Bangladeshi resistance had
stalemated. Indian involvement escalated to full military intervention on
December 3. On December 16 the Pakistani administration crumbled and the
army surrendered. West Pakistan was soundly defeated and Bangladesh emerged
as a sovereign nation.
In January of 1979, Vietnamese forces took control of the capital; Pol Pot and the
remaining Khmer Rouge leadership ﬂed to the Southwest of the country, and the
People’s Republic of Kampuchea was established, marking the end of mass
Table III. Endings by strategic shift
In 1962, though France had considerable military success against the
insurgent FLN, French brutality alienated its domestic citizens and the
international community. This allowed the FLN to achieve most of its
demands when France chose instead to withdraw and negotiate peace,
including complete autonomy. Subsequent violence extended into
post–independence with assaults by Algerian forces against the Harki and
Europeans as the new government consolidated control later the same
With the failure to win militarily, the government began making political
agreements with more moderate insurgents in 1997. This strengthened
moderate positions within both government and insurgent groups, though
atrocity events occurred through 1998. By 1999 enough insurgents had
moderated and accepted political agreements for the atrocities to end,
though lower violence continued thereafter.
(1961-1974) Violence declined when Portugal, after a coup in 1974, ceased to resist
Overwhelming violence perpetrated by the Army failed to defeat the Hutu
rebellion and the civil war stagnated into a bloody stalemate, with voices
on both sides calling for negotiated solutions. International pressure
altered government policies in internment camps and internationally led
mediation produced a peace agreement in 2000.
The U.S. bombing of Cambodia came to a halt in August of 1973 when
the US Congress legislated its conclusion, following the signing of a peace
agreement between the US and North Vietnamese.
In 1953 the military junta declared a general amnesty, which 6,500
guerrilla ﬁghters accepted. Party leadership, surprised by severe
escalation of violence, were willing to compromise their initial goals,
reining in the army, paramilitary groups, and the civilian police. The
National Front government eﬀectively created a political alternative to
violence, establishing a power-sharing agreement between Liberal and
Atrocities declined due to the convergence of several factors: the
withdrawal of the most powerful armed actors, invading national armies
as part of an internationally-mediated peace process; increased UN
peacekeeping presence; and progress in the national political process.
Macias was overthrown in a coup by his nephew. Despite a retaliatory
blood bath killing approximately 400 people over the next two-weeks,
mass violence committed by the regime immediately ended, political
prisoners were freed, forced labor ceased, and amnesty was granted.
The decrease in civilian deaths appears to be partly the result of
international pressure to curb violence against civilians, and partly due to
patterns in the armed conﬂict which ultimately resulted in the
government winning an incomplete victory in the conventional warfare,
before the insurgents shifted to limited, asymmetrical guerilla warfare.
Declines in violence began in 2007 but remained above 5,000 annually
until 2009. The lull lasted only until 2013. It was possible as multiple
factors converged to produce a moderating dynamic: Sunni military
losses, the US ‘surge’ and reoriented counter insurgency; the Awakening
movement; ‘success’ of sectarian cleansing; standing down of the Mahdi
Army; and growing capacity of the Iraqi state and security forces.
Together, these forces created a political opening.
(1948-1953) Civilian fatalities declined as the conﬂict stalemated in 1951, but killing of
POWs continued until the July 1953 armistice.
Atrocities ended abruptly when Portugal’s 1974 military coup led to a
policy of ending its wars against independence. FRELIMO was recognized
as the legitimate government on June 25, 1975, and Portuguese forces
With the loss of foreign support to both sides, FRELIMO initiated the
ﬁrst round of peace talks in 1989 then approved a new constitution
permitting multi-party elections in 1990, appealing to RENAMO
power-sharing demands. Though RENAMO’s violent tactics continued as
talks progress, a ceaseﬁre and the General Peace Accords were signed in
Initially, expulsion of Germans was very violent. This violence decreased
in response to moderating eﬀorts by international actors – particularly
Allied forces and a bilateral agreement with the UK.
Atrocities subsided as the war reached a stalemate, leading in August
1996 to a settlement in which the Russian military withdrew. Chechen
independence was not discussed, but was defacto established until 1999,
when a new war began.
While 2005 does not mark a clear-cut or permanent end to mass violence,
after numerous rounds of ﬁghting and talking, and in the context of
stalemate on the battleﬁeld, moderate voices prevailed in forging a
(heavily internationalized) settlement including deployment of
peacekeeping forces. Parallel violence occurred in western Sudan.
US bombing of Cambodia came to a halt in August of 1973 when the US
Congress legislated its conclusion, following the signing of a peace
agreement between the US and North Vietnamese. Subsequent violence
during the civil war and rise of the Khmer Rouge is treated as a separate
Table IV. Alternate endings
As planned →Strategic shift. Though the speciﬁc campaigns of violence ending in
1976 appear to end as-planned, one could instead look to the longer sequence of
events from 1966-1976, and note that they ﬁnally ended following the death of Mao
Tse-tung and ascendance of Deng Xiaoping, whose approach was broadly more
Strategic shift →Defeat. If the coup is counted as a ’defeat’ of the Macias
government, rather than an internal change of leadership, coding would change.
As planned →Strategic shift. The decrease in violence could be deemed to be the
result of moderation, rather than a fully successful set of ethnic cleansing tactics.
Defeat →Strategic shift. The French suﬀered military losses that pushed it to
engage in a cease-ﬁre agreement, which could be interpreted as moderation rather
than all-out defeat.
Strategic shift →Defeat. Although it did not resemble a traditional military defeat,
the use of force played a signiﬁcant role in defeating terrorist elements (by both US
and Sunni armed actors) and the subsequent stand-down of the Mahdi Army on the
Shi’ite side. (The latter, however, may be considered moderation or even as planned,
given successful ethnic cleansing in Baghdad).
As planned →Strategic shift. The conﬂict ended with Taylor in power, as was his
goal, but there was an international mediation eﬀort that helped usher in the
elections that brought him to power. Hence, this could be coded as ’moderation.’
Strategic shift →As planned. The expulsions of Germans was largely accomplished
by the time that the pressure to moderate policies resulted in changes that reduced
As planned →Strategic shift. The decrease in deaths of German POWs could be
viewed as the result of a Soviet shift toward moderation with the death of Stalin.
As planned →Strategic shift. It is possible to read the decline in atrocities as the
result of international pressure given the global visibility of the violence.
Table V. Foreign armed interventions
Case Type of in-
tervention Context Relationship to
(1979-1988) Unilateral USSR intervention causes overwhelming
number of civilian deaths, and US support
to armed forces
(1954-1962) Colonial French eﬀorts to retain colony cause
overwhelming number of civilian deaths
important factor for
(1961-1974) Colonial Portuguese eﬀorts to retain colony cause
overwhelming number of civilian deaths
important factor for
(1992-2002) UN United Nations Angola Veriﬁcation Mission
(UNAVEM) deployed in 1988; augmented
to a peacekeeping mission in 1994
No relationship to ending
(1971) Unilateral West Pakistan intervention into East
Pakistan; Indian intervention into East
Called for in peace agreement (2000), the
AU Force (AMIB) deployed 2003,
transformed into UN force (2004)
(1975-1979) Unilateral Vietnamese intervention concludes period
Intervention by neighboring states in
support of government (Zimbabwe, Angola
and Namibia) and opposing it (Rwanda,
Uganda); UN peacekeeping force (MONUC
1999 -, MONUSCO 2010 - present),
short-term EU interventionary force (2003)
neighboring forces most
important factor; with
intervention of UN forces
(1945-1955) Colonial French forces defeated
important factor for
Iran-Iraq war provided the context for
atrocities, which continue for a short
period after the war’s conclusion.
Primary ending is ‘as
termination a factor
Intervention by a US-led coalition provides
the context for mass atrocities and causes
highest single month spike in killing. By
some analysis, the drawdown in
international forces enabled a subsequent
spike in atrocities
Drawdown of forces
coincides with decline,
not causal. Moderating
Case Type of in-
tervention Context Relationship to
International armed conﬂict with both
sides committing atrocities
Stalemate in conﬂict
contributes to ending
ECOMOG deployed 1990; UN Observer
Missions (UNOMIL) in 1993; expanded
mandate March 1993
No signiﬁcant impact
(1964-1973) Colonial Portuguese eﬀorts to retain colony cause
overwhelming number of civilian deaths
important factor for
UN ONUMOZ authorized in December 1992 Post-atrocity deployment
Allied USSR forces stationed in Poland following
end of WWII
Primary cause is
pressure; No signiﬁcant
Rwanda (1994) UN UNAMIR in place at onset, and French
No signiﬁcant impact
ECOMOG 1997; observer mission,
UNOMISIL 1998; UN peacekeeping
UNAMSIL in 1999, augmented in 2000,
with UK intervention
important factor for
UN Forces begin deployment in July 1992;
augmented as UNITAF in Dec 1992 – May
1993; given robust mandate in March 1993;
signiﬁcant withdrawals begin in March
(2003-2005) AU/UN AMIS I (May 2004) augmented to AMIS II
(October 2004); UNAMID (deployed Jan
Primary cause is ‘as
played a role in the
UN UNMIS (2005-2011) Post-atrocity deployment
Unilateral Tanzanian intervention to overthrow Idi
US intervention to defeat North
important factor for
Case Type of in-
tervention Context Relationship to
(1965-1973) Unilateral US bombing as expansion of Vietnam War
important factor for
(1996-1997) Unilateral Rwandan intervention to overthrow
important factor for
Table VI. Mass Atrocity-Related Datasets
Dataset Phenomenon Fatality
threshold Intent criteria Perpetrator
criteria Victim criteria Years
whole or part
1955 – 2003
of discrete events
as reported in
Civilians 1989 – 2004
Mass Killing (120
where death is
State agents Discrete group
start date ﬁrst
year over 100
Mass killing (103
conﬂicts, as well
non-state Civilian 1945 – 2010
out to at
non-state Civilian 1960 – 2008
context of a
wider political or
5 or more
soldiers hors de
1995 – 2012,
2013 – present
(with a four
Endings of mass
killing, 65 cases. 5,000 min. Intentional
Civilians 1945 – 2015